National Academies Press: OpenBook
« Previous: INTRODUCTION
Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

2
THE ATMOSPHERE

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Introduction

Humankind lives at the bottom of the sea of air, and climate change is perceived by us mainly as a change in the overall conditions of this sea's lower layers. The atmosphere lies at the heart of decade-to-century climate change: It filters the sun's rays as they reach the surface of the earth and as they are reflected again into outer space, and it is the principal medium of exchange of heat, water, trace-gases, and momentum between the other components of the climate system—oceans, land surface, snow, ice masses, and the biosphere. The atmosphere and ocean are intimately coupled within the climate system, and are governed by similar physical laws. But the atmosphere has been explored in greater detail—in terms of available observations and of existing models—than any other component of the climate system. Thus it is natural to review the results of this exploration first, as we begin our examination of climate variability.

Understanding of natural phenomena proceeds through a sequence of observations, experiments, and models. Given the complexity of the climate system, laboratory experiments can reproduce only very incompletely the system's major aspects, and have not been included in the present volume. Atmospheric observations have led, in past centuries, to very simple, purely descriptive models of atmospheric motions. In the second half of this century, advanced computer models of the atmosphere have simultaneously benefited from an increase in the number and quality of observations, and stimulated vast field programs designed to verify model results and yield the new details necessary for improving the models. The separate treatment of observations and models in this chapter and the next is, therefore, only a matter of expository convenience.

The oldest instrumental records of atmospheric temperature—and, to some extent, precipitation—extend about 300 years into the past. The coverage and density of these measurements have grown more or less continuously, with a dramatic increase occurring in the 1940s and 1950s. This permits us to make a fairly informed assessment of past interannual variability, but we have considerably less confidence about interdecadal changes and only little or indirect information on the century-to-century time scale.

The Atmospheric Observations section below starts with a paper by H.F. Diaz and R.S. Bradley that addresses head-on the question of how different the climate of this century has been from those of previous ones. Proceeding from temperatures (which tend to be more uniform in space and time) to precipitation (which is considerably less so), S.E. Nicholson looks at the socioeconomically critical issue of African rainfall variability on interannual and decadal time scales. J. Shukla provides complementary insight on the initiation and persistence of drought in the Sahel.

The role of snow cover in the radiation balance at the surface makes it an important player in climate variability; this role is discussed by J.E. Walsh. Variability and trends of both liquid and solid precipitation over North America are reviewed by P.Ya. Groisman and D.R. Easterling. Returning to temperature, a careful study of the difference between trends in daily temperature maxima and minima is presented by T.R. Karl and his colleagues. C.D. Keeling and T.P. Whorf then analyze the decadal oscillations in global temperatures and in atmospheric carbon dioxide. These oscillations are at the heart of understanding natural variability on this time scale; they are also covered later in this section in the essay introducing atmospheric modeling.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

The Southern Hemisphere has less instrumental coverage, in space and time, than the Northern Hemisphere. While many of the earlier papers in the section concentrate on the latter, D.J. Karoly describes the observed variability in the atmospheric circulation south of the equator.

Atmosphere-ocean interaction is involved in the last two papers of this section. C. Deser and M.L. Blackmon review atmospheric climate variations at the surface of the North Atlantic, while D.R. Cayan and his associates study a general-circulation model simulation of the Pacific Ocean, driven by observed surface fluxes.

Atmospheric models have a relatively long tradition in the climate community, going back to the 1950s. They span a spectrum, from simple radiative-convective models in one vertical dimension, through energy-balance models in one and two horizontal dimensions, to fully three-dimensional general-circulation models. These models are of interest in their own right as important tools for investigating climate change, either by themselves or coupled to models of other climate subsystems. They also provide an instructive example of the creation of a full suite of models for intercomparison and validation; their development is only now being followed, more or less closely, by the modeling enterprise in oceanography, hydrology, and other disciplines contributing to the climate-change enterprise.

Modeling and observations are closely linked by the reciprocal problems of simulating the observed variability and validating the existing models. The Atmospheric Modeling section is thus appropriately begun by T.M.L. Wigley's and S.C.B. Raper's paper on modeling and interpreting paleoclimate data, with an emphasis on the greenhouse effect. G.R. North and K.-Y. Kim apply classic time-series analysis techniques to climate-signal detection. R.S. Lindzen then considers a few themes in these two papers from complementary viewpoints.

A number of causes for climate variability on time scales of decades to millennia are reviewed by D. Rind and J.T. Overpeck. They emphasize the modeling approach to a study of these causes, while J.M. Wallace addresses similar issues by analyzing the climate record.

The two sections of this chapter are introduced by essays, one by T.R. Karl and the other by M. Ghil. They provide an overview of the current state of the fields of atmospheric observations and atmospheric modeling, reviewing those topics not covered by the workshop and offering a perspective on the papers included. Following each paper, a commentary by the discussion leader and a condensation of the spirited discussion that took place at the workshop shed additional light on our current knowledge in both areas.

The ocean is examined in Chapter 3, and atmosphere-ocean interaction is explored further in Chapter 4. Also of interest are the proxy records of climate variability, such as tree rings and coral reefs, which are covered in Chapter 5. The sponsoring committee's conclusions, which draw on the material in this chapter and in the other three, appear in Chapter 6.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Atmospheric Observations

THOMAS R. KARL

INTRODUCTION

During the 1970s, the 1980s, and into the 1990s atmospheric scientists have accelerated research aimed toward identifying and explaining the presence of decade-to-century scale climate fluctuations. The motivation for special consideration of atmospheric climate fluctuations on these longer scales has its roots in the numerous major climate events whose occurrence is so well documented in the instrumental record. Probably the best-known of these fluctuations occurred about 1970, when rainfall suddenly decreased over the Sahel; the causes of this jump are discussed in this section by Nicholson (1995) and Shukla (1995). Other examples include the Dust Bowl years in North America during the 1930s, the diminished intensity of tropical storms affecting the East Coast of the United States during the 1960s, 1970s, and much of the 1980s, and the wet weather of the 1970s and 1980s over much of the United States.

As access to climate records has broadened during the past few decades, researchers such as Hurst (1957), Mandelbrot and Wallis (1969), Mitchell (1976), Douglas (1982), and Lorenz (1986) have presented evidence suggesting that the notion of a static climate is no longer tenable, even on less-than-geological time scales. We have come to realize that 30 years of data, the length of time that has been used to compute temperature and precipitation "normals" (Court, 1968), is inadequate to define climate. It does not provide us with sufficient information either to minimize adverse climate impacts within such sectors as energy, water supply, transportation, environmental quality, construction, agriculture, etc., or to maximize the availability of the climate-governed resources, such as water and energy, on which both natural and man-made systems depend.

Several climate fluctuations that have affected the United States serve to illustrate this point. Beginning about 1975, the interannual variability of mean winter temperatures and total precipitation averaged across the contiguous United States substantially increased (Figures 1a and 1b); this increase persisted at least through 1985. In contrast, the interannual variability had been very low during the previous 20 years. The mean temperature increased dramatically over the United States during the 1930s, coincident with a large decrease in summer precipitation (Figures 1c and 1d), before returning to more typical conditions. The wetness of the 1970s and the first half of the 1980s, which is clearly evident in Figure 1e, resulted in record-setting high lake levels and caused considerable economic damage and human suffering (Changnon, 1987; Kay and Diaz, 1985). Over the past decade another climate fluctuation has been evident over the United States; as Figure 1f shows, temperatures increased in a jump-like fashion during the early 1980s. Interestingly, the temperature discontinuity that is apparent between January and June is not reflected by the temperatures during the second half of the year.

Due to the limited span of the instrumental climate record, evidence for climate fluctuations on decade-to-century time scales is biased toward higher frequencies. (The situation is much worse in the oceans; as Wunsch (1992) points out, the absence of comprehensive oceanographic data has led many oceanographers to focus on identi-

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 1

Time series of area-averaged seasonal and annual mean temperature and total precipitation for the contiguous United States. The smoothed curve is a nine-point binomial filter. The horizontal line reflects the mean over the period of record.

fying a mean climate state.) Nonetheless, there is evidence that climate fluctuations on longer time scales occur throughout the world (Jones and Briffa, 1995; Diaz and Bradley, 1995). As Karoly (1995) shows, there is a dearth of the data that would permit the identification of decadal-scale circulation variability in the Southern Hemisphere, but analysis of the longer-term surface data available suggests that significant climate fluctuations persist at least through decadal time scales.

The forcing agents of many climate fluctuations may have their origin in either anthropogenic or natural factors, or both. These fluctuations are important in terms both of their socioeconomic and biophysical effects and of our need to distinguish between natural climate variations and anthropogenic climate changes. In analyzing the climate record, the dangers of "data dredging" must be kept in mind. As access to climate data increases, the chance of finding trends and variations that appear to be significant also increases. For this reason decade-to-century-scale climate fluctuations have been referred to as the "gray area of climate change" (Karl, 1988). For instance, Keeling and Whorf (1995) find evidence for decadal fluctuations in the global temperature record that can be reproduced by assuming the existence of two oscillations with small differences in frequency that beat on time scales of about 100 years. The search for an explanation of this statistical result is a good example of the challenge presented by the existence of these decadal fluctuations.

IDENTIFYING CLIMATE FLUCTUATIONS

The instrumental record of atmospheric and related land and marine observations is fragmentary until at least the middle of the nineteenth century. Moreover, virtually all of

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

the observations used to study climate fluctuations are derived from observing programs that were developed not for this purpose, but rather to support day-to-day weather forecasting. Since the day-to-day variability of the atmospheric system in much of the world is far larger than the decadal climate variability, identification of multi-decadal climate fluctuations is often hampered by the poor quality and lack of homogeneity of the observations. For example, Groisman and Easterling (1995), in their study of the variability and trends of precipitation over North America, painstakingly document the many discontinuities and false jumps in the climate record that arise from changes in observing practices. These changes, if ignored, can and do overwhelm important long-term precipitation fluctuations.

Other atmospheric quantities besides precipitation are also affected by problems in measurement. In this section, Robinson (1995) points out the large disparity in snow-cover extent between a NASA data set using microwave measurements and a NOAA data set using visible imagery. Karl et al. (1995) devote considerable discussion to jumps and trends introduced into the surface temperature record by such factors as urban heat islands, changes in irrigation practices, differences in instrumentation, and station relocations. Cayan et al. (1995) use surface marine observations to calculate the latent and sensible heat fluxes that are used as forcing agents of the sea surface temperature field in the North Pacific. Their analysis includes the basin-wide climate jump that began about 1976-1977. To account for known systematic biases related to trends in the wind field and the sea surface and marine air-temperature and moisture lapse rates (Ward, 1992; Cardone et al., 1990; Wright, 1988; Ramage, 1987), the global trend of each of these quantities is removed. As mentioned in Zebiak's commentary on Cayan's paper, it is unfortunate that such adjustments to the data are necessary.

Often the corrections and adjustments that must be applied to a climate record are of such magnitude that it is difficult to be confident that the resulting time series adequately reflects the climate fluctuations. For example, Figure 2 illustrates the significant adjustments required to calculate global temperature fluctuations since the nineteenth century. Through the use of other data bases (e.g., changes in snow cover, alpine glaciers, or sea level) and the isolation of various components of the surface temperature record (e.g., marine air temperature, sea surface temperature, and land temperature), it is possible to gain more confidence in the adjustments applied. Each of the data sets has distinctly different problems related to long-term homogeneity and data quality that require independent adjustment procedures. When these independent data sets provide a physically consistent picture of decade-to-century climate fluctuations, their agreement can be very compelling. In fact, the analysis of quasi-periodic oscillations of surface winds, pressure, and ocean and marine air temperatures

FIGURE 2

Smoothed global surface temperature variations, as derived from the original observations without adjustments for inhomogeneities in the climate record, compared to the same observations after adjustments for inhomogeneities as described by the Intergovernmental Panel on Climate Change (IPCC, 1990, 1992).

by Deser and Blackmon (1995) relies entirely on physical consistency among independent variables.

In order to improve our ability to discern climate fluctuations on decade-to-century time scales within the existing climate record, some federal agencies such as NSF, DOE, and NOAA are supporting data archeology efforts. Data archeology is the process of seeking out, restoring, correcting, and interpreting data sets. Such efforts are critical to identifying and understanding longer-scale climate fluctuations, since they often turn up information that reveals a pattern not otherwise obvious.

The Comprehensive Ocean-Atmosphere Data Set (COADS), which is just one of several major data-archeology efforts, provides a good example of the type of benefits that can be expected from such efforts. Now, several years after the project's inception (Woodruff et al., 1987), scientists are identifying decadal-scale variations and changes in many climate elements previously thought to be too uncertain to document with any confidence (London et al., 1991; Flohn et al., 1990; Parungo et al., 1994).

Interestingly, up to the present, satellite data have not figured prominently in the analysis of multi-decadal climate variability. Their short observing history and a lack of temporal homogeneity have hampered efforts to use them. Certainly, there are notable exceptions, such as NOAA's snow and sea-ice products and the Spencer and Christy (1990) microwave sounding-unit data. However, major temporal inhomogeneities in data sets, like those identified in the International Satellite Cloud Cover Project (ISCCP; see Klein and Hartmann, 1993), threaten the usefulness of much of the multi-decadal satellite data. The two major challenges over the next few decades for the United States will to be to ensure that global-change satellite-research projects such as EOS provide homogeneous data over their planned life-

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

times, and to lay the foundation for converting them into a long-term operational observing systems.

Future observations will require more attention to data quality, homogeneity, and continuity if we are to understand the nature of decade-to-century-scale climate variability. Currently under consideration is a Global Climate Observing System (GCOS)1 that includes observations of terrestrial, atmospheric, and oceanic aspects of climate. Since GCOS would be built around existing observing networks and environmental problems, it is essential that scientists effectively convey long-term climate-monitoring needs to the agencies sponsoring space-based and conventional observing systems. Among the climate-monitoring issues that should be addressed in the near future are stability of network sites, intercomparability of instruments, and increased sampling in data-sparse regions. Two concerns for both present and future observations are the better collection and documentation of metadata on observing instruments and practices as well as processing algorithms, and improved data-archiving practices and data-management systems. All these will increase the value of climate monitoring for decade-to-century-scale research; most essential, of course, is assurance of a long-term commitment to observational objectives.

THE FORCING AGENTS OF CLIMATE FLUCTUATIONS

The climate record provides the key to developing, refining, and verifying our hypotheses regarding the forcing agents responsible for climate fluctuations, be they anthropogenic or natural, internal or external to the climate system, global or regional, or persisting one or many decades. Diaz and Bradley (1995) suggest that the many decade-to-century climate fluctuations evident in both observations and proxy records may indeed have natural origins, since similar fluctuations seem to occur in the climate record both before and after humans became capable of modifying climate.

Often physically based models are the best means of testing our hypotheses about the cause of the fluctuations, as described in the modeling sections of this volume, but in addition they can be effectively used to help discern the physical consistency of apparent climate fluctuations and change. Cayan et al. (1995) demonstrate this approach; they use an ocean general-circulation model to help explain the climate jump over the North Pacific, as documented by Trenberth (1990; Trenberth and Hurrell, 1995, in this volume). Cayan et al. show how the atmosphere and ocean can act together to maintain decadal-scale climate fluctuations on a large spatial scale.

Potentially important factors in explaining climate fluctuations on decade-to-century time scales are land-surface and atmospheric feedback effects. Walsh (1995) provides ample evidence that snow cover has important feedback effects on the climate system on short (daily and interannual) and long (thousands of years) time scales, but on the time scales of interest to us the impact of snow is still not well understood. However, Nicholson (1995) and Shukla (1995) present evidence that the feedback between land-surface characteristics and the atmosphere has led to the prolonged drought in the Sahel. Karl et al. (1995) document an asymmetric increase of the mean maximum and minimum temperatures over many portions of the global land mass. Although they cite a number of potential causes of this multi-decadal trend, such as increases of anthropogenic atmospheric sulfate aerosols (Charlson et al., 1992) and greenhouse gases, empirical evidence suggests that, at least in some regions, observed increases in cloud cover play an important role in modulating the surface temperature. The forcing responsible for the increase in cloud cover remains unknown.

There are many important aspects of climate forcings and associated responses that cannot be covered here. Of particular relevance are the known changes in solar irradiance associated with the sunspot cycle. Recently Friis-Christensen and Lassen (1991) have used the length of the sunspot cycle to explain the decadal fluctuations and trends of land temperatures. Although a linear response of the surface temperature to the sunspot cycle length implies some unlikely responses of the temperature record in the early part of the time series (Kelly and Wigley, 1992), it is clear that changes in solar irradiance must continue to be monitored as a potential source of global climate fluctuations or change.

Recently, Elliott et al. (1991) have found evidence for an increase of tropospheric water vapor since 1973, leading to an enhanced greenhouse effect. However, the many types of observing and data-processing inhomogeneities in the upper-air moisture record make it immensely difficult to separate spurious trends and discontinuities from the true climate signal, even when the signal may be as large as a 10 percent increase in specific humidity.

A critical atmospheric quantity affecting surface temperature is the variability of cloud cover—its type, height, and spatial distribution. Using the ISCCP data base, Hartmann et al. (1992) showed the net forcing of various cloud types as a function of season and latitude. A recent study by Rossow (1995), however, reveals that this data set has serious biases that make it inappropriate for decadal-scale climate assessments. Meanwhile, conventional in situ data, analyzed on a national basis by a number of researchers, show a widespread general increase in total cloud cover

1  

GCOS will develop a dedicated observation system designed specifically to meet the scientific requirements for monitoring the climate, detecting climate change, and predicting climate variations and change (ICSU/UNEP/UNESCO/WMO, 1993).

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

over the past several decades (Dutton et al., 1991; Henderson-Sellers, 1990; Karl and Steurer, 1990; London et al., 1991; McGuffie and Henderson-Sellers, 1989; Parungo et al., 1994). Clearly, a much more focused effort is required to better understand decadal and multi-decadal fluctuations of cloud cover, since they often have very important feedback effects on other climate quantities. For instance, cloud feedbacks work differently during the day from the way they do at night. They therefore differ at high and low latitudes, making generalizations speculative.

Also, there is no discussion in this volume of the world's glaciers, in spite of their known tendency to fluctuate markedly on decade-to-century time scales. Long-term trends of climate change are integrated by mountain glaciers, and during the past century mountain glaciers all over the world have been declining (IPCC, 1990). The World Glacier Monitoring Service (1993) summarized these changes over the past several decades, and the USGS Satellite Image Atlas of the World discusses observed variations of hundreds of glaciers over the past several centuries.

CONCLUSION AND RECOMMENDATIONS

Identification and explanation of the forcings and feedbacks responsible for decade-to-century-scale climate fluctuations are essential to distinguishing between natural and anthropogenic impacts on the climate system, as well as to developing any predictive skill with respect to these phenomena. Further progress in understanding this gray area of climate change will depend upon our ability to address three topics. First, we must be able to document climate fluctuations without spurious discontinuities and trends. Second, we must ensure that we are adequately monitoring potential forcings and feedbacks internal and external to the climate system, so that we will have the data necessary for testing hypotheses about their operation. Last, data analysts must work closely with modelers (and vice versa) to test the hypotheses we formulate on a variety of climate models, ranging from simple one-dimensional climate models to complex coupled ocean-atmosphere general-circulation models. The re-analysis modeling projects of the United States (Kalnay and Jenne, 1991) and the European community are likely to enhance our confidence in both how and why climate has varied on decade-to-century time scales. (It must be emphasized, however, that the value of any re-analysis effort can be jeopardized by the use of data that are biased or, most troublesome, inhomogeneous in time.) Only by advancing in all three of these areas will we be able to overcome our ignorance about multi-decadal climate fluctuations and make predictions with any degree of assurance.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Documenting Natural Climatic Variations: How Different is the Climate of the Twentieth Century from That of Previous Centuries?

HENRY F. DIAZ1 AND RAYMOND S. BRADLEY2

ABSTRACT

Changes in decadal-mean surface temperature and its variance for different land areas of the Northern Hemisphere are examined. In the last 100 years, changes in surface air temperature have been greatest and most positive in the period since about 1970. Interannual variability, particularly at the largest spatial scales has also increased, although it differs according to regions. The most unusual decade of the last 100 years in the contiguous United States may have been the 1930s, although that of the 1980s is probably a close second, and in some regions perhaps the most anomalous. Both the 1930s and the 1980s experienced significant warming together with enhanced climatic variability.

To incorporate a longer-term perspective than is obtainable from the modern instrumental record, we used summer temperature reconstructions based on tree-ring records, and on d18O ratios extracted from different ice-core records. Although none of these records is a simple or even direct temperature proxy, we present them as general indicators of prevailing environmental temperatures. The data were averaged by decades in order to focus on decadal-scale variability. With the exception of the data from tropical ice cores, the proxies indicate that the recent decades were not very unusual, either in regard to the mean or in terms of increased variability. While seasonal and annual temperature changes in the last two decades have been rather large in most areas of the Northern Hemisphere, the available paleoclimate evidence suggests that in many areas there have been decadal periods during the past several centuries in which reconstructed temperatures were comparable to those of the 1970s and 1980s, with climatic variability as large as any recorded in recent decades. Natural variability on decadal time scales is comparatively large—typically about half as large as the interannual variance.

1  

NOAA Environmental Research Laboratories, Boulder, Colorado

2  

Department of Geosciences, University of Massachusetts, Amherst, Massachusetts

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
INTRODUCTION

Considerable effort has been expended over the last 25 years to improve our knowledge and understanding of climate processes and mechanisms associated with changes in climate (Saltzman, 1983; Trenberth, 1992). A major impetus for much of the recent climate research has been the global and regional climate projections made with physical climate models (the so-called general-circulation models, or GCMs) for a doubling of carbon dioxide concentration in the atmosphere.

Most assessments of the climatic impact of increases in the ''greenhouse" gases (CO2, methane, nitrous oxide, chlorofluorocarbons) have concluded, more or less consistently, that global temperatures should rise (under a doubling of the most abundant of these greenhouse gases, CO2) from 1.5°C to 4.5°C (NRC, 1982; Bolin et al., 1986; IPCC, 1990; Schlesinger, 1984, 1991). Such changes, should they occur, will be superimposed on the natural variability of the climate, which is quite large at all time scales (Karl et al., 1989). Depending on a number of assumptions, the amount of global warming that should have been realized by now appears to be more consistent with the lower end of the above estimates (Bloomfield, 1992).

Karl et al. (1991b) performed a variety of tests using output from three different GCMs to ascertain, given their different climate sensitivities, when a statistically significant greenhouse signal might be detected in the central United States. They concluded that it was likely that a greenhouse signal had been masked, to date, by natural climate variability, and that it would likely take another two to four decades before the greenhouse signal in temperature and precipitation might be unambiguously detected in this region.

We should note the possibility that other human-induced factors may be acting to counteract the radiative effects of increased greenhouse-gas concentrations in the atmosphere. Recent work has indicated that sulfate aerosols, which act to increase the planetary albedo (Charlson et al., 1992; Hansen and Lacis, 1990; Wigley, 1989, 1991) may have counteracted to some degree the enhanced infrared warming (see Michaels and Stooksbury, 1992). These and other factors, whether anthropogenic in origin or not (e.g., changes in vulcanism), will undoubtedly "muddy" the climate picture and may make it more difficult to identify unequivocally a contemporary greenhouse signal.

In this paper, we consider two aspects of climate-change detection that we feel bear strongly on the issue of natural versus anthropogenic climate variability. One aspect of the problem is, in effect, at what point one rejects the null hypothesis (no climate change) and accepts the premise that the climate of the last few decades belongs to a different sample population. The difficulty there arises because we are evaluating a relatively short observational record with the knowledge that the climate has fluctuated in the past few centuries by an amount that may be of the same magnitude as the fluctuations observed in the recent record (see Bradley and Jones, 1992).

A second aspect of the problem of climate change versus natural variability, which we consider here, is temporal changes in that variability, i.e., in the variance. Changes in climatic variability are important, since they are likely to have a greater effect on a society's ability to mitigate and adapt to climatic changes than a slow alteration in the mean climate patterns. We have examined a variety of observational records of various lengths (typically 100 to 150 years), which are representative of different spatial scales (from hemispheric to regional basins). We will focus on decadal time scales, since one could argue that climatic changes will be better gauged at time scales that to some extent average out the high-frequency climatic "noise" associated with seasonal variability and other air-sea processes operating on annual time scales (e.g., the El Niño/ Southern Oscillation system). We compare these instrumental series with a suite of high-resolution climate proxy records—namely, tree rings and oxygen isotopes extracted from ice cores—spanning the last 300 to 800 years. Our aim is to give the reader some feeling for the range of this intermediate-frequency (decade-to-century-scale) climatic variability, as well as some appreciation of the uncertainties inherent in the existing records (instrumental, historical, and paleoenvironmental).

We note that the "fingerprint" climate-change-detection technique (Barnett, 1986; Barnett and Schlesinger, 1987; Barnett et al., 1991) provides a very useful methodology for testing a hypothesis (the GCM climate-change projections) against observations. We will attempt here to highlight some aspects of climatic variability at decadal and longer time scales and will re-emphasize the conclusion of Barnett et al. (1991), who noted that the existence of a high degree of "unexplained" interdecadal climatic variability (see also Ghil and Vautard, 1991, and Karl et al., 1991b) will greatly complicate the detection of a greenhouse climate-change signal.

THE MODERN (INSTRUMENTAL) RECORD
Characteristics of Decadal-Scale Changes of Area-Mean Temperature Indices

The available data suggest that during the past century decadal-scale changes in global-scale mean annual temperature are on the order of 0.1°C to 0.3°C (see Folland et al., 1990). Figure 1 illustrates the annual and seasonal temperature changes for the Northern Hemisphere land areas. The plotted data, from Jones et al. (1986a) with subsequent updates, are consecutive 10-year area-weighted averages from 1891-1900 through 1981-1990 of gridded temperature anomalies referenced to the 1951-1970 period. On this "dec-

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 1

Decadal-mean seasonal and annual temperature anomalies (in °C) for land areas of the Northern Hemisphere. Data is from Jones et al. (1986a), with subsequent updates.

adal" time scale, the warming of the 1980s relative to the 1970s (0.32°C for the hemisphere) is comparable to the warming that occurred in the 1920s relative to temperatures in the 1910s (0.26°C). Some seasonal differences can be noted, with northern summer and fall showing the least warming trend and winter and spring showing the greatest warming trend over the last 100 years.

The twentieth century contains many examples of decadal-scale climatic variations of sufficient magnitude to have significantly affected societies. The two-decade-long drought in the African Sahel is perhaps the best-known example. The words "climate change" have become overused, and climatic variations from seasonal to multi-year are often lumped together. Until the mid-1970s, "climate change" was used to refer to climatic changes on time scales of 104 to 106 years (essentially, Milankovitch time-scales). Variations of less than 10 years were generally considered to be part of the natural climate noise associated with interannual variations of the various components of the climate system and their nonlinear interactions, whereas variations on time scales of 10 to 103 years are now thought to be strongly driven by ocean-atmosphere heat exchanges resulting from changes in the ocean's thermohaline circulation.

Because of the changes in the earth's radiative balance imposed by human activities during the past century, climatologists have been trying to detect climatic-change signals that hitherto have typically been associated with time scales of millennia. The climate record available for such studies, even with the aid of high-resolution paleoenvironmental records, is sufficient to define only continental to hemispheric changes on decadal time scales, and perhaps regional (less than about 105 km2) changes on century time scales.

In the United States, perhaps the singular climatic event of this century was the severe, widespread, and persistent drought of the 1930s. This event has been widely chronicled in both scientific (e.g., Skaggs, 1975) and popular literature. Other climatic events that persisted for more than one year and affected relatively large areas of the United States were the drought of the 1950s in south central areas of the United States (Chang and Wallace, 1987), and the drought of the 1960s in the Northeast (Namias, 1966). Periodic drought of variable duration has affected all areas of the contiguous United States to varying degrees (Diaz, 1983). Indeed, multiyear climate anomalies are a characteristic feature of the climate of the United States and of other parts of the world (Karl, 1988; Folland et al., 1990).

In Figure 2, decadal-mean seasonal (December-February for winter, March-May for spring, etc.) and annual temperature anomalies for various areas of the Northern Hemisphere are examined. Although Eurasia and North America (Figure 2a) have broadly similar seasonal trends over this period, differences can be noted in all seasons as well as in the annual averages. Figure 2b illustrates decadal temperature changes for the contiguous United States for the last century. The data used are from the adjusted divisional values (solid curve) described in Karl et al. (1984b). For comparison, the equivalent Jones et al. (1986a) gridded values are given as the dashed curve in Figure 2b. The decadal averages for the lower 48 states exhibit much smaller long-term trends than those noted for the continental-scale regions. Figures 2c and 2d illustrate the decadal temperature changes recorded in subregions of the United States, namely, the eastern and western halves of the United States (divided roughly along the 100th meridian), the state of Colorado, and a single state climatic division in Colorado where the city of Boulder is located. Again we note regional differences that can depart substantially from the temporal behavior of the larger spatial averages.

We have tabulated the mean temperature change from one calendar decade to the next over the 100-year period 1891 to 1990 for each of the above regions. No particular pattern emerges, except that at the largest space scales, the greatest decade-to-decade warming occurs mostly from the 1970s to the 1980s, whereas the time of the largest such

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 2

Decadal-mean seasonal and annual temperature anomalies (in °C). (a) For the Eurasian (solid line) and North American (dashed line) land masses. (b) For the contiguous United States, using data from the National Climatic Data Center described in Karl et al. (1984b) (solid line) and from the gridded temperature anomaly fields of Jones et al. (1986a) (dashed line). (c) For the eastern (solid line) and western (dashed line) United States. (d) For the state of Colorado (solid line) and the Platte River climatic division of Colorado (dashed line).

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

changes varies for the smaller regional averages, with most areas of the contiguous United States experiencing their greatest decadal changes during the first half of this century.

Recent studies have documented some rather pronounced and relatively rapid shifts in the climate system that occurred on decadal time scales. In particular, the atmospheric circulation in the extratropical North Pacific underwent a major shift in the mid-1970s that lasted over 10 years (Trenberth, 1990; Ebbesmeyer et al., 1991). This change was accompanied by a tendency for climate patterns in the tropical Pacific to exhibit conditions reminiscent of El Niño/Southern Oscillation (ENSO) warm-event conditions—weaker trades, warmer equatorial sea surface temperature (SST), anomalous rainfall patterns, etc.—relatively more frequently than during previous decades. Gordon et al. (1992) have also reviewed some of the relatively abrupt changes in water properties and surface climate in the Atlantic Ocean that have occurred in the past several decades, including sudden changes in SST anomalies in the South Atlantic compared to those of the North Atlantic, and the so-called "great salinity anomaly" in the North Atlantic from about the late 1960s to the early 1980s (Dickson et al., 1988). Certainly the sudden decrease of rainfall in the Sahel of Africa, a feature that has lasted for two decades, is a good example of how the climate can undergo significant, sudden, and prolonged change over relatively large spatial scales.

Clearly, increased knowledge of the behavior of climatic variability, from the interannual through decadal and century time scales, is needed to improve our assessments of any future changes in climate from regional to global scales. Indeed, as was noted by Ghil and Vautard (1991), the ability to distinguish a warming trend from natural variability is critical for separating out the greenhouse-gas-induced signal.

As we noted at the outset, an important element of climate that is too often overlooked is the variance or the characteristic variability of climatic means. In determining climate impacts, the importance to society of short- to medium-term climatic instability (i.e., climate fluctuations occurring on interannual to interdecadal time scales) is perhaps equal to or greater than that of slow changes in the background mean. Below we have analyzed some aspects of low-frequency changes in the temperature variance of different regional means.

Changes in Temperature Variability

We have examined contemporaneous variation of two measures of temperature variability, the standard deviation in running 15-year segments, and decadal means of the root-mean-square differences between successive yearly values of seasonal and annual regional temperature anomalies. The latter index is defined as

Since the xi are deviations from a reference mean and have approximately zero mean,

where s2 is the series variance, and r1 is the autocorrelation of the time series with a lag of I year. We will use this measure, rather than the standard deviation, as the key index of interannual variability. Note that this index, Ind V, amplifies changes in the high-frequency part of the variance spectrum.

Figures 3 and 4 illustrate the changes in the interannual variability of the regional groupings discussed earlier. The curves differ from one season to another and from one region to another. For the largest continental-sized regions, there appears to be an increase in interannual variability in the last couple of decades. At smaller spatial scales, there does not seem to be a consistent trend; instead, the interan-

FIGURE 3

Index of interannual variability (in °C) for the Northern Hemisphere land areas. Values correspond to decadal means of the root-mean-square difference between successive seasonal and annual temperature anomalies over the last 100 years.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 4

Indices of interannual variability, as in Figure 3. (a) For the Eurasian (solid line) and North American (dashed line) landmasses. (b) For the contiguous United States, using data from the National Climatic Data Center. (c) For the eastern (solid line) and western (dashed line) United States. (d) For the state of Colorado (solid line) and the Platte River climatic division of Colorado (dashed line).

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

nual variability wanes and waxes over the last 100 years. The corresponding curves for the standard deviation (not shown) reveal similar patterns, with increases in variance at the largest spatial scales with time but no consistent trends in the standard deviation of the smaller regions. As one would expect, there is a well-defined log-linear relationship between the magnitude of the interannual variability and the size of the area comprised by the various indices (see Figure 5). The inherently higher variability at smaller regional scales should be kept in mind when interpreting the paleoclimate record, which is based on limited regional samples.

The question of whether a warming climate will exhibit increased short-term variability is still open. Previous studies have been generally inconclusive on this subject (van Loon and Williams, 1978; Diaz and Quayle, 1980). Certainly, there is no consistent relationship between interdecadal changes in average temperatures and interdecadal changes in variability. Climatic variability may increase, decrease, or stay the same, in response to an arbitrary change in decadal mean temperatures. On the basis of the above results, we can say that, considering surface temperature variations during the last century, there has been a recent increase in variability at large spatial scales (continental to hemispheric-scale averages). This increase in variability is also concurrent with relatively large temperature increases over those areas. However, this is not the case for earlier warming episodes.

THE PRE-INSTRUMENTAL PERIOD

Prior to the late 1800s, reliable quantitative data on climate variations are sparse, and conclusions about the magnitude of these changes on large spatial scales necessarily contain significant degrees of uncertainty. There is ample evidence that multi-decadal temperature changes occurring over regions the size of Europe and China have been at least of order 1°C (Bradley and Jones, 1992). This suggests that it is quite plausible, if not likely, that temperatures as warm as those prevailing since the end of the Second World War may have been experienced at regional spatial scales of about 105 to 106 km2 for periods of one to several decades during some portions of the past thousand years under what amounts to natural (i.e., with negligible human influences) conditions.

Below we present some examples of long-term variability in a set of climate-sensitive paleotemperature records. The climate indices used here are discussed in several chapters of Bradley and Jones (1992). A listing of them, together with the source references and periods of record, appears in Table 1; each of them has been assigned a number for use in later tables comparing their data. Because of its very long record, the instrumental temperature series for central England (Manley, 1974) is considered in this section for comparison with other paleoclimate indices.

Tree-Ring Indicators

There is an extensive body of work relating climate variations to growth changes in particularly climate-sensitive trees (e.g., Fritts, 1976; Fritts et al., 1979; Hughes et al., 1982; Cook and Kairiukstis, 1990). We have selected a suite of climate-sensitive tree-ring records to study possible changes in climatic variability. Two main considerations were applied in the selection of these indices. First, we chose

FIGURE 5

Interannual variability (in °C) as a function of the regional area. Values shown are for northern summer (left panel), and winter (right panel). Area is plotted as log10 values.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

TABLE 1 Temperature Indices Used in This Study

Type of Record

Time Coverage

Reference

Index No.*

Ice-core d18O

Agassiz Ice Cap (Ellsmere Is.)

A.D. 1349-1977

Fisher and Koerner (1983)

1

Devon Island Ice Cap

A.D. 1512-1973

Koerner (1977)

2

Camp Century (Greenland)

A.D. 1176-1967

Johnsen et al. (1972)

3

Milcent (Greenland)

A.D. 1176-1967

Hammer et al. (1978)

4

Quelccaya Ice Cap (Peru)

A.D. 1481-1981

Thompson et al. (1986)

14

Dunde Ice Cap (China)

A.D. 1606-1987

Thompson et al. (1990)

15

Temperature reconstructions from tree rings

Western U.S. (annual) summer

A.D. 1602-1961

Fritts (1991)

5

Eastern U.S. (annual) summer

A.D. 1602-1961

Fritts (1991)

6

Western U.S. & SW Canada (annual) summer

A.D. 1602-1961

Fritts (1991)

7

Northern Treeline (North America) (annual)

A.D. 1601-1974

D'Arrigo & Jacoby (1992)

9

U.S. & SW Canada

(April-September)

A.D. 1600-1982

Briffa et al. (1992b)

8

Northern Scandinavia

(April-August)

A.D. 500-1980

Briffa et al. (1992a)

12

Northern Urals (June-July)

A.D. 961-1969

Graybill and Shiyatov (1992)

13

Tasmania, Australia

(November-April)

A.D. 900-1989

Cook et al. (1992)

16

Patagonia, Argentina

(December-April)

A.D. 1500-1974

Boninsegna (1992) & Villalba et al. (1989)

17

Rio Alerce, Argentina

(December-February)

A.D. 870-1983

Villalba (1990)

18

Instrumental

Central England Temperature (annual)

A.D. 1660-1987

Manley (1974)

 

Dec-Jan-Feb series

 

 

10

June-July-Aug series

 

 

11

* These numbers are used in text and later tables to identify the various indices.

climate-sensitive tree-ring records that have been thoroughly documented in the refereed scientific literature. Second, we chose the longest of those records in order to maximize the temporal coverage, and tried, as far as possible, to select samples from representative geographical areas.

A number of problems are inherent in any climate reconstruction. These problems are discussed in detail in the original published papers; however, we will briefly highlight here the more critical ones as they may affect the results presented below. Because we have analyzed changes in variability, temporal changes in the composition of the tree-ring network used for the reconstructions will, in general, affect the high-frequency variance, and to some degree the low-frequency variance as well. In interpreting the temperature reconstructions shown here, we have taken into account these potential sources of biases as much as possible.

Regardless of whether one uses the width of the annual growth rings or the maximum latewood density to reconstruct a particular climate variable (generally, growing-season temperature and/or precipitation), a process of standardization is required to account for different rates of tree growth as a function of age. This standardization is achieved by fitting a growth curve to the series of annual values, usually a cubic smoothing spline (see Cook and Kariukstis, 1990), and taking residuals about the smoothed curve. The degree of smoothing and the functional form of the growth curve partly determine the spectral properties of the residual series. In order to develop a climate reconstruction from these records, it is necessary to formulate a transfer function to convert the tree-growth index into, say, a temperature index. The procedure usually involves a calibration phase, in which a set of regression coefficients is derived that convert the tree-growth parameter into a climatic estimate, and a verification phase (see, e.g., Cook

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

and Kariukstis). Verification is usually done against an independent set of predictors. The reader is urged to review the reference sources listed in Table 1 for specific details regarding each paleotemperature reconstruction.

Another feature of long-term reconstructions from tree rings that should be borne in mind is that the composition of tree-ring samples that make up each yearly value is variable. In general, an index value for a given site is composed of a number of tree samples from several individual trees. These may vary in age, and in some cases may come from both dead as well as living trees. The number of samples making a particular ensemble average will vary with time, although this is more generally the case near the beginning and end of a particular chronology. These extraneous factors can introduce "noise" and spurious fluctuations into the reconstructions. Nevertheless, the tree-ring indices considered here, while suffering to varying degree from the above-named shortcomings, comprise a high-quality set of proxy temperature records that are of value to compare.

Figures 6 through 8 show decadal-mean values of various tree-ring reconstructions, generally representing summer (or growing-season) mean temperature. For the most part, these tree-ring reconstructions do not sample the 1980s; nevertheless, it is clear from the degree of interdecadal variability that it would be hard to point to any particular recent period as being unique or exceptional. The November-to-April

FIGURE 6

Decadal mean values of reconstructed summer- temperature anomalies for various North American regions. The top three graphs are in standardized units, the bottom two are in °C. See Table 1 for source information.

FIGURE 7

Decadal mean values of reconstructed summer temperature anomalies for Scandinavia and the northern Urals. Units in °C.

temperature reconstruction for Tasmania (Cook et al., 1992) does indicate that the last two decades of that record (1970s and 1980s) exhibit significantly greater tree growth (Figure 8) and hence higher reconstructed mean growing-season temperatures than previously, but it would seem premature to conclude that a "climatic change" has taken place in this region of southern Australia. We also note that, unlike the North American and Scandinavian reconstructions, this one is based on tree samples from only one location.

Some unique features are evident for a few of the temperature indices. For example, the curve of D'Arrigo and Jacoby (1992) is a reconstruction of annual mean temperature based on several sites along the northern North America treeline, stretching from Alaska to the Northwest Territories, plus a site located in northern Quebec Province. This reconstruction exhibits a pronounced warming from a very cold period during the mid-1800s to a maximum around 1940. By comparison, the two other high-northern-latitude sites discussed here—the indices from northern Scandinavia and from the northern Urals (Figure 7)—do not exhibit quite the same type of variations. The reader should note, however, that a much longer time period is sampled by these other two reconstructions, and that a seasonal (April to August for the Tornetrask index, and June to July for the

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

TABLE 2 Century Means of Interannual and Interdecadal Variability for 18 Temperature Indices (index sources identified in Table 1)

 

Index

Century

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

500s (ann.)

x

x

x

x

x

x

x

x

x

x

x

.84

x

x

x

x

x

x

(decad.)

x

x

x

x

x

x

x

x

x

x

x

.20

x

x

x

x

x

x

600s (ann.)

x

x

x

x

x

x

x

x

x

x

x

.79

x

x

x

x

x

x

(decad.)

x

x

x

x

x

x

x

x

x

x

x

.27

x

x

x

x

x

x

700s (ann.)

x

x

x

x

x

x

x

x

x

x

x

.64

x

x

x

x

x

x

(decad.)

x

x

x

x

x

x

x

x

x

x

x

.40

x

x

x

x

x

x

800s (ann.)

x

x

x

x

x

x

x

x

x

x

x

.63

x

x

x

x

x

x

(decad.)

x

x

x

x

x

x

x

x

x

x

x

.35

x

x

x

x

x

x

900s (ann.)

x

x

x

x

x

x

x

x

x

x

x

.70

.27

x

x

,x

x

.87

(decad.)

x

x

x

x

x

x

x

x

x

x

x

.31

.18

x

x

x

x

.41

1000s (ann.)

x

x

x

x

x

x

x

x

x

x

x

.66

1.95

x

x

.53

x

.71

(decad.)

x

x

x

x

x

x

x

x

x

x

x

.25

.53

x

x

.25

x

.28

1100s (ann.)

x

x

x

x

x

x

x

x

x

x

x

.69

1.70

x

x

.62

x

.46

(decad.)

x

x

x

x

x

x

x

x

x

x

x

.43

.56

x

x

.20

x

.14

1200s (ann.)

x

x

1.36

1.25

x

x

x

x

x

x

x

.69

1.85

x

x

.43

x

.59

(decad.)

x

x

.53

.85

x

x

x

x

x

x

x

.22

1.02

x

x

.16

x

.29

1300s (ann.)

.93

x

1.10

1.36

x

x

x

x

x

x

x

.68

2.47

x

x

.42

x

.68

(decad.)

.39

x

.65

.68

x

x

x

x

x

x

x

.30

.66

x

x

.30

x

.19

1400s (ann.)

.70

x

1.38

1.27

x

x

x

x

x

x

x

.68

2.02

x

x

42

x

.65

(decad.)

.47

x

.68

.33

x

x

x

x

x

x

x

.23

.47

x

x

.18

x

.32

1500s (ann.}

.78

.59

1.13

1.18

x

x

x

x

x

x

x

.60

1.75

1.86

x

.42

.96

.62

(decad.)

.85

.69

.65

.67

x

x

x

x

x

x

x

.37

.56

.74

x

.17

.23

.14

1600s (ann.)

.89

.71

1.27

1.43

.45

.32

.31

.63

.22

x

x

.69

1.63

1.80

1.12

.50

.56

.66

(decad.)

.50

.64

.59

.47

.30

.41

.21

.42

.25

x

x

.43

.53

.66

.49

.14

.14

.19

1700s (ann.)

.84

.62

1.05

1.36

.42

.31

.25

.53

.23

1.81

.98

.74

1.85

2.01

1.11

.49

.53

.65

(decad.)

.96

.27

.87

.54

.21

.26

.14

.20

.34

.77

.25

.43

.40

.70

.55

.22

.18

.28

1800s (ann.)

.97

.78

1.25

1.21

.52

.35

.29

.54

.21

1.89

1.18

.84

1.64

1.67

1.10

.45

.53

.75

(decad.)

1.07

.55

.62

.77

.26

.29

.21

.23

.26

.31

.50

.35

.95

.96

.29

.17

.18

.38

1900s (ann.)

1.51

1.02

1.38

1.67

.51

.33

.29

.53

.27

1.67

1.14

.71

1.78

1.24

1.28

.45

.62

.76

(decad.)

.52

.49

.72

.44

.30

.33

.23

.23

.27

.53

.35

.27

.68

1.10

.59

.21

.22

0.33

TABLE 3 Long-term Means of Interannual and Interdecadal Variability for 18 Temperature Indices (index sources identified in Table 1)

Index

 

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

LTM (ann)

.95

.75

1.24

1.33

.47

.33

.29

.56

.23

1.86

1.09

.71

1.91

1.74

1.15

.46

.66

.68

LTM (dec)

.75

.55

.66

.62

.27

.33

.20

.29

.28

.59

.41

.33

.67

.82

.49

.20

.19

.27

Ratio (%)

79

74

54

46

56

100

69

51

122

32

38

47

35

47

42

44

29

39

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

northern Urals set) interval rather than annual periods is considered.

The two summer temperature reconstructions for South America (Figure 8) are of widely different lengths. A comparison of the overlap period indicates that the Boninsegna series (see Table 1) exhibits a warming trend over the period of record. By contrast, the longer Villalba series exhibits some low-frequency variations but little if any trend over both the full data period and that of the Boninsegna period, beginning about the mid-fifteenth century.

Because of the longer record available, a variability index (described in the section on instrumental data above) was computed using the decadal averages, instead of the annual values. Our purpose here is to examine interdecadal temperature variability, with records spanning several centuries, noting that climate-sensitive tree-ring records are particularly useful for studying decadal-scale climatic fluctuations (Fritts, 1991). The changes in reconstructed climatic vari-

FIGURE 8

Decadal mean values of a reconstruction of summer temperatures for Tasmania, Australia (shown as actual temperatures) and two separate reconstructions for regions in Patagonia, Argentina (shown as anomalies). Units in °C.

ability have been summarized on a century-by-century basis in Table 2. For comparison, we also show in Table 3 the mean values of interannual and interdecadal variability for the full-length record. The variability index was calculated by applying equation (1) to both the annual and decadal-average reconstructed series. We fail to see any consistent trends in interdecadal variability associated with these tree-ring temperature reconstructions (see indices 5-9, 12-13, and 16-18 in Table 2). Interdecadal variability is typically about half the interannual values. This implies that substantial low-frequency variance is present in the paleotemperature record, so that recent high values of reconstructed decadal-mean summer temperature may yet represent an oscillation within the range of natural variability.

