Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 27
2 The Empirical and Conceptual Foundations of Indicators ' ndicators describe and summarize: they can be used for diagnosis and warning, and they can be used to monitor change. No indicator needs ~ to do all these things, but if one wants to know whether an indicator is of value, its intended use must be clear. Some indicators are more oriented to describing the state of a system, others to predicting its future state (NRC 1995a). Both description and prediction have their uses. However, it is impossible to imagine a suc- cessful set of indicators that fails to describe current conditions or fails to facilitate prediction. We need to know both where we are and where we are going. Good indicators have three key features. First, they quantify infor- mation so that its significance is more apparent. Second, they simplify information about complex phenomena to improve communication (Hammond et al. 1995~. Finally, indicators are used based on the assump- tion that doing so is a cost-effective and accurate alternative to monitor- ing many individual processes, species, and so on (Landres 1992~. The most difficult conceptual problem in developing indicators is to ensure that they are complete enough to capture the dynamics of key processes without being so complex that their meaning what they indicate is unclear. Not all indicators need to have immediate policy implications, but if they are to be policy-relevant, the relationships between them and the issues relevant to public policy choices should be clear. In addition, so many ecological indicators have been proposed and used that the costs of monitoring all of these indicators would be prohibitive. 27
OCR for page 28
28 ECOLOGICAL INDICATORS FOR THE NATION Most input data to indicators are generated by monitoring environ- mental states over time, either retrospectively or prospectively. The distinction between retrospective and prospective monitoring has been described by the NRC in its review of the EPA's Environmental Monitor- ing and Assessment Program (EMAP) (NRC 1995a): Retrospective or effects-oriented monitoring is monitoring that seeks to find effects by detecting changes in status or condition of some organism, population, or community. Examples include monitoring the body temperature of a person, monitoring the productivity of a lake, monitoring the condition of foliage in forests, and so on. It is retrospec- tive in that it is based on detecting an effect after it has occurred. It does not assume any knowledge of cause-effect relationships, although the intention is usually to try to establish a cause if an effect is found. It is EMAP's general approach. Predictive or stress-oriented monitoring is monitoring that seeks to detect the known or suspected cause of an undesirable effect (a stressor) before the effect has had a chance to occur or to become serious. Exam- ples include monitoring the cholesterol level in a person's blood, moni- toring the stress level along a geological fault, monitoring animal tissues for the presence of known carcinogens or other disease-causing com- pounds, and monitoring with a canary the toxic gas level in a mine. It is predictive in that the cause-effect relationship is known, so that if the cause can be detected early, the effect can be predicted before it occurs. Both retrospective and prospective monitoring are of value in devel- oping ecological indicators. An indicator based on retrospective monitor- ing, which describes the state of an ecosystem, could be useful in assessing the need for environmental management or the effectiveness of environ- mental programs. For example, an indicator of the condition of popula- tions of anadromous fishes, such as salmon and shad, is the degree to which a river is regulated by dams and the kinds of dams that are present. Such an indicator, if developed quantitatively, could be used to trigger action to prevent deterioration, assess future prospects, and suggest appropriate mitigation. Prospective monitoring is common and often required by laws. For example, monitoring concentrations of various gases in the atmosphere and contaminants in water is required by the Clean Air Act and the Clean Water Act. The nation monitors for the presence of microorganisms in drinking water, community composition in some fresh waters (IBI), and the presence of various introduced and native pests. Identifying useful indicators of stressors rather than the stressors themselves is straightforward if the likely stressor and its probable effect are known. Monitoring birds' eggs for DOT by looking at the thickness of their shells (Buckley 1986) is a good example. The stressor DDT and its
OCR for page 29