Oxygen Isotopes

The oxygen-isotope record has not been directly calibrated against temperature, as has been done for the tree-ring record. However, it is well established that the oxygen-isotope ratio (d18O) is dependent primarily on the temperature of formation of the precipitation, with increasingly negative d18O ratios associated with decreasing temperature. The problems lie in the interpretation of a local ice-core record, not only with respect to local temperature variations, but even in terms of the larger-scale temperature patterns. Several factors control the oxygen-isotope composition of the snow that falls on a given ice body, and temperature is only one of them. The sensitivity of these records to annual temperature variations has been discussed in a number of papers, including the source articles referred to in Table 1. Whether or not these oxygen-isotope records accurately represent a "local" temperature record, we consider them to be sensitive indicators of prevailing climatic conditions within a suitably broad source region. Since our stated purpose is to compare recent changes in a suite of climate-sensitive paleoindicators, we have included them in our comparisons.

Figures 9 and 10 illustrate the changes in decadal averages of the oxygen-isotope ratio of glacier ice cores for different locations in northern Canada and Greenland, and for two low-latitude, high-elevation glaciers (the Quelccaya ice cap in the Peruvian Andes and the Dunde ice cap in the Tibetan Plateau region of China (see Table 1 for sources)). The polar ice d18O record, which ends before 1970, shows relatively less warming (trend toward smaller negative values of d18O) than do the tropical ice cores. Furthermore, the warmest decades in the tropical record occur in the most recent time. The relation of this tropical d18O signal to air-temperature changes in the general location of these records may be partially evaluated with independent observations. It is known that significant melting took place at the Quelccaya site during the 1980s (Thompson, personal communication). Increases in tropospheric

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 9

Decadal-mean values of d18O ratios from different polar-region ice cores. (See text for details.)

temperatures and water vapor obtained with tropical radiosondes for the past couple of decades have been documented by Flohn et al. (1992), although the drift toward less negative d18O ratios appears to have begun earlier in the ice-core record than in the instrumental climate record.

Changes in interdecadal variability are summarized in Table 2 (see indices 1-4 and 14-15). There is a suggestion that decade-to-decade changes in d18O ratios have increased in amplitude at most of these sites. In particular, the calculated interdecadal d18O variability at Quelccaya has increased sharply in the last 100 years. This increase is evident even though the greater number of samples in the upper part of the record permitted better averages to be created for the recent decades than for earlier periods. A similar signal is apparent on the Dunde ice cap as well. These changes, however, may reflect a more localized temperature signal, changes in moisture sources for snowfall on the ice caps, or some other unknown cause unrelated to climate. It should also be noted that these two sites are located at very high elevations, and thus differ from most of the other

FIGURE 10

The same as Figure 9, but for two tropical ice cores (Quelccaya ice cap, Peru, and Dunde ice cap. China).

paleoclimate indices, which are derived from locations at much lower elevations.

Central England Temperature

Decadal means of winter (December-to-January) and summer (June-to-August) mean temperature for central England (Manley, 1974, and subsequent updates) from 1700 to present are shown in Figure 11. The recent decades do not appear to be exceptional in comparison with the total record. The variability curves (not shown) indicate little overall change, with a tendency for the summer and winter seasons to have opposite changes in interdecadal variability (when one increases, the other tends to decrease). Jones and Bradley (1992, their Figure 13.1) have compared the central England temperature record with 12 other long-period station records in Eurasia and North America. Although there are obvious differences among the various climatic series, the twentieth century does not stand out as being a period of higher decadal-scale temperature variability. Most records (although not all) do exhibit the general warming trend of the past two centuries that is associated with the rebound from the Little Ice Age (see Bradley and Jones, 1992).

SUMMARY AND CONCLUSIONS

The purpose of this study was to evaluate whether the climate in recent decades was ''changing," both with regard

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 11

Decadal-mean values of central England winter and summer temperature (in °C).

to its mean value as well as with respect to changes in the variance on different time scales. Our findings are mostly mixed, in that the answer depends on the spatial scales selected for analysis.

For the period encompassing the last 100 years, changes in surface air temperature have been greatest and most positive in the period 1970 to 1990. Interannual variability at the largest spatial scales has also increased, although with seasonal differences. By and large, however, increases in variability at these spatial scales are not unique. When one considers smaller regions (in particular different portions of the contiguous United States) the picture changes, so that the 1980s do not necessarily show up as being a unique period in the instrumental climate history of the country.

In order to have a longer perspective on climatic variations, we utilized data from several paleoclimatic reconstructions of growing-season temperature based on tree-ring records and on d18O ratios extracted from different ice-core records. Data from the latter were also taken as a measure of the prevailing air temperature. Annual values were averaged by calendar decades in order to focus on lower-frequency climatic variability. With the exception of the tropical ice cores, the recent decades were not generally unique, in terms of either the average values or increased variability. We note that tropical sea surface temperatures have been generally above the long-term mean since the mid-1970s. It seems plausible that this recent warmth would be reflected in low (less depleted) d18O ratios on the Quelccaya and Dunde ice caps, although more positive d18O ratios were already evident by the 1950s. It should be emphasized again that the exact relationship between actual past climate and these two high-elevation tropical indices is not fully known. Certainly, the period encompassed by the Little Ice Age is reflected in relatively low d18O values, particularly in the Quelccaya record. The development of additional independent ice-core sequences from the Cordillera Blanca in Peru and from extreme western China by L. G. Thompson and colleagues should help in the interpretation of the nature of the climate signals present in these records (e.g., Thompson et al., 1995).

Do we have an answer to the question posed by the title of this study, namely, whether the climate of the twentieth century is different from that of previous centuries? Obviously, the response is not an unequivocal "yes" or "no." As in most instances when one deals with climatic time series, the answer is "It depends." There appear to be differences in regional responses, and they are evident in both the proxy and the instrumental climate records. Insofar as the United States is concerned, the most unusual decade of the last 100 years may have been the 1930s, although the recent period is probably a close second. We can only say that, for most areas of the coterminous United States, the climate of the most recent decades cannot be considered unique, even in the context of the last century.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
Commentary on the Paper of Diaz and Bradley

EUGENE M. RASMUSSON

University of Maryland

When Drs. Diaz and Bradley present a paper, it always provides an enormous amount of information. The central question, in terms of a synthesis of the wealth of information provided in this paper, is the question posed by the title: "How different is the climate of the twentieth century"—more specifically the past few decades—"from that of earlier centuries?" In addressing this question the authors examined temperature, decade-to-decade temperature changes, and changes in interannual and interdecadal variability on various spatial scales.

The problem in any analysis of this type, of course, is that of inadequate data distribution in time and space. Thus, the analysis was largely limited to the land areas of the Northern Hemisphere, where both instrumental and proxy data are most plentiful.

It may be well to note that changes in the spatial distribution of observations from decade to decade may affect both the computed area average and the variability. Figure 10 in my paper, which appears later, illustrates the effects of using different data distributions in synthesizing global averages of SST, where the problem is probably more severe than it is over the land areas.

Figure 1 in this commentary also shows estimates of the variability in SST obtained from two different analysis schemes. Two estimates of variability are obtained. One is from the optimal averaging (OA) analysis technique of Vinnikov et al., and I think Dr. Groisman will tell us about that one. Another is derived from the simple box average that has been used—for example, by the U.K. Meteorological Office—in deriving SST area averages. The curves in Figure 10 of my paper show the low-frequency variations, i.e., periods longer than 30 years. There are some differences, but they are not too great, about 0.15 degrees around 1910 and 1920, with the OA staying a little bit closer to the mean for the entire period. Now, if we remove the low-frequency variations so that only the variations on time scales less than 30 years remain (Figure 1 in this commentary), and look at that difference between the two analysis schemes, we see that the level of higher-frequency variability may also depend on the first guess used. The box method and the OA with the previous month as the first guess are similar, but the OA with climatology as a first guess shows less variability during the earlier decades of the series. Thus, relative to the box method, the OA shows an increasing variability with time. This comparison illustrates that one

FIGURE 1

Time series of high-frequency variability (periods less than 30 years) of globally averaged sea-surface temperature. Top panel: Dashed curve is obtained by box method (Bottomley et al., 1990). Solid curve is obtained from optimal averaging using climatology as a first guess (Vinnikov et al., 1990). Bottom panel: Curves derived as above except that optimal averaging is obtained by using previous month as the first guess.

must be a bit careful in deriving conclusions about changes in variability from area averages. The estimates of variability may be sensitive to changes in data distribution and the particular analysis scheme used.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Data limitations become extremely severe when one attempts to quantify natural variability on decade-to-century time scales from proxy records. What information can we really derive regarding the patterns of large-scale variability from widely scattered proxy measurements? I think that further effort is required in two areas to maximize and better quantify the information content of the entire climate data base, both instrumental and proxy. First, we need more studies along the lines of data system tests. More specifically, using the more dense data distribution of recent decades, we should determine the loss of information as the distribution of observations is decreased. This will help establish confidence limits for estimates of large-scale averages based on the sparse distribution of observations and proxy series obtained during earlier decades and centuries.

Second, we need to place more emphasis on integration of information from various proxy sources, with the goal of obtaining a better picture of the pattern of global variability during earlier centuries. This is admittedly a formidable task when one is dealing with decade-to-century time scales. I do not believe that the concept of an overall integration of information in decade-to-century variations has yet to take root in the proxy community, but I think it should be strongly encouraged as a long-term goal.

Discussion

GROISMAN: Two quick comments, one for Gene and one for Henry. Gene, I just wanted to note that the SST variability does not reflect conditions over the Arctic. There is about a 25 percent difference in amplitude between SST and marine air temperature over the open ocean, the latter being higher, so there may be other reasons for the difference you cite. Henry, the problem with tree rings is that there are very few 1000-year-old trees, so in a time series you can't resolve the same mean variance.

DIAZ: I used only air temperature over land, no SSTs. Also, I started in 1891, when we had fairly good coverage. The changes in data distribution don't have much influence on the large-scale averages. As for the tree rings, we did try to compensate by looking at decadal means and interdecadal variability, and I don't think the middle section has much of a problem.

KEELING: I have a figure that is based on the Jones-Wigley temperature record, with a slightly different smoothing. Some of the early variability and the apparent biennial signal are probably the result of sparse data; beginning in 1951 the record is fairly homogeneous. You can see the 1958, 1961, 1963, 1982 El Niños. I'd like to know whether there is some difference in the quantity of data, or in its processing now that satellite data and ground data are being mixed, that has changed data variability in the last decade.

TRENBERTH: Perhaps when you fit the spline you are taking out more of the variance at the end.

GHIL: Henry, I was puzzled by the apparently much higher interdecadal variability in the second of your first two tree-ring figures.

JONES: It's partly due to the indexing procedure that there is more low-frequency variance in the top curve than the bottom one, though some of it could be real. KARL: It's important to remember that there are many different ways to classify variability. I suspect that I could show more or less of it just by defining it differently. On this particular figure, I think we need to be aware that one is for a longer season than the other, and the shorter period of a time series will give you higher variability.

DIAZ: Another point is that getting longer time series from proxy records may not improve our understanding of climate change mechanisms.

GROOTES: I want to comment on the Quelccaya ice-core records you used. As you go down from the surface, the seasonal cycle disappears. It reappears about 1880 and continues significant to about 1480. Apparently when there are low values the record is smoothed so little that the seasonal cycle is preserved. As temperature changes it becomes more difficult to calibrate the isotope changes; you may be looking at post-depositional changes rather than the signal itself. Tree rings and ice cores both require particular attention to understanding exactly what you're recording, especially if you are talking about temperature.

MYSAK: For anyone who's interested, I'd like to mention that Brian Luckman at the University of Western Ontario has tree-ring records from the Canadian Rockies for about the last thousand years. He's showing a nice cycle of 150 years or so.

MCGOWAN: Are you comparing tree rings taken from similar altitudes at those different locations? Also, how does the variance of instrumentally measured mountain temperatures compare with that of lowland temperatures?

DIAZ: Most of the data sets are from moderate elevations. As for the high/low elevation temperature variances, the trends are reduced with respect to surface, but they're still comparable.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Variability of African Rainfall on Interannual and Decadal Time Scales

SHARON E. NICHOLSON1

ABSTRACT

This paper examines the rainfall fluctuations that have occurred over the continent of Africa during the past century. Major characteristics include anomaly patterns that are continental in scale, with marked teleconnections between particular sectors of the continent, and quasi-periodic fluctuations with dominant time scales of approximately 2.3, 3.5, and 5 to 6 years. In the Sahelian regions these higher-frequency fluctuations are not apparent; instead, most of the variance is found on time scales of 7 years or longer. Both dry and wet episodes tend to persist for one or two decades at a time; decadal means change by a factor of 2. These characteristics are also apparent in the Sahelian climate record for earlier centuries. Elsewhere on the continent, synchronous changes of rainfall occur, but the "wet" or "dry" decades are not as uniformly wet or dry as in the Sahel. Examples are the 1950s, in which above-average rainfall prevailed over most of the continent, and the 1980s, during which most of the continent was abnormally dry. It is likely that the factors producing higher-frequency interannual variability differ from those producing the decadal-scale fluctuations. Both sea surface temperature and land-surface feedback have been suggested as causes of the unusual persistence in the Sahel, but the latter cannot explain the change of decadal means.

INTRODUCTION

The continent of Africa is primarily tropical or subtropical, so the climatic changes that have taken place have been manifested principally in the rainfall regime. The rainfall fluctuations that occur are extreme; those in the semi-arid regions are probably unmatched anywhere in their magnitude and spatial extent. In the Sahel, for example, decadal averages change by a factor of 2 or more. Most major periods of anomalous rainfall, either individual years or decadal-scale episodes, affect most of the continent.

This paper examines the rainfall fluctuations that have occurred during the twentieth century and compares them with historical episodes. The time scales of the fluctuations are summarized, and the decadal-scale fluctuations described in greater detail. The intra-continental teleconnections, and teleconnections to tropical or global rainfall anomalies, are also considered. Finally, a brief review of causal mechanisms is provided.

Undoubtedly, changes in the thermal regime over Africa have occurred, especially along the continent's poleward

1  

Department of Meteorology, Florida State University, Tallahassee, Florida

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

extremes and at higher elevations. These are minor compared to rainfall fluctuations; moreover, they have not been systematically studied. Therefore, this paper will be limited to variability of rainfall.

METHODOLOGY

Most of the description of rainfall variability in this paper derives from previous studies made by the author. Analyses are based on a network (Figure 1) of approximately 1400 stations, the statistics of which are fully described in Nicholson et al. (1988). From these, a spatially averaged data set has been derived, utilizing 90 geographical regions (Figure 1) that are homogeneous with respect to the interannual variability of rainfall.

Regional averages are generally expressed as a standardized annual departure from the long-term mean (i.e., the mean over the entire length of record). In select analyses, the rainfall anomalies are also presented as a percentage departure from the long-term mean. About 75 percent of the station records commence before 1925 and continue to 1984 or later. Hence, the length of record is generally at least 60 years.

Regionally averaged rainfall Rj is calculated as

where Rj is the regional rainfall departure in the year j, and Ij is the number of stations available in year j. In the above, xij is a standardized rainfall departure at station i in

FIGURE 1

Map of rainfall stations in African archive and homogeneous rainfall regions. (From Nicholson et al., 1988.)

year j. It is calculated as

where rij is the annual total at the station in the year j, is the station's long-term annual mean, and si is the variance of annual rainfall at the station.

These regional means are utilized to examine the temporal and spatial structure of rainfall variability over the continent. They also illustrate the magnitude of decadal anomalies. In some analyses 1°-grid averages are utilized instead of the regional averages.

CONTINENTAL-SCALE PATTERNS OF RAINFALL VARIABILITY

Previous studies (e.g., Nicholson, 1986) have utilized a map-classification technique of Lund (1963) to describe the preferred spatial configurations of rainfall anomalies over Africa. This technique is based on linear correlation of map patterns. Most of the rainfall variability over the continent is described by the six "anomaly types" illustrated in Figure 2. Unlike eigenvector techniques, the patterns produced by the Lund method are not orthogonal; hence, these six anomaly types really describe only four basic configurations of anomalies. These include a tendency for above(or below-) average rainfall over most of the continent (Types 2 and 4) and two patterns (Types 1, 3, 5, and 6) illustrating an opposition between equatorial and subtropical latitudes. Approximately 21 of the 90 years between 1901 and 1990 are represented by a pattern of above-average rainfall in the equatorial regions but below-average rainfall in subtropical latitudes, and 12 by the opposite pattern, which is illustrated by Type 3 in Figure 2. Fourteen of the 70 years are best represented by the Type 2 anomaly pattern, with below-average rainfall throughout most of the continent, and 15 by Type 4, with above-average rainfall in most areas.

These patterns are introduced here for two reasons. First, as will later be seen, they are typical of decadal-scale rainfall fluctuations over Africa. Second, they demonstrate that much of the rainfall variability over the continent can be described using a small number of regional time series. The representative regions, shown in Figure 3, include five latitudinal zones running from the southern margin of the Sahara to the Guinea Coast of the Atlantic, eastern equatorial Africa, and two sectors of southern Africa loosely termed the northern and southern Kalahari.

DECADAL-SCALE FLUCTUATIONS
Fluctuations During Recent Decades

Two major, decadal-scale rainfall anomalies have affected the African continent within the current century:

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 2

The most frequently occurring continental-scale rainfall anomaly types for the twentieth century. (From Nicholson, 1986; reprinted with permission of the American Meteorological Society.) The values represent regionally averaged standardized departures.

TABLE 1 Rainfall Anomalies for the Decade 1980-1989

 

1950-1959 Depart.*

(%)

Stdzd. Depart.

(% s)

1970-1979 Depart.

(%)

Stdzd. Depart.

(% s)

1980-1989 Depart.

(%)

Stdzd. Depart.

(%s)

Northern Sahel

37

54

-31

-47

-24

-35

Central Sahel

28

71

-22

-55

-31

-82

Southern Sahel

15

62

-13

-53

-20

-85

Soudano-Guinean Zone

3

22

-5

-36

-8

-56

Guinea Coast

3

15

-6

-31

-7

-35

Eastern Africa

-2

-9

0

4

2

15

Northern Kalahari

8

28

6

20

-5

-21

Southern Kalahari

7

23

10

34

-7

-27

* Anomalies are expressed as a percentage departure from the long-term mean and as a standardized departure (ratio of the departure from the mean to the standard departure).

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 3

Geographical regions described in detail in this study.

relatively wet conditions in the 1950s and relatively arid conditions in the 1980s. The periods of anomalous conditions have not been precisely 10 years in length, nor have they commenced at the onset of each numeric decade (e.g., 1950 or 1960). Nevertheless, these periods and the transition between them are well illustrated from decadal maps and tabulations (Figure 4 and Table 1).

Fluctuations have been particularly large in Sahelian West Africa, where rainfall was about twice as great during the 1950s as during recent years (Table 1). The 1950s (Figure 4a) were also wetter than average over most of the rest of the continent, but abnormally dry in the equatorial regions. The wet episode ended abruptly in the Sahel toward 1960, after which time a change occurred in many other regions.

In the early 1960s, phenomenally wet years occurred in equatorial regions, especially in eastern Africa, while droughts affected much of southern Africa. As an example, Wajir, in northern Kenya with mean annual rainfall of 285 mm, received 612 mm in November 1961; at many Kenyan stations, rainfall that month was five to ten times the long-term mean. The levels of Lake Victoria and other Rift Valley lakes suddenly rose several meters commencing in 1961, abruptly attaining levels unmatched since the end of the nineteenth century. The 1960s as a whole (Figure 4b) was a period of relatively wet conditions throughout the equatorial latitudes and abnormally dry conditions throughout most of the subtropics of both hemispheres. The dry anomaly was particularly strong in the Southern Hemisphere.

The early 1970s were a period of drought over much of Africa (Nicholson, 1986), but later in the decade a series of tremendously wet years occurred in southern Africa, so that for the decade as a whole, a strong opposition is apparent between the two hemispheres (Figure 4c). This is not a common configuration of anomalies, and had previously occurred in only a few isolated years. A plausible explanation is presented below.

The 1980s (Figure 4d), in strong contrast to the 1950s and unlike other decades, was a period of abnormally dry conditions over most of the continent, including most equatorial regions. The decadal anomalies are largest in the Sahel, on the order of + 25 percent in the 1950s and -30 percent in the 1980s (or nearly one standard deviation). In most other areas the departures for the 1950s (Nicholson, 1989) and 1980s (Table 2) are less than 10 percent of the mean.

The continental scale of these anomalies and the abrupt change around 1971 are evident in Figure 5, which shows the linear correlation of the rainfall anomaly pattern of each year since 1901 with anomaly Type 2 in Figure 2. This anomaly type is one of negative departures throughout the continent. Most years from 1950 to 1970 are negatively correlated with it; correlations for 1950-1957, 1961, and 1962 are significant at the 99 percent level. In contrast, nearly every year since 1970 is positively and, in most cases, significantly correlated with the Type 2 pattern.

Teleconnections and Other Characteristics of Rainfall Variability: Comparison of Modern and Historical Periods

The previous section identified several characteristics of African rainfall variability on decadal-scales. One is the tendency of anomalous conditions to persist for roughly a decade. Another is the large magnitude of anomalies in the Sahel. A third is the teleconnections within Africa on decadal-scales, as illustrated by the anomaly types of Figure 2 and the decadal patterns for the 1950s and 1980s (Figures

TABLE 2 Mean Annual Rainfall (mm) at Select Sahelian Stations for Two Recent Periods

 

1950-59

1970-84

Bilma

20

9

Atbara

92

54

Nouakchott

172

51

Khartoum

178

116

Agadez

210

97

Timbuktu

241

147

Nema

381

210

Dakar

609

308

Banjul

1409

791

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 4

Mean rainfall for the periods 1950-1959, 1960-1969, 1970-1979, and 1980-1989 (panels a through d respectively). Rainfall is expressed as a percentage departure from the long-term mean (i.e., the mean for the entire length of record, commencing in 1901 or as soon thereafter as rainfall observations begin). Station data are averaged over 1° squares to facilitate presentation. Shading highlights regions with positive anomalies. (The 1970-1979 map is reprinted from Nicholson, 1994, with permission of Arnold, of the Hodder Headline Group; the 1980-1989 map is reprinted from Nicholson, 1993, with permission of the American Meteorological Society.)

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 5

Correlation coefficient between the anomaly pattern of each year's rainfall (averaged over the regions shown in Figure 1) and the anomaly pattern corresponding to Type 2 in Figure 2. For the 99 percent confidence level, r ranges from about 0.28 to 0.31, depending in part on the number of regions with available data in a given year.

4a and 4d). Overall, the decadal rainfall fluctuations tend to occur quasi-synchronously over the continent. There is a strong tendency for the fluctuations in the subtropical latitudes of both hemispheres to be roughly in phase, as well as a tendency for the fluctuations in the equatorial and subtropical latitudes to be opposed to each other. These same characteristics are evident in the historical past.

Figures 4a and 4d clearly demonstrate that major periods of anomalous rainfall in the Sahel correspond to continental-scale anomalies as well. This is likewise true for historical fluctuations of Sahel rainfall, such as the droughts of the 1820s and 1830s and the 1910s (Figure 6). Both of these periods were times of frequent drought throughout much of the continent. A period of sufficient rainfall from about 1870 to 1895 was part of a continental pattern mirrored by the 1950s pattern (Figure 4a). In the 1950s, lake levels rose dramatically from Chad in the north to Ngami in the south, and the floods of rivers like the Niger and Nile were consistently high. Near Timbuktu, in the Niger Delta where annual rainfall is now 195 mm, wheat production thrived to such an extent that grain was exported to neighboring regions. In the semi-arid regions of North Africa rainfall probably averaged about 25 to 30 percent above the twentieth-century mean during the period 1870-1895 (Nicholson, 1978).

Interestingly, the historical patterns depicted in Figure 6 were constructed prior to the derivation of the rainfall anomaly types in Figure 2; they were based on proxy data and qualitative indicators. While this origin might render the reconstructions questionable, their reliability is supported by the clear correspondence to the major anomaly types apparent in the modern quantitative rainfall record. They also demonstrate other major characteristics of African rainfall variability established with modern records: the decadal-scale persistence of anomalies in the Sahel, and their continental scale.

In certain cases, global teleconnections to decadal-scale African rainfall fluctuations have also been demonstrated. One example is the change from "wetter" to "drier" conditions around the turn of the century. This change was seen throughout the tropics, with a major shift occurring around 1895 (see, e.g., Kraus, 1955a,b). The anomalously dry 1830s was also a period of globally anomalous climate, often considered to mark the end of the Little Ice Age. The anomalous conditions of the 1950s are also noted in other tropical and subtropical regions such as Central America, India, and Australia. These decadal teleconnections have not, however, been studied systematically.

Fluctuations on Time Scales of Centuries

The climatic fluctuations that have affected Africa on time scales of centuries are difficult to establish because

FIGURE 6

African rainfall anomalies for three historical periods. Minus signs denote evidence of drier conditions; plus signs denote evidence of above-average rainfall; zeroes denote near-normal conditions; circled symbols denote regional integrators, such as lakes or rivers. (From Nicholson, 1978; reprinted with permission of Academic Press Ltd.)

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

there are few long-term climate data sets or even proxy records from which past climatic conditions can be reconstructed. Two notable exceptions are Lake Chad, the fluctuations of which have been well documented through both historical and geological methodologies (Maley, 1976), and the Nile, for which quasi-continuous written records extend back to the seventh century (Toussoun, 1923; Riehl et al., 1979). However, neither record lends itself to completely unambiguous interpretation.

The Lake Chad record suggests two long periods of comparatively wet conditions in the West African Sahel, one in the ninth through fourteenth centuries and another from the sixteenth through nineteenth centuries. The first was synchronous with the Medieval Warm epoch and the second occurred more or less during the Little Ice Age of Europe. The more humid climates in the present-day Sahel, which the Lake Chad record implies, are further supported by numerous historical and geological indicators throughout West Africa. Information for other regions of Africa is more scarce, but the Nile record and those of other East African lakes suggest that the first humid period affected much of eastern equatorial Africa as well. The second appears to have affected that region and much of southern Africa (Nicholson, 1979a, 1981). The decline of the recent wetter conditions took place toward 1800, at a time when the Northern Hemisphere was experiencing the end of the Little Ice Age.

INTERANNUAL VARIABILITY OF RAINFALL

Because of the limited availability of meteorological data for time scales of 10 to 100 years, it may be necessary to evaluate mechanisms of variability on these time scales at least partially through analogy with individual years. For this reason, a description of the interannual variability of African rainfall is relevant here. This section examines the temporal characteristics of rainfall variability in various regions in the domains of both time and frequency. Several of the features exhibited will have to be adequately accounted for by any mechanism that is proposed as explaining variability on 10- to 100-year time scales.

Figures 7 and 8 show the standardized rainfall departure series for select regions of Africa. The most striking feature in any of the series is the extreme persistence of rainfall anomalies in Sahelian West Africa (Figure 7). This persistence was first identified by Bunting et al. (1976), then later by Walker and Rowntree (1977), Nicholson (1979b, 1983), and Lamb (1982). In the Sahel as a whole (a combined series of northern, southern, and central Sahel in Figure 7), regionally averaged rainfall exceeded the long-term mean in every year from 1950 to 1967; it has been below the mean in every year since 1970. Rainfall also tended to be above average throughout most of the period 1927 through 1936 and below average in the 1910s and 1940s. To some

FIGURE 7

Rainfall fluctuations in West and North Africa (1901 to 1990), expressed as a regionally averaged standard departure. Location of the regions is indicated in Figure 2.

FIGURE 8

Rainfall fluctuations as in Figure 7, but for regions of southern and eastern Africa, as indicated in Figure 2.

degree, this persistence is also apparent in the Soudano-Guinean and Guinean regions further south.

Such long-term persistence is not evident in any of the other regional time series. For example, the longest run of ''dry" years is 5 years in the northern Kalahari, 6 years in the southern Kalahari, and only 3 years in eastern Africa (Figure 8). The longest run of "wet" years is 5 years in all three regions. However, longer periods of predominantly wet or dry years have prevailed. These include, for example, the "dry" 1950s in East Africa and the drier periods of 1926 to 1935, the 1960s, and the 1980s in the southern Kalahari.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

The contrast between the Sahelo-Saharan region and elsewhere in Africa is underscored by Figure 9. It shows the results of low-pass filtering the rainfall series for the regions in Figure 1 and isolating the percentages of variance of annual rainfall on time scales longer than 7 years. This variance represents a combination of decadal-scale or longer fluctuations and year-to-year persistence of rainfall anomalies. Low-frequency variance is by far the greatest in the Sahelian regions, accounting for as much as 45 to 60 percent of the variance. Elsewhere it is generally between 20 and 35 percent.

Spectral analysis shows that the dominant time scales of rainfall variability throughout the continent are on the order of 2.2 to 2.4, 3.3 to 3.8, and 5 to 6 years (Rodhe and Virji, 1976; Ogallo, 1979; Nicholson and Entekhabi, 1986). These are not the only periodicities that have been detected, but they are the only ones apparent throughout large expanses of the continent and the only ones corresponding to time scales of distinct causal mechanisms such as SST fluctuations or ENSO. The works of Tyson (e.g., 1980), Fleer (1981), and others provide more detail on select regions.

The occurrence of these quasi-periodicities in the regions depicted in Figure 1 were evaluated by Nicholson and Entekhabi (1986). The results are shown in Figure 10. The 5- to 6-year time scale is dominant in equatorial regions; the 3.5-year time scale is dominant in much of southern Africa and in parts of equatorial Africa; the 2.3-year time scale is dominant in the eastern half of equatorial and south-

FIGURE 9

Percentage variance of annual rainfall with periods of 7 years or longer for regions depicted in Figure 1 (in circled areas, variance exceeds 40 percent).

ern Africa. Interestingly, these relatively high-frequency fluctuations (Figure 10) are nearly absent throughout the Sahelo-Saharan region of West Africa, where the spectra are dominated by low-frequency fluctuations (Figure 9). Relatively low-frequency fluctuations have been noted also in parts of South Africa (Tyson, 1980), with time scales of 18 to 22 years being common.

An examination of wet and dry years throughout the continent (e.g., Nicholson et al., 1988) shows other characteristics of interannual variability that must be explained. One is that in many regions interannual variability is associated not with a change in the intensity of the rainy season but by a change in the length of the season or by anomalous rainfall in comparatively dry months. This can be demonstrated for East Africa as well. Throughout most of the region, the main rainy season is April to May, but interannual variability is best correlated with anomalies of the "short rains" of October-November (Figures 11 and 12). The importance of transition-season fluctuations also helps to explain the rainfall's continental-scale coherence over regions (such as the African Northern and Southern Hemispheres) where the rainy seasons are six months out of phase.

Rainfall in East Africa illustrates another interesting characteristic: The factors that produce the climatological-mean

FIGURE 10

Distribution of significant spectral peaks with periods on the order of 2.2-2.4, 3.3-3.8, and 5-6 years. Data are for regions shown in Figure 1, with blanks indicating no significant peak in the regional rainfall. (From Nicholson and Entekhabi, 1986; reprinted with permission of Springer-Verlag.)

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 11

Correlation between the seasonal and annual rainfall departure series for East Africa (representing the countries Kenya, Uganda, and Tanzania). Values approximately equal the percentage variance explained by each season: stippling, 50 to 70 percent; shading, greater than 70 percent.

FIGURE 12

Time series of rainfall departures for individual seasons (solid line) compared with the annual rainfall departure series representing East Africa (Kenya, Uganda, and Tanzania). Data are expressed as a percentage standard departure.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

patterns are not necessarily those that explain interannual variability. Within this region, mean annual rainfall and its seasonality are highly heterogeneous (Figure 13), being associated with several features of the general atmospheric circulation as well as with regional-scale orographic and lacustrine effects. Nevertheless, the patterns of interannual variability are markedly similar throughout the region (Figure 14), so that there is a remarkable resemblance between the time series for the region as a whole, and for those four climatologically homogeneous subregions. The first eigenvector of annual rainfall shows loadings of the same sign throughout the region; the pattern explains 36 percent of the variability of annual rainfall and 52 percent of the variability during the "short rains" (Nyenzi, 1988).

In summary, there are several notable characteristics of rainfall variability that must be explained by mechanisms for producing this variability. These include, first of all, the continental-scale coherence and the seemingly contradictory contrast between the dominant low-frequency fluctuations in Sahelo-Saharan regions and the high-frequency fluctuations elsewhere, in regions that show strong teleconnections to the Sahel on time scales of 1, 10, and 100 years. The characteristic time scales of 2.3, 3.5, and 5 to 6 years must likewise be explained, as must be the seasonality of the changes and the coherent patterns of interannual variability in climatically heterogeneous regions.

MECHANISMS OF RAINFALL VARIABILITY

Most studies of African rainfall have focused on interannual variability. Numerous authors have demonstrated relationships between SSTs and rainfall in various parts of Africa (e.g., Lamb, 1978; Lough, 1986; Semazzi et al., 1988; Wolter, 1989; Folland et al., 1991; Nicholson and Entekhabi, 1987). For parts of eastern and southern Africa linkages to ENSO have also been clearly established (by, e.g., Farmer, 1988; Ropelewski and Halpert, 1987; Ogallo, 1987; Harrison, 1983; Lindesay et al., 1986; Nicholson and Entekhabi, 1986, 1987). For Sahel rainfall, a number of other factors, such as changes in the upper-level winds and in the Hadley and Walker circulations, have also been implicated (Kanamitsu and Krishnamurti, 1978; Newell and Kidson, 1984).

Very little attention has been paid to the causes of the lower-frequency variability. The exception is the work by Folland and collaborators (e.g., Folland et al., 1991), which is limited to the Sahel. Virtually no studies have been carried out on the causes of the continental rainfall fluctuations on decadal time scales.

Rowell et al. (1992, 1995) suggest that the twentieth-century trends in Sahel rainfall can be attributed to interhemispheric differences in sea surface temperature. The interhemispheric contrast increased in mid-century and decreased sharply as the droughts began in the 1970s and 1980s. This mechanism does not, however, account for the extreme persistence of anomalies in the Sahel, or their lack of persistence in regions such as southern Africa, which otherwise appear to have teleconnections to the Sahel in individual years and on decadal-scales.

Land-Surface Feedback in the Sahel

As an example of how this might work, consider a year during which anomalous large-scale circulation triggers drought in the Sahel and in other African regions showing strong teleconnections to the Sahel. When the large-scale forcing returns to normal, so do rainfall conditions in most areas. However, in the Sahel the drought conditions might be reinforced by local, land-surface feedback. In that case they will persist until a year in which a large-scale circula-

FIGURE 13

Maps of rainfall characteristics over East Africa: mean annual rainfall, number of rainfall maxima in the seasonal cycle, and the months of these maxima. (From Nicholson et al., 1988.)

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 14

Time series of rainfall departures for the four regions shown in the inset map (solid lines) compared to the time series for East Africa as a whole (dashed lines). Values are expressed as a percentage standard departure from the long-term mean.

tion anomaly occurs that is sufficiently great to override that local forcing.

In southern Africa, a region with rainfall teleconnections to the Sahel (see above) but without strong interannual persistence of rainfall anomalies, the interannual variability of surface fluxes is smaller and less spatially coherent. The rain-bearing systems in that region are also less sensitive to surface processes than are those affecting the Sahel.

This might suggest the following hypothetical scenario for explaining the unusual pattern of the 1970s: below-normal rainfall in Sahelo-Saharan regions and above-normal rainfall in corresponding sectors of Southern Hemisphere Africa. The factors producing drought commencing in 1968 were reinforced in the Sahel by land-atmosphere interaction that caused the negative anomalies to persist without interruption throughout the 1970s. In the mid-to-late 1970s, some factors in the general atmospheric circulation produced wet conditions in southern Africa and also enhanced rainfall in Sahelo-Saharan regions. In the latter, however, they were insufficient to override the drought-promoting feedback and resulted in less extreme negative anomalies in years such as 1974 to 1978, but not in rainfall exceeding the long-term mean.

It is interesting to note that when only high-frequency fluctuations are considered, the rainfall fluctuations in the Sahel bear a much stronger similarity to those in the Kalahari (Figure 15) during the past 40 years. The correlation between the resulting "Sahelian" series and the rainfall series for the northern Kalahari is 0.5, just slightly lower than the correlation (0.57) between the series for the northern and southern Kalahari in Figure 8.

The results shown in Figure 15 are consistent with the idea that higher-frequency fluctuations, or yearly departures from a decadal or multi-decadal mean, are forced by similar factors in Sahelian and southern Africa. but that additional causal mechanisms of lower-frequency variability must be sought in the Sahel. A similar contrast in forcing is apparent between various parts of the Sahel and between the early

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 15

Rainfall variations in the Sahel and the Kalahari for the period 1950 to 1990. Rainfall in the Kalahari is a standardized departure from the long-term mean. For the Sahel, standardized departures for 1950 to 1969 are calculated with respect to the mean for that period; those for 1970 to 1990 are calculated with respect to the latter period. For both, the variance is calculated with respect to the entire length of the record. This is roughly equivalent to removing the low-frequency variance from the Sahel series. (From Nicholson, 1993; reprinted with permission of the American Meteorological Society.)

and late rainy-season months. High-frequency forcing is also stronger in June-July than in August-September. This observation is relevant because one hypothesized factor in the feedback, the African Easterly Jet (see below), is operative mainly in the latter period.

The potential feedback mechanisms involve the effect of such factors as soil moisture and dust generation on surface fluxes and atmospheric heating. The dynamic link to the Sahel could be the mid-level African Easterly Jet. The African Easterly Jet is a consequence of the meridional temperature gradient in the region, which in turn is produced by the strong thermal contrast between the dry Sahel-Saharan region and the forests and Atlantic Ocean to the south. It is analogous to the mid-latitude westerly jet stream, in that it provides the instability and energy for the development and maintenance of rain-bearing systems in the Sahel. It has a maximum around 650 or 700 mb, a level where the temperature structure is strongly affected by surface-generated dust. The intensity of the dust, in turn, is highly inversely correlated with rainfall in the Sahel.

The contrast in surface fluxes between wet and dry years (Lare and Nicholson, 1993) is sufficiently large to produce the observed changes of jet intensity between wet and dry years (Newell and Kidson, 1984). These in turn are sufficiently large to influence the jet's dynamics, and are thus capable of altering such variables as the number, size, and intensity of disturbances in the region.

SUMMARY AND CONCLUSIONS

Numerous studies have firmly established that:

  • decadal-scale rainfall fluctuations have been quite marked in Sahelian Africa;

  • these fluctuations tend to have a continental spatial scale (although less extreme and persistent than in the Sahel) with distinct teleconnection patterns;

  • the characteristics of modern decadal anomalies are also apparent in earlier centuries; and

  • the 1950s and, to a lesser extent, 1960s were continentally "wet," while the 1980s and, to a lesser extent, 1970s have been continentally "dry."

Among the major areas in which research is needed are the causes of the decadal-scale fluctuations of African rainfall, particularly the continentally coherent patterns; the role land-surface feedback may play in the decadal-scale anomalies in the Sahel; and the teleconnections between African rainfall anomalies on the decadal-scale and global climate anomalies.

Our understanding of decadal-scale fluctuations could be greatly enhanced by comparative analysis of the 1950s and more recent decades. Unfortunately, many of the necessary data are routinely available only for about the last 20 years. However, a great many data, especially upper-air data, do exist for the 1950s and 1960s, but in obscure archives where they are not readily available. Effort should be devoted to creating usable data sets for earlier periods, especially the 1950s.

Our knowledge of African climate fluctuations on the scale of centuries is considerably more vague. In order to establish these regional fluctuations' causal mechanisms and their relationships to global climate fluctuations, they must be more precisely delineated in time and space. This will require the preparation of better proxy data sets. Few attempts have been made to use, for example, tree-ring or varve chronologies. Also, the innovative paleoclimatic techniques that have been developed recently might be appropriate here as well. Creation of these proxy data sets for Africa should be a high priority.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

On the Initiation and Persistence of the Sahel Drought

JAGADISH SHUKLA1

ABSTRACT

On the basis of analysis of data collected over the Sahel, India, and China, it is conjectured that the occurrence of above-normal rainfall over the Sahel during the decade of the 1950s and below-normal rainfall during the decades of the 1970s and 1980s is a manifestation of natural variations in the planetary-scale coupled ocean-land-atmosphere climate system. In the light of the results of general-circulation model sensitivity experiments, and the fact that the Sahel region is particularly susceptible to changes in land-surface conditions, it is concluded that strong local atmosphere-land interactions over the Sahel region have contributed toward further reduction of rainfall. Both natural climate variability and human activities degrade the land surface in a way that exacerbates the ongoing drought, and therefore there is no guarantee that natural variability can reverse the present trend.

INTRODUCTION

The West African famine during the late 1960s and the 1970s focused the world's attention on the Sahel. The Sahel is a 200- to 500-km wide band across the southern reaches of the Sahara desert. It is ecologically fragile, though it supports nomads, herders, and sedentary farmers; it has a mean rainfall of 300 to 500 mm per year, which is extremely variable. During the past 25 years, rainfall over the Sahel has been significantly lower than the long-term mean. In the recorded meteorological data for the past 100 years, there is no other region on the globe of this size for which spatially and seasonally averaged climatic anomalies have shown such persistence. (See the paper by Nicholson in this section.)

In this paper we shall attempt to address the following questions concerning the Sahel drought:

  1. What are the causes of the onset of the drought? In particular, what significant roles did the natural variability of the global climate system and of human activities play in the initiation of the drought?

  2. What mechanisms are responsible for the persistence of the Sahel drought?

  3. What are the prospects for reversal of the current trend? In particular, what roles might the natural variability of the global climate system and human intervention play in reversing the current trend?

It is of course difficult, if not impossible, to give precise

1  

Center for Ocean-Land-Atmosphere Studies, Institute of Global Environment and Society, Calverton, Maryland

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

answers to these questions. We will attempt to address them by interpreting past observational data and the results of controlled sensitivity experiments made with general-circulation models (GCMs).

ONSET OF THE SAHEL DROUGHT

Figure 1 shows the 10-year running-mean seasonal rainfall anomalies over Sahel, India, and China. The Chinese rainfall data are available for only a 30-year period (19511980); therefore, the Indian and the Sahel rainfall data have also been shown for that period.

We interpret these data to suggest that the occurrence of above-normal rainfall during the decade of the 1950s and the early 1960s, and below-normal rainfall during the late 1960s and the 1970s, is a feature common to the vast areas of African-Asian monsoon, not one confined to the Sahel region. We do not know why rainfall over such a large geographic region, covering two continents, was above normal during the 1950s. We also do not know why the transition from above-normal to below-normal rainfall that

FIGURE 1

Ten-year running mean of normalized rainfall anomalies averaged for all the stations over the sub-Saharan region (10° - 25°N, 18°W - 25°E), India, and China. North China refers to the area to the north of 35°N.

took place during the late 1960s occurred over the entire African-Asian monsoon region. However, on the basis of comprehensive GCM sensitivity studies (Folland et al., 1986; Rowell et al., 1992; Xue and Shukla, 1993), it is now generally accepted that the rainfall anomalies, especially over the Sahel, are consistent with, and can be simulated using, the global sea surface temperature (SST) anomalies during these periods.

On the basis of the GCM sensitivity experiments andobservations shown in Figure 1, we conjecture that the onset of the Sahel's rainfall deficit in the late 1960s, along with the above-normal rainfall in the 1950s, should be explained as a regional manifestation of the planetary-scale variability of the global coupled ocean-land-atmosphere climate system. Considering the large spatial and temporal extent of these rainfall anomalies, we cannot conceive of any possible mechanism by which local human-induced changes could initiate such major droughts over two continents. We therefore reject the notion, often mentioned in popular literature, that the Sahel drought was initiated by human activities.

We believe that the pertinent question about the Sahel drought is not what initiated the drought (we conjecture it was natural variability) but why the drought has persisted for 25 years.

PERSISTENCE OF THE SAHEL DROUGHT

We further propose that strong interaction between atmospheric and land-surface processes is one of the primary causes of the drought's persistence: A reduction in rainfall (due to natural variability of the climate system) would change the land surface characteristics (increase in albedo and decrease in vegetation), which in turn would cause further reduction in rainfall. The basic mechanisms of atmosphere-land coupling have the potential to perpetuate an initial drought (or an initial excessive-rainfall) regime until they are reversed by an opposite forcing, which could be produced by natural variability.

The extent to which changes in the land-surface properties can influence the local climate depends on the spatial scale of the changes in the land-surface characteristics and the geographic location of the region. We submit that the Sahel region is particularly sensitive to atmosphere-land coupling processes. The Sahel region is geographically unique in that it is adjacent to the largest desert on the earth's surface, and it represents the largest contiguous land surface without extensive mountain terrains. The absence of large-scale topographic forcing in the region is perhaps one of the main reasons for the lack of zonal asymmetry in the rainfall pattern there. As Charney (1975) pointed out in his seminal paper on the dynamics of Sahel drought, climate in the subtropical margins of the deserts is particularly vulnerable to changes in the land-surface properties.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

During the past 20 years there have been a large number of GCM sensitivity studies (Mintz, 1984; Dirmeyer and Shukla, 1993) that show that increases in albedo and decreases in vegetation cover and soil moisture tend to further reduce rainfall over the Sahel. We do not know of any GCM sensitivity study challenging this basic result. We present here the results of a recent GCM sensitivity study (Xue and Shukla, 1993 and an unpublished manuscript) in which the Center for Ocean-Land-Atmosphere Interactions (COLA) model was integrated with two sets of land-surface conditions over the region south of the Sahara desert. In one set of integrations, referred to as the "desertification" experiment, it was assumed that the area approximately between 10° and 20°N and 15° and 40°E was covered by shrubs above bare soil. In the other set of integrations, referred to as the "reforestation" experiment, the area approximately between 15° and 40°E was covered by broadleaf trees above ground cover. The first set of surface conditions (desertification) was considered to represent the exaggerated state of current conditions. The second set of surface conditions (reforestation) represented a hypothetical situation in which vegetation was maintained in the margins of the Sahara Desert. It was an exaggeration of the conditions during the 1950s, when rainfall over the Sahel was above normal, and the latitude of the 200 mm isoline of seasonal rainfall in the western Sahel was at about 18°N. The model was integrated with identical global SST patterns for each change in the land-surface conditions.

Figure 2a shows the difference between the observed rainfall averaged for 1950 plus 1958 and 1983 plus 1984. Figure 2b shows the difference between the desertification and reforestation experiments for seasonal mean rainfall. The similarity between Figures 2a and 2b is remarkable. Both show negative rainfall departures between 10° and 20°N and a positive rainfall departure to the south of the negative departures. The dipole nature of both the observed and the model rainfall anomalies suggests a southward displacement of the mean rainfall pattern.

Since the only changes in these integrations were those in land-surface conditions, it is reasonable to conjecture that if the natural variability of the global climate system were to produce an initial drought, the strong atmosphere-land interaction over the Sahel region could contribute toward the persistence of that drought.

We further propose that since natural changes in the Sahel rainfall have led to large-scale changes in human activities in the region, it is quite likely that the degradation of the land surface was exacerbated by human activities (overgrazing, deforestation, soil desiccation, etc.), thereby producing further reduction in the Sahel rainfall. The exacerbating effects of human activities could have been especially large during the past 30 years because the population density has been much higher than at the beginning of the century.

FIGURE 2

(a) Observed summer (June, July, August, September) rainfall difference, in millimeters per day, for the average of 1983 and 1984 minus the average of 1950 and 1958. (b) Rainfall difference between desertification and reforestation GCM experiments.

The various conjectures put forward in this paper so far have been synthesized in a schematic diagram shown in Figure 3. It is seen that the atmosphere-land interaction has a positive feedback mechanism, so that initial drought caused by the natural variability of the climate system is perpetuated by increases in albedo and decreases in vegetation, soil moisture, and surface roughness. In addition, the human-induced effects also produce an increase in albedo and a decrease in vegetation, soil moisture, and surface roughness. Thus, both the planetary-scale effects and the human-induced effects are mutually reinforcing the dry conditions, which has led to an unprecedentedly long and severe drought over the Sahel.

THE FUTURE

If we accept the premise presented in the beginning of this paper, that the Sahel drought was initiated by natural

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 3

Schematic representation of the mechanisms affecting Sahel rainfall. Both the natural variability and the human-induced changes are contributing to the reduction in Sahel rainfall.

variability of the global climate system, it is reasonable to assume that at some unpredictable future time the natural variability (global SST patterns, for example) is likely to produce excessive rainfall over the Sahel and thus reverse the current drought. However, it is unclear whether the changes in the land-surface characteristics can also be reversed. Therefore, even if the global circulation shifts in such a way as to be favorable for reversing the current drought, the global natural changes will have to be stronger than before, and more favorable, to overcome the opposing effects of local atmosphere-land interactions. In fact, it is entirely possible that the Sahel drought will continue indefinitely, because the natural variability effects may never be large enough to counteract the persistent effects of the local land-surface changes.

In a hypothetical GCM sensitivity experiment, Xue and Shukla (unpublished manuscript) have shown that rainfall over the Sahel can be increased by replacing the broadleaf shrubs over ground cover and the shrubs over bare soil by broadleaf trees over ground cover over a large region that includes the Sahel. This suggests that although human effects were not responsible for the initiation of the Sahel drought, a suitable combination of large-scale natural variability and human intervention might contribute toward reversing the current drought. Modification of surface albedo, evapotranspiration, or surface roughness would all be of assistance, though of course it would be extremely difficult to sustain vegetation on a scale sufficiently large to halt the spread of the desert.

Discussion

DESER: Do you think local forcing applies to the string of wet years in the 1950s and 1960s too?

NICHOLSON: Absolutely. What I feel is the critical factor involved in the forcing is the fluxes to the atmosphere, including dust. The land surface forcing should be reversible, or symmetric, for the wet and dry periods.

RIND: How successful have people been in quantifying the albedo change between the wet and dry periods, as far as the land surface is concerned?

NICHOLSON: One study made 10 or 15 years ago took the available satellite data and reconstructed the albedo over about 8 years. The small changes found were not consistent with rainfall fluctuations, and I think basically showed that albedo is not one of the major forcing factors. More recently, Vivian Gornitz searched historical archives for material on land-surface changes, and constructed a beautiful map of West Africa showing changes due to human activities. Again, in terms of absolute values of albedo, it's a very small change, and sometimes in the wrong direction for the Charney mechanism. So I think that not albedo but something coupled to soil moisture, vegetation, and dust will be the major factor.

DIAZ: To what degree do you think that time differences in the frequency of the westward-propagating tropical disturbances affect the variability you're seeing?

NICHOLSON: I think they're key to understanding the region. This may be the one place on the globe where the land-surface forcing per se has a major impact on the overall large-scale dynamics. The jet stream at the mid-levels here, which is critical in disturbance development, is due to nothing other than the temperature gradient across the land surface down to the ocean.

Some of the models done years ago, during the GATE days, showed that if you change the relationship between baroclinic and barotropic instability you can change the characteristics of the disturbances. The difference between the wet 1950s and the dry 1970s can be explained by one or two disturbances a month that produce, say, 150 mm of precipitation a day, where the dry periods have nothing over 50. I think the large-scale circulation is clearly coupled to the land surface there. Incidentally, the dust in that region is right at the level of the African easterly jet, about 650700 mb, which is another reason I think it is so important.

CANE: I can see that the land-surface processes can affect what's happening, but I'd like to hear what you see as the feedback

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

mechanism for very-long-period changes. How would you cycle in and out, and what would the trigger be?

NICHOLSON: Well, if you assume some large-scale trigger in the general atmospheric circulation, maybe a major change in SST, you could get a continental-scale pattern like the wet 1950s all over Africa. But in other areas the rainfall goes back to normal, whereas in the Sahel the anomalies persist for 10 or even 20 years. There has to be some other smaller-scale forcing that reinforces the existing conditions until it's overridden by some larger trigger in the atmosphere-ocean circulation.

COLE: Ropelewski and Halpert's analysis of global rainfall suggests that the ENSO teleconnection with African rainfall is strongest in the east. Do you see a strong ENSO signal in the variability in the area you've been looking at?

NICHOLSON: There's a small one in the coastal region, but nothing in the central Sahel.

RIND: Shukla, your hypothesis suggests two questions to me. The first is, can it be documented—through satellite observations, for instance—that the vegetation really changed so extremely in those 10° latitude belts? And the second is, did those British experiments in which SST anomalies related so well to the Sahel changes, also produce the drying conditions in China and India between the 1950s and 1980s?

SHUKLA: We have exaggerated the extent of land degradation in our experiments. But there is a very clear shift in the rainfall between the 1950s and the 1980s, and Tucker and his colleagues have shown from the vegetation data that there is a very clear relationship between the two. If you look at the latitude of an isoline, it is definitely going down; the 200-mm isoline has dropped by 2° or 2.5°. So there are changes at the boundary, albeit of less magnitude than what I have chosen for this experiment.

BERGMAN: Sharon, when I was at the Climate Analysis Center, we found amazing discontinuities in precipitation time series in Western Africa that suggested that the station sites for some locations must have been moved, although there was no record of a change. Did you find this, and were you able to correct for it?

NICHOLSON: There were only a couple of stations in my network that had to be thrown out because of that sort of discontinuity, though we did find differences at the same station because of varied sources. Our approach has been to take aerial averages and hope that the outliers get factored out. But let me make a pitch here for the need to put together an archive of some of the old, forgotten data sources, such as pilot balloon data, to help us understand decadal and longer-scale variability.

RASMUSSON: It is still not clear to me that we are seeing the effect of human intervention. The vegetation seems to move north again in wetter years. Can you see an underlying southward trend'?

NICHOLSON: Shukla and I differ somewhat in our views here. I think that so far human intervention has very little to do with it. Many of the papers on desertification are essentially inferring continent-wide desertification from two data points in West Africa. But I don't think that the vegetation's recovery changes any of the ideas about the relation between low-frequency forcing and land surface.

One other thing I'd like to mention with respect to Shukla's argument is that I think one variable has been left out: dust. If I had to put money on any of those land-surface variables, I'd pick dust. It responds to both rainfall and land-surface factors, it has the best memory in the system, and it has shown the most consistent relationship over time with rainfall fluctuations.

SHUKLA: That is certainly a point that ought to be discussed further, if we had time. I'd like to think a little more before I put my money on dust. But let me add just a couple of things. The population pressure has been higher than ever before in the past 20 or 30 years, and that period coincides with the most severe drought in the 100 or so years for which we have reliable instrumental data. Proxy-data evidence of persistent droughts by no means rules out a human role in the current situation. It is also possible that human intervention, such as large-scale agriculture and afforestation, has the potential for reversing the present tendency, whatever its origin.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Continental Snow Cover and Climate Variability

JOHN E. WALSH1

ABSTRACT

In terms of its interactions with the atmosphere, snow cover has received the most attention through the albedo-temperature associations that have been detected over daily to monthly time scales. For example, snow cover has been shown to contribute, albeit modestly, to the interannual variability of monthly surface air temperatures over land. On decadal and longer time scales, the more important role of snow cover may be its link with the hydrologic cycle over land. Changes in winter precipitation and/or temperature can, through snow cover, result in substantial temporal and spatial changes of soil moisture and runoff to the oceans. Model results are rather consistent in depicting an alteration of the seasonal cycle of temperature when the timing and amount of spring snow melt are modified. Runoff-induced effects on the ocean stratification in the middle and high latitudes have also been hypothesized, although such arguments need to be substantiated by observational data. Greenhouse-induced changes of high-latitude precipitation, especially snowfall, play significant roles in the scenarios of temperature change projected by global climate models. Data on observed snow cover and temperature variations of recent decades are, in some respects, consistent with the projections of the global models.

INTRODUCTION

Snow cover has long been regarded as an indicator and as a possible agent of climate variability over a range of time scales. Over periods of several days to several weeks, the largest changes in the earth's surface properties result from variations in snow cover on land and on sea ice. Snow cover clearly influences the local values of near-surface atmospheric variables over these time scales. Over time scales of a thousand years or longer, the advance and retreat of the ice sheets depend ultimately on changes in the rates at which snow accumulates and ablates over the continents. However, over the interannual-to-century time scales on which this workshop will focus, the climatic roles of snow cover are poorly understood. Aside from the areal snow coverage of the past 15 to 20 years, the variability of snow cover over decadal time scales is not well documented. Yet model results are sufficiently suggestive of climatic roles of snow cover that one cannot ignore snow cover in projections of climate change over decade-to-century time scales.

1  

Department of Atmospheric Sciences, University of Illinois, Urbana

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

This paper begins with a brief discussion of the physical basis of associations between snow cover and atmospheric variability. Manifestations of these associations in observational data and in model experiments will then be surveyed. While this survey contains several illustrations of shorter-term interactions between snow cover and the atmosphere, the emphasis will be on interactions within the land-atmosphere system over the interannual, decadal, and century time scales. Because sea ice is the focus of another paper in this volume, we limit the discussion to interactions involving snow on land.

PHYSICAL BACKGROUND

A major difficulty in quantifying the climatic role of snow cover is that the distribution of snow is primarily a consequence of the large-scale pattern of atmospheric circulation, which determines the broad features of the distributions of temperature and precipitation over land. Thus, even in the absence of any causal role of snow cover, there can be large statistical correlations between anomalies of snow cover and atmospheric circulation. However, the pattern of atmospheric circulation is itself determined by the distribution of diabatic heating-radiation, conduction and convection of sensible heat, and latent heating. By modifying the exchanges of energy and moisture between the surface and the atmosphere, snow cover alters the distribution of diabatic heating in the atmosphere. For example, the albedo of fresh snow is 0.80 to 0.85 in solar wavelengths, whereas the albedo of bare land or ice-free ocean is typically between 0.05 and 0.30. Snow cover can therefore reduce the solar energy available to the surface by 50 percent or even more, depending on the age and depth of the snow, the vegetative cover, and cloudiness. If this energy reduction is distributed through the lowest 2 km of the atmosphere, it can be equivalent to a cooling of 3°C to 7°C in middle latitudes under clear skies during March (Namias, 1962). Snow cover is also an effective insulator of the underlying surface and an effective radiator of infrared energy. Finally, melting snow represents an effective sink of (latent) heat for the atmosphere and an effective source of moisture for the soil. The subsequent evaporation of this moisture may prolong the tendency for snow to delay the sensible heating of the soil, thereby modifying the phase of the seasonal cycle of surface temperature. The latter hypothesis has provided the basis for several experiments with global climate models (see below).

A striking feature of the present-day distribution of snow cover over land is the virtual absence of seasonal snow in the Southern Hemisphere, where the only large area of land at latitudes in which snow can easily accumulate is the glaciated Antarctic continent. The following discussion of snow cover is therefore limited almost exclusively to the Northern Hemisphere.

SNOW COVER AND ATMOSPHERIC VARIABILITY: SHORT-TERM RELATIONSHIPS

Although the subject of this workshop is climate variability over decade-to-century time scales, we first review relevant studies of snow-atmosphere interactions over shorter time scales. The relevance of these studies stems from the fact that they illustrate interactions that may lead to climate changes over the longer time scales if one component (e.g., snow, atmospheric temperature) is systematically perturbed by some other climate forcing mechanism, either internally or externally. In this sense, the short-term relationships may be regarded as the ''building blocks" of long-term change.

Observational Studies

There is little doubt that snow can have substantial impacts on local surface temperature. Analyses of rates at which relatively warm moist air is cooled as it is advected over snow indicate that the loss of heat by conduction to the surface can reduce the surface air temperature by 4°C to 5°C per day (Treidl, 1970). More recently, Petersen and Hoke (1989) showed that the accurate specification of snow cover reduced the error of a regional numerical weather-prediction model's 48-hour forecast of surface temperature (by 8°C to 9°C); the corresponding forecast of precipitation type (rain instead of snow) was also correct over a larger portion of the model domain when the snow cover was accurately prescribed. The radiative impact of the surface albedo enhancement by snow can depress daytime surface temperatures by 5°C to 10°C during spring, as shown by Dewey's (1977) diagnosis of errors in statistical forecasts that ignored snow cover (Figure 1). If this approach is extended to monthly temperature specifications based on upper-air geopotential, specification errors of 5°C to 7°C are found equatorward of the normal snow boundary during months with extremely large positive snow anomalies (Namias, 1985). The errors decrease to 1°C to 3°C when all months over approximately 30 years are included in the statistical sample (Walsh et al., 1985). These impacts of snow are generally larger in the spring, when insolation is stronger. They are, nevertheless, indicative of the changes of mean surface temperatures that could result locally from a systematic advance or retreat of snow cover over decade-to-century time scales, i.e., from a change of the normal position of the snow margin.

It should be noted that the suppression of air temperature by positive anomalies of snow cover is generally confined to the lowest 100 to 200 mb of the atmosphere (Namias, 1985). Because the wintertime troposphere over land areas is characterized by relatively strong static stability even without a suppression of the near-surface temperature, it is unlikely that conditions are favorable for the vertical propagation of the thermal anomalies produced by snow

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 1

(a) Snow depth (cm) at 1200 UTC on 5 March 1977, and (b) 5-day average of the error (°C) of the daily maximum temperature forecasts for 5-9 March 1977 derived from the National Meteorological Center's Model Output Statistics. (From Dewey, 1977; reprinted with permission of the American Meteorological Society.)

cover. However, Lamb (1955) showed that there is a detectable decrease in the 1000 to 500 mb thickness as a layer of air passes over a large area of snow cover, implying that snow cover can contribute to the maintenance of a trough of cold air, which in turn helps to maintain the snow cover.

Some of the earliest work on the role of snow cover in seasonal atmospheric variability was directed at the Asian monsoon. Blanford (1884), Walker (1910), and others postulated a link between Himalayan winter snow cover and the strength of the Indian summer monsoon by reasoning that extensive snow cover could retard the heating of the Asian landmass. This link has been substantiated in recent years with the aid of satellite-derived measurements of snow cover (Hahn and Shukla, 1976; Dey et al., 1984; Dickson, 1984; Bhanu Kumar, 1988). On the basis of data for the 1967-1980 period, linear correlation coefficients between December-to-March snow extent in the Himalayas and June-to-September monsoon rainfall were found to be approximately -0.6 (Dey et al., 1984; Dickson, 1984).

While the monsoon-snow correlations seem to provide the basis for useful monsoon predictions at ranges of several months, there are several important caveats. First, snow cover over the Himalayan region was not mapped consistently during the first part (1967-1972) of the satellite record (Ropelewski et al., 1984). Second, as shown in Figure 2 (from Bhanu Kumar, 1987), the agreement between the interannual fluctuations of snowfall and monsoon variables has been noticeably poorer since 1980. Inclusion of data from 1981 to 1985 lowered the snow-rainfall correlations from -0.60 to -0.38 (r2˜0.14); the 95 percent significance threshold for a sample of N = 19 years is approximately 0.45. Third, Indian monsoon rainfall appears to correlate as highly with snow cover over the remainder of Eurasia as it does with Himalayan snow cover (Dickson, 1984). The physical linkage between snowfall and monsoon rainfall may therefore be more complicated than implied by the proposed effect of snow on the timing of the heating of the Himalayan and Tibetan region. For this reason, the modeling studies (e.g., Barnett et al., 1989) described below provide potentially important diagnostic information.

Other observationally derived linkages between snow cover and the atmospheric circulation over seasonal time scales have been addressed by Afanas'eva et al. (1979), who examined the position of the Planetary Upper-air Frontal Zone (PUFZ) over Eurasia during the autumn and spring. Time variations of the PUFZ and the snow boundary correspond closely (r = 0.78). However, as noted above, such correlations are at least partially attributable to the fact that the position of major upper-air features (e.g., the jet stream) is a primary determinant of the position of the snow boundary. Interannual variations of snow cover over Eurasia were also examined by Toomig (1981), who found that the annual value of the absorption of solar radiation at Soviet stations is a strong function of the springtime absorption, which in turn depends on the albedo during spring. Toomig also reported modest correlations (r ˜ 0.45) between the surface albedo of early spring and the surface air temperatures through July at several stations. A similar lag relationship

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 2

Variations of Eurasian winter snow cover and the corresponding summer monsoon rainfall over India. (From Bhanu Kumar, 1988; reprinted with permission of the Journal of the Meteorological Society of Japan.)

was suggested by Lamb (1972), who argued that the exceptionally cold summer of 1968 in northeastern Europe and northwestern Asia was attributable in part to the previous winter's extensive snow cover. Because the snow did not retreat until June and July, the ground remained cold and wet, thereby favoring the persistence of a trough. The attendant cloudiness and frequent cyclones produced above-normal rain and occasional snowfall throughout the summer over the northern Soviet Union. Wahl (1968) suggested that persistent snow may have played a similar role in the unusually cold summers of the 1830s in northern North America.

Feedbacks involving snow cover have been suggested by Namias (1962, 1981), who has argued that extensive snow cover over eastern North America should enhance the coastal baroclinicity and favor strong cyclones along the East Coast. These systems, in turn, reinforce the cold trough and contribute to its persistence over periods of several weeks or longer, as in the winters of 1960-1961 and 1976-1977. Dickson and Namias (1976) have attempted to link extensive snow/extreme cold over eastern North America with anomalies in the winter climatic regime of Greenland and northwestern Europe. According to the authors' hypothesis, the southward displacement and intensification of cyclones on the East Coast of the United States lead to a reduction of cyclone activity over the Iceland-Greenland region. This weakening of the Icelandic low reduces the advection of mild air into the northeastern North Atlantic and northern Europe, thereby increasing the frequency of polar outbreaks over the European land areas. The frequent recurrence of this abnormal winter regime during the 1960s may have contributed to the relatively extensive European snow cover during that decade. However, the role of snow cover in modifying the atmospheric circulation in such regimes remains speculative because other hypotheses (involving sea surface temperature anomalies, tropical forcing, etc.) can be formulated. Controlled model experiments are needed to explore the physical basis of seasonal anomalies in these specific cases, as has been done for anomalies such as the North American drought of 1988 and the ENSO event of 1982-1983 (e.g., Trenberth and Branstator, 1992; Palmer and Mansfield, 1986).

Iwasaki (1991) has recently reported a tendency for large-scale anomalies of snow cover to persist from December through February, but he found no indication of anomaly persistence from February to March in the longer data record. Iwasaki also reported an apparent lag relationship, in which winters with extensive snow cover over eastern Eurasia tend to be followed by winters with extensive snow cover over North America. Although this lag relationship is statistically significant (r = 0.52), at least three counterexamples occurred during the 1980s.

Modeling Studies

As the preceding subsection indicates, various roles have been attributed to snow cover in the evolution and persistence of monthly and seasonal atmospheric anomalies. Most of these roles involve feedbacks, which are notoriously difficult to unravel from observational data. Consequently,

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

model experiments provide a potentially useful vehicle for diagnosing the effects of snow cover. However, the mixed nature of the model-derived conclusions pertaining to the impacts of snow cover will become apparent as the survey proceeds.

In a study of snow-induced effects in non-winter months, Yeh et al. (1983) used a simplified version of the Geophysical Fluid Dynamics Laboratory's general-circulation model (GCM), containing idealized geography and a limited computational domain. The model contained no diurnal cycle, and its cloudiness was prescribed to be zonally uniform and seasonally invariant. The complete removal of the snow cover in mid-March was found to reduce the water available to the soil through snow melt, thus decreasing the soil moisture during the spring and summer in the region of snow removal. The drying of the soil resulted in an increase of surface temperature at high latitudes by 2°C to 8°C for the subsequent 3 to 4 months. The temperature increase extended into the upper troposphere, thereby reducing the meridional temperature gradient and the zonal wind in high latitudes.

A similar conclusion about the snow-hydrology-temperature linkage was obtained from a more realistic GCM by Yasunari et al. (1989). In this experiment, the 5° × 4° version of the Japanese Meteorological Institute GCM was run for 6 months beginning March 1. The experimental runs were identical to the control runs except for the addition of 5 cm (water equivalent) of snow in the snow-covered portion of the 30° to 60°N zone of the Eurasian continent. The results, shown in Figure 3, contained evidence of both (1) an albedo feedback, which suppressed temperatures over lower latitudes (e.g., Tibetan Plateau) by 2°C to 3°C during spring, and (2) a snow-hydrology-temperature linkage, which suppressed temperatures over middle latitudes by 2°C to 3°C during the summer months of June to August. During the summer, the anomalous Eurasian heat sink also appeared to induce a stationary Rossby wave

FIGURE 3

Latitude-time (month) sections of anomalies ("Heavy Snow" run minus "Control" run) of snow mass (cm liquid equivalent) and surface air temperature (°C) averaged over the Eurasian continent. (From Yasunari et al., 1989; reprinted with permission of the World Meteorological Organization.)

pattern extending from eastern Asia to northern North America.

The most thorough investigation of spring-summer feedbacks involving snow cover is Barnett et al.'s 1989 study of Eurasian snow impacts on a low-resolution (T2 1) version of the European Centre for Medium-Range Weather Forecasting's model. In Barnett et al.'s first experiment, snow extent corresponding to observed extremes was prescribed and interactions between snow and the surface hydrology were suppressed in order to isolate the albedo effect. The atmospheric response to the snow anomalies was local and confined primarily to air temperature and upper-air geopotential (but not sea-level pressure). All significant signals vanished when the snow disappeared in the spring, and the albedo effect had no sustained impact on the development of the Asian monsoon.

In Barnett et al.'s second experiment, rates of snowfall over Eurasia were doubled and halved so that the subsequent melt and evaporation could induce changes in the regional hydrology. The two sets of simulations showed statistically significant differences extending through the subsequent two seasons. The results derived from the doubled snowfall were characterized by significantly lower surface and tropospheric temperatures from May through July, higher pressures over Asia and lower pressures over North America, weaker zonal winds over the Arabian Sea, weaker surface convergence over southern Asia, and a weaker monsoon over southeast Asia. The Indian monsoon, however, was not substantially weaker in the "heavy snow" simulations, although this result may be partially attributable to the model resolution. The sea-level pressure signal over Asia and North America is stronger in the model than in the real world; the exchange of mass between the two continents may have been exaggerated in the model because the sea surface temperature (SST) distribution was prescribed climatologically in all the model runs.

In general, the physical mechanisms underlying Barnett et al.'s model response are similar to those of Yeh et al. (1983). The doubling of snowfall by Barnett et al. also produced a general weakening of the wind in the Southern Hemisphere and along the equator. While this response is similar to that which occurs prior to the warm phase of an ENSO event, the model with prescribed SST cannot sustain an ENSO event. In order to address the snow-ENSO link in more detail, Barnett et al. performed additional experiments with a coupled ocean-atmosphere model. The conclusion was that "the snow/monsoon signal has all the characteristics necessary to trigger the Pacific portion of an ENSO event, but the signal is too small by a factor of at least 2. In balance, it appears that snow-induced monsoon perturbations may be one of the (multiple) triggers that can initiate an ENSO cycle" (Barnett et al., 1989, p. 683-684). The elimination of major model biases and the improvement

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

of the model resolution are clearly high priorities in the context of this large-scale signal involving snow cover.

The studies summarized above suggest that snow cover may play a major climatic role regionally and perhaps even globally. However, the most recently published investigation of snow-climate interactions provides some indications of a negative feedback involving snow and temperature anomalies, at least during late winter and early spring. Cohen and Rind (1991) used the Goddard Institute for Space Studies 9-layer model with 8° × 10° resolution to examine the sensitivities to snow cover during March. The initial (March 1) boundary conditions corresponded to the observationally derived "maximum" and "minimum" snow cover and depth over North America and Eurasia. A key feature of this experiment was that the snow cover was interactive rather than prescribed. In the results of a five-case sample of March simulations, the positive snow anomalies caused only a short-term local decrease in surface temperature. There was no non-local signal, and even the local signal became weak after about 7 days. In the model physics, the reduction in the surface absorption of solar radiation and the increased consumption of latent heat for melting the snow contributed to lower temperatures. However, the remaining terms in the surface energy budget (e.g., long-wave radiation, sensible heat flux) adjusted so that they offset the cooling. Thus a negative feedback limited the impact of the snow anomalies to a slight cooling of about 1°C—a smaller effect than indicated by the observational studies and by the model experiment of Yasunari et al. (1989). This negative feedback that limits the albedo effect of snow was also found in Barnett et al.'s (1989) first experiment, and the weak signal is not inconsistent with the results of Robock and Tauss (1986), who used a simple, linear, steady-state model. Thus the relatively negligible impact of snow cover in the Cohen and Rind experiment may be due to the fact that their simulations did not extend into the spring season, when the hydrologic role of snow anomalies can become more important. The compatibility of the results from these various model experiments, nevertheless, requires further attention.

SNOW COVER AND ATMOSPHERIC VARIABILITY: DECADE-TO-CENTURY TIME SCALES

The importance of snow cover for climate change over decade-to-century time scales depends on several factors. First, changes in other components of the climate system are likely to alter the large-scale distribution of snow cover. Relatively small shifts in the atmospheric circulation pattern can have major effects on the snow distribution in mountainous areas, e.g., the Rocky Mountains (Changnon et al., 1993). The extent to which snow cover will change in response to a changing climate is not well known; for example, a general warming may increase snow melt and decrease the fraction of precipitation that falls as snow, but (according to models) it is likely to increase precipitation in high latitudes-where most precipitation falls as snow. In view of these competing effects, it is conceivable that snow extent could decrease while high-latitude snow volume increases. Second, changes of snow cover can trigger a host of potential feedbacks involving air temperature, soil moisture, cloudiness, the phasing of the seasonal cycle, and other variables. The magnitudes and relative importance of these feedbacks are poorly known. The individual feedbacks are notoriously difficult to isolate in observational data. Although various model experiments have addressed individual feedbacks, the isolation of individual feedbacks can be a nontrivial undertaking even in model simulations.

In the section below, we address the first of the two factors listed above by surveying recent analyses of observational data pertaining to snow cover. We then address the issue of snow-related feedbacks in climate change by surveying the recent model experiments that may be most relevant to changes over the decade-to-century time scales.

Observational Studies

Observational studies of historical variations of snowfall, snow depth, and snow water equivalent are confounded by measurement difficulties pertaining to small-scale variations of these variables. Large spatial gradients of all three variables are found over areas containing even moderate topographic features. Moreover, snow gauges are known to "undercatch" snow; the degree of undercatch varies with the type of gauge, and the type of gauge has changed during the period of record at nearly all stations. While satellite data have provided essentially continuous global coverage since the early 1970s, the useful information is generally limited to areal extent. The critically needed mapping of snow depth or water equivalent is not yet achievable over large areas, although algorithms for snow depth in vegetation-sparse areas have been used with some success (Chang et al., 1987).

The most comprehensive analysis of satellite-derived data on snow coverage was made by Robinson and Dewey (1990), who found that the Northern Hemisphere snow cover of recent years is less extensive than that of 10 to 20 years ago. Perhaps coincidentally, the increase of surface temperature over the last few decades is larger over land than over the oceans. The high-latitude land areas, which are generally snow covered during winter, show the strongest warming (Figure 4; see color well); this warming has been strongest in spring and winter. In a recent analysis of North American data, Karl et al. (1993) identified several regions that, because of their high variability of snow cover, have exerted the primary influence on North American snow variations. These regions, which vary seasonally, are ones in which the inverse snow-temperature relationship is

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

strong. Karl et al. also used station data to extend their analysis to the multidecade-to-century time scale. Their findings included a 4 to 5 percent increase of both solid and total precipitation rates over northern Canada during the past four decades, century-scale increases of precipitation over southern Canada and the contiguous United States, and a decrease in the proportion of precipitation falling as snow in southern Canada. These findings are generally consistent with the changes that have been hypothesized to accompany a greenhouse warming. Karl et al. also found that the unprecedented warmth of the 1980s in Alaska was accompanied by a 10 percent increase of annual precipitation in that region. Leathers and Robinson (1993) have examined the winter region (the central United States) in more detail, finding that the concurrent 500 mb height anomalies are generally collocated with the snow/temperature anomalies during December but not during January and February.

The general coincidence of the "marginal snow zone" and the areas of strongest warming over the past several decades (Figure 4) deserves further comment with regard to the possible nature of the forcing. The "land-leading-ocean" feature is characteristic of large-scale forcing such as global warming or the response to major volcanic events. By contrast, natural low-frequency variations will generally manifest themselves in an "ocean-leading-land" pattern because the ocean is the low-frequency source of the forcing on the near-surface atmospheric temperature. The extent to which a large-scale forced temperature response is amplified by the retreat of snow over land is one of the key unknowns in the interpretation of a pattern such as that in Figure 4.

Decadal-scale summaries of snowfall in China have been compiled by Li (1987), who found a decrease of snowfall over China during the 1950s followed by an increase during the 1960s and 1970s; the late 1970s had the largest running-mean values of the 30-year period of record (Figure 5). Li noted a general correspondence between the time series of global mean temperature and mean snow depth in China (as well as an apparent association between years of heavy

FIGURE 5

Ten-year running mean of horizontally integrated snow depth and amount in China. (From Li, 1987; reprinted with permission of IAHS Press.)

snow and ENSO events). Similar compilations have been made for other regions such as the Swiss Alps (Lang and Rohrer, 1987), although the representativeness of such time series is a key issue. From the standpoint of areal coverage, the snowfall records that are potentially most valuable are those of the former Soviet Union. The status of these data is uncertain.

The longest time series of snow cover are generally those coming from single stations (e.g., the Tokyo record dating back to the 1630s (Lamb, 1977)) or the records of annual precipitation derived from polar ice cores (e.g., the Greenland Ice Sheet Project's work in Antarctica and Greenland, as in Alley et al., 1993). The cores represent potentially valuable records of regional snowfall if the spatial representativeness of the point data can be established. The South Pole data, for example, have been used to deduce an increase of snowfall from the 1700s to the early twentieth century (Giovinetto and Schwertfeger, 1966; Lamb, 1977). Analysis of GISP's Greenland cores has only recently begun (Mayewski et al., 1993).

Another type of data for which analyses are in the early phases is the runoff (stream-flow) data for the high-latitude rivers that are fed primarily by snow melt. Mysak et al. (1990) have proposed a mechanism by which an interdecadal (approximately 20-year) cycle results from a feedback loop involving high-latitude precipitation, runoff, arctic sea-ice export to the North Atlantic, and ocean salinity/temperature anomalies. The linkages involving snowfall and runoff have yet to be thoroughly evaluated. In view of the potential implications of this cycle for the global ocean circulation, quantitative analyses of high-latitude snowfall and runoff are being assigned high priority in the upcoming ACSYS (Arctic Climate System) component of the World Climate Research Program. The extension of such analyses to include soil moisture and land surface temperatures also appears to merit high priority in the context of possible greenhouse-induced changes over decadal time scales.

Finally, glaciers are ultimately attributable to continental snowfall over time scales of decades to millennia. The advance and retreat of glaciers have long been regarded as proxy indicators of climate change over these time scales. In the context of global change, the key properties of glaciers are extent, ice volume, and mass balance. Temporal changes in these glacial properties can be complex functions of temperature, precipitation, the seasonality of temperature and precipitation, and topography. Nevertheless, the advance and retreat of glaciers are known to be consistent with century-scale changes of regional climate, e.g., the Little Ice Age.

Field measurements of glaciers are now assembled and reported by the World Glacier Monitoring Service in Zürich. These data, which are summarized by Haberli et al. (1989), include direct measurements of the mass balance of approximately 75 glaciers in the Northern Hemisphere. Statistics on

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 6

Five-year moving averages of percentage of glaciers having positive mass balance. (From Wood, 1988; reprinted with permission of the University of Colorado.)

glacial changes have been compiled for about 500 glaciers worldwide (e.g., Wood, 1988). Figure 6 (from Wood, 1988) shows the trends in glacier mass balance over approximately 20 years, expressed as the percentage of observed glaciers having a positive mass balance. While this percentage for all glaciers is close to 50 percent, the percentage for an individual region such as the European Alps can undergo multi-year excursions well above or below 50 percent.

Modeling Studies

Aside from CO2-doubling experiments, there have been very few three-dimensional model simulations of the land-atmosphere system over decadal time scales. In this respect, one may argue that less attention has been given to the simulation of land-atmosphere interactions than to the simulation of large-scale ocean variations over decade-to-century time scales. Examples of the latter are the experiments of Bryan (1986), Weaver et al. (1991), and Yang and Neelin (1993). Moreover, most models containing an interactive land surface treat the surface physics quite crudely: the "bucket" method is typically used to handle soil moisture, multi-level soil treatments are generally not included, and runoff often "disappears" from the system.

Most global climate models do include a thermodynamically and hydrologically interactive snow cover. In its assessment of climate models, the IPCC (1990) found that the snow cover simulated by several models was "broadly realistic," leading to the statement that "the simulated snow extent should thus not distort the simulated global radiative feedbacks" (IPCC, 1990, p. 113). Even if one gives the benefit of the doubt to that assertion, the snow cover simulated by all models appears to contain significant errors, particularly over eastern Asia. More comprehensive diagnostic assessments of the simulated snow cover (extent and water equivalent) in the context of both the present climate and projections of climate change are needed.

A more sobering finding concerning model simulations of the feedback between snow and radiation is contained in the IPCC's (1992) update, which cites Cess et al.'s (1991) comparison of the snow feedback (under perpetual April forcing) in 17 global climate models. The temperature sensitivity or feedback parameter, λ, associated with snow cover was found to vary from negative values in some models to a wide range of positive values (Figure 7). The sensitivity parameter in clear-sky regions ranges from values corres-

FIGURE 7

Snow feedback parameter for 17 general-circulation models. Solid circles denote global values, open circles denote values for clear-sky regions. (From Cess et al., 1991; reprinted with permission of the American Association for the Advancement of Science.)

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

ponding to negligible snow feedback to values representing a twofold amplification. Cloudiness complicates the interpretation of the snow feedback even further. The IPCC concludes that "the snow-albedo-temperature feedback processes in models are somewhat more complex" (IPCC, 1992, p. 117) than the conventional interpretation of the positive snow-albedo-temperature enhancement of a perturbation. Since these models also generally lack realistic surface hydrologies, one must view with caution their simulations and/or projections of atmosphere-snow interactions.

The model experiments of Yeh et al. (1983), Barnett et al. (1989), Yasunari et al. (1989), and Cohen and Rind (1991), all of which were summarized above, are potentially relevant to decade-to-century-scale climatic change even though the simulation periods are subdecadal. This relevance stems from the fact that climate changes over decadal time scales may perturb the large-scale snow distribution by amounts comparable to the changes of snow prescribed in the modeling experiments. However, biases introduced by the parameterizations of the snow and surface physics must be addressed before the results of the experiments can be viewed with confidence. A step in this direction has been made by Washington and Meehl (1986), who found that the inclusion of a simple temperature dependence in the parameterization of snow albedo can change substantially the globally averaged surface temperature increase caused by a doubling of CO2. Ingram et al. (1989) and Covey et al. (1991) also examined the effects of high-latitude surface albedo parameterizations, although their experiments focused on sea-ice-albedo. Using an earlier generation of the National Center for Atmospheric Research's model, Williams (1975) found that changes in surface albedo (and SST) influenced a simulated ice-age circulation more than did the orographic changes caused by the glaciation. Further surface-sensitivity experiments with more current atmospheric models and with more realistic treatments of the surface physics and hydrology are needed.

An issue in need of particular attention is the apparent paradox involving recent observational data (on snow and air temperature) and modeling studies of the feedback between snow, soil moisture, and temperature. The modeling studies cited above indicate that a positive anomaly of snow cover tends to depress surface air temperatures through its enhancement of soil moisture for up to several months after the snow melt. The model results seem inconsistent with the observed tendencies toward higher springtime temperatures and increases of snowfall in northern land areas over the past several decades. However, an earlier retreat of snow during the late winter or early spring may be a consequence of the warming, which is most pronounced in the winter and spring seasons, especially if the increases of snow depth have occurred primarily in the northernmost land areas (as the data suggest they have). The earlier retreat of snow creates the possibility that the albedo effect may offset, or even dominate, the tendency of greater soil moisture to delay the seasonal warming. Since this scenario is especially likely if the upper layers of soil dry rapidly, it is clearly important that models accurately resolve the hydrological and thermal changes in the upper layers of the land surface during the snow-melt period. The potentially high climatic leverage of the snow-melt period should make it a focus of observational and modeling studies, whether the time scales of interest are decadal or century or even longer.

CONCLUSION

Several conclusions emerge from the results surveyed here. The first, which pertains to the strategy for diagnosing the role of snow cover in climate variability, is that both modeling and data analysis are essential and complementary diagnostic tools. Because snow is involved in a variety of interactions within the climate system, and because the distribution of snow is determined largely by other components of the climate system, controlled experimentation with numerical models is a key element of the diagnostic strategy. However, the biases and other limitations of models are such that numerical experiments yield convincing conclusions only when the observational data give some credibility to the model results.

The diagnostic studies of the past few decades appear, at first glance, to have produced a somewhat inconclusive picture of the climatic role of snow cover. However, several scientific conclusions do emerge when the results are viewed in an aggregate sense:

  1. Anomalies of snow cover are clearly associated with significant local anomalies of air temperature, at least in the lowest 100 to 200 mb. The duration of these local anomalies generally ranges from several days to several weeks and is often limited by the fluctuations of the snow anomalies themselves over weekly and monthly time scales.

  2. Over the past two to three decades, changes of surface air temperature are broadly consistent with changes of snow coverage, especially in winter and spring.

  3. There is little or no evidence that the albedo effect of wintertime snow anomalies produces meaningful signals in the large-scale atmospheric circulation.

  4. The snow-hydrology-soil moisture feedback appears to be capable of producing a meaningful response in the atmosphere during spring and summer. The scale of this response is at least regional and possibly larger. Regions susceptible to snow-hydrology effects include eastern Asia and the northern portions of Eurasia and North America.

  5. The mechanism(s) by which snow influences the large-scale atmosphere are sufficiently complex that rela-

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

tively sophisticated (e.g., nonlinear) models are required for the diagnosis of snow-atmosphere interactions.

With regard to (4) and (5), the validity of parameterizations involving snow cover in global climate models may be a major constraint on further progress toward an understanding of snow-atmosphere feedbacks. Priority should be given to the parameterization of quantities and processes such as snow surface albedo, the influence of snow on the boundary layer structure, and the disposition of the liquid water produced by snow melt.

ACKNOWLEDGMENTS

The preparation of this paper was supported by the National Science Foundation through Grants DPP-9214793 and ATM-9319952. We thank Norene McGhiey for typing the manuscript.

Commentary on the Paper of Walsh

DAVID A. ROBINSON

Rutgers University

Rather than commenting on Dr. Walsh's paper, I am going to show you a little about my work on hemispheric snow cover over the last several decades and also on the century time scale I am beginning to put together. This is at Dr. Walsh's suggestion, I hasten to add.

Figure 1 gives you an idea of the annual cycle of snow cover over Northern Hemisphere lands. The area covered by snow ranges from 40 to 50 million km2 during the winter over the Eurasian and North American continents to several million km2 in summer, primarily on top of the Greenland ice sheet.

FIGURE 1

Monthly snow cover over Northern Hemisphere lands (including Greenland) between January 1972 and August 1991. The median area of cover is the horizontal line within the 12 monthly boxes, and the interquartile range (ICR) is between the top and bottom of the box. Whiskers show the extreme values between + 1 and + 1.5 times ICR and between - 1 and - 1.5 times ICR, and asterisks show values outside these ranges. Values are calculated from NOAA weekly snow charts.

In recent years there has been a dearth of snow cover over Northern Hemisphere lands (cf. Figure 3 in my paper in this volume). It has been true for the last 5 to 6 years over both Eurasia and North America. So it is seen hemisphere-wide, with many record lows set in the last couple of years, particularly 1990. I remind you that this is based on 20 years of observation. Nonetheless, this record, produced weekly by NOAA, happens to be the most consistent long-term satellite-derived data set available for any climatic variable. Charts go back to 1966, but most of us who have looked at the data feel that it is relatively homogeneous and credible only from 1972 on.

Looking at seasonal variations in snow cover over the past 20 years (cf. my Figure 4 in this volume), you see that there has been very little variation over the winter seasons in Eurasia, and even less in North America. This agrees with what Dr. Walsh showed earlier for sea ice. Fall cover has varied a bit more, but spring and summer show the greatest year-to-year variations in cover. In recent years snow cover has decreased in March through June over both North America and Eurasia. These spring decreases in recent years have been most pronounced over northeastern Asia and down the leeward side of the Canadian Rockies into the northern portions of the United States.

So that is what we have seen hemispherically over the past 20 years. Unfortunately, we have no data sets going back longer than that on a hemispheric scale, but there have been some recent efforts to try to extend local and regional data sets back to the turn of the century. For instance, at Denison, Iowa, over the course of this century there appears to be an increase in the number of snow-covered days with =1 inch (2.5 cm) of snow on the ground during the fall, winter, and spring.

As for other data sets to examine, we are beginning to sort out daily data from a set containing several hundred Russian stations. In some cases, the records go back to the turn of the century. We have also been examining station data in western China extending back into the 1950s.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
Discussion

NORTH: The warming pattern John mentioned, with the leading in the middle of the continental areas and the lagging in the ocean, seems to be a very robust feature. It seems to hold for transient runs for all models, and even toy models have it. I think it's something that needs to be put into the filter when you're trying to do the signal processing I mentioned earlier. You get an increase on the order of 30 percent in the signal-to-noise ratio.

This same land-leading-ocean feature will appear when you have a large-scale forcing. But when it's natural variability at very low frequencies, the ocean tends to lead—for example, with Mt. Pinatubo. It would be interesting to look at the phase relations during the 1940s, when the cooling was so dramatic in the Northern Hemisphere, to see whether the sea ice was leading.

DELWORTH: In the GFDL model for natural variability, for time scales less than 50 years, the continental-region temperatures actually correlate most with the hemispheric means. It's complex, but I think you have to go to longer time scales to get the deep ocean involved.

SHUKLA: I wonder about the apparent paradox of a higher snow amount's being succeeded by warming and reduction in area in the following season.

WALSH: If the snow is disappearing earlier because of the greater springtime warmth, the perturbation of soil moisture due to the extra snow may just be swamped. The additional solar radiation available may be a larger effect than the soil moisture. Also, we don't know that the time scales of the soil perturbations in the model agree with the real world's.

ROBINSON: You might also have different effects deep within the pack and near the periphery.

RASMUSSON: One of the contributors to the recent warming over Eurasia was the exceptionally warm winter of 1982-1983. We were mystified by it, since it was not supposed to be associated with the 1982-1983 El Niño event, but Alan Robock's work on volcanic eruptions suggests that it could have been a response to El Chichon. Also, Dave, I wondered whether you had been able to clear up the snow-cover data problems you were having.

ROBINSON: Yes; I found that two different land masks had been used in the 1970s and 1980s, both of which were wrong. Using GIS techniques, I've come up with a land mask and a procedure that yield consistent and reasonably accurate numbers.

RIND: One caveat: The model run you looked at did not have one additional forcing, the reduction in ozone in the lower stratosphere. That is most effective at polar regions, and would also induce a cooling trend at the surface.

BRYAN: It seems to me that the resemblance to a greenhouse effect that appeared in the slide you showed for the 30-year average may be at least partly an artifact of the averaging—a combination of a very cold event at the beginning of the period and warming over land masses toward the end. I think it could be a little misleading to compare it with the slow warming observed in the greenhouse model experiment.

WALSH: I agree with your comment on the land, but the changes were actually negative over the high-latitude oceans during that period, so I don't think the start can have been unrepresentative.

NORTH: The Hanson-Lebedeff paper of 1988 or so, which was for a longer period leading up to the 1940s, showed the same kind of pattern.

TRENBERTH: I feel obliged to mention that the Southern Hemisphere has snow cover and sea ice too. I know Claire Parkinson has done some work there; can you comment, John?

WALSH: There's essentially no snow there outside the Antarctic continent. The one thing we know from the 20-year satellite record of the South Orkneys is that those data are not representative of the larger-scale changes.

ROBINSON: Ken Dewey and Randy Cerveny are doing a little work with snow over Patagonia and other areas.

JONES: Returning to an earlier point, I wanted to say that the temperature variations at the two Antarctic inland stations and Vostok over the past 35 years are similar to what we've seen around the coast. Since Greenland is a much smaller ice sheet than Antarctica, we may be able to infer that the temperature changes in its interior would also be similar to those measured around the coast.

KUSHNIR: Just a correction to an earlier suggestion: The coldest temperatures in the Atlantic were in the 1970s, not the 1960s, so John may indeed be picking up a trend.

GROISMAN: I'd like to note that it's important that John found the changes in sea ice in summer. Wintertime ice can go back to its old limits, but the new ice is most easily melted off in the spring.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Decadal Variations of Snow Cover

DAVID A. ROBINSON1

ABSTRACT

In situ and satellite observations of snow-cover duration show a considerable amount of year-to-year variability on local to hemispheric scales. This variability is often embedded in longer-term fluctuations. In situ data from the central United States indicate multi-decadal fluctuations in the duration of snow over the past century, and hemispheric satellite observations suggest variations lasting for several years or more.

Such analyses have become possible in recent years as a result of the recovery, digitization, and validation of historic in situ observations of snow cover and the availability of two decades of satellite observations of continental snow cover. Efforts are needed to recover additional historic data, to improve the recognition of snow using microwave satellite data, and to create global snow products using all available data sources. With these data available for analysis, knowledge of the spatial and temporal kinematics of snow cover will continue to improve. This will contribute to a better understanding of the role of snow in the climate system and to the utility of snow as an indicator of climate change.

INTRODUCTION

Snow cover is a critical influence on the earth's climatic energy and hydrologic budgets. In many regions it may play an influential role in determining the magnitude of any human-induced climate change, and might be a useful indicator and monitor of such change. To understand better the importance of snow in the climate system, it is essential that accurate information on the temporal and spatial dimensions of snow cover be available. Such data have been obtained through in situ and satellite observations, and recent efforts have begun to locate, assimilate, and validate these data. Once available, they are being employed in empirical and modeling investigations of snow-cover kinematics and the dynamical aspects of snow within the climate system. An excellent survey of the latter is provided by Walsh (1995) earlier in this section. This paper will concentrate on the distribution of snow cover in space and time over the past two decades and the past century. Much of the discussion will focus on the newly available in situ and satellite data sets. Given the lack of attention paid to secular snow cover data until recently, only limited analyses of these data have taken place. A few examples of these efforts

1  

Department of Geography, Rutgers University, New Brunswick, New Jersey

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

will be presented. For information on snowfall and its long-term variability across the North American continent, the paper later in this section by Groisman and Easterling (1995) is recommended.

IN SITU OBSERVATIONS

In situ snow-cover data are gathered mainly over land. Only a few short-term studies have measured snow on sea ice or ice sheets (e.g., Hanson, 1980). Most observations on land are made on a once-per-day basis. The general practice is to record the average depth of snow lying on level, open ground that has a natural surface cover. At primary stations, the water equivalent of the snowpack may also be measured. In some regions, snow courses have been established where snow depth, water equivalent, and perhaps other pack properties are measured along prescribed transects across the landscape. Observations are often made only once per month, and the number of courses is extremely limited in North America. More frequent and abundant course data are gathered in the Commonwealth of Independent States, and currently this information is being recovered through a cooperative effort between the U.S. National Snow and Ice Data Center (NSIDC) at the University of Colorado and A. Krenke of the Russian Academy of Sciences.

Current station observations of snow cover are of a sufficient density for climatological study in the lower elevations of the middle latitudes. Elsewhere, while data of a high quality are gathered at a number of locations (Barry, 1983), the spatial and temporal coverage of the information is often inadequate for climate study. There is no hemispheric snow-cover product based entirely on station reports. The U.S. Air Force global snow-depth product depends heavily on surface-based observations as input into a numerical model that creates daily charts with global coverage, but it must rely on extrapolations and climatology in data-sparse regions (Hall, 1986; Armstrong and Hardman, 1991). There have been a number of regional snow-cover products over the years that are based on station data. Of greatest longevity are the Weekly Weather and Crop Bulletin charts, which have been produced since 1935. These, and the daily NOAA charts, are produced for the conterminous United States mainly from first-order station observations. Therefore, neither has a particularly high resolution, and observations may be influenced by urban heat-island effects.

In a number of countries, there are numerous stations with relatively complete records of snow extending back 50 years or more (Barry and Armstrong, 1987). Until recently, most data have remained unverified and disorganized (Robinson, 1989). As a result, few studies have dealt with long-term trends or low-frequency fluctuations of snow over even small regions (e.g., Arakawa, 1957; Manley, 1969; Jackson, 1978; Pfister, 1985). Through the cooperative efforts of a number of scientists and data centers, this situation has begun to be rectified. Examples include the exchange of data through the US/USSR Bi-Lateral Environmental Data Exchange Agreement and between the Lanzhou Institute of Glaciology and Geocryology and both Rutgers University and the NSIDC. These and other data are in the process of being quality controlled, and routines to fill in gaps in snow-cover records are being developed (Hughes and Robinson, 1993; Robinson, 1993a). Clearly, there is a need to continue efforts to identify, assimilate, in some cases digitize, and in all cases validate station and snow-course observations from around the world. These data must also must be accompanied by accurate and complete metadata.

Lengthy in situ records continue to be analyzed for individual stations, and data from networks of stations have begun to be studied on a regional level. For example, marked year-to-year variability in snow-cover duration is recognized over the course of this century at Denison, Iowa (Figure 1). Snow at least 7.5 cm deep has covered the area for as much as 80 percent of the winter, but in a number of years no or only a few days have had a cover this deep. Overall, the duration of winter and spring cover was at a maximum in the 1970s, and fall cover peaked in the 1950s. Other periods of more frequent winter cover include the 1910s and the late 1930s to early 1940s. The 1920s, middle 1940s, and late 1950s to early 1960s were periods with less abundant winter cover. Missing data around 1950 prohibit a direct assessment of winter cover at Denison, but adjacent stations suggest cover was scarce at this time. All three seasons show a greater abundance of snow-cover days in the past 40 years than in the first half of the century.

Efforts are under way to develop gridded snow files for a large portion of the central United States, using data from several hundred stations. Raw and filtered winter records from four of these stations are shown in Figure 2. They are for days with snow cover =7.5 cm, and all indicate long-term fluctuations on the order of one to several decades. The Nebraska and Kansas stations show maximum durations during the past several decades, with a similar early maximum at Oshkosh, Nebraska, in the 1910s and early 1940s. Late 1920s, early 1950s, and 1970s maxima were observed at Dupree, South Dakota, the latter two ending abruptly shortly thereafter. The North Dakota station had maximum winter snow cover in the 1930s, around 1950, and in the late 1970s. The range in filtered values over the period of record was approximately two weeks in Kansas and Nebraska and seven weeks in South Dakota and North Dakota.

SATELLITE OBSERVATIONS

Snow extent is monitored using data recorded in short-wave (visible and near-infrared) and microwave wave-

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 1

Time series of fall (September-November), winter (December-February), and spring (March-May) days with =7.5 cm of snow cover at Denison, Iowa. Missing years are marked with an X.

lengths by sensors on board geostationary and polar orbiting satellites. Retrieval techniques, the strengths and limitations of each spectral region for sensing snow, and the snow products derived using short-wave and microwave input are discussed in this section. The secular remote sensing of snow over Northern Hemisphere lands will be the principal focus; only a few efforts have addressed this over Southern Hemisphere lands (Dewey and Heim, 1983) or Arctic sea ice (Robinson et al., 1992).

Short-Wave and Microwave Snow Charting

Short-wave data provide continental coverage of snow extent at a relatively high spatial resolution. Snow is identified by recognizing characteristic textured surface features and brightness. Information on surface albedo and percentage of snow coverage (patchiness) is also gleaned from the data. Shortcomings include (1) the inability to detect snow cover when solar illumination is low or when skies are cloudy, (2) the underestimation of cover where dense forests mask the underlying snow, (3) difficulties in distinguishing snow from clouds in mountainous regions and in uniform, lightly vegetated areas that have a high surface brightness when covered with snow, and (4) the lack of all but the most general information on snow depth (Kukla and Robinson, 1979; Dewey and Heim, 1982).

Microwave radiation emitted by the earth's surface penetrates winter clouds, permitting an unobstructed signal from the surface to reach a satellite. The detection of snow cover from microwave data is possible mainly because of differences in emissivity between snow-covered and snow-free surfaces. Estimates of the spatial extent, as well as of the depth or water equivalent, of the snowpack are derived from equations that employ measurements of radiation sensed by multiple channels in the microwave portion of the spectrum (e.g., Kunzi et al., 1982; McFarland et al., 1987). Estimates of snow cover have been made using microwave data since the launch of the Scanning Multichannel Microwave Radiometer (SMMR) in late 1978. The spatial resolution of the data is approximately several tens of kilometers. Since 1987, close to the time of SMMR failure, the Special Sensor Microwave Imager (SSM/I) has provided data. Both sensors, having nearly the same spectral characteristics, have similar levels of success in monitoring snow extent (cf. the SMMR analyses below).

As with short-wave products, the microwave charting of snow extent is not without its limitations. The resolution of the data makes the detailed recognition of snow cover difficult, particularly where snow is patchy, and it is difficult to identify shallow or wet snow using microwaves. Also, the lack of sufficient ground-truth data on snow volume, wetness, and grain size makes an adequate assessment of the reliability of microwave estimates uncertain. The influence of a forest canopy on microwave emissions in snow-

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 2

Time series of winter days with =7.5 cm of snow cover at four stations on the U.S. Great Plains. Also shown smoothed with a nine-point binomial filter, with only those points plotted where nine years are available for filtering (i.e., plotted year ±4 yr).

covered regions, which is currently not well understood, must also be taken into account when estimating snow cover. Because of the region-specific differences in land cover and snowpack properties, no single algorithm can adequately estimate snow cover across Northern Hemisphere lands.

Hemispheric Snow Products: NOAA Weekly Charts

In 1966, NOAA began to map snow cover over Northern Hemisphere lands on a weekly basis (Matson et al., 1986). That effort continues today, and remains the only such hemispheric product. NOAA charts are based on a visual interpretation of photographic copies of short-wave imagery by trained meteorologists. Up to 1972, the sub-point resolution of the meteorological satellites commonly used was around 4 km; since then it has become, and remained, close to 1.0 km. Charts show boundaries on the last day that the surface in a given region was seen. Since May 1982, dates when a region was last observed have been placed on the charts. An examination of these dates shows the charts to be most representative of the fifth day of the week.

It is recognized that in early years the snow extent was underestimated on the NOAA charts, especially during fall. Charting improved considerably in 1972 with the improvement of sensor resolution, and since then charting accuracy

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

has been such that this product is considered suitable for continental-scale climate studies (Kukla and Robinson, 1981). The NOAA charts are most reliable where skies are frequently clear, solar illumination is high, snow cover is reasonably stable or changes slowly, and pronounced local and regional signatures are present owing to the distribution of vegetation, lakes, and rivers.

The NOAA charts are digitized on a weekly basis using the National Meteorological Center Limited-Area Fine Mesh grid. This is an 89 ´ 89 cell Northern Hemisphere grid, with cell resolution ranging from 16,000 to 42,000 km2. If a cell is interpreted as being at least 50 percent snow covered, it is considered to be completely covered; otherwise it is considered to be snow free. Inconsistencies in the designation of a cell as land or water have occurred in the past during the digitization process. This has recently been resolved for the approximately 100 cells in question through the use of digital map files analyzed on a geographic information system (Robinson, 1993b).

A new routine for calculating monthly snow areas from the NOAA data has also recently been developed (Robinson, 1993b). This has eliminated previous inconsistencies resulting from undocumented changes in the methods used by NOAA to calculate the monthly values (Robinson et al., 1991). The new Rutgers Routine calculates weekly areas from the digitized snow files and weights them according to the number of days of a given week that fall in the given month. A chart week is considered to center on the fifth day of the published chart week (cf. above). No weighting has been employed in the NOAA routines.

Hemispheric Snow Products: NASA Microwave Files

Monthly charts of Northern Hemisphere continental snow extent have been produced from SMMR data by NASA scientists (Chang et al., 1990). The only such time series available to date, it covers the interval from November 1978 through August 1987. A single algorithm is used to estimate snow depth on a 0.5° × 0.5° grid. This theoretical algorithm uses the difference in brightness temperatures of 18 and 37 GHz SMMR data to derive a snow-depth/brightness temperature relationship for a uniform snow field. A snow density of 0.3 g/cm3 and a snow-crystal radius of 0.3 mm are assumed; fitting the differences to the linear portion of the 18 and 37 GHz responses permits the derivation of a constant that is applied to the measured differences. This algorithm can be used for snow up to 1 m deep.

The monthly depth estimate for a given grid cell is calculated by averaging depths reported for the five or six pentad charts centered in a given month (SMMR data are gathered every other day and three of these passes are used for each pentad chart; there is a one-day gap between each pentad). If the average is =2.5 cm, the cell is considered to be covered with snow for the whole month. This methodology biases the snow areas to the high side, especially in those areas and periods where snow cover fluctuates.

Northern Hemisphere Continental Snow Cover: 1972-1992

According to the NOAA snow charts, the extent of snow cover over Northern Hemisphere lands varies from 46.5 million km2 in January to 3.9 million km2 in August (Table 1); most of August's snow lies on top of the Greenland ice sheet. The past two decades of monthly data are close to normally distributed, and monthly standard deviations range from 0.9 million km2 in August to 2.9 million km2 in October. The annual mean cover is 25.5 million km2 with a standard deviation of 1.1 million km2. The snowiest year was 1978, which had a mean of 27.4 million km2; 1990 was the least snowy at 23.2 million km2. Eight of the monthly minima for the two decades occurred in 1990.

With only two decades of reliable hemispheric information, it is impossible to identify anything in the way of trends or cycles in the temporal or spatial distribution of snow. What has become recognizable during the period of record is a tendency for multi-year departures in snow cover, in which less pronounced month-to-month and season-to-season departures are embedded. Twelve-month running means illustrate the periods of above-normal cover that occurred in the late 1970s and mid-1980s, and the lower and below-normal extents during the mid-1970s and early 1980s (Figure 3). The most significant lengthy departure during the past 20 years began in the late 1980s and continued into 1992. Of the 65 months between August 1987 and December 1992, only eight had above-normal snow cover. Three of these eight were September, November, and December 1992.

Spring cover has shown pronounced deficits over the past five years in Eurasia and six years in North America; areas in these springs have been at or below lows established before this period (Figure 4). During the same interval, both continents have had low seasonal cover in the fall and summer, although frequently neither continent has been at or approached record low levels. Winter cover has been close to average over the past six years.

The NOAA snow estimates are considered the most accurate figures available. Despite the positive methodologically induced bias of the monthly NASA estimates (cf. the previous section), they range from less than I million up to 13 million km2 below the NOAA area estimates for those nine years for which both agencies provide estimates. These absolute differences are greatest in the late fall and early winter. In a relative sense, microwave areas are between 80 and 90 percent of short-wave values in winter and spring, 20 to 40 percent of the short-wave estimates in summer, and 40 to 70 percent of short-wave areas in fall. Possible explanations for the significant disparities in the latter two seasons are wet and shallow snows. These are difficult if

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

TABLE 1 Monthly and Annual Snow Cover (million km2) over Northern Hemisphere Lands, Including Greenland

 

Maximum (yr)

Minimum (yr)

Mean*

Median

Standard Deviation

Jan

49.8 (1985)

41.7 (1981)

46.5

46.1

1.8

Feb

51.0 (1978)

43.2 (1990,92)

46.0

45.6

2.0

Mar

44.1 (1985)

37.0 (1990)

41.0

40.8

1.9

Apr

35.3 (1979)

28.2 (1990)

31.3

31.4

1.8

May

24.1 (1974)

17.4 (1990)

20.8

20.6

1.9

Jun

15.6 (1978)

7.3 (1990)

11.6

11.4

2.1

Jul

8.0 (1978)

3.4 (1990)

5.3

5.5

1.2

Aug

5.7 (1978)

2.6 (1988,89,90)

3.9

3.8

0.9

Sep

7.9 (1972)

3.9 (1990)

5.6

5.6

1.1

Oct

26.1 (1976)

13.0 (1988)

17.6

17.5

2.9

Nov

37.9 (1985)

28.3 (1979)

33.0

32.8

2.3

Dec

46.0 (1985)

37.5 (1980)

42.5

42.7

2.3

Annual

27.4 (1978)

23.2 (1990)

25.5

25.4

1.1

* Means are for the period January 1972 through May 1992; extremes are for January 1972 through December 1992.

not impossible to monitor using microwaves. Depth may be the more important of these two variables, given the better agreement in spring, although it has been suggested that unfrozen soil beneath the pack is a major contributor to the underestimates during fall (B. Goodison, pers. commun.). The 85-GHz channel on the SSM/I has shown promise in improving the monitoring of shallow (85 cm) snow cover (Nagler and Rott, 1991).

FIGURE 3

Twelve-month running means of snow cover over Northern Hemisphere lands (including Greenland) for the period January 1972 through December 1992. Running means are also shown for Eurasia and North America (including Greenland). Values are plotted on the seventh month of the twelve-month interval.

CONCLUSIONS

Significant strides have been made in recent years in the recovery, digitization, and validation of in situ observations of snow cover and their assimilation into regional networks. The availability of this information, in some cases going back to the turn of the century, complements the more recent global monitoring of snow from satellites. Analyses of both in situ and satellite observations show a considerable amount of year-to-year variability in the duration of snow cover on local to hemispheric scales. However, it is beginning to be recognized that this variability is often embedded in longer-term fluctuations. In situ data from the central United States points to multi-decadal fluctuations in the duration of snow over the past century, and the hemispheric satellite observations suggest variations lasting on the order of several years.

Efforts must continue to ensure the recovery and digitization of all available historic in situ observations, from both stations and snow courses. On the satellite side, the NOAA weekly product must continue to be produced in its present form. To abruptly or, perhaps more seriously, subtly alter the manner in which these charts are produced would severely weaken what is at present the longest and most consistent satellite-derived data set for any surface or atmospheric variable. Coincident with this must be efforts to improve the hemispheric monitoring of snow from microwave data using composited regional algorithms. Ultimately the goal must be to integrate all surface and satellite sources in a series of products reporting snow extent, depth and water equivalent, and regional surface albedo. These data should cover not only Northern Hemisphere lands but also lands in the Southern Hemisphere, ice sheets, and sea ice. Means

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 4

Seasonal time series of snow cover over Eurasia and North America (Greenland is excluded).

of integrating satellite and integrative products with earlier in situ observations must also be improved to provide the longest, most consistent records possible.

With adequate data available to analyze, knowledge of the spatial and temporal kinematics of snow cover over recent decades or even the past century will continue to improve. This will contribute to a better understanding of the role of snow in the climate system and to the utility of snow as an indicator of climate change.

ACKNOWLEDGMENTS

Thanks to G. Stevens at NOAA for supplying digital snow data and to A. Chang at NASA for providing microwave data. Thanks also to A. Frei, M. Hughes, and J. Wright at Rutgers for technical assistance. This work is supported by NOAA under grant NA90AA-D-AC518, and by the Geography and Regional Science Program of the National Science Foundation under Grant SES-9011869.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Variability and Trends of Precipitation and Snowfall over North America

PAVEL YA. GROISMAN1 AND DAVID R. EASTERLING2

ABSTRACT

Biases and large-scale inhomogeneities in the time series of measured precipitation and snowfall over the continent are discussed and analyzed. After adjustments and selection of the ''best" network, reliable "first guess" estimates of North American snowfall and precipitation have been obtained. Century-long time series of unbiased annual precipitation over the regions to the south of 55°N, and 40-year time series of unbiased area-averaged annual precipitation and snowfall for all of North America, have been developed. Analysis of the trends shows that:

  1. During the last hundred years, annual precipitation has increased in southern Canada (south of 55°N) by 13 percent and in the contiguous United States by 4 percent to 5 percent; the main domain of this century-scale precipitation increase, however, is eastern Canada and the adjacent northern regions of the United States.

  2. An increase of up to 20 percent increase in annual snowfall and rainfall has occurred during the last four decades in Canada north of 55°N.

INTRODUCTION

During the last century, global surface air temperature has increased approximately 0.5°C (over Canada it has been about 1°C); the seven warmest years in the instrumental records of global temperature occurred during the last decade (Budyko and Izrael, 1987; Gullett and Skinner, 1992; IPCC, 1990, 1992). With such a large change in the earth's heat balance, we suspect that changes in the water balance of North America may have also occurred. Given the potential for even greater changes over the next decades, there is an urgent need to obtain reliable results on North American precipitation changes during the last century.

Numerous studies have been made of precipitation over North America and the problems associated with precipitation measurement. The results of these studies can be summarized by saying that no definite century-scale precipitation trends have been found for the United States. (Canadian precipitation cannot be analyzed for the entire century,

1  

State Hydrological Institute, St. Petersburg, Russia; current address: Department of Geosciences, University of Massachusetts, Amherst, Massachusetts

2  

NOAA National Climatic Data Center, Asheville, North Carolina

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

due to lack of data in the northern part of the country.) Fluctuations on decadal time scales are evident, however, which led Klugman (1983) to claim "evidence of climatic change" in the U.S. seasonal precipitation during the 19481976 period. Diaz and Quayle (1980) found large-scale changes in contemporary precipitation (1955-1977) as compared to precipitation at the beginning of the century (1895-1920), but these changes were shown to be mainly statistically non-significant. Vining and Griffiths (1985) found an increase over time in the decadal variance of annual precipitation at 10 long-term U.S. stations for the period 1900-1979. Bradley et al. (1987), analyzing the area-averaged percentile of the U.S. annual precipitation, found a precipitation decrease during the period from the 1880s to the 1930s and a general increase thereafter.

There are several reasons for precipitation undercatch by the standard gauges currently being used worldwide (Bogdanova, 1966; WMO, 1991; WWB, 1974). The main factor that reduces the amount of measured precipitation as compared to "ground truth" is wind-induced turbulence over the gauge orifice (Sevruk, 1982). The absence of a wind shield for the gauge (or its poor design) makes the problem even more severe. Field experiments show that an appropriate wind shield (such as the modified Nipher shield used in Canada) reduces wind-related bias in the gauge catch to manageable levels that can then be easily adjusted. All elevated snow gauges in Canada have wind shields, but only about 200 of the approximately 6,000 U.S. gauges that report daily precipitation have a wind shield (Karl et al., 1993a). Furthermore, the Alter shield in use at U.S. stations provides less protection than the Canadian shield (Goodison et al., 1981). As a result, the existing U.S. rain-gauge network measures rainfall with a bias of 3 to 10 percent (Golubev et al., 1992; Sevruk and Hamon, 1984) and snowfall with a bias of up to 50 percent or more (Goodison, 1978; Larkin, 1947; Larson and Peck, 1974). Moreover, in Alaska total biases up to 400 percent in measurements of water equivalent of snowfall have been reported (Black, 1954).

Rain gauges at some of the U.S. primary stations were equipped with the Alter wind shields toward the end of the 1940s. The addition of the wind shield introduced an inhomogeneity into the time-series precipitation data at these stations. Although the stations with Alter shields constitute a small percent of the U.S. stations, they comprise 40 percent of the U.S. National Weather Service stations transmitting meteorological observations over the Global Telecommunication System (GTS) for publication in the Monthly Climatic Data for the World.

Canadian methods of making precipitation measurements differ significantly from those used in the United States. This is illustrated by the jump in measured precipitation values often registered by stations on opposite sides of the border between the two countries (WMO, 1979; WWB, 1974). Although the difference in liquid precipitation is small—the Canadian gauge typically measures a few percent more rain than the standard U.S. gauge (Sanderson, 1975)—there is a considerable difference in the solid precipitation measurements (Goodison et al., 1981; Sanderson, 1975). Currently, 85 percent of the Canadian meteorological network uses a 10:1 ratio as a measure of the water equivalence of fresh-fallen snow. Since the 1960s, 15 percent of this primary network has used an elevated snow gauge with a modified Nipher wind shield. This gauge catches solid precipitation effectively even during high-wind conditions (Goodison, 1978; Goodison et al., 1992). Its record differs systematically from the snow-stick measurements continued at other stations (Goodison et al., 1981; Groisman et al., 1993; Karl et al., 1993a). This difference in snow-measurement methods causes both space and scale inhomogeneities, which cannot be avoided by users of such international publications as World Weather Records and Monthly Climatic Data for the World, since Canada transmits only the data from its primary network over the GTS.

When point precipitation measurements are expanded to area-averaged values in rough terrain, the point values are often less than the "ground-truth area mean", because most meteorological stations are located in valleys, but the vertical gradients in precipitation have been neglected. Figure 1 shows two maps of annual precipitation over North America. The first map (la), which was constructed during the World Water Balance studies (1974), incorporates estimates of topographic effects on precipitation. However, the second map (lb) simply uses data from 1,900 stations, without any consideration of topographic effects on precipitation. Precipitation in the mountainous West is obviously underestimated on the second map, and even Appalachian-ridge precipitation estimates are noticeably biased.

The area-averaged annual precipitation value for the contiguous United States derived from Figure 1a is about 880 mm, while the value from Figure 1b is 700 mm. When we use data from all 6,000 U.S. cooperative stations, this latter annual total increases to 735 mm (NCDC, 1991) but still remains quite low compared to the value reflecting topography. The same problem has been revealed to exist for Canada by Hare (1980), who showed that annual totals over the country should be increased by 20 percent, on average, to obtain a reliable water balance (i.e., runoff = precipitation - evaporation).

Precipitation measurements are very sensitive to changes in the environment surrounding the gauge, in the gauge type, and in the methods of measurement. A small move of the gauge can result in a twofold change in the values of measured precipitation if the gauge exposure is changed (Karl et al., 1993b; Sevruk, 1982). After spatial averaging, it can be expected that the random errors connected with such shifts will cancel out. However, when systematic changes occur countrywide, they cannot be assumed to cancel out. Several such changes have occurred during the

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 1

Mean annual precipitation over North America: left panel, from WWB (1974); right panel, as estimated using the meteorological network.

last century in the Canadian and the U.S. meteorological primary networks (Figure 2). Area-averaged precipitation time series are especially sensitive to such large-scale inhomogeneities because their noise component has been decreased by averaging. Therefore, when studying trends in these time series, we must be sure that the observed large-scale decadal changes represent true climatic changes rather than artificial changes introduced into the network. Any analysis of precipitation over North America should be preceded by a thorough analysis of the history of observations in the network. This analysis should be followed by selecting a subset of the "best stations" and applying adjustments to the initial data to compensate for all known large-scale inhomogeneities. Our analyses in the following sections give special attention to the selection of the precipitation data and to the adjustments to be made to them.

DATA SELECTION AND PROCESSING
Networks Used

In our studies of precipitation over North America, we selected several networks, each with a special purpose in mind. Figure 3a shows the network that was used to study snowfall. It consists of 223 U.S. first-order stations (including Alaska) and 1,107 Canadian stations selected using the requirement that each station have data available for at least 15 years out of the last two decades. The period 1950-1990 was chosen because by then all U.S. first-order stations had been relocated to airports with relatively open sites, and because the networks for northern Canada and Alaska had become dense enough to assure adequate spatial sampling only after World War II. A reliable analysis of changes in snowfall and total precipitation over the entire continent could thus be made only for the last 40 years.

Since the 1960s there have been parallel observations of snowfall (by stick measurements) and its water equivalent (measured by elevated snow gauges) at 335 Canadian stations. These observations were used to evaluate the adjustments made by Groisman (1992) and Karl et al. (1993a) to homogenize time series of solid precipitation in Canada.

Figure 3b shows the network that was used to study century-scale changes of annual precipitation over the contiguous United States and southern Canada. It consists of the same Canadian stations, 45 cooperative Alaskan stations, and 593 "best" cooperative stations selected from the U.S. Historical Climatology Network (Karl et al., 1990). The U.S. stations were selected on the basis of the length of their records (they all began their observations in the nineteenth century) and their having had few station moves or other

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 2

Sketch of the history of precipitation measurement in North America related to precipitation time-series homogeneity. Major large-scale changes caused inhomogeneities in these time series. (a) Canadian primary network; (b) U.S. primary network.

obvious sources of inhomogeneities. The data are therefore free of the large-scale inhomogeneities shown in Figure 2, but other inhomogeneities may still exist.

The Canadian network, which seems quite large in Figure 3b, declined rapidly in size during the first half of the century (see Table 1). In southern Canada it contained only 190 stations by 1940, and 26 stations in 1900. As a consequence, we selected a special network of Canadian stations that were operating in the 1920s (Figure 3c) to use in to studying Canadian precipitation prior to World War II. The area-averaged precipitation time series based on this network were combined with the corresponding time series developed from the main network shown in Figure 3b.

Data Processing

The precipitation data for the 593 contiguous U.S. stations and the snowfall data were used without adjustments.

FIGURE 3

Stations used in this paper for analysis of: (a) snowfall (squares depict the primary Canadian stations where the elevated snow gauges were installed in the 1960s); (b) precipitation; (c) accuracy of precipitation changes in southern Canada before World War II.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

TABLE 1 Number of Stations with Available Data on Annual Precipitation for Southern Canada

 

Year

Data Set

1895

1910

1925

1940

1955

1990

Total

Main data set used in the study

22

35

119

190

315

765

995

Alternative set of Canadian stations with observations prior to 1925

54

86

323

261

162

24

385

All Canadian stations with observations prior to 1925

70

117

439

384

277

133

523

The Canadian rainfall data were corrected for the gauge change that occurred in the mid-1970s. All rainfall data before 1975 were increased by a factor of 1.025 to approximate the results of Goodison and Louie (1986).

To obtain the water equivalence of snowfall at the Canadian stations, we were able to derive scale adjustments from the set of parallel observations made at the Canadian stations depicted in Figure 3a (full squares). These adjustments reduce snow-stick measurements to match the records of the elevated snow gauge. However, these records still need some corrections for the type (or absence) of wind shield (Goodison, 1978). Larger corrections are required for the U.S. rain gauges (Larson and Peck, 1974; Goodison and Metcalfe, 1989; Peck, 1991) because these gauges do not have wind shields. We hope to be able to use such corrections in the future, but for the moment we can only conclude that the discontinuity across the border of the two countries in measured water equivalent of snowfall is caused by significantly biased measurements in the U.S. data. Therefore, area-averaging of annual precipitation over the regions encompassing both Canada and the United States was done separately for each country. Only snow-stick measurements were used to represent the Canadian solid precipitation. Each area-mean time series was adjusted to its unbiased precipitation value (using the method described by Groisman et al. (1991), which relies on the WWB (1974) estimates of long-term annual precipitation shown in Figure 1a); then time series from the two parts of the region were combined. Furthermore, we prefer to consider the southern Canadian regions separately, due to the instability of their data coverage.

As noted earlier, Table 1 shows the rather variable number of stations providing annual-precipitation data for southern Canada (south of 56°N) for the two non-interlaced networks shown in Figures 3b and 3c. Figure 4 shows time series of annual snowfall and total precipitation area-averaged over Canada south to 55°N, based on these two independent data sets. There is little difference between them during the twentieth century, which confirms the accuracy of the area averaging and the representativeness of the network used. The only noticeable difference is in the 1890s, when precipitation data were available for only 20 to 30 stations over the main network. Hence, we combined these two networks to develop a century-long area-averaged time series of southern Canadian precipitation. The time series for the period 1891-1960 were developed from the network of stations shown in Figure 3c, together with those stations from the main Canadian network that were operating prior to 1925. These time series were combined with the time series for the period 1921-1990 that was based on the data of the 1,107 stations.

Snowfall Scale Correction

In developing the data set used to construct unbiased area-averaged snowfall time series (Figure 5), we made one

FIGURE 4

Annual snowfall and total precipitation over southern Canada (south of 55°C) estimated by the network depicted in Figures 3b (dashed line) and 3c (solid line).

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 5

Annual snowfall over North America during the last 40 years. The applied scale-correction factors for network elevation are: for Alaska—1.23; for Canada, Zone 55-70°N—1.14, Zone 45-55°N—1.19, and Zone 35-45°N—1.47.

important scale correction that was not performed in the work of Karl et al. (1993a). Bogdanova (1977) showed that in mountains the mean long-term ratio of solid to total annual precipitation rises at a mean rate of 27 percent per 1000 m. We used this to adjust the snowfall time series for network elevation. Without this adjustment, the biases in the regionally averaged snowfall values would be very large. Table 2 provides the mean elevation of the network depicted in Figure 3b, together with true mean surface elevation for large regions of the United States and Canada (the mean elevation of the networks depicted in Figures 3a and 3b scarcely differ). From this table it is clear that with our network it would be possible to estimate occurrence of

TABLE 2 Mean Elevations of Meteorological Stations in the United States (Historical Climatology Network) and Canada Versus Real Surface Elevations*

 

Elevation, m

Region

Network

Surface

Western United States**

790

945

Eastern United States

145

235

Canada

305

490

Alaska west of 141°W

100

510

* Real surface elevations were estimated from NGDC (1988) data with 5-degree resolution

** "Western" and "Eastern" here refer to west or east of a line drawn from New Orleans to Detroit

traces of snowfall only in the valleys and on the plains of western United States, and completely miss its mountainous portion. So our snowfall adjustment for each region was evaluated in the following way:

  1. The mean long-term ratio of solid to total annual precipitation (rat1) was estimated for a plane with the elevation of the network depicted in Figure 3a.

  2. This ratio was increased using a coefficient of 27 percent km-1 to obtain its value (rat2) for a plane with the elevation corresponding to the true surface elevation of the region.

  3. The value of the estimated area-averaged snowfall was then multiplied by an adjustment scale coefficient equal to rat2 divided by rat1.

These coefficients were used for every graph in Figure 5. They are, however, conservative estimates of the adjustment needed if indeed the total precipitation area-averaged over the whole region depends on true surface elevation rather than on the elevation of the network used ( Karl et al., 1993a).

In the following section, area-averaging by the polygon method (Kagan, 1979; Thiessen, 1911) and scale correction (Groisman et al., 1991) were used in to develop regional, unbiased century-scale annual precipitation time series for the contiguous United States and southern Canada, together with 40-year time series of snowfall and total precipitation over all of North America. So that missing data would not affect the results, we averaged the anomalies from the reference period 1971-1990 and then reconstructed the area-averaged precipitation, adding the corresponding area-

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

averaged value of the mean precipitation for the same reference period. (While processing the data of the Canadian network that was operating in the 1920s, we used the reference period 1921-1940.) These time series provide our first approximation of contemporary changes of North American precipitation. The first experiments using individual gauge corrections (Groisman, 1991) showed that this approximation does not change substantially on a continental scale after the intra-annual variability in gauge errors has been taken into account.

CURRENT CHANGES IN PRECIPITATION AND SNOWFALL
Snowfall Changes During the Last Four Decades

Figure 5 shows the zonally averaged snowfall over Alaska, Canada north of 55°N, and the zones 45°N to 55°N and 35°N to 45°N for the last four decades. The basic statistics of these time series appear in Table 3.

When the spatial correlation function is known, the accuracy of area averaging over these regions using the networks depicted in Figure 3 can be estimated. The method of estimation, which was developed by R.L. Kagan (1979), can be applied to every isotropic two-dimensional field. In North America we used this method only for northern Canada. Since even today the meteorological network in northern Canada is relatively sparse, we had to estimate the representativeness of the network beforehand in order to analyze the changes over this area.

The spatial correlation function for northern Canadian snowfall can be approximated by the relationship r(R) =, where R is the distance; R0 = 710 km, the radius of correlation; and C = 0.84. The value of (1-C)/C, which equals 0.19, is a measure of the error of measurements combined with microscale snowfall variability near the stations (Kagan, 1979). For the network selected, the mean-square error of spatial averaging of annual snowfall over northern Canada was estimated to be 17 percent of the theoretical variance of the area-averaged snowfall time series in the first half of 1950s; during the period 1956-1990 the error was about 10 percent. The number and spatial distribution of snowfall-measuring stations for southern Canada and the United States during the last four decades were adequate, so there was no need to verify the representativeness of the area-mean values.

The accuracy of area averaging of the data in northern Canada is high enough to confirm the strong positive linear trend in annual snowfall revealed in the zone (19 percent over four decades). This trend, in conjunction with the trend in rainfall data (not shown), confirms the systematic increase in annual totals of precipitation during the entire period of instrumental observations.

Snowfall in the zone 45°N to 55°N is closely connected to temperature changes on continental and hemispheric

TABLE 3 Statistical Characteristics of Time Series (1950-1990) of Annual Snowfall and Total Precipitation Zonally Averaged over North America (see Figures 5 and 6)

Annual Snowfall

Mean and Standard Deviation, in cm

Region

Mean

Std. Deviation

Linear Trend ± its Standard Error, in % per Decade

Alaska

190

35

2.7 ± 2.5

Canada, zone 55-70°N

195

17

5.1 ± 0.8

Canada/U.S., zone 45-55°N

260

27

-0.7 ± 1.4

U.S./Canada, zone 35-45°N

140

18

1.8 ± 1.7

Total Precipitation

Mean and Standard Deviation, in mm

Region

Mean

Std. Deviation

Linear Trend ± its Standard Error, in % per Decade

Alaska

510

59

2.7 ± 1.5

Canada, zone 55-70°N

495

33

4.2 ± 0.6

Canada/U.S., zone 45-55°N

940

41

0.7 ± 0.6

Canada/U.S., zone 35-45°N

905

66

1.9 ± 0.9

United States, zone 28-35°N

1175

137

3.4 ± 1.5

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

scales; the correlation between this time series and the annual mean surface air temperature over the Northern Hemisphere is about -0.7 (Groisman et al., 1993). Using this relationship, and taking into account the intrinsic coherence of annual snowfall totals in southern Canada (which encompasses the main part of the 45°N to 55°N zone) with regional maximum temperature (Groisman, 1992), we could expect a future decrease in snowfall for this zone if the global warming projected by IPCC (1990) actually occurs.

We realize that the large adjustment coefficient for zonal snowfall in the 35°N to 45°N zone (1.47), while providing a more realistic mean value, cannot be considered a satisfactory solution to the problem of snowfall changes over this zone. Unfortunately for the current concerns about potential climatic change, we still do not have a reliable source of data on the snowfall over the main part of the United States.

Annual Precipitation Changes During the Last Four Decades

Figure 6 shows the unbiased annual precipitation totals for the last four decades, area-averaged over the same four zones used in Figure 5 plus the southern part of the United States (a zone from 28°N to 35°N and to the east of 105°W). The basic statistics of these time series appear in Table 3. It can be seen from this table that annual precipitation has increased during these last decades over the entire continent, but especially significantly over the southeastern United States (14 percent per 40 years) and northern Canada (17 percent per 40 years).

Century-Long Precipitation Changes

Table 4 (adapted from Groisman, 1992) provides the same characteristics as in Table 3 for the century-scale precipitation changes over southern Canada and the contiguous United States. This table confirms the main conclusions derived from Table 3, but shows that, as with the U.S. precipitation, it is unwise to discuss the statistical significance of precipitation trends on the basis of the 41-year record alone (on the century scale the linear trend is not statistically significant).

Given the major terms of the water-balance equation, we can assume that, in general, the variability of annual precipitation drives the variability of annual stream flow (Linsley et al., 1958). Streamflow will thus have approximately the same large-scale features and areas of coherent change as the precipitation used to estimate it. The stream flow data are independently collected and do not exhibit the errors and data problems mentioned in previous sections (they have their own problems and errors, but do not suffer from the rain-gauge problem; being natural area averages, they give representative estimates of changes in the hydrological cycle of the regions under consideration. Therefore, in our study of the spatial correlation of the precipitation field over North America, we divided the area into several homogeneous regions on the basis of the results of a principal component analysis of stream flow data for the United States made by Lins (1985). In his analysis, Lins used a dense network of small watersheds not disturbed by anthropogenic activity. He evaluated five principal components,

FIGURE 6

Annual precipitation over North America during the last 40 years.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

TABLE 4 Statistical Characteristics of the Century-long Time Series of Annual Precipitation, Area-Averaged Over Canada South of 55°N and the Contiguous United States

Mean and Standard Deviation, in mm

Region

Mean

Standard Deviation

Linear Trend ± its Standard Error, in % per 100 yrs

Southern Canada

925

56

13.3 ± 1.7

Contiguous United States

870

66

4.5 ± 2.6

each with a distinctive regional pattern: south-central United States, northwestern United States, southwestern United States, north-central United States, and northeastern United States. The generalized map of these regions expanded on the Canadian territory up to 55°N is shown in Figure 7.

In order to identify the independent components of the precipitation pattern, we applied the spatial averaging to the regions depicted in Figure 7. Time series of adjusted annual precipitation area-averaged over these five regions are shown in Figure 8; their correlation matrix appears in Table 5, and their main statistical features in Table 6. From Table 5 we can see that the time series from the different regions are not closely correlated (the north-central region correlated best with other regions, since it has a common border with each of them). Hence, we can analyze and describe each time series separately.

The debiasing procedure used (Groisman et al., 1991) does not affect trends, but it does change the scale of the time series. For Canada and eastern United States, the corresponding scale factors are not large; they vary from 1.05 (south-central region) to 1.25 (northeastern region). However, in the western United States they become larger (1.5 for the southwestern region and 1.7 for the northwestern United States). This shows that even annual totals are not well measured by the existing meteorological network.

An increase of more than 10 percent per century in annual precipitation over southern Canada (Table 4) is reflected by a corresponding precipitation increase over the northeastern

TABLE 5 Correlation Matrix of the Area-Averaged Century-long Time Series of Annual Precipitation Depicted in Figure 8

U.S. Region*

N-West

N-Central

N-East

S-West

S-Central

Northwestern

1.00

0.44

0.26

0.23

0.16

North-central

 

1.00

0.29

0.28

0.38

Northeastern

 

 

1.00

0.08

0.11

Southwestern

 

 

 

1.00

0.11

South-central

 

 

 

 

1.00

* See Figure 7 for a depiction of these regions

TABLE 6 Statistical Characteristics of the Area-Averaged Century-long Time Series of Unbiased Annual Precipitation Depicted in Figure 8

 

Mean and Standard Deviation, in mm

Linear Trend ± its Standard Error, in % per 100 yrs

Region*

Mean

Standard Deviation

Northwestern

825

70

5.5 ± 2.9

North-central

845

65

12.1 ± 2.4

Northeastern

1265

90

11.3 ± 2.2

Southwestern

515

90

0.5 ± 6.1

South-central

1115

110

5.4 ± 3.4

All Five Regions Together

925

55

7.6 ± 1.9

* See Figure 7 for a depiction of these regions

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 7

Regions over North America (south of 55°N) with spatially homogeneous patterns of stream flow (based on principal-component analysis by Lins (1985)) extended to southern Canada.

FIGURE 8

Century-long annual precipitation changes over the United States and southern Canada, area-averaged over the five regions depicted in Figure 7.

and north-central regions. Linear trends account for up to 20 percent of the variance of annual precipitation over each of these two regions. However, the trends of 5 to 6 percent per century documented for northwestern and south-central regions are not statistically significant at the 0.05 level, and there have been no systematic changes over the southwestern region during the last 100 years. Therefore, we can conclude that during the last century a large-scale increase has occurred over all of North America, but that the major source of this increase was changes in eastern Canada and adjacent regions of the United States (the northeastern quarter of the country).

CONCLUSIONS

Standard rain gauges installed at the U.S. and Canadian meteorological networks undercatch "true" precipitation, especially in its solid form. The biases induced by this undercatch have altered during the last century, due to changes in instrumentation and measurement practices. This introduces large-scale inhomogeneities in the time series of the precipitation field over the continent. Climatological studies that do not take into account these inhomogeneities will give misleading results. The low-elevation locations of the existing meteorological network necessitate further research to determine reliable estimates of snowfall changes over the United States. Problems also exist in estimation of total precipitation in the mountainous west of the continent. The scale corrections, which transform area-averaged precipitation data into the unbiased values of "ground true" regional precipitation, increase precipitation estimates in that region by up to 70 percent. These uncertainties in the data restricted our analysis to spatially averaged values of annual precipitation and snowfall over large regions.

After adjustments and selection of the "best" network, reliable "first guess" estimates of North American snowfall and precipitation were obtained. Century-long time series of unbiased annual precipitation over the regions to the south of 55°N, and 40-year time series of unbiased area-averaged annual precipitation and snowfall for all of North America, were developed. Analysis of the trends shows that:

  1. During the last hundred years, annual precipitation has increased in southern Canada (south of 55°N) by 13 percent, and in the contiguous United States by 4 to 5 percent.

  2. An increase of up to 20 percent in annual snowfall and rainfall has occurred during the last four decades in Canada north of 55°N.

  3. In the zone 45°N to 55°N, century-scale total precipitation has increased, while a redistribution between the solid and liquid forms has occurred coherently with variation of the zonal maximum temperature (and global surface air temperature).

  4. The main domain of the century-scale precipitation increase over North America to the south of 55°N (to the north of this latitude the data for the first half of the century are absent) is eastern Canada and adjacent northern regions of the United States.

ACKNOWLEDGMENTS

This paper was completed while the first author worked as a National Research Council senior research associate

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

at the U.S. National Climatic Data Center. Support for David R. Easterling was provided under Department of Energy (DOE) Interagency Agreement DE-A105-90ER60952 and NOAA's Climate and Global Change Program.

Commentary on the Paper of Groisman and Easterling

DAVID A. ROBINSON

Rutgers University

I think Dr. Groisman's presentation has been an excellent example of the advantages and pitfalls of dealing with observational data and of putting data together on a variety of temporal and spatial scales. And I can assure you that the paper contains some tremendous information that, unfortunately, Dr. Groisman had time only to gloss over. I would like to break the study down into a couple of segments to foster further discussion.

The first section I shall simply entitle ''obtaining numbers". This includes the problems Dr. Groisman presented that relate to measurement techniques, whether for obtaining liquid rainfall or, more important, getting adequate snowfall measurements. Next, there are the problems in developing an observational network that is homogeneous through time—which, as Dr. Groisman found, was impossible to do for the entire North American continent. He could look only at the southern reaches of Canada and over the United States for the entire century. When it came to looking into the Arctic, he could go back only about 40 years.

Third is the paper's serious criticism, which may have passed people by too quickly, that it is difficult to believe anything that comes out of the primary stations in the United States. The figure that shows the changes at these stations (relocation, gauge changes, wind shields) really substantiates that. So we essentially have different observational networks that vary through time.

Last, there is the whole idea of metadata. We must have enough information on the history of a station or network that we can go back and reconstruct a homogeneous, standardized time series over decade-and century-long periods. While North American metadata is certainly better than African, there are gaps and lapses in the station histories. For precipitation, the impact of station moves is less significant than for temperature. But the gauges that measure precipitation, the wind shields, the 8-inch gauge versus the tipping bucket, and so forth are as important as any problems in measuring temperature in the United States.

It occurs to me to wonder how Dr. Groisman would rank these concerns in terms of importance for obtaining homogeneous precipitation records. To me, the most significant problem is perhaps the switch to the tipping bucket with the introduction of automated surface observing systems at primary U.S. observing stations over the next few years. Some time down the road when we are looking for decadal variations in precipitation, we may have to examine the metadata for the 1990s very carefully to come up with a homogeneous time series.

Another question I have is how deeply inhomogeneities interfere with obtaining true signals. You have to do so much with the data to be able to compare it at different times at an individual station, or among stations or countries. How deeply does this limit our ability to recognize climate change? Tom Karl et al. stated in their 1991 Science paper that it would take several decades to recognize a change in precipitation over central North America that might be an anthropogenic signal. I wonder how much these inhomogeneities in the different gauges and the different national recording procedures would extend or contribute to that lag, particularly when it comes to a noisy record like precipitation.

When it comes to point-to-area transformations, the question is where to draw the line. Dr. Groisman has done a lot of extrapolation, particularly in the western reaches of North America, in order to take into account elevational factors or orographic factors for precipitation. In some cases factors as great as a 70 percent increase were applied to the precipitation results of station networks. At some point one must decide whether to stay with point data, or at least data from a similar elevational area, or to stretch and extrapolate the data in order to obtain a more spatially complete and, in terms of absolute values, a more accurate record. Dr. Groisman did make it clear that the extrapolation does not interfere with any type of trend analysis; my question relates only to the absolute values of precipitation and snowfall generated.

Now for evaluating the numbers. Let us begin with the climatologies of snowfall and precipitation. It was interesting that at a given point in time when there was a larger mean, the standard deviations increased. When it comes to climate change, it is important to know whether we see a change in the mean or a change in the variability of climate.

Next Dr. Groisman looked at time series: the 40-year variety, the 100-year variety, precipitation, snowfall, Alaska, Canada, and the lower 48. There are so many interesting things in these time series-for instance, the major increase in snowfall in the Alaskan time series in the 1980s. Is

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 1

Smoothed time series of winter (December to February) snowfall in cm at Napoleon, North Dakota. Data from 1902 through 1987 have been smoothed with a nine-point binomial filter; the only points plotted are those where 9 years are available for filtering (i.e., plotted year ± 4 yr).

this associated with what has been happening in the North Pacific? We know that the North Pacific was rather cool in the 1980s compared to most of the rest of the globe.

I would like to show a few time series of snowfall. The stations are from a network of about 1,000 stations—about 300 of which go back to the turn of the century—over the United States, which I have assembled and quality-controlled in cooperation with NCDC. Figure 1 looks at Napoleon, North Dakota for the past century. An important point here is that these are stick measurements of snowfall and snow depth. Thus, they have no biases due to a change of gauge. We have not yet applied filters and looked for decadal variations. But you can see the suggestion of an increase in snowfall at Napoleon during the century.

Figure 2 is a record of New Brunswick, New Jersey snow going back to 1860. This is not total snowfall, but the number of 5-inch-plus and 10-inch-plus snowfall events. You can see quite a bit of variation across time. I do not see anything in the way of trends or cycles. This small study was inspired by the fact that until March 18, 1992, it was the third least snowy winter in New Brunswick in the last 135 years. Then it snowed 10 inches in 4 days, shooting the near-record to pieces.

It is interesting to note from this figure that the 1973-1974 to 1990-1991 period was the longest interval in which not a single year missed having a 5-inch-plus snow event. The winter of 1991-1992 ended that string. Another item of interest is that in the past 135 years, no year has had more than five 5-inch-plus snow events in central New Jersey.

Another way of examining this time series is to sum the number of 5-inch (12.7 cm) snow events by decade, as in Table 1. Here you can see that the 1980s had the most 5 inch-plus snowfall events of any decade on record. I have shown this chart to people throughout New Jersey, and no one ever believes me. However, the subject of one's

FIGURE 2

Number of =5-inch to <10-inch ( =12.7-cm to <25.4-cm) and =10-inch snowfall events per winter (October through April) from 1859/1860 to 1990/1991 at New Brunswick, New Jersey. (A snowfall event may be longer than one day, but may not be interrupted by a day without snowfall; no such events were observed for winter 1991/1992.)

perception of climate change is best left for discussion at another time.

Finally, Dr. Groisman talked about associations between snowfall, precipitation, and other climatic variables, and noticed that from 45°N to 55°N there was an inverse rela-

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

TABLE 1 Snowfall Events* in New Brunswick, N.J. by Decade, 1860-1989

 

Number of Snowfall Events*

Decade

=5" to <10"

10"

=5"

1860s**

18

4

22

1870s

13

5

18

1880s

13

3

16

1890s

12

4

16

1900s

15

6

21

1910s

12

6

18

1920s

14

1

15

1930s

10

2

12

1940s

10

5

15

1950s

7

2

9

1960s

11

7

18

1970s

13

2

15

1980s

19

4

23

Total

167

51

218

* Snowfall event may be longer than I day, but may not be interrupted by a day without snowfall.

** Decade of the 1860s = winters 1859-60—1868-69.

tionship between temperature and snowfall. When examining Northern Hemisphere snow and temperature together, as shown in Figure 3, we have found that both are 12-month running means. The snow, which is from the NOAA

FIGURE 3

Twelve-month running means of snow cover, in millions of km2 and surface air temperature, in degrees Celsius, over Northern Hemisphere lands for January 1973 through August 1990. Temperatures are expressed as departures from the reference period, 1951-1970. Temperature data from P.D. Jones. From Robinson et al. (1991).

hemispheric snow charts, is a weekly product that we have averaged into monthly values. The temperature is Phil Jones's data set. Notice the striking relationship between hemispheric temperature—and that is from pole to equator—and Northern Hemisphere snow cover. We are beginning to examine snow and temperature by region and by season. So we see the same general association at this point, at least in a hand-waving sense, as Dr. Groisman found with the snowfall.

Discussion

RIND: You mentioned the relationship between precipitation over North America and global mean temperature. I wanted to add that the warming patterns for the 1930s and the 1980s were very different, just as you said the precipitation patterns were. The 1930s had very large warming at high latitudes with little effect at low latitudes, like an ocean heat-transport change pattern. The 1980s had very uniform warming as a function of latitude, more like what the models are projecting for CO2 and trace-gas forcing. They may be good examples of two different types of processes.

WALSH: That increase of snowfall in northern Canada is interesting in relation to the borehole measurements mentioned yesterday. If snowfall increases, it tends to insulate the ground during the coldest part of the year, which can complicate interpretation of borehole measurements. Also, perhaps we should take a closer look at how subsurface temperatures relate to snowfall to see what the implications might be of a continued increase in snowfall in permafrost areas.

TRENBERTH: The summer and winter correlations between precipitation and temperature tend to be very different.

GROISMAN: That's true. We simply wanted to create annual precipitation time series we could relate to some of the early data.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Asymmetric Trends of Daily Maximum and Minimum Temperature: Empirical Evidence and Possible Causes

THOMAS R. KARL1, PHILIP D. JONES2, RICHARD W. KNIGHT1, GEORGE KUKLA3, NEIL PLUMMER4, VYACHESLAV RAZUVAYEV5, KEVIN P. GALLO1, JANETTE A. LINDESAY6, AND ROBERT J. CHARLSON7

ABSTRACT

An examination of monthly maximum and minimum temperatures indicates that the rise of minimum temperature (0.84°C) has occurred at a rate three times that of rise of maximum temperature (0.28°C) over the period 1951 to 1990. This conclusion is based on data from over 50 percent (10 percent) of the Northern (Southern) Hemisphere land mass, accounting for 37 percent of the global land mass. The decrease of the diurnal temperature range is approximately equal to the increase of mean temperature. The asymmetry is detectable in all seasons in most of the regions studied.

The decrease in the daily temperature range is at least partially related to an observed increase in cloudiness. However, an empirical analysis of nearly 50,000 daily observations across 14 stations throughout the United States indicates that the daily temperature range is also sensitive to relative humidity, global radiation, surface stability, thermal advection, and snow cover; earlier work has shown that the temperature range is also sensitive to surface boundary conditions such as soil moisture. Locally, the diurnal range can also respond to increasing urbanization, irrigation, and desertification, but this effect appears to be small in our analysis.

Both general circulation models (GCMs) and radiative-convective models (RCMs) indicate that a decrease in the daily temperature range is likely with increases of CO2. The magnitude

1  

NOAA National Climatic Data Center, Asheville, North Carolina

2  

Climatic Research Unit, School of Environmental Sciences, University of East Anglia, Norwich, Norfolk, United Kingdom

3  

Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York

4  

Bureau of Meteorology, Melbourne, Australia

5  

Research Institute of Hydrometeorological Information, Obninsk, Russia

6  

Climate Research Group, University of Witswatersrand, South Africa

7  

Department of Atmospheric Sciences and Institute for Environmental Studies. University of Washington, Seattle, Washington

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

of the range decrease simulated by GCMs, however, is only 10 percent of the magnitude of the simulated mean average temperature increase, whereas the observational record shows nearly equal changes of the daily temperature range and the mean temperature. The clear-sky daytime forcing by sulfate and carbonaceous aerosols, and other indirect effects such as increased cloudiness and albedo, operate in a sense that may reduce the daily temperature range; however, the decadal changes of this geographic and seasonally varying forcing are not well known. At present we do not have an adequate explanation for the observed decrease in the temperature range, or for the lack of daytime warming, yet these changes are clearly important, both scientifically and practically.

BACKGROUND

The mean monthly maximum and minimum temperature is derived from an average of the daily maximum and minimum temperatures. The monthly mean diurnal temperature range (DTR) is defined as the difference between the mean monthly maximum and minimum temperatures. The dearth of appropriate data bases that include information on the daily or mean monthly maximum and minimum temperature has previously impeded our ability to investigate changes in these quantities. The problem has historical roots. It arises because the climatological data that are made available internationally by national meteorological and/or climate data centers are usually derived from the monthly climate summaries (CLIMAT messages) on the Global Telecommunications System (GTS), which do not include information on the maximum or minimum temperatures. The GTS is the principal means by which near-real-time, in situ, global-climate data are exchanged. Moreover, the problem has been exacerbated because the World Meteorological Organization's retrospective data collection projects such as World Weather Records and Monthly Climatic Data of the World have always been limited to mean monthly temperatures. This has forced climatologists interested in maximum and minimum temperatures either to develop historical data bases on a country-by-country basis (Karl et al., 1991 a) or to try to work with the hourly GTS synoptic observations. The former is a slow, painstaking process, and the latter has been limited by poor data quality and metadata (information about the data) and records of short duration (Shea et al., 1992).

The first indication that there might be important large-scale characteristics related to changes of the mean daily maximum and minimum temperatures was reported by Karl et al. (1984a). Their analysis indicated that the DTR was decreasing at a statistically significant rate at many rural stations across North America. Because of data accessibility problems, subsequent empirical analyses continued to focus on data from North America over the next several years (Karl and Quinn, 1986; Plantico et al., 1990). By 1990, however, a United States/People's Republic of China (PRC) bilateral agreement organized by the U.S. Department of Energy and the PRC Academy of Sciences provided the opportunity to analyze maximum and minimum temperatures from the PRC. Also about this time, the Intergovernmental Panel on Climate Change (IPCC) made arrangements with the Australian National Climate Centre to analyze maximum and minimum temperature data from southeastern Australia. The IPCC (1990) reported a significant decrease in the DTR from both of these regions. Meanwhile, work from another data exchange agreement, a bilateral between the United States and the former Soviet Union, came to fruition as a data set of mostly rural maximum and minimum temperatures was developed for the former Soviet Union. Karl et al. (1991a) reported on the widespread decrease of the DTR over the former Soviet Union, China (not including Tibet), and the contiguous United States, as was highlighted by the IPCC (1992).

Additional data from other countries and updates to previous analyses have now been analyzed. Data from the eastern half of Australia, Canada, Alaska, Sudan, Japan, some Pacific island stations, South Africa, and a few other long-term stations in Europe have now been included. Figure 1 shows the area of the globe that has now been analyzed for differential changes of the maximum and minimum temperature. This area now covers over 50 percent of the Northern and 10 percent of the Southern Hemisphere land masses, but still only about 37 percent of the global land mass.

FIGURE 1

Areas of the globe (indicated by shading) that have been analyzed in this paper for changes of mean maximum and minimum temperature.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
OBSERVED CHANGES OF MEAN MAXIMUM, MINIMUM, AND DIURNAL RANGE
Spatial and Seasonal Patterns of Contemporary Trends

Daily maximum and minimum temperatures from over 2,000 stations were available for analysis in the countries shaded in Figure 1 during the period 1951 to 1990 (except for Sudan and the former Soviet Union, which had data through 1987 and 1989, respectively). Selected subsets of these data were averaged within various regions of each country. The definition of each region represents a compromise between climatic homogeneity and ensuring an adequate number of stations within its boundaries to limit sampling error. The base period for calculating departures from the average included the years 1951 to 1990 (or slightly fewer years in some countries, as mentioned before). The regions are delineated in Figure 2 where, as in other large-scale studies of the change of the mean annual temperature (Jones et al., 1986a,b; Jones, 1988), the average of the trends of the mean annual maxima and minima reveals a general rise of temperature. A decrease of the minimum temperature within any region is uncommon; it is somewhat more frequent for the maximum temperature, as seen over the United States and China. The differential rate of warming between the maximum and minimum temperatures is apparent, with only a few regions reflecting an increase of the DTR. These weak exceptions occur in central Canada and south-easternmost Australia.

There are some seasonal variations of the rates of decreasing DTR, but they vary from country to country (Table 1). In Japan the decrease is not evident during summer, and it is not as strong during this time over China. In the United States the decrease is weak during the spring but quite strong during the autumn. Alaska and Canada have strong decreases during the spring, with somewhat lesser decreases during the autumn. Over the former Soviet Union the decrease in the DTR is significant throughout the year, but somewhat weaker during the winter. Over Sudan, the rate of the DTR decrease is strong in all seasons except during the summer rainy season, where rains have been very sparse over the past few decades. Over South Africa, decreases in the DTR are greatest in the Southern Hemisphere spring, and DTR actually increases during autumn. In the eastern half of Australia the decrease of the DTR is apparent throughout the year, but is weakest during the Southern Hemisphere summer.

When considered collectively, 60 percent of the trends in Table 1 reflect statistically significant decreases of the DTR. A test for a change point in the trend (Solow, 1987) indicates that for most seasons and areas there is insufficient evidence to suggest a statistically significant change point in the rate of the decrease. The Table 1 temperature trends can be area-weighted to reflect the overall rate of DTR decrease. Table 2 shows the decrease both north and south of the equator, but without any pronounced seasonal cycle in the Northern Hemisphere. The area of available data in the Southern Hemisphere is too small to permit any general statements about trends in that portion of the globe, but the decrease in the Northern Hemisphere is quite apparent. The rate of the decrease in the DTR (-1.4°C/100 years) is comparable to the increase of the mean temperature (1.3°C/ 100 years).

For all areas combined (Figure 3), a noticeable difference between the rates of warming of the minimum and the maximum temperature began in the 1960s. The minimum temperature has continued to warm relative to the maximum through the 1980s. The time series ends in 1989, the year after the major North American drought, because data from the former Soviet Union were not available past 1989. The

TABLE 2 Trends of Temperature (°C/100 years) for Annual and Three-Month Mean Maximum Temperature, Minimum Temperature, and Diurnal Temperature Range for the Areas Denoted in Figure 1 (less Pakistan, northern Finland, and Denmark)

Northern Hemisphere 1951-1990 (50% of land area)

Seasons

Max

Min

DTR

D-J-F

1.3

2.9

-1.5

M-A-M

2.0

3.2

-1.3

J-J-A

-0.3

0.8

-1.1

S-O-N

-0.4

1.3

-1.7

Annual

0.5

2.0

-1.4

Southern Hemisphere 1951-1990 (10% of land area)

Seasons

Max

Min

DTR

D-J-F

1.6

22

-0.6

M-A-M

1.7

25

-0.8

J-J-A

1.0

1 3

-0.4

S-O-N

0.8

2 1

-1.3

Annual

1.3

2.0

- 0.8

Globe 1951-1990 (37% of land area)

Seasons

Max

Min

DTR

D-J-F

1.3

2.9

-1.6

M-A-M

1.9

3.1

-1.2

J-J-A

-0.2

0.8

-1.1

S-O-N

-0.3

1.4

-1.7

Annual

0.7

2.1

-1.4

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 2

Spatial patterns of annual trends of mean maximum temperature, minimum temperature, and diurnal temperature range (mostly 1951-1990) in °C per 100 years. Diameter of circles is proportional to the trend; solid (open) circles represent negative (positive) trends. Circles pertain to regions within each country except for the island stations, e.g., the South Pacific and Hawaii.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

TABLE 1 Trends of temperature (°C/100 years) for Annual and Three-Month Mean Maximum Temperature, Minimum Temperature, and Diurnal Temperature Range, Based on a Weighted Average of the Region (by country) in Figure 1.

Alaska 1951-1990

(39 stations)

Japan 1951-1990

(66 stations)

Seasons

Max

Min

DTR

Seasons

Max

Min

DTR

D-J-F

6.0

8.8

-2.8

D-J-F

-0.5

-0.2

-0.2

M-A-M

3.1

6.3

-3.2

M-A-M

-0.4

-0.7

0.3

J-J-A

0.9

2.4*

-1.5

J-J-A

0.5

0.0

0.4

S-O-N

-1.4

0.4

-1.9*

S-O-N

-0.3

-0.5

0.2

Annual

2.1

4.5

-2.4

Annual

-0.2*

-0.4

0.2

Canada 1951-1990

(227 stations)

People's Republic of China 1951-1988

(44 stations)

Seasons

Max

Min

DTR

Seasons

Max

Min

DTR

D-J-F

1.8

2.1

-0.2

D-J-F

0.5

3.5

-3.0

M-A-M

3.7

3.8

-0.1

M-A-M

-0.8

1.4

-2.2

J-J-A

0.5

1.4

-0.9

J-J-A

-1.8

-0.8*

-1.0

S-O-N

-2.2

-1.2

-1.0

S-O-N

-0.6

1.0

-1.6

Annual

0.9

1.5

-0.6*

Annual

-0.7*

1.3

-2.0*

United States (contiguous) 1951-1990

(494 stations)

South Africa 1951-1991

(12 stations)

Seasons

Max

Min

DTR

Seasons

Max

Min

DTR

D-J-F

-2.3

-0.7

-1.5

D-J-F

0.8

2.0

-1.2

M-A-M

2.3

2.5

-0.2

M-A-M

2.2

1.7

0.5

J-J-A

-0.3*

1.0*

-1.4

J-J-A

1.3

1.3

0.0

S-O-N

-1.7

1.3

-3.0

S-O-N

-0.7

1.8

-2.4

Annual

-0.6*

1.0

-1.5

Annual

0.9

1.7*

-0.8*

Sudan 1951-1987

(15 stations)

Eastern Australia 1951-1991

(44 stations)

Seasons

Max

Min

DTR

Seasons

Max

Min

DTR

D-J-F

-1.2

2.7

-3.9*

D-J-F

1.8

2.3

-0.4

M-A-M

0.4

3.3

-2.8

M-A-M

1.6

2.8

-1.2

J-J-A

2.8

2.1

0.7

J-J-A

0.8

1.4

-0.5

S-O-N

1.4

2.5

-1.1

S-O-N

1.3

2.2

-0.9

Annual

0.9

2.7

-1.7

Annual

1.4

2.2

-0.7

USSR (former) 1951-1990

(165 stations)

 

 

Seasons

Max

Min

DTR

 

 

 

 

D-J-F

2.8

4.2

-1.3

 

 

 

 

M-A-M

2.5

3.8

-1.2

 

 

 

 

J-J-A

-0.4*

0.9*

-1.3

 

 

 

 

S-O-N

0.6

2.2

-1.6

 

 

 

 

Annual

1.4

2.8

-1.4

 

 

Key: Trends significant at the .01 level (two-tailed t-test) are bold and italicized

Trends significant at the .05 level are italicized

Trends with significant change points are denoted with an asterisk

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 3

Time series of the temperature anomalies of the annual mean maximum temperature, minimum temperature, and diurnal temperature range for 37 percent of the global land mass (areas shaded in Figure 1). Smooth curve was obtained with a nine-point binomial filter with "padded" ends.

end of the time series is significantly affected by the drought. Furthermore, variance of the time series is significantly affected when such large regions drop out of the analysis, which is why the series ends prematurely. Nonetheless, Figure 3 reflects a gradual decrease of the DTR through much of the past several decades.

Longer-Term Variations

Unfortunately, the availability of maximum and minimum temperature data covering the globe is currently limited prior to 1951. In the United States a network of approximately 500 high-quality stations has remained intact since the turn of the century, and in the former Soviet Union a fixed network of 224 stations (165 if only rural stations are used) is available back to the 1930s. The time series from these two countries shown in Figure 4 reflect significant decadal variations in the DTR, which are particularly evident during the dry 1930s and the early 1950s in the United States. The general decrease of the DTR did not begin in the United States until the late 1950s, and the DTR decreased rather dramatically in the mid-to-late 1970s over the former Soviet Union as part of substantial increases in the minimum temperature. The decrease of the DTR in these two countries is a phenomenon of recent decades. Data are also available further back in time for smaller areas and countries, notably Japan, eastern Australia, and South Africa. Figure 5 indicates that the decrease in the DTR in eastern Australia has occurred rather gradually since the decline in the late 1940s. In South Africa the decrease is predominantly due to the sharp decline in the early 1950s.

A very long record of maximum and minimum temperatures was available from the Klementinum Observatory in Prague, Czech Republic, and another from a benchmark station from northern Finland. Figure 6 portrays a remarkable increase of the DTR at the Klementinum Observatory from the early twentieth to the mid-twentieth century, with a substantial decrease since about 1950. The increase coincides with the increase of global mean temperature since the turn of the century, and the decrease occurs when the

FIGURE 5

Time series of the variations of the diurnal temperature range for Japan, eastern Australia, and South Africa. Smooth curve was obtained with a nine-point binomial filter.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 4

Time series of the variations of the annual mean maximum, minimum, and diurnal temperature range for the contiguous United States and the former Soviet Union. Smooth curve was obtained with a nine-point binomial filter. Trends since 1951 are depicted by the dashed line.

mean temperature reflects little overall change. In the first half of the nineteenth century the DTR averages about 0.5°C lower than in the latter part of the century. The DTR at Sodankylä, Finland, also displays a gradual decrease since 1950, but contrary to the Klementinum Observatory's pattern, the decrease is evident back to the turn of the century. The mean temperature at Sodankylä reflects little or no change. The high-frequency variability of the DTR at both stations is less than that of their respective mean temperatures, but the converse is true for low-frequency variations.

If the data from the Klementinum Observatory truly reflect the regional change of the DTR in central Europe, then the recent decrease of the DTR in this area is less persistent and less substantial than the increase prior to 1950. In light of the variations at the Klementinum Observatory, what makes the results from Figure 2 so remarkable

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 6

Time series of the variations of the annual mean temperature and the annual diurnal temperature range from two long-term stations: the Prague Klementinum-Observatory, Czechoslovakia, and Sodankylä, Finland, which has been designated as a climate reference station. Smooth curve was obtained with a nine-point binomial filter.

is the fact that so many areas share an overall decrease of the DTR.

Recently, other investigators have also compiled information on the change of the DTR over other portions of the globe. Frich (1992) provides evidence to indicate that there has been a general decrease over Denmark since about 1950, based on an analysis of several long-term stations, most located in rural areas. Bücher and Dessens (1991) analyzed a long record of maximum and minimum temperatures from the Pic du Midi de Bigorre Observatory in the Pyrenees at a height of over 2,800 m. Their analysis revealed a significant decrease of the DTR since the late nineteenth century, but an inhomogeneity in the record prevented a continuation of the analysis beyond 1970. Kruss et al. (1992) reported on changes of maximum and minimum temperature between the two 30-year periods 1931 to 1960 and 1961 to 1990 over Pakistan. Despite considerable missing data, they managed to obtain at least 20 years of data from each of the two periods for 35 stations across Pakistan. Their analysis revealed a mix of decreasing and increasing changes of the DTR. In our analysis of Pakistani data we could only manage to identify five stations with adequate data to analyze year-to-year changes, and these were all located in the northern half of the country. These stations also depicted a decrease of the DTR.

Relation to Variations of the Seasonal and Annual Extremes

For a variety of practical considerations, it is important to know whether the decrease in the mean DTR translates to a decrease in the extreme temperature range. Karl et al. (1991a) provide evidence to suggest that, indeed, over the United States and the former Soviet Union (the only areas for which they had access to daily data) there was often a significant and substantial decrease in the seasonal and annual temperature extremes similar to the decrease in the seasonal and annual mean DTR. This similarity is also reflected in the time series of monthly extreme maximum and minimum temperature over Sudan (Jones, 1992).

Data Quality

The reliability of the data used to calculate the changes of the DTR is a critical question. The data presented and discussed here have been subjected to various degrees of

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

quality assurance. The degree to which precautionary measures have been taken to minimize data inhomogeneities varies considerably from country to country. In the United States, a fixed network of stations in the Historical Climatology Network is used (Karl et al., 1990), which largely consists of rural stations that have been adjusted when necessary for random station relocations, changes in instrument heights, systematic changes in observing times (Karl and Quinn, 1986; Karl et al., 1986), the systematic change in instruments during the mid—and late 1980s (Quayle et al., 1991), and increases in urbanization (Karl et al., 1988). The potential warm bias of the maximum introduced by the H083 series of thermometer (Gall et al., 1992) is not a factor in this network, since the H083 instrument was not used in the rural cooperative network.

In the former Soviet Union, the fixed network of 165 stations consists of rural stations (1990 populations less than 10,000 and local surroundings free of urban development). The data from the former Soviet Union have not been adjusted for any random or systematic inhomogeneities. Station histories, however, indicate that there have not been systematic changes in that network's operation over the course of the past 50 years.

In Canada, the results reported here are derived from a set of 227 rural stations (those in areas with population less than 10,000). These data were selected from a network of 373 principal stations, but a large number of urban areas and stations that relocated to airports were eliminated from the analysis.

In Alaska, a network of 39 stations was used. It included most stations that had been operating in that state since the early 1950s, with the exception of the stations in the major cities of Juneau, Fairbanks, and Anchorage. Once again no attempt was made to adjust for station relocations, and the stations consist of a mix of instrument types with some changes at specific sites.

Station histories from China do not reflect any systematic changes in instrumentation, instrument heights, instrument shelters, or observing procedures relative to the maximum and minimum temperature. Our analysis is based on a subset of the more than 150 stations available to us. No attempt was made to correct for random station relocations in the fixed network of 44 stations we finally selected from the larger network. The potential impact of urbanization precluded the use of many stations; we eliminated all stations that were in or near cities with populations over 160,000.

All stations in Australia have recently undergone a thorough homogeneity analysis (Plummer et al., 1995), but the results were not available for this analysis. Instead, stations were selected on the bases of length of record and distance from major areas of urbanization. All of the Australian stations used are in small towns or rural areas, many in post office ''back yards."

Fewer than half of the 154 stations available from Japan were used in this analysis. As with China, many stations were eliminated because of their proximity to major urban areas. An inspection of the station histories reveals a number of network "improvements" related to the automation of the temperature measurements in recent years. A full assessment of the homogeneity of the data awaits a detailed analysis. The station networks from Sudan and South Africa include some stations from urban areas, but countrywide decreases of the DTR are not overwhelmed by these stations. Incomplete information was available regarding systematic changes in instrumentation at these locations over the past several decades, but the data were inspected for station relocations and adjustments made when necessary on the basis of temperature differences from neighboring stations. In total, four stations in South Africa were adjusted using the procedures outlined by Jones et al. (1986a).

INFLUENCES ON DIURNAL TEMPERATURE RANGE
Local Effects

As more data become available from a variety of countries, it becomes difficult to dismiss the general decrease of the DTR over the past several decades as an artifact of data inhomogeneities. Observing networks are managed differently in each country. If local effects are significantly influencing the DTR, then at least three possible sources of change need to be explored: urbanization, irrigation, and desertification. Evidence to support or refute the impact of these human-induced local and regional effects are discussed in the subsections below.

Urban Heat Islands

It is well known that the urban heat island often tends to manifest itself most strongly during the nighttime hours (Landsberg, 1981). In mid-latitude North American cities the urban-rural temperature difference usually peaks shortly after sunset, then slowly decreases until shortly after sunrise, when it rapidly decreases and, for some cities, actually vanishes by midday. In many cities increases in urbanization would differentially warm the minimum relative to the maximum temperature. A number of precautions have been taken to minimize this effect.

In the contiguous United States, the corrections for urban development recommended by Karl et al. (1988) have been applied to the data, and any residual heat-island effect in this analysis should not be an issue. In Canada, only stations with population less than 10,000 were used in the analysis, and the average population of the cities in the proximity of the observing stations was slightly over 1,000. If the Canadian stations behave similarly to stations in the United States, the effect on the diurnal temperature range may be

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

exaggerated by about 0.1°C. An effect of similar magnitude may exist for Alaska. In the former Soviet Union, the population limit for inclusion of any station into the network was 10,000; in addition, however, no station could be within 1 km of any multi-story urban development. If the impact of the effect of urbanization on the DTR in the former Soviet Union is anything like that in the United States, the residual urban heat island effect on the DTR should be at least an order of magnitude smaller than the observed decrease of the DTR (nearly 1.5°C/100 yr).

In China and Japan, a number of tests were conducted to identify the impact of urbanization on the DTR. Three networks of stations were categorized on the basis of population. For China the categories included stations in proximity to cities with populations over 1 million, under 160,000, and in between these two threshold values. Categories in Japan were based on threshold values of 500,000 and 50,000. In the three population categories, proceeding from high to low, China had 23, 42, and 44 stations while Japan had 17, 71, and 66 stations. Figure 7 shows that the decrease of the DTR actually is more marked in China for stations in the lowest population category than for stations in the medium category, while the trend of the average temperature continues to decrease. This unexpected finding suggests that urban-

FIGURE 7

Temperature trends calculated for the People's Republic of China and Japan. Abbreviations: Wi, winter; Sp, spring; Su, summer; Au, autumn; MAX, maximum; MIN, minimum; AVG, average; and DTR, diurnal temperature range. The three bars in each of the seasonal and annual-average groupings reflect stations near cities in the high, medium, and low categories (left to right) as defined in the text.

ization effects in China are dissimilar to those in the United States, since they seem not to have affected the maximum and minimum temperature trends in Chinese cities of 500,000 or less. In Japan, however, the impact of urbanization on the DTR is evident even in the lowest population category (less than 50,000), and is even more apparent for the average temperature. On the basis of these analyses, it seems that the impacts of urbanization in China (Wang et al., 1990) are unlikely to significantly affect the trends reported in Tables 1 and 2.

A previous paper by Jones et al. (1990) investigated the impact of increasing urbanization in the land data base used by the IPCC (1990, 1992) to calculate changes of global temperature. The conclusion from the work of Jones et al. (1990) was that any residual urban bias in the land-based average temperature records was about 0.05°C during the twentieth century. A comparison of the average temperature trends derived from the stations used in Tables 1 and 2 with the stations used by Jones (1988) revealed differences in trends from country to country, but virtually identical trends of temperature (within 0.02°C/100 yr) were found when all areas depicted in Figure 1 were considered. This similarity suggests that the degrees of urban-induced bias in these two data sets are of comparable magnitude over the past 40 years, despite the use of substantially different station networks.

Irrigation

It can be argued that increases in irrigation may account for the decrease in the DTR. The evaporation associated with soil moisture would convert sensible to latent heat and thus significantly reduce daytime temperature. In order to test this hypothesis, the correlation coefficient (both the Pearson product-moment and the Spearman rank) was calculated using the values of the trends of the DTR and the change in land area under irrigation from 1950 to 1987 (U.S. Department of Commerce, 1950, 1988) for each of the regions in the United States delineated in Figure 2. No relationship was found between the change in the DTR and the increase of irrigated lands, and in fact many of the largest decreases of the DTR were associated with areas with the smallest increases of irrigation. Considering that the United States has had significant decreases of the DTR over the past several decades, and has had relatively large increases of irrigation relative to other countries over the past 40 years, it seems unlikely that increases of irrigation can be regarded as a serious explanation for the decreases of the DTR.

Desertification

The converse of the theoretical effects of irrigation would result from increased desertification. This might arise from

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

poor land practices such as over-grazing or deforestation. Given the mid- and high-latitude bias of the results reported here, it seems unlikely that desertification would have a significant impact on the results reported in Tables 1 and 2. Moreover, this effect would tend to make the reported decreases underestimates, especially during the warm season, as desertification would increase the maximum and decrease the minimum.

Climatic Effects

Since it seems unlikely that any of the human-induced local effects can provide a satisfactory answer to the widespread decrease of the DTR, a number of climatic variables that differentially affect the maximum and minimum temperature were analyzed to discern which of them most strongly affect the DTR. Over 50,000 days of climatic observations were selected from the stations listed in Table 3 during periods of consistent measurement procedures for the variables defined in Table 4.

The selection of variables to be studied was based on a priori information. For example, increases of the DTR over land have previously been related to snow cover ablation, in simulations by the U.K. Meteorological Office's general circulation model (GCM) with doubled CO2 (Cao et al., 1992). It is well known that the ability of the surface boundary layer to absorb, radiate, transform, and mix sensible heat differentially affects the maximum and minimum temperature. Relative humidity and cloudiness are two important climate variables that influence these surface-layer properties. In this analysis cloud-related information was contained in two climatic variables, the sky cover (in tenths) and an index of the ceiling height. The ceiling height (CIG)

TABLE 3 Stations and Years Used in Identifying the Sensitivity of the Diurnal Temperature Range to Various Climatic Variables

Stations

Years

Sacramento, CA

1961-69

Tallahassee, FL

1962-70

Indianapolis, IN

1966-74

Worcester, MA

1961-69

Bismarck, ND

1973-81

Scotts Bluff, NE

1971-79

Reno, NV

1961-69

Oklahoma City, OK

1975-83

Pittsburgh, PA

1961-69

Columbia, SC

1971-79

San Antonio, TX

1973-81

Seattle/Takoma, WA

1971-79

Spokane, WA

1966-75

Green Bay, WI

1971-79

TABLE 4 Definitions and Abbreviations of the Climatic Variables Used to Test the Sensitivity of the Diurnal Temperature Range

Variables

Abbreviations

Diurnal temperature range (daily max-daily min)

DTR

Snow cover (binary, if snow depth ≥2.54 cm)

SNOW

Mean relative humidity (0600 LST* & 1500 LST)

RH

Mean wind speed (0600 LST & 1500 LST)

WS

Mean sky cover (0600 LST & 1500 LST)

SKY

Mean ceiling (0600 LST & 1500 LST)

CIG

Total daily top-of-the-atmosphere solar radiation

TRAD

Day-to-day temperature differences

ΔTMP

(|TMP0 − TMP−1 | + |TMP0 − TMP+1|)

 

* local standard time

was broken down into seven categories. The cloud ceiling is defined as height above ground of the lowest cloud layer that covers 50 percent or more of the sky. The wind speed is an effective measure of the degree of mixing within the surface boundary layer, since it affects and interacts with the frequency and/or intensity of inversions and superadiabatic lapse rates.

In addition, the DTR is affected by the seasonal and latitudinal changes of incoming solar radiation as well as by the magnitude of day-to-day temperature differences. TRAD (top-of-the-atmosphere solar radiation) can also be regarded as a surrogate for the temperature, especially when it is used in conjunction with the other variables listed in Table 4. Karl et al. (1986) demonstrate the impact of the interdiurnal temperature difference on the maximum and minimum temperature. Large interdiurnal temperature differences lead to a large temperature range, even in the absence of a diurnal temperature cycle. These day-to-day changes of temperature are largely controlled by the thermal advection associated with synoptic-scale cyclones and anticyclones.

All of the variables in Table 4 were used in a multiple-regression analysis. Variables were regressed against the square root of the DTR, as opposed to the actual DTR, because the DTR is bounded by zero. Without the transformation, non-normal residuals result in multiple linear-regression analyses, making it more difficult to interpret the results. Figure 8a indicates that the partial-correlation coefficients of each variable with respect to the DTR are often significantly different from the simple linear correlation coefficients, making it difficult to speculate on the effect of changes in any one variable without knowing the changes in (or assuming constancy of) the other variables. Given the huge sample size, very low correlations have high statistical significance (even considering the day-today persistence of the DTR). On a local basis, a generalized

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 8

Relationship between various climate variables and the diurnal temperature range. (a) Simple and partial correlation coefficients (removing the effects of all other variables) between each variable (defined in Table 4) and the diurnal temperature range. (b) Partial correlation coefficients of each variable for cases partitioned by the total daily solar radiation at the top of the atmosphere: 170 W m−2 (low), ≥271 W m−2 (high), and remainder (medium).

multiple linear-regression model (one model for all stations and all days) based on the seven climatic variables in Figure 8a explains about 55 percent (53 percent without the square-root transformation) of the daily variance of the DTR. Although the explained variance is substantial, it is apparent that other factors may also need to be considered in order to explain the variations of the DTR (e.g., better representations of the atmospheric stability, external forcing factors, more precise techniques for calculating the mean quantities used in the analysis) and their relationships are not adequately expressed by a linear equation.

On a variable-by-variable basis, the signs of all the partial correlations make qualitative physical sense. It is interesting to note that the partial correlation of ΔTMP with DTR is greater than that of the simple correlation coefficient (Figure 8a), because the correlation between TRAD and ΔTMP mask the importance of ΔTMP in influencing DTR. The decrease of the partial correlation relative to the simple correlation for the variables RH, CIG, and SKY is to be expected, because changes among these variables are all related to each other. The decrease in the partial correlation of SNOW is particularly noteworthy, especially since Cao et al. (1992) attribute the ablation of snow cover in their model to an increase in the DTR. The empirical results in Figure 8a suggest that SNOW is only weakly related to DTR (7.5 percent of the cases, nearly 4,000, had snow cover), especially by comparison with the other variables.

In order to investigate the linearity, or lack thereof, of the relationships implicit in Figure 8a, the data were partitioned by TRAD. Figure 8b provides strong evidence to suggest that the relationships change with the amount of TRAD. In particular, the partial correlation coefficients of RH, WS, and SKY become stronger as TRAD (and thus temperature) increases. This probably has more to do with a reduction of the maximum temperature than an increase of the minimum. During daylight hours high values of the RH, WS, and SKY are indicative of higher albedos, higher potential evapotranspiration, higher atmospheric water-vapor absorption of incoming radiation, and larger-than-normal mechanical mixing. These factors act to reduce the maximum temperature that would otherwise result from high-intensity TRAD, which would be manifested as sensible heat within the surface boundary layer. The non-linearity of RH as TRAD increases is substantially greater than that of WS (Figure 8b). As the TRAD reaching the surface increases, the temperature increases, and comparatively more of the TRAD can be used for evaporation than for raising surface temperature, as would be anticipated by integrating the Clausius-Clapeyron equation to obtain saturation vapor pressure as a function of temperature.

The reduction of the partial correlation of TRAD with the DTR when the sample is partitioned with respect to TRAD (Figure 8b) relates to the balance between long nights and short days. During the late autumn and early winter in the northern half of the United States (areas which include the lowest partition of TRAD), a moderate increase in the TRAD (by inter-seasonal and latitudinal variations) generally results in a higher DTR. Contrarily, in the warm half of the year, TRAD is more than ample, so the relation between TRAD and DTR is near zero. In fact, a further selection for very high values of TRAD (short nights) leads to small negative partial correlations between TRAD and the DTR.

The reduction of the partial correlation of ΔTMP as TRAD increases is related to the decrease in intensity of the day-to-day changes of temperature during the warm season. Rossby waves and extra-tropical cyclones have lower amplitude, speed, and intensity during the warm season.

A change in sign of the partial correlation coefficient of

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

SNOW with the DTR (Figure 8b) suggests that the length of night relative to the TRAD is a significant factor in the impact of changes of snow cover on the DTR. In the northern United States, around the winter solstice, TRAD is relatively low. Snow on the ground at this time of year is important because of its excellent insulating properties (it reduces heat flow from the soil), which help lower the nighttime minimum. During the daytime the TRAD is already low, so the amount of solar radiation reflected by the snow cover is no longer so important. As a result, snow cover at this time of year at these latitudes leads to an increase in the DTR. This is not the case as the season progresses or the latitude decreases, as is reflected by the negative partial correlations (lower values of DTR with snow cover) associated with the highest category of TRAD, where there were still nearly 1,000 cases of snow cover. The data suggest that snow-cover ablation will not necessarily lead to an increase of the DTR.

From the above analysis it is apparent that there are many factors, often intricately related, that affect the DTR. Many of these variables are very much related to a greenhouse effect, not necessarily one anthropogenically induced. Overall, however, the two variables related to changes in cloudiness, sky cover, and ceiling height explain the greatest portion of the variance of the DTR. Changes in cloudiness should be one of the first considerations in searching for an explanation of the observed decrease of the DTR (Figure 2). Indeed, when large continental scales are considered, the relationship between cloud amount (or sky cover) with the DTR is quite impressive (Figure 9). Plantico et al. (1990) have already demonstrated that the decrease in the DTR

FIGURE 9

Seasonal relationships between U.S. area-average cloud cover and the diurnal temperature range.

over the United States is strongly linked to an observed increase in daytime and nighttime cloud cover and to a lowering of cloud ceilings.

Is there a general increase in cloud cover over much of the globe? Empirical evidence by Henderson-Sellers (1986, 1989, 1992) and Jones and Henderson-Sellers (1992) suggests that this may be the case over Canada, the United States, Europe, the Indian subcontinent, and Australia. Analyses of cloud cover changes over China from a network of 58 stations are inconclusive, but there is evidence for a decrease in the sunshine (a 2 to 3 percent decrease in sunshine from the 1950s to the 1980s). The quality of the cloud data (and perhaps the sunshine data) is questionable because the correlation between monthly anomalies of sunshine and cloudiness at many sites is not high. Analyses of changes in cloudiness over the former Soviet Union by Kaiser and Razuvayev (1995) reveal a general increase of cloud cover (about 5 percent) over the period 1936 to 1986, with cirrus increasing in frequency by nearly 20 percent. They also found considerable interannual and interstation variability, so the quality of these data could also be called into question. Nonetheless, the trend over the former Soviet Union is consistent with a decrease in the DTR. Frich (1992) and Bücher and Dessens (1991) also found that decreases in the DTR in Denmark and at the Pic du Midi de Bigorre Observatory occurred concurrently with an increase in cloudiness. An analysis of changes in cloud cover over Japan, using many of the same stations selected for the analysis of the maximum and minimum temperatures, indicates that cloud cover may have increased on an annual basis by nearly 1 percent since 1951, but there is no apparent response in the DTR. Changes in sunshine in Japan are not altogether consistent with the increase in cloud cover, perhaps because a new instrument was introduced into the network in 1986.

LARGE-SCALE ANTHROPOGENIC EFFECTS
Greenhouse Gases

Interest in the possible change of the DTR with increasing anthropogenic greenhouse gases has prompted several modeling groups to publish information from their models regarding the change in projected DTR when CO2 is doubled. Table 5 summarizes the results of these models. For the GCM experiments, the magnitude of the decrease in DTR is small relative to the overall increase of mean global temperature. Moreover, these experiments reflect a level of CO2 increase well in excess of present-day values, so even smaller changes would be expected in the observed-temperature record. Outside of the Rind et al. (1989) study, only Cao et al. (1992) focus specifically on the changes of the DTR over land areas. Cao et al. show that their model can reasonably simulate the range of the present-day diurnal

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

TABLE 5 Summary of Modeling Results Showing the Relationship Between Doubled CO2 Concentrations and Changes in the Diurnal Temperature Range and the Maximum and Minimum Temperatures

General-Circulation Models

 

 

Resolution

Model

Author

Horizontal

Vertical

Ocean

ΔTeq

ΔDTReq

CCC

Boer (pers. comm.)

T32

10

Mixed layer

3.5

−0.28

GISS

Rind et al. (1989)

8° × 10°

9

Mixed layer

4.2

−0.7

 

 

 

 

 

 

(summer, USA)

 

 

 

 

 

 

−0.1

 

 

 

 

 

 

(annual, USA)

UKMO

Cao et al. (1992)

8° × 10°

11

Mixed layer

5.2

−0.17

UKMO

Cao et al. (1992)

5° × 7.5°

11

Mixed layer

6.3

−0.26

Radiative Convective Model (Cao et al., 1992)

Type

 

ΔMaximum Teq

 

ΔMinimum Teq

 

ΔDTReq

Fixed absolute humidity

 

N/A

 

N/A

 

−0.05

No surface turbulence

 

2.5

 

2.9

 

−0.4

No evaporation

 

2.3

 

2.9

 

−0.6

Full surface exchange

 

1.5

 

2.2

 

−0.7

Key: ΔTeq is the equilibrium global temperature change (°C) for doubled CO2 concentrations

ΔDTReq is the equilibrium global change of the DTR (°C) for doubled CO2 concentrations

CCC is the Canadian Climate Centre

GISS is the Goddard Institute for Space Studies

UKMO is the U.K. Meteorological Office

temperature cycle, although the modeled diurnal range in mid-latitudes is generally less than the observed.

Cao et al. (1992) also conducted a number of experiments with a one-dimensional radiative-convective model (RCM) to show that the decrease in the diurnal temperature range with doubled CO2 in that model is due primarily to a water-vapor feedback. A reduction of only 0.05°C in the DTR was observed when the absolute humidity was held constant. Table 5 indicates that increased sensible-heat exchange and evaporation are also important factors leading to a reduction in the DTR in RCM simulations with enhanced CO2.

Interestingly, the ratio of the DTR decrease relative to the increase of the mean temperature is closer in the RCM to the observed ratio over the past several decades than it is in the GCM (Table 5). The RCM omits the positive feedbacks to the DTR from reductions in cloud and surface albedo included in the GCM simulations (Cao et al., 1992). These feedbacks tend to increase the DTR because of reduced atmospheric (cloud cover) and surface (snow cover) albedo. Other GCMs simulate both increases and decreases in cloudiness from global warming—decreases in much of the troposphere, but increases in the high troposphere, low stratosphere, and near the surface in high latitudes (Schlesinger and Mitchell, 1985). A tendency for a general increase in cloud cover over land (which now seems likely from the observational evidence) could help explain the large discrepancy between the observed data and the model projections of the ratios of the decrease in the DTR range relative to the mean temperature increase. (This explanation assumes, of course, that the recent warming is induced by increases in anthropogenic greenhouse gases. On the other hand, it leaves questions regarding the cause of the apparent change in cloudiness and how it has affected the mean temperature.)

The ability of present-day GCMs to adequately simulate changes in the DTR resulting from enhanced CO2 is also affected by surface parameterizations of continental-scale evaporation. As Milly (1992) points out, present-day GCMs can overestimate the surface evaporation because of their failure to properly account for the cooling that occurs with the evaporation. Milly raises concerns about the veracity of the results from studies of soil-moisture changes induced by an increase of greenhouse gases. Accurate projections of the changes in the surface boundary-layer DTR with increases of anthropogenic greenhouse gases will be strongly dependent on adequate simulation of these processes.

Given the dependency of the DTR on surface-layer processes, interactions with the land surface, and cloudiness (which are all areas of significant uncertainties within pres-

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

ent-day GCMs), it may not yet be possible to adequately project changes of the DTR with enhanced concentrations of greenhouse gases.

Tropospheric Aerosols

It has recently been shown that increases in sulfate aerosols over and near industrial regions can have a significant impact on the earth's surface temperature (Charlson et al., 1992). Charlson et al. (1991, 1992) provide evidence to indicate that the anthropogenic increase of sulfate (and carbonaceous) aerosol is of sufficient magnitude to compete regionally with present-day anthropogenic greenhouse forcings. This forcing, which is confined primarily to the Northern Hemisphere, is a combination of direct aerosol forcing (especially over land) and indirect aerosol forcing leading to increases in cloud albedo (especially over the marine environment). Charlson et al. (1992) conclude that at present not enough is known to allow the effects of sulfate aerosols on the lifetime of clouds to be estimated. Increased aerosols (cloud condensation nuclei) could lead to smaller droplet sizes, causing a decrease in the fallout rate, which in turn could lead to an increase in cloud cover. Charlson et al. (1991) show a pattern of the geographic regions where direct aerosol forcing should be greatest. It is difficult to identify a direct relation between the pattern and magnitude of the decrease in the DTR (Figure 2) and the anthropogenic radiative forcing (Charlson et al., 1991). For example, on an annual basis over the United States the clear-sky daytime forcing varies from essentially zero (over several western states) to about -4 W m-2 (over southeastern states). It also varies with season by roughly a factor of 2 to 3, being greatest during summer.

Recently Penner et al. (1992) have argued that atmospheric aerosols from biomass burning also act to increase the planetary albedo, both directly by clear-sky planetary albedo increases and indirectly through increases in cloud albedo. Since biomass burning is most extensive in subtropical and tropical areas, this effect may be directly relevant to only a small portion of the data analyzed here.

Two tropospheric-aerosol forcings that tend to increase the clear-sky albedo have been identified. Has either of these forcings acted to reduce the maximum temperature and thereby the DTR? In the United States and northern (and perhaps eastern) Europe, where we detected a significant decrease in the DTR, there has actually been a net decrease in sulfur emissions over the past several decades, which would appear to eliminate sulfate aerosols as a cause of the DTR in these areas. However, there are reasons why such a conclusion may be premature. First, there is considerable geographic, seasonal, and secular variation of anthropogenic sulfur emissions. For example, Figure 10 shows the emissions of SO2 gas (precursor of the sulfate aerosol) doubling over the southeastern United States from 1950 to ca. 1980,

FIGURE 10

Time-dependent emissions of SO2 gas for (a) the northeastern United States and (b) the southeastern United States.

and then decreasing by 25 percent since then, while emissions have remained nearly constant in the northeastern United States. Also, the direct climate forcing due to tropospheric aerosol loading is influenced by (1) the absence of cloud cover, (2) the emission rate, and (3) the residence times of sulfate aerosol in the atmosphere, which are known to be short (about a week, according to Slinn (1983)). Second, in the United States at least, the effective heights, or stack heights, of the sulfur emissions have increased as a consequence of rural electrification in the southeast and, later, of the U.S. Clean Air Act. This increase in stack height may have had an important effect on the lifetime of SO2 and the sulfate aerosols (because of their relatively short lifetimes) and, consequently, on the aerosol concentration. Third, the indirect effects of sulfate aerosols on the DTR must also be considered, e.g., more cloud cover or a changing distribution of cloud characteristics.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
CONCLUSIONS

Strong evidence exists for a widespread decrease in the DTR over the past several decades in many portions of the globe. The climatic factors that can affect the DTR are many, but indications are that cloud cover, including low clouds, has increased in many areas where DTR has decreased. These increases in cloud cover could be indirectly related to the observed global warming and increases of greenhouse gases (or perhaps even to the indirect effects of increases in aerosols), could be simply a manifestation of natural climate variability, or could reflect a combination of all three.

Obtaining a robust answer regarding the cause(s) of the decrease in the DTR will require efforts in several areas. First, an organized global effort is needed to develop relevant and homogeneous time series of maximum and minimum temperature, along with information on changes of climatic variables that influence the DTR, such as cloudiness, stability, humidity, thermal advection, and snow cover. Second, improvements in the boundary-layer physics and treatment of clouds within existing GCMs is critically important. Third, the treatment of both anthropic tropospheric aerosols and greenhouse gases must be realistically incorporated into GCMs with a diurnal cycle. Fourth, measurements need to be made to help clarify the role of aerosols. These include (1) data on diurnal, seasonal, and geographic changes in solar radiation, in areas influenced by aerosol forcing and areas free of it; (2) simultaneous data on key aerosol properties (mass-scattering efficiency, wavelength dependence, backscatter fraction/asymmetry factor), concentration, and composition; (3) data related to changes in all the relevant cloud parameters and their coupling to aerosol changes (if any); and (4) improved chemical models to allow more accurate calculation of the sulfate aerosol effects and to provide an estimate of the time course of this aerosol forcing. Finally, imaginative climate-change detection studies that link the observed climate variations to model projections will be required to convincingly support any relationship between anthropogenic changes and the DTR.

It will be difficult to satisfactorily explain the observed changes of the mean temperature until an adequate explanation for the observed decrease in the DTR can be determined. Moreover, the practical implications of projected temperature changes and whether they are likely to continue will be even more difficult to assess.

ACKNOWLEDGMENTS

This work was supported by a U.S. Department of Energy/National Oceanic and Atmospheric Administration (NOAA) interagency agreement and NOAA's Climate and Global Change Program.

We thank the following scientists for providing us with additional data: Reino Heino for the Sodankylä data, and Takehiko Mikami for the Japanese data.

Discussion

GOODRICH: Tom, were you able to use the ISCCP data sets to test your hypothesis that the decrease in the annual range of cloudiness might be associated with temperature?

KARL: I think there might be problems with both the longevity and the quality of those data, and the solar radiation data network is not as comprehensive as one might wish. We really need to get cloud and temperature experts together to see whether relating the two makes sense.

CHARLSON: To determine how cloud cover affects heat balance you would really need to know cloud liquid-water content and mean droplet size, as well as cloud area. All three govern the system albedo. The ISCCP data don't have those, and I'm not sure our current remote-sensing systems have that capability either. If we found a correlation, we wouldn't know whether it was caused by cloud microphysics or dynamics.

SHUKLA: Tom, one of your most dramatic differences was that between the rural and urban areas of Japan. Would it be naive to ask whether radiative forcing associated with urban pollution could be involved?

CHARLSON: Urban haze around most cities actually extends for hundreds of kilometers. Also, neither the soot aerosols nor the light-absorbing gases in smog cause any great amount of heating.

KARL: Various studies have shown that maximum temperature is not much affected by urbanization. But it is a significant point to keep in mind, especially given how it dominates the record in Japan.

KEELING: Many of you have probably heard of the Arctic and Antarctic borehole data that suggest that the heat flux from the interior of the earth is not steady state. Is it possible that these borehole temperatures are showing the effects of nighttime radiative warming?

CHARLSON: The next set of NCAR model calculations will be simulating diurnal temperature range, and a model in Hamburg is doing that too. It will be interesting to hear their results.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

RIND: Our GCM looked at diurnal temperature change for doubled CO2 . We found that for a 4° warming we got about a 10 percent change, or 1°C. That was primarily during the summer, so we did not get the one-to-one correspondence you see in the data. I also wanted to mention that the ISCCP data set can't get optical thicknesses for nighttime clouds, since they can't measure the albedos.

KARL: I should perhaps add that while most of the CO2-doubling models show a decrease in diurnal temperature range of maybe 10 percent of the equilibrium temperature, the simple radiative convective model of Cao et al. shows fairly large decreases. I really think that there's some question whether a GCM can be expected to reproduce the diurnal cycle, given the type of land-surface physics involved.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Decadal Oscillations in Global Temperature and Atmospheric Carbon Dioxide

CHARLES D. KEELING AND TIMOTHY P. WHORF1

ABSTRACT

Since 1958, global air temperature has been found to correlate with the concentration of atmospheric CO2 on interannual time scales, from biennial to decadal. The decadal variations also correlate approximately with the 11-year sunspot cycle. Although CO2 data are lacking before 1958, global temperature data are available back to the 1850s. No consistent solar correlation is evident in this longer record. Instead, the record shows a decadal pattern that can be expressed by two oscillations having periods of approximately 9 and 10 years. These oscillations show reinforcement in the 1880s and 1970s and interference in the 1920s. Thus, decadal oscillations appear to have been relatively prominent in the latter half of the nineteenth and twentieth centuries, while being weak or absent from about 1905 to 1940. In the absence of an identified mechanism that would explain this pattern, the results must be treated as tentative.

INTRODUCTION

Several years ago we detected a possibly cyclic variation in the concentration of atmospheric CO2 (Keeling et al., 1989). The period of variation, of the order of 10 years, appeared to be synchronous with a decadal oscillation in air temperature when both records were band-passed to remove high- and low-frequency fluctuations and the influence of fossil-fuel combustion on the CO2 concentration. As shown in the upper panel of Figure 1, the adjusted 32-year CO2 record shows maximum concentrations near 1961, 1971, and 1981, the same years in which decadal maxima in air temperature occur. The maxima in CO2, especially the latter two, nearly coincide with maxima in the number of sunspots (lower panel) as though both CO2 and temperature were influenced by changes in solar irradiance. The investigation was not extended back to early times, because of a lack of CO2 data.

Globally averaged surface air temperature anomalies have been determined back to 1854, however (Jones and Wigley, 1991). To extend our analysis, we have accordingly computed the difference between two smoothing splines fit to this temperature record, where the differing stiffnesses of the splines result in band-passed data similar to those shown in Figure 1. The results, plotted in Figure 2 together with the record of sunspots, suggest that the positive correlation of global temperature with sunspot numbers is transi-

1  

Scripps Institution of Oceanography, University of California San Diego, La Jolla, California

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 1

Decadal variability of anomalies in global averaged atmospheric CO2 concentration (in parts per million by volume (ppm)) and in global surface temperature (in °C) compared with decadal variations in sunspot numbers, modified from Keeling et al. (1989, Figure 66). Upper curves: Approximately 10-year oscillations in atmospheric CO2 concentration (dashed curve) and global temperature (solid curve), calculated as differences between splines of different stiffness. For CO2 the stiffer spline has a standard error (s of 0.245 ppm with respect to monthly averaged data (1958-1989); the less stiff spline has an error of 0.200 ppm. For temperature over the same time period the stiffer spline has a s of 0.088°C. Lower curve: Sunspot numbers (National Geophysical Data Center, 1989), shown as line segments connecting monthly averages. The CO2 record in 1968 and 1969 was modified from that reported by Keeling et al. (1989) as a result of a reassessment of the original calibrating data.

FIGURE 2

Comparison of a 12-month running mean of sunspot numbers (upper curve) with a decadal band-pass filter of global surface temperature in °C (lower curve). The latter is the same as that shown below in panel 2 of Figure 4.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 3

Global surface temperature anomaly from 1855 through 1989, in °C (Jones et al., 1986c, and Jones, private communication). Monthly data are shown as dots. The solid curve is a spline fit through this monthly data with a s of 0. 10°C.

tory. The two decadal-scale peaks immediately prior to 1961 (in 1943 and 1952) came well after the sunspot maxima of 1938 and 1948; a series of decadal oscillations appeared before 1905, but the peak temperatures occurred near times of minimum sunspots, rather than near maxima. From about 1905 to 1940, instead of decadal oscillations, the temperature record shows oscillations with periods of only about 6 years.

Although these 6-year and decadal variations may reflect internal oscillations of the coupled atmospheric-oceanic circulation system (James and James, 1989), or simply a noisy record that falsely appears to be periodic from time to time (Gribbin, 1979), the present evidence does not, in our view, conclusively rule out the possibility of a persistent decadal periodicity in temperature. It is our purpose here to describe this possible decadal signal to help decide whether its recent appearance, where it correlates with atmospheric CO2, is short-lived, or whether instead it might reflect an oscillatory feature of the long-term temperature record, present in spite of the lack of evidence for a simple mechanism (such as solar forcing) that could produce it.

ANALYSIS OF OSCILLATIONS IN AIR TEMPERATURE

To investigate decadal-scale variations in temperature we have made use of a compilation of monthly averaged Northern and Southern Hemisphere temperature anomalies both over land and in surface seawater, originally summarized by Jones et al. (1986c) but later supplemented by additional data for the years 1854-1860 and 1985-1989, inclusive. The globally averaged data that we used directly in our computations are listed in Appendix A below.

The large scatter in monthly averaged global temperature data (the dots plotted in Figure 3) is not a strong encouragement to look for cyclic phenomena. Nevertheless, if the data are fit to a flexible smoothing spline (Reinsch, 1967) to suppress the high-frequency scatter (the solid curve in Figure 3), persistent periods of warmer and cooler conditions show in the record. The average spacing of warming and cooling events is between 3 and 4 years, depending on how many of the minor fluctuations are considered significant. A large number of the warmer periods occurred close to the times of El Niño events. It seems likely that the latter reflect phenomena that originate mainly in the Indo-Pacific region of the tropics (Quinn et al., 1987; Quinn and Neal, 1992) as a result of interactions of the atmosphere and oceans that arise from imbalances in the coupled system unrelated to any possible external forcing. If only the stronger warm events are counted, the recurrence interval becomes about 9 years (Quinn and Neal, 1992), so that El Niño events cannot be ignored in considering the possible causes of decadal variations in temperature. However, if they are so related, one would not expect them to be closely periodic.

To obtain a closer look at possible decadal-scale fluctua-

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

tions in temperature, we have fit the individual data points of Figure 3 to three additional smoothing splines of successively greater stiffness (Figure 4, panel 1), chosen to produce increasing degrees of low-pass filtering of the monthly data (Enting, 1987). The choices were subjective, but are not critical to the outcome, because distinct decadal patterns can be produced over a considerable range of stiffnesses. These three spline fits were run on both monthly and yearly averaged data, and the resulting curves showed essentially no difference. The stiffest spline (the curve labeled ''1") shows only a tendency for global air temperature to rise irregularly since 1855. The two looser splines (curves "2" and "3"), whose difference is shown in panel 2, are nearly identical to the spline curves whose difference produced the approximately decadal band-pass of temperature that is shown in Figure 1. The difference between curves " 1" and "2" (Figure 4, panel 3) approximates a band-pass centered near 20 years, while the difference between curves "1 " and "3" (Figure 4, panel 4) approximates a broader band-pass showing temperature fluctuations on both decadal and bidecadal time scales.

To examine further the spectral character of the global temperature record of Jones and co-workers, we have computed a series of spectra of their record by the maximum entropy method (MEM) (Press et al., 1989). This method of spectral analysis includes both the Thomson multitaper method and the fast fourier transform (FFT) method. Its results are particularly suitable for trying to resolve closely spaced frequencies (Press et al., 1989; Berger et al., 1990). We afterwards established the amplitudes and phases of the identified spectral peaks by least-squares fits of sinusoidal functions, since the maximum entropy method, as discussed by Sonett (1983), is not quantitatively reliable with respect to amplitudes and cannot establish phase relationships.

FIGURE 4

Global temperature anomaly from 1855 through 1989, based on annual averages of the data in Figure 3 (Use of monthly data showed negligible difference at these decadal frequencies.) Panel 1: Three superimposed spline fits. The dotted curve "1" is a very stiff spline (s equal to 0.106°C); the dashed curve "2" and solid curve "3" are looser spline curves (s's of 0.095°C and 0.075°C). Panel 2: Curve "3" minus curve "2". Panel 3: Curve "2" minus curve "1". Panel 4: Curve "3" minus curve "1".

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Specifically, we first carried out an MEM analysis of the original monthly "raw" data (shown as dots in Figure 3), with only a linear trend subtracted. The data set consisted of 1620 points. Spectral estimates were obtained for order M (number of poles; see Press et al. (1989)) equal to 750. The periods of the 24 most prominent spectral peaks found are listed in the first column of Table 1 for frequencies greater than 0.6 cycles per year. Since the use of such a large number of poles invites criticism that we have exceeded the limits of the MEM method, we carried out a second analysis in which we first removed the lowest and highest frequencies from the record by subtracting the very stiff curve "1" shown in Figure 4 from the flexible spline shown in Figure 3. The resulting detrended and slightly smoothed temperature record is shown in Figure 5, panel 1. It can be seen by comparison with Figure 3 that almost all of the oscillatory character in the original record with periods between 1 and 30 years is retained in panel 1 of Figure 5. We then calculated the MEM spectrum on the basis of this slightly

TABLE 1 MEM Spectra of Monthly Global Surface Temperature, 1855-1989

Raw Data Spectrum*

(years)

Smoothed

(years)

Amplitude for 24-Harmonic Fit**

(°C)

Non-parametric Multitaper

(years)

41.3

31.40

0.017

 

22.10

21.69

0.039

20.5

15.28

15.59

0.038

15.5

10.07

12.15

0.017

 

9.18

10.22

0.036

 

7.509

9.29

0.037

9.1

6.643

7.610

0.014

 

6.042

6.670

0.026

 

5.218

6.043

0.037

6.1

4.740

5.227

0.040

5.2

4.382

4.758

0.041

4.8

4.128

4.436

0.023

 

3.756

4.170

0.026

4.0

3.546

3.746

0.022

3.8

3.283

3.548

0.034

3.5

3.122

3.290

0.025

 

2.867

3.127

0.022

 

2.716

2.866

0.027

 

2.318

2.727

0.009

 

2.241

2.320

0.023

 

1.998

2.242

0.017

 

1.885

1.997

0.020

 

1.796

1.884

0.014

 

 

1.797

0.017

 

* From Ghil and Vautard (1991, Figure 2). Their plot also shows an unlabeled, strong peak near 10.2 years.

** From a fit of sine and cosine functions to the smoothed data. The amplitudes are peak to trough.

smoothed record with data points spaced three months apart, beginning with January 1855. The data set consisted of 540 points. Given the degree of smoothing of this record, more closely spaced data would not have contributed any significant additional information, as perhaps they could have done if the raw data had been used. As shown in the second column of Table 1, the power spectrum of the slightly smoothed record, with M set equal to only 250, gives almost the same frequencies for peaks as the spectrum based on the raw data with M set equal to 750.

A check was also made by creating a subset of points, reducing the raw data by a factor of 3 (selecting values for every third month only, beginning with the first). With a linear trend removed as before, we obtained nearly the same spectrum with M set equal to 275 as in the previous two cases.

We then computed the spectrum using the slightly smoothed data spaced three months apart, but with M reduced from 250 to 150. Figure 6 shows this spectrum (indicated by a dashed line) has fewer and less sharp peaks than that computed with M equal to 250 (solid line). In particular, in the decadal-frequency region a single peak with a period of 9.7 years replaces a pair of peaks with periods of 9.3 and 10.2 years. This single peak, having a frequency of approximately the mean of the two peaks in the prior case, shows relative stability in frequency despite the change in M value.

As demonstrated by a synthetic example of a sample of points from the sum of two sinusoids (Press et al., 1989), the MEM method at low pole numbers does not resolve closely spaced oscillations, whereas when MEM is used with high pole numbers, spectral lines may split where there is no firm basis for expecting multiple peaks. These are typically exaggerated in regions of the spectrum having low power spectral density. In our particular example of global temperature, both spectra in Figure 6 indicate high power in the decadal region, whether that power is represented by a single oscillation or a pair. We chose to accept the spectrum showing a pair of peaks, because the spline fit to the temperature data in the decadal band (shown in Figure 4, panel 2) also indicated that the decadal signal vanished in the middle of the record, and this is better represented by a pair of decadal oscillations.

Apart from the changes in frequency caused by the merger of spectral lines as the M value decreased, we did not find noticeable changes in the position of the spectral peaks as we varied our method of analysis. We found, however, that the peaks shifted somewhat in frequency if we used only limited portions of either version of the record—for example, by deleting the data before 1875 that exhibit more high-frequency oscillations than the subsequent record. Nevertheless, for peaks with periods of 6 years or greater, all of the spectra examined were similar to the two versions summarized in Table 1.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Figure 5

Global surface temperature anomaly in °C from 1855 through 1989, together with spectral oscillations derived from it determined from the maximum entropy spectrum. Panel 1: Smoothed and detrended temperature anomaly obtained by subtracting from the smooth curve of Figure 3 a spline fit to the yearly averaged anomaly data with a s of 0.106°C, as in curve 1 of Figure 4. Panel 2: Sum of 24 spectral oscillations derived from a fit to the curve in panel 1, which are consistent with the spectral peaks as listed in Table 1, column 2. Panel 3: Sum of decadal oscillations. Panel 4: Sum of bidecadal oscillations. Panel 5: Sum of eight oscillations with periods of 6.67 calendar years or greater (no exclusions). The source of the temperature data is the same as in Figure 3. The time of maximum interference of the decadal oscillations in panel 3 is indicated.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Figure 6

Two maximum-entropy spectra of the global surface temperature anomaly, expressed as power (logarithmic scale) versus frequency, in cycles per year. The temperature record used to determine the spectra was expressed as the difference between the flexible spline plotted in Figure 3 and the very stiff spline from curve "1" of Figure 4, panel 1. The solid curve indicates a spectrum with the number of poles, M, equal to 250; the dashed curve shows a spectrum with M equal to 150. The periods of the prominent peaks of the former, in years, are shown in standard characters, those of the latter by figures in italics.

Another study of nearly the same temperature record by Ghil and Vautard (1991) used a combined singular spectrum and multi-taper analysis that provides an F-test for the significance of the peaks. Their study confirms the significance, at the 97 percent confidence level, of the peaks that we found near 6, 9, 15, and 21 years, and showed evidence of several of the other peaks listed in Table 1.

Accepting the peak periods revealed by the second spectrum of Table 1 for further study, we have fit the slightly smoothed temperature record of Figure 5, panel 1, using a non-linear least-squares method (Bevington, 1969, subroutine CURFIT) to a series of sine and cosine functions having the periods indicated for peaks in the MEM spectrum, omitting only very weak peaks and any peaks with periods less than 1.7 years.

The reconstructed time series obtained by summing these 24 computed sinusoidal spectral oscillations reproduces the main peaks and troughs of the original record (Figure 5, panel 2). If individual warming and cold events are scrutinized, admittedly the fit is far from perfect. Allowing, however, that the lowest period of oscillation in the reconstruction is 1.8 years, one cannot expect a wholly consistent representation of times and amplitudes of short-term interannual fluctuations such as those related to El Niño events. As we verified by carrying out a further reconstruction using 48 harmonics that included oscillations with periods as low as 1.0 years, the correctness of the phasing and amplitudes for short-term interannual fluctuations can be improved by including higher-frequency oscillations in the fit to the temperature data. The amplitudes of the lower-frequency oscillations, which are those of importance to the present study, were only negligibly altered by the addition of higher frequencies in the fit.

The amplitudes of the sinusoidal functions are listed in Table 1. Of the 24 spectral oscillations in the reconstruction, only the two noted earlier, having periods of 9.3 and 10.2 years, occur near to the decadal time scale. Summed, these produce oscillations with an average period of 9.7 years and a maximum amplitude, peak to peak, of 0.15°C. They beat with a recurrence period of 102 years (half of the period of the long-term oscillation produced by the difference of the two shorter-period oscillations) and exhibit interference

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

near 1925 (Figure 5, panel 3). The pattern of interference and reinforcement is distinct because the amplitudes of the separate oscillations, as found by the least-squares fits to the sinusoidal functions, are nearly equal.

Two additional strong spectral oscillations are found in the temperature spectrum at approximately the bidecadal time scale, with periods of 15.6 and 21.7 years. Summed, they have a maximum amplitude of 0.14°C and beat with a period of 55 years (Figure 5, panel 4), which is close to half the beat period of the decadal harmonics. They also exhibit interference in the 1920s, at almost the same time as the decadal oscillations. Furthermore, their summed oscillations, with an average period of 18.1 years, essentially impose variable amplitudes on the decadal oscillations. The addition of weaker harmonics with periods of 6.7, 7.6, 12.2, and 31 years to form a broad low-pass filter of the detrended temperature record does not cancel out this beat pattern (Figure 5, panel 5). The decadal oscillations are still present, with the strongest temperature maxima where the bidecadal oscillations enhance the decadal signal. Between 1905 and 1940, however, only oscillations with periods close to 6 years appear, as was the case for the spline-fit differences of panels 2 and 4 of Figure 4.

As shown in Figure 7, the sinusoidal oscillations of Figure 5, which are derived from spectral analysis, show essentially the same patterns as those previously derived from spline fits. The decadal and bidecadal oscillations are equally prominent, whether they are obtained by spectral analysis or by the use of smoothing splines. Thus, the amplitudes and phases obtained by spectral analysis do not appear to be falsified by the restricted number of degrees of freedom of the spectral compositing.

We conclude that the decadal oscillations that have appeared in the temperature record since the late 1950s may be characteristic of the longer-term record as well. If so, the bidecadal oscillations seen in Figures 5 and 7 may also be characteristic of the longer record, especially since they appear to be an amplitude modulation of the decadal oscillations. Both pairs of oscillations are intermittent, however, being essentially absent for about 35 years near the middle of the temperature record.

Figure 7

Spectral oscillations of the global spectrum of global surface temperature anomaly (dashed curves), compared with the same temperature record that has been band-pass filtered using spline fits (solid curves). Panel 1: Sum of decadal oscillations, as in Figure 5, panel 3, and curve "3" minus curve "2" of Figure 4. Panel 2: Sum of bidecadal oscillations, as in Figure 5, panel 4, and curve "2'' minus curve "1" of Figure 4. Panel 3: Sum of eight oscillations, as in Figure 5, panel 5, and curve "3" minus curve "1" of Figure 4.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
DISCUSSION

In our quest to understand decadal oscillations that appear in the record of atmospheric carbon dioxide measurements, we have been led to investigate seemingly similar oscillations in global air temperature. Employing a much longer temperature record than that available for atmospheric CO2, we have found that decadal variations in global air temperature are characteristic of the past 135 years, except for an interval near the middle of the record. This pattern can be reproduced by assuming the existence of two oscillations with slightly different frequencies that beat on the time scale of about 100 years.

Is it possible that these decadal oscillations are caused by a cyclic process? Perceptions of cyclic behavior in the climatic record have in the past involved so many exceptions and inconsistencies that the subject does not enjoy good repute. Compelling evidence that cyclic behavior is actually caused by an identifiable forcing agent such as solar irradiance is difficult to find.

Decadal oscillations have not been emphasized in previous discussions of global temperature, but the existence of bidecadal oscillations has been alleged by Newell et al. (1989) on the basis of marine air-temperature data. An inspection of their Figure 2 indicates that the signal that they found is similar in phase and amplitude to that shown in our Figure 4, panel 4. There is no indication in their analysis of a weakened signal near 1925, however, like that found by our analysis of global air temperature. We are not aware of other finds of either decadal or bidecadal oscillations.

Of possible importance in deciding the significance of the oscillations in temperature that we report here is the indication that the decadal signal appears to be replaced by shorter oscillations, with periods of close to 6 years, from about 1905 to 1940. Although the spectral evidence for this, as shown in the low-passed record of Figure 5, panel 5, is based only on spectral peaks with periods greater than 6.5 years, the strong 6.0-year peak in the MEM spectrum produces an oscillation (not shown) with almost exactly the same phase. Thus a 6.0-year oscillation appears to be a significant contributor to the overall pattern of variability in the global temperature record, although not as prominent as the decadal signal.

The 6.0-year oscillation, and two other adjacent oscillations with periods of 4.8 and 5.2 years (see Figure 6), could well be associated with the low-frequency component of the El Niño/Southern Oscillation (ENSO) phenomenon, as pointed out by Ghil and Vautard (1991) in their spectral analysis of temperature. When these three oscillations are included in a low-frequency band-pass of temperature, warm periods in this band-passed record include most of the strong and many of the moderate El Niño events.

The 4.8- and 6.0-year oscillations, when summed, show an interference pattern with a beat period of 22.4 years. The timing of these beats shows a possible relationship to the 21.7-year spectral oscillation, inasmuch as the warm phases of this bidecadal oscillation tend to occur near the times of reinforcement of the two higher-frequency oscillations. If these higher-frequency oscillations are related to global aspects of the El Niño phenomenon, is it possible that the bidecadal oscillations that we have been discussing are also related? In the absence of a mechanism to explain any of these oscillations, and not even clear proof that they exist, it is not possible to resolve this question, but the topic appears to us to deserve further investigation.

Does the record of atmospheric CO2 provide any insight into possible causes of the interannual variations in global temperature? At first glance one would not expect so, because interannual variations in CO2 are presumably a consequence of climatic factors that do not depend on the earth's carbon cycle. Nevertheless, two aspects of the CO2 record are worth noting.

First, atmospheric CO2 anomalies show a striking relation to global temperature anomalies with respect to El Niño events (Keeling et al., 1989). Second, on all interannual time scales that can be investigated in the 32-year CO2 record up to 1990, CO2 variations tend to lag temperature variations by approximately a half-year (Kuo et al., 1990). The existence of a lag of that length on the decadal time scale is evident in the plot of Figure 1.

This lag, as shown by Keeling et al. (1989), cannot easily be explained as an oceanic phenomenon, because temperature-induced exchange of CO2 gas at the air-sea interface on a decadal time scale should result in a lag in atmospheric CO2 variation of about 1.5 years, whereas a change in upwelling of cold water should produce atmospheric CO2 variations of the opposite phase with respect to temperature, i.e., a lag of about 5 years. On the time scale of El Niño events, Keeling et al. suggest that atmospheric CO2 responds to global temperature, because vegetation on land tends to release CO2 during anomalously warm periods. With respect to this terrestrial response, a short lag in CO2 is to be expected.

Perhaps variations caused by terrestrial vegetation and driven by climatic change also produce decadal variations in CO2. Indeed, a decadal signal in CO2 might be simply a modulation in the magnitude or frequency of variations on El Niño time scales. If this should be the case, a nearly constant phase lag in CO2 with respect to temperature would be expected over a broad range of interannual frequencies, as observed.

In support of such an hypothesis are several papers' assertions of possibly non-random, long-term behavior in El Niño events. Their focus has been mainly on the decadal time scale (see, for example, Hanson et al., 1989; Michaelsen, 1989; and Enfield and Cid, 1990) but there is a hint of a decadal connection in the study of Barnett (1989),

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

as interpreted by Enfield and Cid. To these speculations may be added our suggestion, above, of a possible association of spectral temperature oscillations on El Niño time scales with a bidecadal oscillation.

In conclusion, we suggest that the atmospheric CO2 record may be of some use in the study of decadal variability in climate. Conclusive evidence of decadal periodicities in climatic parameters will probably be established, however, only if and when a mechanism is found that is physically reasonable and successfully predicts changes in climate.

ADDENDUM

In the oral presentation of this paper, we presented an hypothesis that forcing of ocean temperatures by oceanic tidal dissipation might explain the observed decadal oscillation in global temperature and its disappearance in the early part of this century. We were attracted to the coincidence of the times of prominent decadal temperature signal with the times when the strongest equilibrium oceanic tidal forcing was produced by the gravitational forces of the sun and moon. At these times, within 10 years of 1881 and 1974, episodes of unusually strong tidal forcing occurred at times of the strongest tides of the year, because the sun and moon tended then to be both in unusually close alignment with the Earth and at nearly closest approach. In contrast, in the 1920s, when the decadal signal was weakest, only weaker tidal forcing occurred at times of the strongest tides. Because strong tidal forcing occurs at intervals of 9 years when the alignments of the sun, moon, and earth are nearly optimal for strong tidal forcing, and otherwise at 3- and 6-year intervals, the decadal signal itself and the 6-year signal in the middle of the record period also correspond to the periodicities of tidal forcing.

It was not possible to prepare an appropriate presentation of the tidal discussion for this volume. An article describing the tidal hypothesis is in preparation.

ACKNOWLEDGMENTS

We are grateful to scientists who gave generously of their time to discuss the subjects of solar phenomena and climatic variation with us in the course of preparation of this article. We specifically thank Philip Jones and Thomas Wigley, Michael Ghil, James Hansen, Harry van Loon and Gerry Meehl, David Parker, and Henry Diaz and Thomas Karl. We are also grateful for discussions with Tim Barnett, Robert Bacastow, Daniel Cayan, and Hugh Hudson, colleagues of ours at the University of California at San Diego. We further thank Drs. Jones and Hansen for supplying us with their temperature data sets, and David Parker for supplying us with extensive global and regional sea surface temperature data in graphic form. Computer time was provided by the San Diego Supercomputer Center. Financial support came from the National Science Foundation via Grant ATM-91-21986 and from the U.S. Department of Energy via Grant FG03-90ER-60940.

APPENDIX A TABULATION OF TEMPERATURE DATA

Table A1 presents the global-average anomalies of air temperature used in our study. The anomalies are listed for each calendar month, followed in the last column by the annual average. They were obtained by averaging Northern and Southern Hemisphere anomalies. The hemispheric data for 1855 through 1989, expressed both over land and in surface seawater, were supplied to us by P.D. Jones (personal communication of April 1990). These data were supplied to us as monthly averages, whereas for 1987 through 1989 only annual averages were available with respect to temperatures in surface seawater. For these final three years of our data set we estimated global averages for each month from the monthly anomalies supplied to us for temperatures over land.

For the Northern Hemisphere we multiplied the monthly land values for each given year by the ratio of the annual global mean for that year to the mean for land. For the Southern Hemisphere we assumed that no intra-annual variations existed over the oceans, and that temperatures over land contributed 30% to the hemispheric averages. For the full data set we accepted the premise of Jones et al. (1986c) that sea surface and marine air temperatures follow each other closely on interannual time scales, so that combined land and surface sea-water-temperature data portray global-average interannual variations in surface air temperature.

Recently P.D. Jones (personal communication of April 1993) kindly supplied us with updated monthly data for all years of our study. We have repeated our calculations with these data and find only very small changes in spectral amplitudes and frequencies, and thus in the time plots shown in this article.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

TABLE A1 Global Surface Temperature Anomalies Used in Our Analysis, in Units of .01°C

 

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

 

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

 

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
Discussion

CICERONE: I think you're touching on something very fundamental: the possibility that there may be an external source of climate variability.

TRENBERTH: It seems to me that the key question is what physical mechanism is operating here. Have you any thoughts on that, Dave?

KEELING: Well, there was a paper by John Loder and Christopher Garrett in JGR in 1978 that suggested that an 18.6-year cycle in SST can be seen on the east and west coasts of North America, of opposite phase. They found that the amount of energy involved in the tidal currents on the continental shelves might be sufficient to cause temperature changes. Peter Holloway and his colleagues have done some nice studies off the northwest coast of Australia. They find that the breaking of the internal waves eventually causes turbulence that results in so much mixing that you get an almost linear decrease in temperature from the sea surface downward. I am inclined to think that the tides are causing some sort of turbulence that affects the upper ocean.

DOUGLAS: The amplitude of the 18.6-year tide is less than a centimeter. It's very difficult to detect in the tide-gauge records. Where does all the energy go?

KEELING: I think the semi-diurnal and diurnal tides would have to carry most of the energy dissipation. What strikes me is that if you look at the correlation of cooling with the tides themselves, you see that the greatest amount of cooling tends to occur in conjunction with a strong tide.

GHIL: I find it interesting that, using an entirely independent technique, Dave is getting results that are nearly identical to those that I and Robert Vautard published in Nature in March 1991. We had peaks at similar locations, including a 15-to-16-year one.

KEELING: The 15-year one seems to me to be the best candidate for some sort of tidal connection because it is not explained by any kind of solar activity.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Observed Variability of the Southern Hemisphere Atmospheric Circulation

DAVID J. KAROLY1

ABSTRACT

The dominant modes of interannual variation of the Southern Hemisphere (SH) circulation are described, using results from several studies of grid-point analyses of the SH troposphere. The two leading modes are primarily zonally symmetric, representing out-of-phase variations of height between middle and high latitudes in one case (called the high-latitude mode) and between the tropics and middle latitudes in the other (called the low-latitude mode). Wave-train patterns of anomalies are described also. These include a wave-number-three mode in winter with large amplitude at high latitudes, a wave-number-four mode in middle latitudes in summer, a meridional wave-train structure originating over Australia, and a monopole or a meridional wave train over the Pacific Ocean in winter. The two most important processes leading to interannual variations of the SH circulation appear to be the Southern Oscillation and the interaction between waves and mean flow in the SH storm track, which leads to the High Latitude mode.

Since upper-air data and grid-point analyses are available for only a short period, surface data must be used to describe decadal variability of the SH circulation. The barotropic vertical structure of low-frequency variations in the SH means that surface pressure variations are representative of variations throughout the troposphere. Relatively little attention has been paid to decadal variability in the SH. It appears from indices of the two dominant modes of interannual variations, determined using the surface data, that the modes show substantial decadal variability over the last 90 years. Few observational data exist that can be used to directly indicate climate variability on century time scales in the SH. The only possible observational evidence for this comes from proxy data, such as tree rings.

1  

Cooperative Research Centre for Southern Hemisphere Meteorology, Monash University, Clayton, Victoria, Australia

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
INTRODUCTION

There has been relatively less study of the variability of the Southern Hemisphere (SH) atmospheric circulation than of that of the Northern Hemisphere (NH). There are a number of reasons for this; the lower population and greater ocean areas in the SH mean that observational records are sparser and there have been fewer people to analyze them. In this brief and incomplete review, I will concentrate on studies of the variability of the SH tropospheric circulation, starting at interannual time scales and moving onto longer time scales. This progression is justified by the available observational data and the number of relevant studies, as outlined below.

Detailed and regular observations of the troposphere over much of the SH did not start until the International Geophysical Year in 1957-1958, when the number of SH upper-air sounding stations increased markedly. Operational subjective analyses of the SH circulation have been prepared since that time by the Australian and South African weather services. The introduction of meteorological satellites in the late 1960s led to the development of the first operational objective numerical-analysis scheme for the SH at the Australian Bureau of Meteorology in 1972. Hence, objective numerical analyses for the SH are available for only the last 20 years, and upper-air sounding data are available for only about 30 years.

Hemispheric numerical analyses have been used to investigate the interannual variability of the SH circulation, particularly over the last decade (Trenberth, 1979; Rogers and van Loon, 1982; Mo and van Loon, 1984; Mo and White, 1985; Kidson, 1988a,b; Karoly, 1989a; and others). In addition, radiosonde station data have also been used for similar investigations by Szeredi and Karoly (1987a,b). The results of these studies of the dominant modes of variability of the SH circulation on interannual time scales are summarized in the next section.

Since these upper-air data are available for a short period only, it is not possible to use them to describe variations on decadal time scales or longer. There are longer records of surface meteorological observations for SH stations, some extending back to before the turn of the century. The equivalent-barotropic vertical structure for the low-frequency tropospheric variations (Szeredi and Karoly, 1987a) permits these surface records to be used to provide some information on decadal variability in the SH. However, there have been no comprehensive studies of decadal time-scale variability for the SH. More recently, Jones (1991, hereafter referred to as J91) has reconstructed from station data a gridded mean sea-level pressure data set for the SH that extends back to 1911. J91 and Allan and Haylock (1993) have used this data set to consider decadal variations of the SH circulation. These and a few other studies of long surface records are described briefly below, although the others have generally concentrated on variations of El Niño and the Southern Oscillation.

In an attempt to provide some evidence of the range of SH variability on century time scales, some climate reconstructions based on proxy data are discussed below. Although these proxy data are available for very few sites in the SH, they constitute the only source of data on natural climate variability on century time scales for the SH. More extensive descriptions of proxy data for the SH are provided in Chapter 5 of this volume.

INTERANNUAL VARIATIONS FROM UPPER-AIR DATA

Relatively less attention has been paid to low-frequency variations of the SH circulation than to those of the NH, probably because of the shorter time period for which analysis data sets are available. Trenberth (1979), Kidson (1988a), and Shiotani (1990) have used zonal-mean data to show that there are large low-frequency variations of the zonal-mean circulation in the SH involving variations of the zonal wind and associated dipole variations of geopotential height. Mo and White (1985, hereafter referred to as MW) carried out a teleconnection analysis for the SH (similar to that of Wallace and Gutzler (1981) for the NH) using 8 years of monthly mean SH analyses. From their grid-point correlation maps, they identified three modes of variation: a zonally symmetric dipole mode at middle to high latitudes, a zonal wave-number-three mode in winter, and a continent-ocean contrast in summer. Szeredi and Karoly (1987b, hereafter referred to as SK) used the longer period of upper-air station data available for the SH (up to 30 years) to investigate low-frequency variations. They found two dominant modes of monthly variation in the station data that were primarily zonally symmetric, involving opposing departures of height between middle and high latitudes in one case and between the tropics and middle latitudes in the other. The sparse and irregular station network in the SH meant that only the largest-scale modes could be identified.

More recently, the technique of unrotated and rotated principal-component analysis (PCA) has been applied to grid-point analyses of the SH to clarify the typical modes of low-frequency variations of the SH circulation in a number of different studies. These may be separated into two groups according to the time scale of interest: Periods longer than about 50 days, including interannual variations, have been considered by Rogers and van Loon (1982), Kidson (1988b, hereafter referred to as K88), and Karoly (1989b), while another group concentrated on intraseasonal variability. The results are similar for different studies of the same time-scale fluctuations, and differences can usually be explained in terms of differences in analysis method, such as PCA of the correlation (Karoly, 1989b) or covariance matrix (K88), use of different variables (300 hPa height by

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Karoly (1989b), 500 hPa height in most other studies), differences in time filters used, or seasonal stratification. The main data sets used have been the Australian SH analyses, available since 1972 (Karoly, 1989b; Rogers and van Loon, 1982; MW) and the European Centre for Medium-range Weather Forecasts (ECMWF) analyses (K88 and many intraseasonal studies), which are available only for 1979 and later but are likely to be of higher quality than the Australian analyses.

In the following discussions, the typical modes of interannual variations are described first, using results from Karoly (1989b). A rotated PCA has been carried out on monthly anomalies of 300 hPa height using 15 years of Australian analyses (1973-1987). Three different types of horizontal structure have been identified for the low-frequency variations in the SH. The leading modes are approximately zonally symmetric, occur in both summer and winter, and explain about a quarter of the variance. They are associated with opposite departures of height for middle compared with high latitudes and for the tropics compared with middle latitudes. The other stable modes are hemispheric-scale wave-like variations, either zonal wave trains around the hemisphere or regional meridional wave trains, both of which seem to be less important than the symmetric modes. All the modes have an equivalent-barotropic vertical structure in the extratropics, and generally are more horizontally extensive than the typical modes of low-frequency variation in the Northern Hemisphere. The equivalent-barotropic vertical structure of low-frequency variations in the SH was identified using radiosonde data by Szeredi and Karoly (1987a).

Zonally Symmetric Modes

The two leading modes from the rotated PCA of monthly anomalies of 300 hPa height are mainly zonally symmetric, representing opposite height variations over different latitude bands. These modes are found in both summer and winter. One mode represents a north-south out-of-phase variation of height between Antarctica and middle latitudes; it will be referred to hereafter as the high-latitude mode. The horizontal structure of this mode is shown in Figure 1a for winter, using correlations of the time series of PC scores for this mode with monthly SH height anomalies. The proportion of the total monthly variance explained by this mode is about 12 percent. Although it is primarily zonally symmetric, there are substantial asymmetries, with larger correlations and gradients over the Indian Ocean and south of Australia. The middle latitudes show evidence of zonal asymmetry resembling a wave-number-three pattern. This high-latitude mode has an equivalent-barotropic vertical structure with very similar variations in both 500 hPa height and MSLP.

This mode of low-frequency variation in the SH has

FIGURE 1

Rotated principal-component patterns for monthly 300 hPa height anomalies in winter representing the two leading modes, (a) high-latitude mode and (b) low-latitude mode. Correlations of the time series of the PC scores with monthly anomalies are shown.

been identified by SK using station data; by Rogers and van Loon (1982), MW, and K88 using shorter time periods of analyses; by Shiotani (1990), Kidson (1988a), and Trenberth (1979) using zonal-mean fields; and by Trenberth and Christy (1985) using zonal-mean surface pressure. SK found that it was the second mode of monthly variations of the station data, explaining a smaller fraction of variance than

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

here, whereas K88 found that it was the leading mode of interannual variations of 500 hPa height in the ECMWF analyses. Kidson (1988a) has shown that this mode occurs on all time scales from intraseasonal to interannual.

The other mode found in both seasons represents an out-of-phase north-south variation of height between low and middle latitudes; it is shown in Figure 1b for winter. This mode, hereafter referred to as the low-latitude mode, explains about 13 percent of the variance in winter. Strong departures from zonal symmetry occur in the South American region, with the nodal line running almost parallel to the coast. More intraseasonal variation than interannual variation is found for this mode in winter, but in summer it is mainly associated with interannual variations.

Structures similar to this low-latitude mode have been identified by SK using station data and by Kidson (1988a) using zonal mean wind fields. This mode in the station data explains a larger fraction of variance than here, probably because of the more northerly extent of the analysis region in SK. K88 did not identify this mode, probably because his analysis using the covariance matrix of 500 hPa height variations is restricted to patterns that have large-amplitude anomalies at that height, whereas they are generally small in low latitudes.

MW presented a correlation analysis of the monthly zonal mean height and MSLP fields from their data set (refer to their Figures 1 and 2). They found structures in the zonal-mean fields that are similar to the zonally symmetric modes described here. The two leading modes for the zonal-mean fields, which explain about 70 percent of the variance, have see-saw structures over latitude bands similar to the high-latitude and low-latitude modes shown in Figure 1.

Wave Trains

In addition to the zonally symmetric structures, wavelike patterns were also found in the PCA. They explain less variance and are less stable than the zonally symmetric modes, but are found in one-point correlation maps and composite analyses. In contrast to the NH wavelike modes, the modes found in the SH exert influence over hemispheric scales. They include zonal wave-number-three patterns and meridional wave trains in winter, and wave-number-four patterns or continent/ocean contrasts in summer.

Two stable zonal wave-number-three patterns were found for winter 300 hPa height; they are shown in Figure 2. The centers of action are located at around 60°S for the first version and about 50°S for the second version. The main difference between the two patterns is that one is shifted by about one quarter wavelength relative to the other, so that the extrema of one occur along the nodal lines of the other. The second version shows a stronger relationship (out-of-phase relative to the main centers) with lower latitudes, while the first version shows evidence of a weak

FIGURE 2

Rotated PC patterns of winter 300 hPa height in winter describing the two versions of the wave-number-three mode, as in Figure 1.

wavenumber-four structure at around 30°S. About 7 percent of the variance is explained by the first version, while the second explains about 5 percent of the variance. Together, wave-number-three patterns represent about 12 percent of the variance, a total similar to that of each of the zonally symmetric modes.

Using one-point correlation maps, MW identified a zonal wave-number-three pattern in the SH winter. The two orthogonal versions of the mode found here, with large amplitude at high latitudes, indicate that zonal wave-number-three

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

anomalies are a common low-frequency pattern in the SH, but that they do not have a strongly preferred phase. Mo and van Loon (1984) describe a change in the amplitude of wave-number-three on long time scales.

As well as the zonal wave-number-three pattern in winter, two versions of a zonal wave-number-four pattern were found in summer that are phase-shifted with respect to each other. Together these modes explain about 13 percent of the summer height variance. They both have centers of action around 45°S and show a continent-ocean contrast. MW identified a teleconnection pattern in the SH summer associated with the continent-ocean thermal contrast; this pattern appears to be related to this zonal wave-number-four pattern found in the upper troposphere.

A meridionally oriented standing-wave pattern resembling a wave train has been found in winter. The pattern for this winter wave train is shown in Figure 3a. Five centers of action can be seen extending from eastern Australia southeastward to high latitudes and equatorward into the southern Atlantic. This mode, which explains about 6 percent of the variance, is not clearly evident in the lower-level variables, where only parts of the wave-train structure are reproduced. Using low-pass filtered ECMWF analyses, however, K88 has identified a similar wave-train pattern of height variations.

The third principal component of 300 hPa height in winter explained a relatively large fraction of the variance, but was not stable; it displayed different structures when different analysis methods were used. The pattern for this mode, shown in Figure 3b, essentially represents monopole height variations over the Pacific Ocean, although there is some indication of opposite variations further equatorward and poleward, suggesting a meridional wave train. Although the feature over the central Pacific was relatively stable, the weaker features in this pattern were quite variable.

Mechanisms

Although the above studies have described low-frequency variations in the SH, much less attention has been paid to the mechanisms associated with these observed variations. Karoly (1989a,b) and SK have shown that the interannual variations of the low-latitude mode in summer are are related to the Southern Oscillation and the associated variations of the temperature of the tropical troposphere. In winter, the Southern Oscillation is associated with a pattern of large-amplitude anomalies over the Pacific Ocean, as shown in Figure 3b. Hence the Southern Oscillation is associated with two of the dominant modes of interannual variability in the SH.

The other common interannual variations in the SH are linked to the high-latitude mode. K88, Karoly (1990), and Shiotani (1990) have shown that the zonally symmetric variations of the high-latitude mode are associated with

FIGURE 3

Rotated PC patterns for 300 hPa height in winter for the two meridional wave-train modes.

changes in the observed transient-eddy activity and momentum flux in the SH storm track. Trenberth (1984) showed that the large circulation anomalies during the SH winter of 1979, during the Global Weather Experiment, were related to mean flow and eddy variations associated with this high-latitude mode. A general-circulation modeling study of Zwiers (1987) found that the largest interannual variations of the model SH circulation were zonally symmetric opposite variations between high and middle latitudes, very similar to those of the observed high-latitude mode. Simplified

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

atmospheric-modeling studies by James and James (1989) and Robinson (1992) have shown that it is also the dominant mode of low-frequency variations in extended simulations with simplified general-circulation models with zonally uniform boundary conditions. Hence, this zonally symmetric high-latitude mode appears to be a ubiquitous mode of low-frequency variation of the SH, associated with interaction between the mean flow and the transient eddies in the SH storm track.

Little study has been made of the zonal wave-train modes, but it is likely that the quasi-stationary interannual wave-number-three modes are associated with the three SH land masses and continent-ocean thermal contrasts. In addition, the zonal wave-number-three Rossby wave is almost stationary in the mean SH zonal wind in mid-latitudes.

INTERDECADAL VARIATIONS ASCERTAINED FROM SURFACE DATA

Since upper-air observations are available for the SH for about three decades only, it is not possible to use such data to describe the interdecadal variations of the SH circulation. The only analyses of longer-term variations of the SH troposphere using upper-air station data have concentrated on temperature trends (Angell, 1986; Karoly, 1987). These have shown a small warming trend in the SH lower troposphere over the last three decades.

Some evidence of longer-term changes in the SH circulation is found when the mean sea-level pressure analyses prepared by the South African Weather Bureau for 19511958 (Taljaard et al., 1961) are compared with analyses from the Australian Bureau of Meteorology from 1972. Although some differences are due to different analysis methods, some pressure changes appear to be real (Mo and van Loon, 1984; J91).

The equivalent-barotropic vertical structure of low-frequency variations in the SH (Szeredi and Karoly, 1987a) means that longer records of surface observations may be used to describe interdecadal variations. Surface-pressure observations are available for some SH stations for periods extending back to the last century, and sea surface pressure observations over limited SH ship tracks are available for a similar period. It should be noted that these records are often incomplete and are representative of only limited SH regions.

J91 has used station data to reconstruct gridded SH mean sea level pressure (MSLP) fields for most of this century. The quality of this MSLP data set has been assessed by J91 and Barnett and Jones (1992). It is considered to be of reasonable quality over the region 15°S to 60°S from 1951 on, but the reconstructions back to 1911 are useful only near the continents and over the southwestern Pacific Ocean. Jones has used these data to assess changes since 1951 in the strength of the three subtropical anticyclones and the southern high-latitude westerlies; time series of indices of the strength of the South Pacific anticyclone and the high-latitude westerlies from J91 are shown in Figure 4. They

FIGURE 4

Seasonal and annual time series of two indices of the SH circulation: upper panel, South Pacific anticyclone high (average pressure in the region 25°S to 35°S, 80°W to 110°W); lower panel, strength of the high-latitude westerlies (zonal-mean pressure difference 40°S to 60°S). The filtered curve is from a 13-term Gaussian filter designed to suppress variations on time scales less than 10 years. (From Jones, 1991; reprinted with permission of the Royal Meteorological Society.)

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

show large interannual variations, as well as some fluctuations on decadal time scales. The interannual variations of the South Pacific anticyclone are associated with the Southern Oscillation, and they are larger than those of the anticyclones over the other two SH ocean basins. The index used for the strength of the southern westerlies is also a measure of the high-latitude mode described earlier. It shows pronounced interannual and longer-term variations.

Allan and Haylock (1993) have used the J91 MSLP data set to identify the SH circulation variations associated with the long-term winter rainfall decrease over southwestern Australia. In the Australasian region, they describe variations of MSLP on decadal time scales as well as a longer-time-scale trend, both of which are associated with the regional rainfall variations.

Several indices of the large-scale surface-pressure field in the SH have been used to investigate variations of the SH circulation (Troup, 1965; Trenberth, 1976; Pittock, 1984). The index most frequently used is the Southern Oscillation Index (SOI), the difference between normalized mean sea-level pressure anomalies for Tahiti and those for Darwin. This index represents the magnitude of the pressure oscillation between the eastern Pacific Ocean and Indian Ocean regions associated with the Southern Oscillation. Pittock (1984) defined the Trans-Polar Index (TPI) as the pressure anomaly at Hobart (43°S, 147°E) minus that at Stanley (52°S, 58°W), on the opposite side of the hemisphere. The TPI represents variations of the amplitude of zonal wave-number-one asymmetries in the pressure field. It probably captures variations in the high-latitude mode to some extent, because of the difference between the latitudes of the two stations. Pittock (1984) has shown that the two dominant modes of common interannual variations of rainfall in Australia, New Zealand, South Africa, and South America are associated with variations of the SOI and the TPI, and that there is little relationship between these two indices. Pittock considered only 30 years of data, 1931 to 1960, and did not directly examine interdecadal variations of these indices and hence of the SH circulation. However, in commenting on several other studies that have considered relatively long time series of the SOI, he noted that there were significant differences in the relationship between the SOI and Australian rainfall between the periods 1930 to 1950 and 1950 to 1980, with a much weaker relationship in the 1930-to-1950 period.

Several recent studies (Hamilton and Garcia, 1986; Elliott and Angell, 1988; Allan et al., 1991; Quinn and Neal, 1992) have examined decadal variability of the Southern Oscillation; it is covered in some detail in Cane et al. (1995), Keeling and Whorf (1995), Rasmusson et al. (1995), and Trenberth and Hurrell (1995), all in this volume. It is sufficient to say here that decadal variations in the Southern Oscillation must be associated with significant decadal variations in the SH circulation.

Time series of the SOI and the TPI since 1896 are shown in Figure 5. There is negligible correlation of the interannual variations between these two time series over the past century, confirming the conclusion of Pittock (1984) that they represent independent modes of variation of the SH circulation. The SOI and TPI show large interannual fluctuations as well as variations on longer time scales. The TPI exhibits decadal variations that are a larger fraction of its interannual variability than does the SOI. The decadal time-scale variations of these two indices must be associated with variations of the SH circulation on these time scales and, almost certainly, with regional climate fluctuations as well.

LONGER-TIME-SCALE VARIATIONS FROM PROXY DATA

It is not possible to use the available surface meteorological observations for the SH to describe variations at century or longer time scales in the SH. The only source of data at these longer time scales is proxy data from tree rings or other sources from a very few sites in the SH. Cook et al. (1991) and Villalba et al. (1989) have used tree rings to reconstruct temperature variations at some SH sites, while Lough and Fritts (1990) have used tree rings to identify

FIGURE 5

Variations of two surface-pressure indices of the SH circulation, with the solid line joining annual values and the dashed line 5-year running averages. Upper panel, Southern Oscillation Index (SOI), normalized Tahiti—Darwin pressure anomalies. Lower panel, Trans-Polar Index (TPI), Hobart-Stanley pressure differences.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

SOI fluctuations over several centuries. The temperature reconstructions show marked variability on decadal time scales. The tree-ring temperature reconstruction for Tasmania by Cook et al. (1991) shows a pronounced warming trend over the last few decades, which is consistent with sea surface temperature variations in the SH and in this region over the same period (Folland et al., 1990). The tree-ring series from Tasmania is discussed further in this volume by Cook et al. (1995).

Other proxy evidence of climate variability in the SH has been obtained from coral growth rings (Isdale, 1984), geological lake-level records (Kotwicki and Isdale, 1991), ice cores (Thompson et al., 1984), river-flow records (Wells, 1987; Whetton et al., 1990) and from a variety of sources (Markgraf et al., 1992). In general, however, it is difficult to directly relate these regional proxy data directly to large-scale climate variability in the SH. They provide an indication of the range of climate variability in the source regions for the data. Further work is required to determine the representativeness of some of these regions for the climate of the whole SH.

SUMMARY

The dominant modes of intermonthly and interannual variation of the SH circulation have been described using results from several studies of grid-point analyses of the SH tropospheric circulation. The two leading modes are primarily zonally symmetric, representing out-of-phase variations of height between middle and high latitudes in one case (called the high-latitude mode) and between the tropics and middle latitudes in the other (called the low-latitude mode). These two modes, which explain almost a quarter of the variance of monthly mean 300 hPa height in the SH in both summer and winter, have large interannual variations. The dominance of these zonally symmetric modes by comparison with wavelike or regional patterns in the SH is in contrast to what is typical in the NH. Wave-train patterns of anomalies are also found in the SH circulation. These include a wave-number-three mode in winter with large amplitude at high latitudes, a wave-number-four mode in middle latitudes in summer, a meridional wave-train structure originating over Australia, and a monopole or meridional wave train over the Pacific Ocean in winter. The two most important processes leading to interannual variations of the SH circulation appear to be the Southern Oscillation and the wave, mean-flow interaction in the SH storm track, which leads to the high-latitude mode.

Since upper-air data and grid-point analyses are available for only a very short period in the SH, surface data must be used to describe decadal variability of the SH circulation. Because of the barotropic vertical structure of low-frequency variations in the SH, the surface-pressure variations are representative of variations throughout the troposphere. Relatively little attention has been paid to decadal variability in the SH. It appears from indices of the two dominant modes of interannual variations, determined using surface data, that the modes show substantial decadal variability over the last 90 years.

There are few observational data that can be used directly to indicate climate variability on century time scales in the SH. The only possible observational evidence for this is proxy data, such as tree rings.

ACKNOWLEDGMENTS

This brief review represents my own perspective on studies of observed variability of the SH circulation. It is incomplete, and I apologize for any omissions. I wish to thank Phil Jones for providing Figure 4 and the station data for generating the time series of the TPI in Figure 5, Rob Allan for allowing access to his detailed bibliography on ENSO, and my many colleagues around the world who have provided stimulating discussions on the SH circulation over a number of years. This review was supported in part by a Dedicated Greenhouse Research Grant from the Australian National Greenhouse Advisory Committee. I am grateful to the U.S. National Research Council for the invitation to prepare this brief review.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
Commentary on the Paper of Karoly

David E. Parker

U.K. Meteorological Office

I was particularly interested in Dr. Karoly's video, which raises several questions. I think it would be possible to verify and improve his land-based analyses, because we have large numbers of marine data that we at the Met Office, and Henry Diaz and others at NCAR, are hoping to put together in a blend of our marine data bank and the COADS data base. The COADS data and ours include pressures and winds, so that they can be checked against one another. They should be a good method of filling in this area of the world, which is not very well covered with observations.

Some of you may also have read Neil Ward's work (J. Climate, 1992) in which he corrects marine winds. The measured winds appear to have become stronger relative to the geostrophic winds calculated from the pressures. So far, this work has gone back only to 1949, although similar work could be carried out for earlier years. It is very labor-intensive, but one of my recommendations is that the Southern Hemisphere surface data base be improved. An improved surface data base, especially in the Southern Hemisphere, might help us better understand the mechanisms of climate. For example, Section C of the 1992 IPCC Science Supplement shows what happens to estimates of trends in oceanic latent heat fluxes if we apply appropriate corrections to the winds.

Unfortunately, many of the upper-air data from Southern Hemisphere rawinsonde stations are missing, e.g., in Brazil, Antarctica, and southern tropical Africa. I would therefore recommend that everything be done to maintain the network, especially in developing countries and on remote islands. The other problem with the upper-air data is that changes of instrument type and changes of operating practice must be compensated for. To do this, we need adequate station histories. These, as well as the data, are being collated in the Comprehensive Aerological Reference Data Set (CARDS) project under Tom Karl and others at Asheville.

Ultimately, of course, the adjusted data should be put together in a fixed-model-based reanalysis, to yield self-consistent analyses. This is being done by the European Centre for Medium-range Weather Forecasting, and data back to 1958 are being analyzed at NOAA/NMC.

And last, the South Pacific Convergence Zone is an intriguing phenomenon, and vital to our understanding of the Southern Hemisphere climate. So, extra effort should be made to measure and understand it.

Discussion

GHIL: You called your first two EOFs high latitude and low latitude, but what matters is where the nodal line is. Perhaps you have pools of air trapped behind the high-latitude and subtropical jets that happen to have different temperature anomalies.

KAROLY: The interaction between the storm tracks and the main jets is a complicated process that can lead to long-lived low-frequency anomalies. The general symmetry of the anomalies is perhaps more characteristic of the Southern Hemisphere because of the topography. But I agree that it could be described in terms of symmetric air-mass contrast between warmer and colder regions.

GROOTES: Ice-core records for different parts of Antarctica might be able to give you an idea of whether this kind of circulation persisted.

SHUKLA: In the video you showed, almost all the major fluctuations around Australia were coming from the south. Was that a data problem?

KAROLY: It looks that way until the 1950s, but then till the 1970s or 1980s they seem to come more from the north and after that there is no obvious signal. I imagine it's a data problem, but the only way to know is to look at the individual station data. I'm hoping to persuade Phil Jones to do some analysis of this.

TRENBERTH: The dominant mode in the Southern Hemisphere is a double jet structure, with the jets alternating in strength. This mode has been simulated by the Canadian Climate Center and by the Japanese, and in fact turns out to be the dominant global mode. It is entirely derived from natural variability, and seems to have no connection to any external forcing.

KAROLY: It's not clear that internal fluctuations in the atmosphere alone can provide enough longevity to extend it into the decadal range. There might be a link with the Antarctic circumpolar current or the distribution of ocean temperature in the Southern Hemisphere that could provide a longer duration. The Max Planck Hamburg model produces a 50-year episode of this mode, as I remember. It is internal to these sorts of models.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Surface Climate Variations over the North Atlantic Ocean During Winter, 1900-1989

CLARA DESER1 AND MAURICE L. BLACKMON2

ABSTRACT

The low-frequency variability of the surface climate over the North Atlantic during winter is described, using 90 years of weather observations from the Comprehensive Ocean-Atmosphere Data Set. Results are based on empirical orthogonal function analysis of four components of the climate system: sea surface temperature (SST), air temperature, wind, and sea level pressure. An important mode of variability of the wintertime surface climate over the North Atlantic during this century is characterized by a dipole pattern in SSTs and surface air temperatures, with anomalies of one sign east of Newfoundland, and anomalies of the opposite polarity off the southeast coast of the United States. Wind fluctuations occur locally over the regions of large surface-temperature anomalies, with stronger-than-normal winds overlying cooler-than-normal SSTs. This mode exhibits variability on quasi-decadal and biennial time scales. The decadal fluctuations are irregular in length, averaging about 9 years before 1945 and about 12 years afterward. There does not appear to be any difference between the wind-SST relationships on the different time scales.

Another dominant mode of variability is associated with the global surface warming trend during the 1920s and 1930s. The patterns of SST and air temperature change between the periods 1900-1929 and 1939-1968 indicate that the warming was concentrated along the Gulf Stream east of Cape Hatteras. Warming also occurred over the Greenland Sea and the eastern subtropical Atlantic. The warming trend was accompanied by a decrease in the strength of the basin-scale atmospheric circulation (negative phase of the North Atlantic Oscillation). In marked contrast to the dipole pattern, the wind changes occurred downstream of the largest SST anomalies. Hence the gradual surface warming along the Gulf Stream may have been a result of altered ocean currents rather than of local wind forcing.

1  

Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, Colorado

2  

NOAA Climate Monitoring and Diagnostics Laboratory, Boulder, Colorado

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
INTRODUCTION

Climate variability on decadal and longer time scales is a subject of increasing interest and relevance. Concern over anthropogenic effects on global climate provides a strong impetus to describe and understand the natural modes of variability of the climate system. In this study, we focus on the low-frequency climate fluctuations over the North Atlantic Ocean since the turn of the century.

The North Atlantic is a region of particular importance to the global climate system. The formation of bottom water at high latitudes of the North Atlantic drives a trans-equatorial thermohaline circulation. Changes in the rate of deep-water production south of Iceland (and hence in the strength of the thermohaline circulation) can have a profound effect on global climate, as can be inferred from paleoclimate data (Broecker et al., 1986; Lehman and Keigwin, 1992) and as is shown by the modeling studies of Rind et al. (1986) and Manabe and Stouffer (1988). Recent ocean modeling experiments indicate that self-sustained oscillations of the thermohaline circulation can occur on decadal and longer time scales (Weaver and Sarachik, 1991).

Low-frequency climate fluctuations in the North Atlantic since World War II have been investigated in several studies. Levitus (1989) used subsurface measurements of temperature and salinity to document a major shift in the ocean circulation in the North Atlantic between the late 1950s and the early 1970s. Knox et al. (1988) showed that the Northern Hemisphere atmospheric circulation experienced an abrupt transition during the early 1960s (see also Flohn, 1986; Shabbar et al., 1988). One of the best-documented examples of low-frequency variability at high latitudes of the North Atlantic is the Great Salinity Anomaly, a freshwater mass that was observed to travel around the subpolar gyre from 1968 to 1982 (Dickson et al., 1988). The freshening of the surface waters was sufficient to halt temporarily deep-water formation in the Labrador Sea (Lazier, 1980). Mysak et al. (1990) discuss the relationship between the Great Salinity Anomaly and interdecadal oscillations in the Arctic climate system.

Studies dealing with the longer (100-year) marine records have tended to emphasize globally averaged surface-temperature variations (cf. Paltridge and Woodruff, 1981; Jones and Kelly, 1983; Barnett, 1984; Folland et al., 1984; Jones et al., 1986c). These studies find that the dominant signal in worldwide temperatures is a warming trend from about 1920 to 1940, with the largest amplitudes at high latitudes.

Bjerknes (1959, 1961, 1962, 1964) investigated air-sea interaction in the North Atlantic on time scales ranging from interannual to interdecadal. Using data from 1890 to 1940, Bjerknes provided compelling evidence that interannual fluctuations in SST are largely governed by wind-induced changes in latent and sensible heat fluxes at the sea surface. However, the long warming trend during the first quarter of this century appears to be linked to a change in the ocean circulation rather than to air-sea energy fluxes (Bjerknes, 1959). Similar conclusions regarding interannual and interdecadal SST variations are reached by Kushnir (1994).

The purpose of this study is to describe the variability of the winter climate over the North Atlantic Ocean, using 90 years of surface marine data. The North Atlantic is the only ocean basin with data coverage that is sufficiently dense for a regional study of climate change since 1900. We focus on the winter season because (1) the atmosphere-ocean coupling is most vigorous and (2) the SST variations reflect significant heat-content anomalies in the upper ocean. We are particularly interested in whether the climate system exhibits preferred time scales of variability, and whether the atmosphere-ocean relationships change with spectral frequency. Our results are based on an objective (empirical orthogonal function) analysis of four components of the climate system: SST, air temperature, wind, and sea level pressure. We will show that the surface climate over the North Atlantic exhibits coherent decadal fluctuations that resemble the variations on interannual time scales. The decadal fluctuations appear to be closely related to sea-ice extent in the Labrador Sea. We will also show that the Gulf Stream was apparently involved in the long-term warming trend during the 1920s to 1930s.

The data and methods are described; the two leading modes of surface-temperature variability during the period 1900 to 1989 and their relation to the atmospheric circulation are documented; and the results are discussed.

DATA AND METHODS
Data

The surface wind, sea level pressure, SST, and surface air temperature data used in this study are from the Comprehensive Ocean-Atmosphere Data Set (COADS), an extensive compilation of weather observations from merchant ships (Woodruff et al., 1987). We have used data gridded by 4° latitude/longitude squares for the period 1900 to 1989. The COADS data are uncorrected for changes in instrumentation, observing practice, ship type, etc. Spurious trends resulting from these changes have been reported in surface wind speed and SST over the tropical oceans (Ramage, 1987; Wright, 1988; Cardone et al., 1990). Our approach to the issue of whether the routine ship-based measurements are sufficiently accurate and homogeneous to allow the detection of real climate signals is to demonstrate physical consistency among independent variables: wind patterns are compared to pressure distributions, and SST fields to surface air temperatures.

Methods

The monthly mean data were converted to anomalies by subtracting long-term monthly mean values for the period 1900 to 1989. These monthly anomalies were then averaged

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

into seasonal anomalies (November to March) to form winter-mean departures from normal. The winter-mean value was considered missing if two or more months in a winter were missing. Missing data have not been replaced.

Empirical orthogonal function (EOF) analysis was used to identify objectively the dominant modes of variability in the North Atlantic climate system during the period 1900 to 1989. EOF analysis was performed separately on the SST, air temperature, sea level pressure, and zonal wind fields. EOFs were calculated using only those grid squares with at least 60 years of winter-averaged data. A minimum of 50 grid squares containing data in each winter was required for reconstructing the principal components (time series of the EOFs).

RESULTS
EOF Modes of Surface Temperature Variability

Figure 1 shows the first two EOFs of SST over the North Atlantic, superimposed on the long-term mean SST distribution. The EOFs are based on un-normalized winter-mean (November to March) anomalies during the period 1900 to 1989. The first EOF (Figure 1a; hereafter referred to as E1 SST), which accounts for 45 percent of the variance, has uniform polarity over the entire basin. The largest loadings occur along the Gulf Stream (as indicated by the tightly packed isotherms in the climatology). The time series of E1 SST (Figure 1b) exhibits a sudden transition from below-normal values to above-normal values around World War II. It is well known that the technique for measuring SSTs aboard ships changed during the early 1940s, and that the bucket temperatures used before World War II were about 0.3° lower than the engine intake samples used later (cf. Wright, 1986; Folland et al., 1984). Over the Gulf Stream this difference can reach approximately 0.8°, due to the rapid evaporative cooling from the bucket samples (Bottomley et al., 1990). Thus, one may be tempted to disregard El SST; however, air temperatures exhibit a similar EOF pattern with a more realistic time series (see below).

The second EOF (Figure 1c; hereafter E2 SST), which

FIGURE 1

(a) EOF 1 of North Atlantic SST anomalies based on un-normalized winter (November-March) means, 1900 to 1989 (bold contours). This mode accounts for 45 percent of the variance. Also shown is the climatological SST distribution (thin contours). (b) Time series of EOF 1 (dashed curve) and smoothed with a 5-point binomial filter (solid curve). The circles denote the winter anomaly, plotted in the year in which November occurs. (c) As in (a) but for EOF 2. This mode accounts for 12 percent of the variance. (d) As in (b) but for EOF 2.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 2

(a) EOF 1 of North Atlantic surface air temperature anomalies based on un-normalized winter (November-March) means, 1900 to 1989 (bold contours). This mode accounts for 21 percent of the variance. Also shown is the climatological air temperature distribution (thin contours). (b) Time series of EOF 1 (dashed curve) and smoothed with a 5-point binomial filter (solid curve). (c) As in (a) but for EOF 2. This mode accounts for 17 percent of the variance. (d) As in (b) but for EOF 2.

accounts for 12 percent of the variance, exhibits a center of action east of Newfoundland at the boundary between the subtropical and subpolar ocean gyres. A weaker center of opposite polarity is located off the southeastern United States. The time series of E2 SST (Figure 1d) exhibits quasi-biennial and quasi-decadal fluctuations, as well as longer-term trends.

The first two EOFs of air temperature, shown in Figure 2, are similar to those of SST, but the order of the modes is reversed. E1 of air temperature (Figure 2a), which accounts for 21 percent of the variance, exhibits centers of action off Newfoundland and the southeastern United States; the correlation between the time series of E1 of air temperature and E2 of SST is 0.76.3 E2 of air temperature (Figure 2c), which accounts for 17 percent of the variance, exhibits maximum amplitude along the Gulf Stream and near western Europe. The time series of E2 (Figure 2d) is dominated by a long warming trend from the 1920s to the 1940s, followed by a cooling trend during the 1950s and 1960s. These trends follow those of global air temperature (Folland et al., 1984). Shorter-period fluctuations are also apparent.

The power spectra of E2 of SST and E1 of air temperature, based on the lag-correlation method, are shown in Figure 3. Both time series exhibit spectral peaks at approximately 10 to 15 years and approximately 2 to 2.5 years, consistent with our visual impression of the time series. E2 of SST also exhibits enhanced power at the lowest spectral estimate. The decadal and biennial peaks in El air temperature and the biennial peak in E2 SST are statistically significant at the 95 percent level (a priori) above the red-noise background spectra. The power spectra of E2 of air tempera-

3  

All of the correlation coefficients cited in this study are statistically significant at the 99 percent level or higher, based on a one-tailed Student's t-test and taking into account the autocorrelation in the time series according to Leith (1973).

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 3

(a) Power spectrum of EOF 2 of SST. The thin dashed lines represent the background red-noise spectrum and its 95 percent confidence limit. (b) As in (a) but for EOF I of air temperature.

ture and E1 of SST (not shown) are red, reflecting the dominance of the long trends before and after World War II.

The similarity of the leading EOFs of the independently measured fields of surface air temperature and SST lends credence to the patterns.

We have examined the sensitivity of the leading EOF patterns of air temperature to temporal filtering. In view of the power spectrum of E1 of air temperature, we pre-filtered the data to emphasize (a) the biennial time scale and (b) the decadal time scale, and then computed EOFs. The first EOF of each of the sets of filtered data (not shown) is nearly identical to the first EOF based on unfiltered data. For the biennial time scale, the first EOF accounts for 26 percent of the variance; on the decadal time scale, it accounts for 29 percent. In view of the low-frequency behavior of E2 of air temperature, we computed EOFs based on lowpass filtered data (half-power point at 5 years). The first EOF of the low-pass filtered data (not shown), which resembles the second EOF of the unfiltered data, accounts for 29 percent of the variance of the low-pass filtered data. The correlation coefficient between the time series of E2 (unfiltered) and El (low-pass) is 0.88. Thus, although the first and second EOFs of unfiltered air temperature are not well separated according to the criterion of North et al. (1982)— they explain 21 percent and 17 percent of the variance, respectively—the sensitivity experiments described above suggest that they are distinct modes of variability.

RELATION OF SURFACE TEMPERATURE VARIABILITY TO THE SURFACE ATMOSPHERIC CIRCULATION
Dipole Pattern

How are the dominant modes of variability in surface temperature related to anomalies in the surface atmospheric circulation? Figure 4a shows the patterns of SST and surface wind anomalies regressed on the time series of E2 of SST (a similar picture is obtained if E1 air temperature is used in place of E2 SST). Note that the polarity of E2 SST is opposite from that shown in Figure 1c and that the SST pattern has been smoothed in space with a 3-point binomial filter. The relationship between the surface wind and SST anomalies is primarily local: stronger-than-normal westerlies and trade winds are coincident with cooler-than-normal temperatures, and southerly wind anomalies are associated with warmer-than-normal temperatures. The negative SST anomalies east of Newfoundland are located slightly upstream of the largest wind anomalies where the air-sea temperature differences are largest. The local nature of the wind/temperature relationships suggests that changes in the air-sea fluxes and wind-induced vertical mixing processes contribute to the formation of the SST anomalies. Ocean modeling experiments by Luksch et al. (1990) and Alexander (1990) confirm this interpretation. A somewhat surprising result is that the wind/temperature relationships are similar on decadal and biennial time scales (not shown).

Is the wind pattern shown in Figure 4a a preferred mode of variability in the atmosphere? Figure 4b shows the second combined EOF of sea level pressure and zonal wind, based on winter means for 1900 to 1989. This mode accounts for 15 percent of the variance in the combined fields. The meridional wind pattern was obtained by regressing the meridional wind anomalies on the time series of the com-

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 4

(a) SST (bold solid and dashed contours) and wind anomalies (arrows) regressed on the time series of EOF 2 SST. Also shown is the climatological SST distribution (thin contours). Note that EOF 2 of SST here has polarity opposite to that shown in Figure 1c; it has been smoothed in space with a 3-point binomial filter. (b) Combined EOF 2 of sea level pressure and zonal wind, based on winter means of 1900 to 1989. The meridional wind field was obtained by regressing meridional wind anomalies on the time series of the combined EOF. This mode explains 15 percent of the variance in the combined pressure and zonal wind fields.

bined EOF. The wind and pressure distributions in Figure 4b are consistent with the geostrophic relation, lending credence to the patterns. The wind patterns in Figures 4a and 4b are similar, indicating that E2 SST is coupled to a dominant mode of variability in the atmosphere. The power spectrum of the second combined EOF of pressure and zonal wind, shown in Figure 5, exhibits peaks at biennial and decadal time scales, similar to those found for E1 air temperature and E2 SST. The circulation mode represented by the combined EOF corresponds to the West Atlantic pattern as defined by Wallace and Gutzler (1981), although the northern center of action in the sea level pressure field is centered about 5° of longitude east of the corresponding feature in the West Atlantic pattern.

Time series of the various surface parameters to the east of Newfoundland (the center of actions of E1 air temperature and E2 SST) are presented in Figure 6. These records have been low-pass filtered with a 3-point binomial filter. Air temperature and SST were averaged over the region bounded by [52°-40°N, 50°-30°W]; zonal wind was aver-

FIGURE 5

Power spectra as in Figure 3a, but for EOF 2 of the combined fields of sea level pressure and zonal wind.

aged over the region within [48°-38°N, 52°-22°W] and the geostrophic zonal wind was calculated from sea level pressure differences between [40°-32°N, 52°-22°W] and [52°-44°N, 52°-22°W]. Note that the air and sea temperatures and the winds and pressures were measured independently, so the comparisons provide a means of assessing data quality. All four parameters exhibit prominent decadal variability in the area east of Newfoundland; the amplitude of the decadal surface temperature (wind) fluctuations is a b o u t 1 . 5 ° C (1.5 m s-1). The correlation between the air temperature and SST time series is 0.88, and that between the measured and geostrophic wind series, 0.80. The high correlations attest to the reality of the decadal signal. The squared coherence between the SST and zonal wind time series (not shown) exceeds the 99 percent significance level at the decadal period, with a phase lag of 180° between the two parameters.

It is apparent from the time series in Figure 6 that the time scale of the quasi-decadal fluctuations was longer after about 1940 than it was before then. This is borne out by calculations of the power spectra for the pre- and post-1944 periods. During the period 1900 to 1944 the decadal spectral peak is centered at 9 years, whereas during 1945 to 1988 it is centered on 12 years (not shown). Thus, the decadal fluctuations are by no means regular.

Gulf Stream Pattern

The dominant features of the time series of El of SST and E2 of air temperature are the cold period of 1900 to 1929 and the warm period of 1939 to 1968 (Figures 1b and 2d). These cold and warm epochs may also be identified

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 6

Time series of winter anomalies east of Newfoundland of (a) SST (solid) and surface air temperature (dashed) and (b) zonal wind (solid) and geostrophic zonal wind calculated from sea level pressure differences (dashed). All curves are smoothed with a 5-point binomial filter. See text for definition of indices.

on the basis of globally averaged temperatures (cf. Folland et al., 1984) and North Atlantic basin-averaged temperatures (cf. Bottomley et al., 1990). The surface temperature, wind, and sea level pressure differences between the warm and the cold period are shown in Figure 7. A minimum of 20 winters was required for the period average in each grid square. The surface wind and sea level pressure fields (Figure 7a) are dynamically consistent, with negative pressure anomalies along 40°N and easterly (westerly) wind anomalies to the north (south). The maximum sea level pressure difference is approximately 3 mb, and the largest wind-

FIGURE 7

Difference between the periods 1939 to 1968 and 1900 to 1929 of winter (a) sea level pressure and surface wind, (b) SST, and (c) surface air temperature. In (a) the contour interval is 0.5 mb, with negative contours dashed. The lowest pressure anomaly is -3mb. Wind scale is indicated in lower right. In (b) light shading indicates values between 0.8°C and 1.0°C; heavy shading indicates values greater than 1.0°C. Contour interval is 0.2°C. In (c) light shading indicates values between 0.6ºC and 0.8°C; heavy shading indicates values greater than 0.8°C. Contour interval is 0.2°C.

speed anomaly is around 1.5 m s-1. This circulation pattern, which resembles the negative phase of the North Atlantic Oscillation (cf. Barnston and Livezey, 1987), represents an overall weakening of the basin-wide wind system.4 Similar

4  

The North Atlantic Oscillation is the dominant mode of variability of the surface atmospheric circulation, accounting for 33 percent of the variance of the winter sea level pressure field during 1900 to 1989 (not shown).

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

results were obtained by Parker and Folland (1988) from historical sea level pressure analyses. The SST (Figure 7b) and surface air temperature (Figure 7c) anomaly patterns are grossly similar: Both exhibit maximum positive anomalies (about 1°) along the Gulf Stream east of Cape Hatteras, and positive anomalies of smaller amplitude in the eastern ocean. There is some indication of a poleward amplification of the air-temperature anomalies, but the data are sparse north of about 55°. (Stations on Greenland, Iceland, and Scandinavia experienced a warming of approximately 2° during the 1920s and 1930s (Jones and Kelly, 1984).) Although the SST (and to a lesser extent the air-temperature) records suffer from discontinuities around World War II (cf. Jones et al., 1986c), the similarity between the air and sea temperature changes lends credence to the patterns in Figure 7. The ''corrected" SST data set of Bottomley et al. (1990) yields a similar pattern, although the warming along the Gulf Stream does not extend as far eastward as in the COADS (not shown).5

The wind/temperature relationships are different from those associated with the dipole pattern discussed above. While there is some evidence of wind anomalies overlying SST anomalies in the eastern portion of the basin, the largest surface-temperature anomalies occur in the west, upstream of the atmospheric-circulation anomalies. Whether the temperature anomalies along the Gulf Stream are connected with changes in the strength or position of the current is unknown. However, it is worth noting that a northward shift of the Gulf Stream of only 50 km would be sufficient to produce the observed warming. An alternative explanation for the warming along the Gulf Stream is a reduction in frequency or intensity of cold air outbreaks. This would occur if the temperatures over eastern North America were elevated or the offshore winds were reduced.

To examine the robustness of the patterns and to test whether the patterns are artifacts of data discontinuities around World War II, we computed the linear least-squares trends during the period 1917 to 1939. The data were first smoothed in time with a 3-point binomial filter to enhance the low-frequency signals. The surface wind and sea level pressure fields (Figure 8a) exhibit a cyclonic circulation trend in the central North Atlantic, similar to the results from the epoch analysis. Over the western North Atlantic, the trend analysis shows easterly wind anomalies, which are consistent with the pressure field, whereas the epoch analysis shows weak westerly anomalies. The trends in SST (Figure 8b) and air temperature (Figure 8c) are similar to the results from epoch analysis; the largest warming occurs

FIGURE 8

Linear least-squares trends during 1917 to 1939 of winter (a) sea level pressure and surface wind, (b) SST, and (c) surface air temperature. In (a) contour interval is 0.05 mb per year, with negative contours dashed. The wind scale is given in the lower right. In (b) and (c) the contour interval is 0.02°C yr-1. Light shading indicates values between 0.04°C yr-1 and 0.06°C yr-1; heavy shading indicates values greater than 0.06°C yr-1.

along the Gulf Stream, over the eastern subtropical Atlantic, and near Iceland. The warming along the Gulf Stream is not as continuous, and the warming in the eastern Atlantic not as pronounced, in the trend analysis as in the epoch analysis. As in the epoch analysis, the largest SST trends along the Gulf Stream occur remotely from the wind trends.

The time series of air temperature along the Gulf Stream [44°-36°N, 76°-42°W] and sea level pressure in the region bounded by [46°-34°N, 48°-20°W] and at Ponta Delgada, Azores [40°N, 28°W] are shown in Figure 9. Surface warming along the Gulf Stream occurred during the period 1920 to 1950 (amplitude about 2.5°C). Note that the temperatures

5  

Geographically and seasonally varying corrections are incorporated in the Bottomley et al. (1990) SST data set. The corrections are based on a simple model of air-sea heat transfer from an uninsulated bucket and are designed to minimize the difference in the annual cycle of SST before and after 1941.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Figure 9

Time series of winter anomalies of (a) surface air temperature along the Gulf Stream [44°-36°N, 76°-42°W], and (b) sea level pressure in the central North Atlantic [46°-34°N, 48°- 20°W] and at Ponta Delgada, Azores [40°N, 28°W]. The curves in (b) are smoothed with a 3-point binomial filter. The solid (dotted) curve denotes pressures from Ponta Delgada (COADS).

since the mid-1950s have been relatively constant. The COADS and Ponta Delgada sea level pressure time series show remarkable agreement: Both indicate that the warming trend was accompanied by a decrease in the strength of the Subtropical High of approximately 4 mb. The pressure time series give the impression that the downward trend in the atmospheric circulation began about 1905 and lasted until 1962 or so, although the steepest continuous decline occurred during 1920 to 1940.

Bjerknes (1959) found a similar pattern of warming along the Gulf Stream during the first quarter of this century (he used differences between 1890 to 1897 and 1925 to 1932 to depict the warming trend). The warming was accompanied by a strengthening of the subtropical anticyclone, in contrast to the results of this study. We have attempted to reproduce the results of Bjerknes (1959) by using data from the COADS. While a warming along the Gulf Stream is hinted at in the COADS, it is much weaker than that depicted in Bjerknes (0.2°C versus 2°C; not shown). The pressure and wind patterns derived from the COADS broadly support Bjerknes's results.

DISCUSSION

One important mode of variability of the wintertime surface climate over the North Atlantic during this century is characterized by a dipole pattern in SSTs and surface air temperatures, with anomalies of one sign east of Newfoundland, and anomalies of the opposite polarity off the southeast coast of the United States. This pattern has been noted in the post-World War II data by Wallace et al. (1990), Cayan (1992b), and Kushnir (1994). Wind fluctuations occur locally over the regions of large surface-temperature anomalies, with stronger-than-normal winds overlying cooler-than-normal SSTs. The atmospheric-circulation anomalies resemble the West Atlantic pattern. This mode (both for surface temperatures and for surface winds) exhibits variability on quasi-biennial and -decadal time scales. The decadal fluctuations are irregular in length, averaging about 9 years before 1945 and around 12 years afterward. There does not appear to be any difference between the wind-SST relationships on the different time scales.

The local nature of the atmosphere-ocean relationships exhibited by the dipole pattern suggests that surface wind anomalies contribute to the formation of SST anomalies by altering the fluxes of latent and sensible heat at the ocean surface and the strength of vertical mixing in the upper ocean. Support for this interpretation is given by the ocean modeling studies of Alexander (1990) and Luksch et al. (1989).

Further research is needed to understand the origin of the quasi-decadal cycles in the North Atlantic ocean-atmosphere system. One possibility is that 90 years is not long enough to establish the significance of decadal cycles, and those that have occurred may be random in the sense that they are due to internal atmospheric processes. The decadal SST variations could then be interpreted as simply a reflection of low-frequency modulation of the high-frequency (synoptic and monthly) wind forcing. It should be noted that long (100-year) integrations of atmospheric GCMs, when subject to fixed SST boundary conditions, exhibit prominent low-frequency (decadal and longer) fluctuations (James and James, 1980; Feldstein, 1992).

A more inviting possibility is that the quasi-decadal time

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

scale is a property of the coupled ocean-atmosphere system. We note that decadal and longer time scale SST variations along the boundary between the sub-polar and sub-tropical ocean gyres are obtained in ocean GCMs as a result of self-sustained oscillations of the thermohaline circulation (cf. Weaver and Sarachik, 1991). In addition, atmospheric GCM experiments indicate that the local atmospheric response to an SST anomaly off Newfoundland is such that weak (strong) winds coincide with high (low) SSTs, reinforcing the original SST anomaly (Palmer and Sun, 1985; Lau and Nath, 1990). The implication of these studies is that observed mid-latitude climate anomalies may result from a positive feedback between the atmosphere and the ocean.

A third possibility is that the decadal atmospheric fluctuations are a response to decadal SST variations outside the North Atlantic. This scenario is difficult to test due to the lack of long marine records in the other ocean basins. Our preliminary results indicate that for the three most recent decadal oscillations, SSTs in the North Pacific are not coherent with those in the Atlantic.

One intriguing link we have found is that between sea ice in the Davis Strait-Labrador Sea region and quasi-decadal fluctuations. Figure 10a shows the time series of winter sea-ice concentration anomalies in the Davis Strait/Labrador Sea region from Agnew (1991), based on data from Walsh and Johnson (1979). The circles denote the winter (December-February) anomaly, plotted in the year in which January occurs. The solid curve shows the data smoothed with a 3-point binomial filter. Decadal variability is evident in the sea-ice record, with peaks occurring in the winters of 1957/ 1958, 1971/1972, and 1983/1984. Figure 10b shows the sea-ice record superimposed on the inverted time series of the second EOF of winter (November to March) SST (see Figure 1). Both the sea-ice and the SST time series have been detrended by subtracting least-squares parabolas based on the period 1950 to 1988, and smoothed with a 3-point binomial filter. It may be seen that the maxima in sea-ice concentration precede the minima in SST by one to two years. The correlation between the two time series is -0.26 with no lag, -0.62 when sea ice leads SST by 1 year, -0.76 when sea ice leads by 2 years, and -0.62 when sea ice leads by 3 years. The strong lag correlations indicate that, on the decadal time scale, winters of heavy sea ice in the Labrador Sea precede winters of colder-than-normal SSTs east of Newfoundland. It is plausible that the sea-ice anomalies in the Labrador Sea are advected southeastward, resulting in colder-than-normal SSTs east of Newfoundland the following (or second) year. Thus, the quasi-decadal cycle in SSTs east of Newfoundland may result in part from low-frequency Arctic sea-ice variations. Mysak and Manak (1989) and Mysak (1991) have also discussed the possible link between sea ice and SST anomalies in the northern North Atlantic, and Mysak et al. (1990) have postulated the existence of an interdecadal Arctic climate cycle involving

FIGURE 10

(a) Time series of winter sea-ice-area anomalies (104 km2) in the Davis Strait/Labrador Sea region from Agnew (1991), based on data from Walsh and Johnson (1979). The circles denote the winter (December to February) anomaly, plotted in the year in which January occurs. The solid curve shows the data smoothed with a 3-point binomial filter. (b) Sea-ice anomalies from (a) (solid curve) superimposed on the time series of EOF 2 of North Atlantic winter (November to March) SST (dashed curve). Note that the SST time series has been inverted. Both the sea-ice and SST time series have been smoothed with a 3-point binomial filter and detrended by subtracting a least-squares parabola based on the period 1950 to 1988.

runoff, sea ice, and SSTs. Further analysis is needed to elucidate the role of Arctic sea ice in decadal climate variability over the North Atlantic.

Another dominant mode of variability of the wintertime

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

surface climate over the North Atlantic during this century is associated with the global surface-warming trend during the 1920s and 1930s. The patterns of SST and air temperature change from 1900 to 1929 and from 1939 to 1968 (or, equivalently, the trends between 1917 and 1939) indicate that the warming was concentrated along the Gulf Stream east of Cape Hatteras. Warming also occurred over the Greenland Sea and the eastern subtropical Atlantic. The warming trend was accompanied by a decrease in the strength of the basin-scale atmospheric circulation (negative phase of the North Atlantic Oscillation). In marked contrast to the dipole EOF pattern, the wind changes occurred downstream of the largest SST anomalies. Hence the gradual surface warming along the Gulf Stream may have been a result of altered ocean currents rather than of local wind forcing.

A similar idea has been put forth to explain the cooling in the North Atlantic from the 1950s to the early 1970s (Levitus, 1989; Greatbach et al., 1991; Kushnir, 1994). The cooling trend was accompanied by a decrease in the transport of the Gulf Stream, as diagnosed from subsurface data by Greatbach et al. (1991). The decrease in Gulf Stream transport was in turn traced to bottom-pressure torque effects rather than wind effects (Greatbach et al., 1991).6

In a recent coupled atmosphere-ocean GCM experiment, Manabe and Stouffer (1988) showed that an intensified oceanic thermohaline circulation is associated with surface warming at high latitudes of the North Atlantic, and with an intensification and poleward shift of the Gulf Stream. The model's atmospheric response to an intensified thermohaline circulation is a weakening of the basin-wide wind system; this is consistent with the reduced baroclinicity. It is tempting to speculate that the climatic trends observed over the North Atlantic during the 1920s and 1930s were due to an intensification of the thermohaline circulation. However, we note that SSTs in the South Atlantic also warmed during this period (cf. Bottomley et al., 1990), contrary to the cooling observed in Manabe and Stouffer's experiment. Further modeling studies are needed to elucidate the role of the ocean circulation in low-frequency climate variability.

ACKNOWLEDGMENTS

We thank Ludmilla Matrosova for her help with the data processing. We are indebted to Harry van Loon for stimulating discussions and to Michael Alexander for his thoughtful review of the manuscript. James Hurrell kindly provided the Bottomley et al. (1990) data set. This work was supported by a grant from NOAA's Climate and Global Change Program.

6  

It should be noted that the spatial pattern of cooling during the 1950s to 1970s was somewhat different from the pattern of warming during the 1920s and 1930s (see Kushnir, 1994).

Commentary on the Paper of Deser and Blackmon

LAWRENCE A. MYSAK

McGill University

Dr. Deser presented a very nice study. I enjoyed it, but I do not have a lot of comments at this stage. Some postwar data sets that have been analyzed show two to three cycles of interdecadal variability in the empirical orthogonal functions, for example. In Dr. Deser's analyses we have now seen decadal-scale variability over a 90-year record, which I for one really appreciate.

I think it is important to emphasize that only the winter data was analyzed. Dr. Deser and I discussed earlier whether this decadal variability shows up in other seasons, such as summer, and thus is a year-round signal. I think the answer is that it is there in the summer, but it is not as strong. It is definitely much more intense in winter. I should perhaps mention here that winter is defined as being from November to March, a five-month period.

The two time scales that I found particularly fascinating in the paper were the decadal and the biennial. While I will not discuss the biennial any further here, I just want to point out some recent work of one of our Ph.D. students at McGill, J. Wang, who also noticed a quasi-biennial cycle in the ice cover in the Labrador Sea and the Baffin Bay region. In addition, he found quite a strong correlation between anomalies in sea-ice extent and the North Atlantic oscillation (which is defined as a winter index).

The decadal time scale Dr. Deser discusses clearly falls between two familiar time scales: the North Atlantic oscillation, which has a period of 7 to 8 years, and the interdecadal cycles in the Arctic, which I shall discuss in my own presentation. Also, many recent modeling studies of the thermohaline circulation in the North Atlantic have shown the existence of internal oscillations with decadal to interdecadal periods.

I should like to list some relevant references, which I believe were not mentioned in the paper. Also, I might remark that many of these papers are by various people at this workshop. One piece of work is the master's thesis

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

done in 1987 by Gordon Fleming at the Naval Postgraduate School. He was supervised by John Walsh, who is here. One of the interesting calculations in the thesis is the cross-correlation analysis of the COADS SST in the Labrador-Greenland-Barents Sea region and the position of the ice edge. Fleming found the two to be inversely correlated, not unlike Dr. Deser's findings for the Davis Strait. However, I believe Fleming was able to show that in some regions the SST actually leads the ice edge and therefore could be used as a predictor.

A recent paper by me and Manak (Atmosphere-Ocean, 1989) studying 32 years of sea-ice concentration data clearly showed the existence of a decadal oscillation of the sea-ice cover in the Labrador Sea and in the Barents Sea. The signal in the former region is the same as that shown in Dr. Deser's Figure 10a. In the Greenland Sea, variations in the sea-ice extent seemed to have a much longer time scale, because there the sea ice was dominated by a big ocean event during the 1960s and early 1970s that we have now associated with the Great Salinity Anomaly.

In that paper Manak and I suggested something that might set up an oscillation of 10 to 12 years, namely, the circuit time of the subpolar gyre in the northern North Atlantic. While 12 to 14 years was the circuit time for the GSA, there was nothing to reinforce it within the subpolar gyre region. Thus we abandoned the idea that there might be self-sustained oscillations of this period in this region.

Other papers by Mysak et al. (Climate Dynamics, 1990) and Marsden et al. (JGR, 1991) showed that salinity and sea-ice extent are highly correlated in the Greenland and Labrador seas, and that there is advection of both ice and salinity anomalies from the Greenland Sea into the Labrador Sea, with an advection time of a few years.

In terms of modeling studies, I first want to mention M. Ikeda (Atmosphere-Ocean, 1990), who developed an ocean model for the decadal oscillations in the Barents-Greenland seas. He emphasized that there were feedbacks between the atmosphere and the ocean in the Barents Sea. In 1991, Weaver and Sarachik published in Atmosphere-Ocean the first model evidence of a decadal oscillation in the thermohaline circulation in the North Atlantic. The physical location of the oscillation is somewhat further south than the Greenland-Labrador Sea, because it is centered between the subtropical and subpolar gyres. In more recent modeling studies with an improved geometry, Weaver and his coworkers have found that the decadal oscillation now extends to 20 years.

There has also been the more recent modeling work by Delworth and his colleagues at GFDL, who find a somewhat longer-time-scale fluctuation in the thermohaline circulation in a coupled atmosphere-ocean model. The period is 30 to 50 years, and the oscillation is found to have a dipole-like structure in the subsurface waters in the central North Atlantic. As for recent data studies looking at decade-to-century time scales, a paper in Climatic Change (Stocker and Mysak, 1992) presents a spectrum of the Koch ice index. (The Koch ice index is the number of weeks per year that ice affects the coast of Iceland.) A spectrum of 360 years of this index shows peaks at 90, 27, and 14 years.

Finally, a well-known time series of SST anomalies in the North Atlantic (45°N to 55ºN) is presented by Bryan and Stouffer in J. Marine Systems. This time series also shows interdecadal variability in the northwest Atlantic. In addition, in the Centre for Climate and Global Change Research's Report No. 91-1 (1991), I suggested that heavy ice conditions in the Greenland Sea lead cold SST anomalies in the northwest Atlantic (i.e., east of Newfoundland) by 4 or 5 years. Similarly, warm SST anomalies there were preceded by generally light ice conditions in the Greenland Sea during the previous 5 years.

Let me now refer you to two figures, Figures 10 and 12a in Stocker and Mysak. The first is a spectrum of the Koch ice index as defined above. During heavy ice conditions in the Greenland Sea, you have a high index; during mild conditions, a low index. This spectrum clearly shows a significant peak at 27 years, which indicates an interdecadal climate cycle in the northern North Atlantic. The second, a spectrum of the famous Central England temperature, shows peaks at around 15 and 7 years and an interdecadal peak at 24 years.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
Discussion

NORTH: I wonder whether EOFs are the best way of describing the data. True, they are objective, but can you tell from analyzing them whether the Gulf Stream's temperature is changing or whether it is just shifting location?

DESER: If you look at the anomalies during the cold 30 years, or one of those decades, the warming first appears near Cape Hatteras. Then you see anomalies along the Gulf Stream east of Cape Hatteras, then east of Newfoundland, then in higher latitudes. A northward shift of about 50 km would be sufficient to produce the observed warming, and seems to me to be the simplest explanation.

MCWILLIAMS: The Gulf Stream warming pattern in your last figure is much broader than 50 km. While a mechanistic interpretation is tempting, an advective supply of heat from the Gulf Stream would be related to wind driving, whereas instead the forcing of the anti-cyclonic gyre is actually weakening.

GHIL: I just wanted to comment on the causes of the decadal variability. It seems to me that certain phase differences between fluctuation or oscillation are clear indications of coupling. A coupled atmosphere-ocean phenomenon seems the most plausible to me.

CAYAN: Have you looked at SST minus air temperature together with your Gulf Stream pattern? I wonder whether part of the reason for the change could be a change in the air mass distribution over the Gulf Stream. In extratropical regions the sensible flux is driven a lot by the temperature lapse rate, and to some extent by the wind.

DICKSON: I should like to show a slide, my Figure 14, that suggests two things. The first is that temperature and salinity show similar patterns, and I think were both involved, meaning that both the circulation and the ecosystem might have had large changes. The second is that the century-long trend covered a much larger area of the North Atlantic than we've been talking about—right up to the ice in the north, in fact.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

Seasonal-to-Interannual Fluctuations in Surface Temperature over the Pacific: Effects of Monthly Winds and Heat Fluxes

DANIEL R. CAYAN1, ARTHUR J. MILLER1, TIM P. BARNETT1, NICHOLAS E. GRAHAM1, JACK N. RITCHIE1, AND JOSEF M. OBERHUBER2

ABSTRACT

Monthly heat fluxes and wind stresses are used to force the Oberhuber isopycnic ocean general-circulation (OPYC) model of the Pacific basin over a two-decade period from 1970 to 1988. The surface forcings are constructed from COADS marine observations via bulk formulae. Monthly anomalies of the fluxes and stresses are superimposed upon model climatological means of these variables, which were saved from a long spin-up. Two aspects of this work are highlighted, both aimed at a better understanding of the atmosphere-ocean variability and exchanges and at diagnosing the performance of the OPYC model in simulating monthly to decadal-scale variability. The first is the evaluation of the data used to force the model ocean, along with its relationship to other observed data. The second is the diagnosis of the processes revealed in the model that are associated with sea surface temperature (SST) variability, including their seasonal and geographic structure.

Although both random and systematic errors arise from the marine data and the bulk formulations, large signals in the air-sea fluxes are nonetheless consistent with the large-scale atmospheric circulation anomalies over the Pacific. This signal is large in a composite prepared from months with similar circulation modes. Also, latent and sensible heat-flux anomaly patterns correspond well to those of SST anomaly tendencies. Considering short-period variations, SST anomaly tendencies have typical magnitudes of 0.3°C mo-1. These are associated with monthly mean flux anomalies having typical magnitudes of 50 W m-2 and are consistent with observed mixed-layer depths. Decadal anomalies have much smaller magnitudes, perhaps reduced by two orders of magnitude, and it is here that the signal-to-noise problem is more severe. The forcing terms are generally products of variables, so realistic means and fluctuations of these variables are crucial for a successful simulation.

The 19-year simulation of the Pacific basin by the monthly marine data-forced OPYC model displays good skill in reproducing SST variability. These results represent the first

1  

Scripps Institution of Oceanography, University of California, San Diego, La Jolla, California

2  

Max-Planck-Institut für Meteorologie, Hamburg, Germany

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

hindcast of which we are aware that uses both observed total heat-flux and wind-stress anomalies as forcing for such a long time interval. There is close agreement between the model SSTs and those observed in many regions of the Pacific, including the tropics and the northern extratropics. Besides performing credibly on the monthly time scale, the model captures the essence of low-frequency variability over the North Pacific, including aspects of a marked basin-wide change that occurred in 1976-1977. In the model's detailed heat budget, the anomalous air-sea heat fluxes, entrainment, and to a lesser extent horizontal advection, force thermal-anomaly changes in the mixed layer. Each of these components was apparently involved in the 1976-1977 decadal SST shift.

INTRODUCTION

Our purpose is to describe monthly to decadal variations in surface variability over the Pacific basin. The results presented combine a historical set of monthly marine observations and a two-decade numerical simulation (Miller et al., 1994a,b) that uses the Oberhuber (1993) isopycnic ocean general-circulation model.

The first part of the paper is an analysis of observed monthly surface marine atmospheric variability in relation to sea surface temperature (SST) over the Pacific Ocean. This observational evidence provides background to the second section of the paper, in which monthly wind stresses and heat fluxes are used to force an ocean general-circulation model (OGCM). In the observational section, we concentrate on the variability of the latent and sensible heat fluxes, since they provide a good example of the characteristics and problems involved in bulk formulations using the marine data. The fluxes exhibit monthly variability with large-scale organization, and their effect on the ocean can be detected in the monthly fluctuations of the SST anomaly field.

The second part of the paper reports on applying the marine data parameterizations to a simulation over two decades (1970-1988) of the Pacific Ocean basin using the OPYC OGCM. This run is forced by monthly mean wind stress and fluxes derived from surface marine observations, which are introduced in the first section. In assessing this run, we compare the model SST anomalies with observations, particularly their low-frequency variability. Further insight is provided by the model upper-ocean heat budget, which is not well described by observations.

Aside from diagnosing mechanisms causing monthly SST variability, one motivation for this simulation was to better understand a strong regime-like change in SST that occurred in the North Pacific basin during the mid-1970s (Douglas et al., 1982; Nitta and Yamada, 1989; Trenberth, 1990; Miller et al., 1994a,b; Graham, 1994; Trenberth and Hurrell, 1995, in this volume). Although thorough observational documentation of decadal-scale variability is lacking, glimpses from recent historical episodes suggest that this "gray area" (Karl, 1988) of the variability spectrum contains important climate effects. The actual shift was identified in fall and winter of 1976-1977 when SSTs in the central North Pacific cooled markedly and SSTs along the west coast of North America warmed (Venrick et al., 1987; Trenberth, 1990; Ebbesmeyer et al., 1991). In the atmosphere, the shift involved a basin-scale deepening of the wintertime Aleutian Low System, and appears to have been at least partially instigated by forcing from the tropical Pacific (Graham, 1994; Trenberth, 1995, in this volume), although an alternative theory involves sea-air feedback and advection of thermal anomalies by the North Pacific subtropical gyre (Latif and Barnett, 1994). Accompanying changes in many other physical and biological variables in the North Pacific basin and around its margin were also noted (Venrick et al., 1987; Ebbesmeyer et al., 1991).

The bulk formulae parameterized heat fluxes, calculated from standard marine surface meteorological observations, are the only means of estimating a multi-year time history of the heat exchange over broad regions of the oceans. Similarly, SST has been measured routinely by merchant ships for several decades, and, in the absence of a comprehensive temperature-versus-depth set in the upper ocean, SST is used to infer the variability of the upper-ocean heat content. Although the ocean heat-content structure can be complex, during winter in the extratropics the upper ocean is quite well mixed and SST is a good indicator (White and Walker, 1974).

As is brought out in the section below, there is a rich variability, superimposed upon the climatological mean, that involves both atmospheric and oceanic fluctuations. Three tests of the surface heat budget are discussed here. The first two, which study the data directly, are to relate monthly anomalies of the latent and sensible fluxes to the anomalous atmospheric circulation and to anomalous tendencies of the SST field. The third is to incorporate the monthly bulk-formula wind stress and heat flux as forcing of an extended OGCM run and to compare the simulated versus the observed SST fields. These results represent the first hindcast of which we are aware that uses both observed total-heat-flux and wind-stress anomalies as forcing for such a long time interval.

OBSERVATIONAL EVIDENCE FOR ANOMALOUS AIR-SEA HEAT EXCHANGE
Data

The primary data employed in the observational section of this study, as well as in the model forcing, are gridded

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

monthly mean ship-observed marine surface data for 1946-1986 from the Comprehensive Ocean-Atmosphere Data Set (COADS) (Slutz et al., 1985; Woodruff et al., 1987). From this set, several variables are employed, including the scalar wind or wind speed w, west-to-east and south-to-north horizontal surface wind components u,v, sea level pressure (SLP), specific humidity qa, air temperature T, and sea surface temperature SST. Derived variables include the products {w?T}, {w?q}, {wu}, and {wv}, where the monthly averages of products of the individual observations are indicated by { }. These products are involved in parameterizing the sensible and latent heat flux, and the horizontal wind-stress components, respectively.

To reduce errors in the data fields and condense the number of data in the analyses, the COADS 2°-gridded "Monthly Summaries Trimmed" data were averaged onto 5° squares, centered at 5° latitude-longitude intersections. This averaging was applied to the anomalies of the 2° squares, rather than their values, to avoid spurious 5° anomalies caused by poor sampling within the domain. Long-term means employed were calculated over 1950 to 1979. Details are provided in four papers by Cayan (1990; 1991, hereafter designated DC1; 1992a, designated DC2; and 1992b, designated DC3).

Bulk Formula Parameterizations

Latent and sensible heat exchange at the ocean surface is an important mechanism for venting ocean heat absorbed from solar radiation to the atmosphere. Globally, approximately 60 percent of the solar radiation absorbed at the earth's surface is released by latent and sensible heating, primarily from the ocean (Sellers, 1965). Gill and Niiler (1973) invoked scaling arguments in the governing equations to infer that, over large scales (= 1000 km meridionally and = 3000 km zonally), anomalous open-ocean surface temperature changes are dominated by changes in heat fluxes through the surface, rather than by advective influences. Since they depend on both oceanic and atmospheric conditions, the latent and sensible fluxes vary strongly over space and time. In the extratropics, latent and sensible fluxes are greatest in fall and winter when the near-surface vertical gradients of humidity and temperature are largest and wind speeds are highest (Esbensen and Kushnir, 1981; Isemer and Hasse, 1985, 1987).

The bulk aerodynamic formulae used for the latent and sensible heat fluxes are

and

where Ql is the latent flux, Qs is the sensible flux, L is the latent heat of evaporation of water, and cp is the specific heat of air at constant pressure. Again, w is the wind speed, ?q is sea surface saturation minus air specific humidity, and ?T is SST minus air temperature. CE and CH are the transfer coefficients for latent heat and sensible heat, respectively, and are taken to be the modified Bunker coefficients given by Isemer and Hasse (1987). Unlike their mean values (Blanc, 1985), the anomalies of the fluxes are not particularly sensitive to the choice of exchange coefficients. A comparison of monthly anomalies obtained with these exchange coefficients to those from two other schemes yielded strong agreement (DC2).

The sense of the latent and sensible fluxes adopted throughout the paper is positive for heat entering the ocean from the atmosphere.

Errors in the Latent and Sensible Fluxes

Because of uncertainties in the bulk formulae and because the marine data have inaccuracies and are non-uniformly measured, the parameterized flux estimates contain errors. Although it is not possible to determine precisely the spatial and temporal characteristics of errors in the bulk-formulae-derived heat fluxes, estimates of their magnitudes have been attempted. In many regions the aggregate of the errors in the fluxes considerably exceeds the 10 W m-2 tolerance that has been targeted for climate studies (Taylor, 1984). Nevertheless, it will be shown that important climate variability (monthly to decadal time scales) is still contained in the flux estimates. A review of individual monthly maps of the fluxes finds that several have anomalies with magnitudes of 50 W m-2 or greater, a "signal" which is well beyond reasonable error estimates. From our observational and modeling experience, the character of this error, as much as its magnitude, is crucial. For example, a 10 W m-2 error is probably acceptable if it has a high-frequency, random variability, but it is intolerable if it has low-frequency (decadal-scale) changes.

Both random and systematic errors in marine observations and errors in the bulk formulae contaminate the monthly mean fluxes (Weare, 1989; Taylor, 1984; DC1). Random measurement errors and errors from incomplete sampling of the natural weather fluctuations are reduced by averaging several observations together. Because the mean of the fluxes was removed and the resulting anomalies are generally only a fraction of the mean, the non-time-varying biases are reduced considerably. The largest random errors in the flux estimates appear to be caused by sampling variability, and to a lesser extent by observation errors (DC1). This error was estimated by a Monte Carlo sampling exercise (DC1) where the mix of observations entering particular flux estimates within well-sampled regions was repeatedly re-sampled. The uncertainty of the monthly mean for well-sampled 5° squares (50 or more independent samples per month) is reduced to approximately 40 W m-2 or less for

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

the latent flux and to 20 W m−2 or less for the sensible flux. The density of observations is relatively high over the extratropics of the North Pacific and North Atlantic, with many 5°squares having 100 or more observations per month. However, the density is marginal-to-poor over the tropics and in the Southern Hemisphere, with fewer than 20 observations per month at many 5° squares. In these sparsely sampled regions, random errors in the monthly flux estimates are likely to be substantial. Consequently, in forcing the OGCM in the region between 20°N and 20°S and data-poor regions south of 20°S, we have resorted to a Newtonian damping scheme for the net flux anomalies.

Systematic errors are caused by biases in the bulk formulae and in the observations, and are not automatically reduced by averaging. Time-varying biases in the fundamental observations caused by changes in instrumental practices are probably a significant source of error. These errors are difficult to quantify, but the marine data have "global" (over the available COADS-covered region) trends in w and ΔT, which translate to changes of approximately 0.8 m s−1 and −0.2°C over 1950 to 1986. Because these seem to be at least partially caused by instrument and calibration changes over the period, the fluxes were adjusted to remove the effects of the global average linear trends. This adjustment reduced the linear change in the latent and sensible fluxes over 1950 to 1986 by about 20 W m−2 and 5 W m−2, respectively (DC1 and DC2). This adjustment has only minor influence on the high-frequency variability, but it does significantly affect the multi-year numerical model run, because the effects are cumulative. More discussion of the systematic errors in the marine data is provided below.

Anomalous Variability of the Heat-Flux Components

The dependence of the flux anomalies on the mean conditions and the anomalies of the fundamental surface variables was explored in DC1. For the monthly latent flux anomaly

where an overmark indicates the long-term monthly mean and a prime indicates the monthly anomaly. For monthly means, most of the variation comes from w and Δq, not from the exchange coefficients (DC1). The last term on the right-hand side of (3) does not contribute strongly to latent flux anomalies, because w and Δq monthly anomalies are not locally well correlated over most of the oceans (DC1). A similar relationship involving ΔT instead of Δq describes the monthly sensible flux anomalies.

Equation (3) shows that the joint behavior of mean values and anomalies of wind speed w, the ocean-surface saturation humidity-air humidity difference Δq, and the sea-air temperature difference ΔT must be included to determine the flux anomalies. For atmosphere-ocean models, this interplay between the mean and anomaly components is a challenge. It is not enough to simulate anomalies of w, Δq, and ΔT; their mean fields must also be properly represented.

There are strong seasonal and geographical modulations of the flux anomalies. In the extratropics the variance is greatest in fall and winter when w, Δq, and ΔT have largest mean values and strongest variability. The largest contributions to the extratropical variance of Q1 anomalies involve . Anomalies of Qs become important north of about 35°N, where their variance is dominated by the term involving . In the extratropics, the variance of each of these components is maximum along the western side of the basin, because strong ocean boundary currents and large air-mass differences between the upwind land mass and the sea cause great contrasts. Latent flux anomalies in the tropics often involve greater contributions from than do those in mid-latitudes, although the eastern tropical Pacific has relatively large contributions from .

The Δq and ΔT monthly anomalies are correlated (cooler air is usually drier), especially in the extratropics, so the latent and sensible anomalies usually reinforce each other, yielding relatively large net heating anomalies. In the extratropics, flux anomalies are linked to the local wind direction, but in the tropics they are usually not (DC2). North of about 15°N, the largest positive anomalies are associated with northerly to northwesterly winds; they probably result from meridional advection of atmospheric humidity and temperature, and also from changes in the strength of the westerly winds. In the tropics, there is little relationship between wind direction and the flux anomalies, since horizontal gradients of humidity and temperature are weak and the wind direction is relatively steady.

How does the variability of the anomalous latent and sensible fluxes compare with that of the radiative fluxes? The balance of the heat-flux anomaly terms is distinctly different from that of the mean terms. Combined anomalies of Ql and Qs often exceed 50 W m−2 over regions several hundred kilometers in extent (DC1 , DC2, and DC3). These estimates indicate that monthly anomalies of Ql and Qs are usually larger than those of the radiative components. Comparing the variances of Ql and Qs, with those of bulk-formulae-estimated net solar radiative flux QSW and infrared flux QIR, DC1 found that: (1) Ql and Qs, anomalies dominate the monthly anomalous surface heat budget during winter in the extratropics north of 30°N, (2) Ql anomalies dominate from about 15°N to about 30°N, and (3) Ql and QSW anomalies are about equally large from 15°N to 15°S.

Flux Anomalies and Atmospheric Circulation

Ql and Qs anomalies have regional-to-basin-scale coherence both the North Atlantic and North Pacific. The first four empirical orthogonal functions (EOFs) of the sum of the Ql and Qs anomalies (Q'l+s) account for about half of the total variance of this field in each basin (DC1). That these

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

patterns relate to the anomalous atmospheric circulation is evidence for a realistic flux-anomaly signal. The dominant atmospheric-circulation anomaly modes in the North Atlantic and North Pacific, represented as EOFs of the sea level pressure, produce systematic flux-anomaly patterns (DC2). The first SLP EOF in the North Pacific features a large pressure anomaly in the central North Pacific, and is associated with the well-known ''Pacific-North American" (PNA) pattern. Three correlation fields linking the PNA to surface variables are shown in Figure 1: the correlations between winter-month PNA time amplitudes and gridded fields of w (wind speed), Δq, and Ql+s anomalies. This set of maps, and others in DC2, show that the anomaly fields of w, Δq, and latent-plus-sensible flux are remarkably consistent with major SLP anomaly patterns over the North Pacific and North Atlantic. Negative SLP anomalies favor positive w anomalies to their south and negative w anomalies to their north, associated with shifts in storm tracks and changes in the mean wind field. Negative SLP anomalies favor positive Δq (and ΔT) to their west, and negative Δq to their east, probably because of meridional advection of air temperature and humidity. The flux anomalies are a hybrid of these patterns, with enhanced sea-to-air fluxes to the southwest and diminished fluxes to the east of negative SLP anomalies.

For the strong Aleutian low phase of the PNA, the anomalous wind-speed field is dominated by a single zonally oriented high-wind patch to the south of the Low in the central North Pacific, with out-of-phase tendencies to the north and the south. Δq and ΔT anomalies are arranged in pockets to the southwest (positive anomalies) and to the east (negative anomalies) of the low. Farther afield, a positive anomaly center appears in the Gulf of Mexico, reflecting the well-known downstream teleconnection of deep troughs and cold dry air outbreaks leading to positive ΔT and Δq anomalies over the southeast United States. The pattern has a broad center of positive anomalies to the southwest of the Aleutian Low, and a region of negative anomalies east of the Low along the West Coast. In the section of this paper that reports on the ocean model, it is shown that low-frequency changes in the PNA between the 1960s and 1980s were associated with marked variations in heat flux and wind stress, which produced striking variability in the upper-ocean thermal structure.

Flux Anomalies and SST

Does the heat flux drive the SST, or does the SST drive the heat flux? The relationship between SST and the climatological mean of total latent and sensible heat flux appears to support two different points of view. On one hand, the fluxes may be driven by the SST: High ocean-to-atmosphere fluxes are found in the tropics where there is warm water and high saturation-vapor pressure, and low fluxes occur

FIGURE 1

The correlation coefficients (× 100) between time series of the amplitudes of Pacific-North American EOF and our all-grid-point time series of w' (above), Δq' (middle), and Q'l+s, (below) in the North Pacific for the winter months, 1950-1986. Solid contours show PNA pattern. Significant correlation regions are superimposed upon the EOF patterns; shading indicates regions where the magnitude of the correlation is ≥0.3. Significant positive/negative correlations are shown by stippling, hatching in upper two frames; and by hatching, stippling in lower frame.

in high latitudes and along the eastern side of the ocean basins where surface temperatures are cool. Concerning the anomaly relationships, there is evidence for the SST-forcing-flux mechanism in the warm-season extratropics and in the tropics (Cayan, 1990; Liu and Gautier, 1990).

On the other hand, in the extratropics the fluxes appear

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

to force the temperature (Gill and Niiler, 1973; DC3): SST is coolest in late winter following the highest ocean-to-atmosphere fluxes. For the flux-forcing-SST case to be valid, the SST anomaly must play a negligible role in determining the flux, so that the flux is driven by atmospheric conditions—i.e., w, Ta, and q variations dominate the variations of SST. Indirectly, this scenario has been investigated by relating SST anomalies to the large-scale atmospheric circulation (Namias, 1972; Davis, 1976; Wallace et al., 1990). While all of these studies established a definite connection between the monthly or seasonal mean circulation and the SST anomalies, they could only infer the role of anomalous heat flux in forcing SST anomalies. When atmospheric fields are related to contemporaneous monthly SST anomalies, it is difficult to determine whether SST is forcing the atmosphere, both fields are being forced by some other agent, or the atmosphere is forcing the SST.

In testing the phase relationship between anomalies of the estimated flux and SST, the dominant linkage exhibited in the extratropics is that the latent and sensible flux anomalies force the SST anomalies, rather than SST anomalies' forcing the flux anomalies (DC3). Historical time series of flux data provide the basis for directly testing the forcing. Correlations between bulk formulae fluxes and anomalous temperature variations have been documented over the North Pacific by Clark et al. (1974) and Frankignoul and Reynolds (1983), and in the western tropical Pacific by Meyers et al. (1986). As was reported in DC3, the latent and sensible heat-flux anomalies proved to strongly affect monthly changes in SST anomalies over a large portion of the world oceans, in particular the North Atlantic and the North Pacific. A simple thermodynamic model was adopted to relate the flux anomalies to the tendency (time rate of change) of the SST anomalies. More discussion of the full mixed-layer thermodynamic equation is given below, and a comprehensive treatment is provided by Frankignoul (1985). Note that the temperature equation predicts the tendency of the SST anomaly , not the anomaly itself; the tendency is represented here by its finite-difference forms, . The flux parameterizations do not contain knowledge of the SST tendency, so the relationship of fluxes to SST-tendency anomalies is an independent test of the influence of the fluxes.

Having a higher-frequency character than the anomaly itself, has nearly as many independent samples of as there are months; four decades of records contain approximately 120 independent December-January-February samples.

Spatial Distribution of Flux versus SST Tendency Anomalies

The geographical distribution of the correlations between Q'l+s and observations for winter months is examined in Figure 2. The 0.3 level is used as a threshold of statistical significance. Meaningful correlations on this map are almost everywhere positive, as they were for zonal averages of the two fields (DC3). Strongest correlations (= 0.5) are found mostly between about 25°N and 40°N within the anticyclonic subtropical gyres of the two oceans. For the North Pacific, strongest correlations are east of 180° and extend to the California Current. Though significant in many locations, correlations are weaker in the western North Pacific and from the tropics to 30°S. In the central North Atlantic, the flux and anomalies have strong correlations in the central subtropical gyre as well as in the high-variance western North Atlantic region. Seasonally, most of the regions have strongest correlations in winter.

Near the equator between 5°N and 5°S in all three basins, Q'l+s and are not well correlated. Correlations between zonal average flux and SST anomalies (DC3; not shown) indicate that in the tropics, the flux and SST anomalies (not the tendencies) tend to be negatively correlated. This suggests that the flux is driven by SST, presumably because equatorial SST anomalies are governed more by internal ocean processes than by the air-sea heat exchange.

versus Fluxes during Strong Atmospheric-Circulation Modes

The organization of the latent-plus-sensible flux anomalies (Q'l+s) by the anomalous atmospheric circulation can be used to test the consistency of large-scale links between the flux and SST tendency anomalies . If the SST tendency anomalies are caused by the fluxes, they should have corresponding patterns. In DC3, the organization by the circulation was exploited by using the dominant SLP EOF modes as an index to compare flux-anomaly and corresponding SST-anomaly tendency patterns.

To identify strong winter-circulation months, extreme positive and negative EOF amplitudes were chosen for each SLP EOF (DC2 and DC3). Using this criteria, several (10-30) months of each of the extreme EOF modes (positive and negative amplitudes) were selected. Composites of Q'l+s and were formed by averaging the fields during the respective extreme months. For brevity, the composites were expressed as the difference between averages of positive (strong) and negative (weak) phase months of the SLP EOFs.

In both northern oceans, the patterns of are well aligned with the flux anomaly patterns. In the North Pacific, composite differences and Q'l+s differences corresponding to positive-minus-negative extremes of the PNA pattern are shown in Figure 3. Remarkably, the flux and signatures of the PNA are marked out more than halfway around the hemisphere. Major centers of are closely matched to those of Q'l+s, which confirms that

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 2

Correlation coefficients (×100), mapped for global ocean Q'l+s vs. at each grid point, for the winter months, 1946-1986. Contours at 0.3, 0.5, and 0.7. Light and heavy shadings indicate correlations =0.3 and =0.5.

FIGURE 3

Difference of composites of Q'l+s (W m-2; left) and (°C mo-1; right) associated with positive (deep Aleutian Low phase) vs. negative extremes of EOF I of North Pacific SLP. Shading indicates regions where the difference exceeds the 95 percent confidence limit using a two-tailed t-test.

they are causally linked. At the centers, differences exhibit magnitudes of 0.2°C mo-1 to 0.8°C mo-1. Corresponding flux differences range from about 30 to 120 W m–2 Over the North Pacific, the PNA incorporates a swath of positive Q'l+s differences across mid-latitudes (centered at 40°N) from Asia to 140°W, and an arc of negative Q'l+s differences extending from the subtropics at 20°N, 160°E to the eastern North Pacific border, from California to the Gulf of Alaska. Downstream, the strong downstream arm of the PNA produces negative Q'l+s differences in the Gulf of Mexico and the eastern seaboard of the United States, and positive anomalies in the subtropical North Atlantic (20°N, 50°W) and east of Newfoundland. Throughout the North Pacific and North Atlantic, the PNA-associated differences mirror the flux differences, having centers co-located with the flux differences. Magnitudes of the corresponding flux and SST anomaly tendencies are consistent with typical observed mixed-layer depths, approximately 100 to 250 m in winter and 10 to 40 m in summer. Similar seasonal behavior and the relation between the flux and SST tendency are replicated in the OGCM, as presented in the following section.

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×
OCEAN MODEL

In this section, we report on an extended run of the OPYC OGCM, which was forced by the wind stress and fluxes derived from the surface marine observations (see Miller et al., 1994a,b). The OPYC model, developed by Oberhuber (1993), consists of eight isopycnal interior ocean layers fully coupled to a surface bulk mixed-layer model, the latter also including a sea-ice model. The interior layers (fixed potential density) and the mixed layer (variable density) have prognostic thicknesses that vary as a function of space and time; the interior layers may have zero thickness, and the mixed layer has a minimum thickness of 5 m. The interior potential densities are specified in such a way as to yield increased vertical resolution (thinner layers) in the thermocline and less in the deep ocean (thick layers). The low horizontal resolution perforce disallows mesoscale eddy variability, which is probably unimportant in the evolution of large-scale SST anomalies. The ocean model solves the full primitive equations for mass, velocity, temperature, and salt for each layer in spherical geometry with a realistic equation of state. The domain is the Pacific Ocean from 70°S to 65°N and 120°E to 60°W. The grid resolution is 77 by 67 points. Resolution is enhanced near the equator and near the eastern and western boundaries. Open-ocean resolution in the middle latitudes is 4°, which is suitable for examining the large-scale variability. The conditions on horizontal solid boundaries are no-slip for velocity and thermally insulating for temperature. The model's Antarctic Circumpolar Current, however, has periodic boundary conditions and unrealistically connects to itself from 60°W to 120°E (half the global circumference). The surface boundary conditions for interior flow are determined by the bulk mixed-layer model, which is forced by the atmosphere. Frictional drag acts between each layer but most strongly along the bottom boundary, which has realistic topography. Horizontal Laplacian friction and diffusion with variable coefficients are also included.

A full discussion of the dynamics is given by Oberhuber (1993), who used the model in Atlantic Ocean modeling studies, and by Miller et al. (1992, designated MOGB below), who used an earlier version of this model for tropical Pacific Ocean circulation studies. Besides the differences in geometry and the forcing functions to be described, the present model differs from MOGB in the following ways: A horizontal finite-differencing scheme based on Bleck and Boudra (1981) that conserves both enstrophy and potential vorticity is implemented in this version of the model. In addition, for a more realistic mixed-layer depth (MLD) in the middle latitudes, the mean turbulent kinetic energy (TKE) input into the mixed-layer equation for entrainment velocity has been altered to provide for month-to-month variability, and is lower than that of the MOGB model. The net result is to reduce the MLD in the middle and high latitudes of the North Pacific relative to that of MOGB, with the present values being more typical of those observed.

Concerning the OPYC ocean general-circulation model, it is important to point out some potential defects that may obscure our interpretations (Miller et al., 1994a). First, horizontal advection may be underestimated because the model SST climatology is too weak, especially in the Northwest Pacific. Second, the Kuroshio is not well resolved, so the model cannot generate strong western boundary currents. This will impair its ability to advect SST anomalies to nearby open ocean regions. Third, a warm bias in the model winter SST field (approximately 4°C too warm) in the northern North Pacific results in an overly stable stratification that diminishes the effects of entrainment. Last, salinity variations were not included, and these may be important, particularly in high latitudes. Although the OPYC OGCM's rather low resolution has excluded the effects of oceanic mesoscale variability on SST anomaly generation, we expect that such anomalies would have spatial scales too small to provoke important reactions in the large-scale atmospheric fields (Klein and Hua, 1988; Halliwell and Cornillon, 1989; Miller, 1992).

Forcing Functions and Model Runs

Since no ocean model is perfectly realistic, any model forced by observed total heat fluxes (without any feedback) will establish an oceanic temperature climatology that will depart from the real observations. To circumvent this problem, we forced the model with observed anomalies of heat fluxes superimposed upon the model climatology rather than with the total observed flux fields. This scheme preserves the model ocean climatology, as follows. We first establish the model climatology by forcing with observed monthly long-term mean wind stresses, TKE input, and total heat fluxes derived from bulk formulae using monthly long-term mean atmospheric observations combined with model ocean temperature. After the oceanic system has reached an acceptably equilibrated state (gauged by the absence of drift in mean SST and MLD), monthly means of pertinent fields are saved for use in the later experiments.

Using the same strategy as that just discussed, the oceanic salinity field is stabilized to climatological average observations. However, the equilibrium salinity is very sensitive to the evaporation-minus-precipitation (E - P) field, which is not available at present from observations. Hence, we use Newtonian relaxation to the observed annual mean surface salinity field compiled by Levitus (1982) rather than attempting to specify E - P.

Spin-up

During a 35-year-long spin-up period, the model was forced by monthly long-term mean fields of atmospheric

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

wind stress, interpolated from analyses made by the European Centre for Medium-range Weather Forecasting (ECMWF), and total surface heat fluxes (latent, sensible long-wave radiation, and insolation), derived from bulk formulae using model SST and observed atmospheric fields of air temperature (from ECMWF analyses), humidity (COADS), wind speed (ECMWF), and cloudiness (COADS) as inputs. These long-term mean fields, described in Oberhuber (1988), are distinct from the flux and wind-stress anomaly fields, which are formed from individual monthly COADS means as discussed below. Salinity in the surface mixed layer is Newtonially relaxed to Levitus's annual mean observed salinity field so that no evaporation or precipitation fields are needed in the continuity equation.

During the spin-up period, year-to-year changes in SST and mixed-layer depth were monitored to help determine whether the run had developed a reasonable seasonally varying state. Over most of the North Pacific, the SST drift was only a few hundredths of a degree from year 34 to year 35. After the spin-up period was complete, we stored the monthly mean fields of total surface heat flux () and SST () from the final year of spin-up. The field was used as the mean part of the forcing for all the subsequent runs discussed below. To verify that was sufficient for mean forcing and to extinguish initial drifts or adjustments in SST, we ran the model for an additional 5 years with as the surface heat-flux forcing. Since there was little change in mean SST in the North Pacific, the 1970-1988 simulation, outlined below, was commenced from the end of that 5-year period.

Monthly Forcing Fields

Our interdecadal forcing experiments were forced by monthly-mean-plus-anomaly fields of wind stress, total heat flux, and TKE. The wind-stress field is composed of the mean ECMWF analyses, , plus monthly mean anomalies, τ', derived from the COADS observations. In the extratropics, the COADS wind-stress anomalies are monthly means of the products wu and wv from individual observations. Drag coefficients were taken from Isemer and Hasse (1987); like the heat-flux exchange coefficients, they are weakly dependent upon wind speed and ΔT. However, since the COADS observations are sparse in the low latitudes, we used anomalies of Florida State University (FSU) wind-stress analyses (Goldenberg and O'Brien, 1981) in the ± 20° latitude zone. In an overlap region, of approximately 10° at 20°N and 20°S, the COADS and FSU flux anomaly fields were smoothly merged.

The TKE variations from month to month were estimated using the observed monthly mean wind speed. We invoked the Weibull parameterization of Pavia and O'Brien (1986) for relating the monthly mean of the wind speed cubed, {w3}, to monthly mean wind speed, {w}, of COADS. Choosing a representative value of this parameterization from the North Pacific regions shown by Pavia and O'Brien yields {w3} ≈ 1.9{ w}3. Comparison of 10 years (1970-1979) of {w}3 with {w3} estimated from individual COADS observations exhibits fair agreement: {w3}/{w)3 ranged from 1.3 to 2.5 over most 2° squares of the North Pacific. After computing TKE from the monthly mean cubic wind speed and removing the long-term monthly mean, we added the resulting anomalies to the mean ECMWF TKE fields that had been extracted from the model spin-up period.

For the total heat-flux anomalies, we adopted a similar strategy. We used Q' as determined from the COADS observations (after DC1 and the discussion above) poleward of 20° latitude. As described in the Observational Evidence section above, the COADS latent and sensible heat fluxes were formed from monthly averages of products of individual observations. Latent and sensible heat exchange coefficients taken from Isemer and Hasse (1987) are weakly dependent on wind speed and ΔT. In the tropics, ship weather reports are sparse, so we could not easily apply the COADS flux anomalies. However, since there is some evidence that the role of heat fluxes in the tropics is to damp SST anomalies (Liu and Gautier, 1990; Cayan, 1990), we used a Newtonian damping scheme for 20°S to 20°N (in this zone the bulk-formulae flux anomalies are excluded). The Newtonian scheme was formulated as Q' = αT', where T' is the model SST anomaly. The spatially variable parameter α was determined by analyzing the total heat-flux output of the ECHAM T42 model (Roeckner et al., 1992) run that was forced from 1970-1985 with observed SST anomalies. The parameter field α varies geographically over the domain; it was determined by regressing the observed T' with the ECHAM model's Q' output. Typical values of α vary from −10 W m−2 °C−1 in the eastern tropical Pacific to near −40 W m−2 °C−1 in the western Pacific.

A point to be made is that the heat-flux parameterizations do depend on the SST, but they do not "build in" the observed SST anomaly variability in the resultant simulation. Actually, the latent and sensible flux formulations depend on the wind speed and the difference SST − Tα, which is not well correlated to the SST anomaly. Also, in the governing thermodynamic equation (see equation 4 in the next section), it is the temperature-anomaly tendency, not the anomaly itself, that responds to the flux forcing. Thus, in the simulation the model temperature is free to drift (possibly away from the observed SST state), governed only by the observed heat flux and wind forcings.

Adjustment of the Marine Weather Data

One of the biggest problems with using marine observations is that they contain artificial long-period fluctuations (Cayan, 1990, 1992a; Michaud and Lin, 1992), which cause spurious low-frequency behavior in the fluxes. Specifically,

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

the marine data contain substantial trends in the most important fundamental variables: wind speed, ΔT, and possibly Δq, even where these data are averaged over the largest spatial scales (DC1). The wind speed trends have been noted by several previous authors (Ramage, 1987; Cardone et al., 1990; Posmentier et al., 1989) and appear to be instrumental artifacts instead of natural variability. Although we were unable to identify a simple instrumental bias responsible for negative trends in ΔT and Δq, we did not trust wholesale decreases in oceanic ΔT and Δq, so we also filtered the ultra-low/large-scale variability from these.

We made two attempts to remove time-varying changes in the fundamental variables involved in the flux calculations. As described in DC1 and above, our first attempt calculated the area-averaged linear trend of w and ΔT, where the area considered was the entire COADS date coverage over the global ocean. Then, the equivalent latent and sensible flux changes were computed for each month of each 5° grid point, by adding the linear change in w and ΔT to their respective long-term monthly means, inserting these into the bulk formulae, and computing the fluxes over 19501988. By subtracting the long-term monthly mean value from each month of the 1950-1988 time history, a "correction" to the latent and sensible fluxes was obtained to reduce the apparent time-varying bias. This version of the fluxes still left apparently spurious low-frequency changes. Further inspection indicated three problems. First, the time-varying bias for ΔT, which generally decreases over time, was not linear, but contained low-frequency undulations over the period since 1950. Second, inspection of Δq, which had not been corrected, revealed similar low-frequency fluctuations. (The wind-speed trend, on the other hand, was quite nearly a linear increase of about 1 m s−1 over 1950-1988.) Third, changes in the net outgoing terrestrial flux from ΔT trends were not removed in the original correction. (This was because the initial flux-anomaly study focused on variations in the latent and sensible heat components, not the net flux.)

In our second attempt to adjust the flux (Miller et al., 1994a), we more accurately removed the basin-scale low-frequency variation in (generally decreasing) ΔT, by including a similar adjustment for variation (also generally decreasing) in Δq, and retaining the adjustment for w increases. The flux "correction" was calculated in the same manner as before, by perturbing the long-term mean data entering the bulk formulae with the time-varying adjustments. However, in this case, the adjustments to ΔT, Δq, and w were determined from a basin-scale EOF analysis, rather than from a linear trend. The EOF analysis was performed on smoothed versions (13-month running mean filter applied twice) of three COADS data fields encompassing the Pacific basin from 30°S to 60°N and 130°E to 75°W. For each of the three smoothed variables, the spatial pattern of the first EOF exhibited one sign over virtually the entire field. The time coefficients for each of the first EOFs contained the time variation that was portrayed by the linear trend (w increasing, ΔT and Δq decreasing). The first EOFs of the smoothed ΔT, Δq, and w fields accounted for 55 percent, 43 percent, and 73 percent of their variability, respectively. The projection of the first EOF back onto the smoothed data constituted the time-varying adjustments that were then applied in computing the flux corrections.

These adjustments were applied to the respective bulk formulae for the latent, sensible, and net terrestrial (infrared) heat fluxes. Since the first EOF of the smoothed wind speed was very nearly linear and showed relatively little spatial variation, we did not readjust the COADS wind-stress anomalies, which were corrected using the original COADS global-average linear trend of 0.8 m s−1 over 1950-1988. These second versions of the corrected fluxes were employed in the "R-run" simulation experiments discussed below.

Mechanisms of SST Variability

To help determine the physical processes forcing SST anomalies, we examine the time variability of the anomalies of the terms in the surface mixed-layer heat budget, along with MLD anomalies and SST anomalies. The surface heat budget is governed by

Here the SST (Ts) tendency is balanced by heat-flux input, horizontal advection, vertical mixing/entrainment, and diffusion. Q is the net surface heat flux, u is the horizontal velocity, we is the entrainment velocity, To is the temperature of the ocean just underlying the mixed layer (whose depth is H), and κ is the horizontal diffusivity. Monthly means of these quantities are computed during integration, and the climatological monthly means are computed and subtracted from the stored fields to obtain the anomalies.

The mean and standard deviations of the terms contributing to the heat budget (Equation 4) of the mixed layer are illustrated in Figure 4 for the North Pacific basin, the California Coast, and the central North Pacific. In addition to the heat-flux terms, the mean tendency of SST, the mean and standard deviation of SST, and the mean and standard deviation of MLD are shown for each region. All three regions are heated in spring and summer (April through September) and cool in fall, winter, and early spring (October-March), as indicated by . In the climatological mean, the shallowest and deepest MLDs occur in the central North Pacific where the wind forcing undergoes a large annual cycle. This heating and cooling is mirrored by the shoaling and deepening of the mixed layer, which ranges between about 30 m and 150 m. SST and MLD exhibit anomalous year-to-year variability that is a significant fraction of the long-term mean, with typical standard deviations

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

FIGURE 4

Long-term mean and standard deviation of model heat budget terms , SST (°C), and mixed layer depth (m) for the North Pacific basin, the California Coast, and the central North Pacific regions. Units are 0.1°C mo-1 for heat-budget terms and .

of 1°C and 20 m, respectively. Of the forcing terms in the heat balance, the heat flux has the largest mean and standard deviation (which is in line with previous studies, e.g., Frankignoul, 1985; Haney, 1985; Luksch and von Storch, 1992), followed by the entrainment. The heat flux is in phase with the climatological mean SST tendency, swinging from positive in spring-summer to negative during fall-winter.

Entrainment acts only in one direction: to deepen the mixed layer and to cool the temperature. Entrainment exhibits its maximum effect in fall. Presumably this is because the combination of strong winds and a sharp gradient at the bottom of the mixed layer has its greatest potential in fall. Advection, although a small component, acts in phase with the net heat flux in its mean behavior. Diffusion, which acts to dissipate temperature extremes, is out of phase with the SST and about 90° phase-shifted (but usually of opposite sign) from the heating terms.

The spatial pattern of the primary linkages between heat-budget components and are shown by the maps of the local correlations over the Pacific basin in Figures 5a through 5c, which appear in the color well. Note that is approximated here by the centered difference between the SST means for the months before and after the month considered. The correlations, calculated separately for heat flux, entrainment, and advection, were obtained for the set of all months of 1970 through 1988. Correlations between heat flux and (Figure 5a) are relatively high (> 0.6) at nearly all grid points north of 20°N. There is a strong similarity of the model correlations to those derived from observations (Figure 2), with the

Suggested Citation:"THE ATMOSPHERE." National Research Council. 1995. Natural Climate Variability on Decade-to-Century Time Scales. Washington, DC: The National Academies Press. doi: 10.17226/5142.
×

strongest correlations (exceeding 0.8) in the lower middle latitudes of the eastern North Pacific and weakest correlations to the northwest, east of Kamchatka. The heat-flux correlations are small in the subtropics, but this is where we prescribed the fluxes to be in phase with the SST via the Newtonian damping scheme. In the tropical Pacific, the correlations are negative and moderately strong, indicating that ocean dynamics processes play a significant role in the model's SST anomaly variability. (Note, however, that the tropical heat flux vs. relationship may be artificial because the fluxes are prescribed using the Newtonian damping scheme.) Interestingly, in data-available regions of the Southern Hemisphere, the correlations again are relatively strong and positive, implying that heat flux is a substantial component of the model's anomalous temperature budget. For entrainment vs. (Figure 5c), fairly high (˜ 0.5) correlations occur at scattered locations poleward of about 10°, and right along the equator. Concerning advection (Figure 5b), greatest correlations (> 0.5) with are found in the subtropics and tropics, with strongest centers developed south of the equator between 5°S and 25°S along South America and between 15°S and 15°N north of Australia in the western tropical Pacific. In the North Pacific extratropics, the correlations are generally not very strong, but there is a mild positive correlation region (˜ 0.4) south of the Aleutian Islands, extending into the Gulf of Alaska.

SS