THE EMPIRICAL AND CONCEPTUAL FOUNDATIONS OF INDICATORS 29 metabolite DDE was known; at least one of its effects interference with avian reproduction was known; and a good indicator eggshell thick- ness was relevant and useful. Such indicators must be developed on a case-by-case basis. No systematic way exists to predict the existence and effects of stressors before they become a problem. In many cases, however, an adverse effect is identified whose cause is not known (e.g., recent observations of deformed frogs), or a variety of adverse effects are known to be caused by a mixture of factors whose individual roles are unknown (e.g., declining salmon populations in the Pacific Northwest, various ecosystem changes in Chesapeake Bay). In these cases, stressor-indicators cannot be developed until the stressors are known. Several possible stressors could be monitored, but there are so many that this approach is unlikely to be cost-effective or efficient. For example, although the developmental condition of frogs might well be a good indicator of some as-yet unidentified stressor, and it might be a good retrospective indicator of something important beyond the valuable information it provides about frogs themselves, the value of using deformed frogs as prospective indicators cannot be assessed without knowing why they are deformed. Some factors are both stressors and effects. For example, soil erosion is a cause of stream sedimentation, and so erosion rate is a stressor- indicator of the condition of stream communities. But soil erosion also is an indicator of the effects of overgrazing or deforestation on terrestrial ecosystems. For these reasons, the committee focuses its recommenda- tions on national indicators that inform about the status and trends in ecosystem extent, condition, and functioning rather than focusing specifi- cally on indicators of the stressors themselves. Useful ecological indicators are based on clear conceptual models of the structure and functioning of the ecosystems to which they apply. The models can be empirical or theoretical, quantitative or qualitative, but, as the following discussion emphasizes, some type of model is essential to ground and rationalize all indicators. SCIENTIFIC UNDERPINNINGS OF INDICATORS All indicators are grounded in substantive knowledge and some sort of scientific logic. Consideration of the major varieties of such logical arguments helps assess the degree to which any given indicator is reliable. Natural History All indicators begin with data taken from the world. Whether these data concern the nesting season of ravens or the ozone concentration of
OCR for page 30
30 ECOLOGICAL INDICATORS FOR THE NATION the upper atmosphere, they are needed because it is impossible to under- stand the environment without observing it. Indicators based on natural history are descriptive. Even when they measure a rate, such as stream flow, they are not capable, by themselves, of predicting future conditions. To become predictive, natural history observations must be incorporated into some model of how the relevant environmental processes operate. Paleabiology By taking a long view of things one can determine whether current conditions and trends are within the range of variations that have occurred in the past or whether they are in some way anomalous. Are current changes in temperature and precipitation unusual? Is the current relationship between diversity and land area similar to that prevailing millions of years ago? Patterns that are repeated many times during geological history can be incorporated into the tools used to predict the future. The past can be a key to the future if the Earth has performed experiments that have yielded reasonably consistent results. Experiments Based on observations of the environment, scientists develop hypoth- eses about the relationships and processes that influence it. These hypotheses are tested experimentally in the field and in the laboratory and modified appropriately. Where they can be used, experiments are powerful tools in the development and testing of hypotheses. Analytical Predictions Mathematical models of environmental processes generate predic- tions that can be tested observationally and experimentally in the field. They also may generate predictions of future events that cannot be tested directly, but these predictions may suggest which events should be moni- tored to determine whether the predictions are borne out. Analytical models have been used to explain the existence of the ozone hole over the Southern Hemisphere and to predict future ranges of species under dif- ferent scenarios of global climate change. Computer Simulations The properties of many complex ecological systems cannot be deter- mined analytically, but they can be studied by using computer simulation models. These models allow investigators to manipulate key variables to
OCR for page 31
THE EMPIRICAL AND CONCEPTUAL FOUNDATIONS OF INDICATORS 31 assess the sensitivity of the ecosystem's behavior to changes in the values of these different inputs. Global climate models and models of the dynamics of complex ecosystems, which are examples of such models, have played important roles in assessing the likely consequences of human-induced changes in the environment. USING MULTIPLE APPROACHES The most important point to make about these different sources of scientific data is that no one source by itself is sufficient to guide the design and implementation of useful ecological indicators or to formulate sound environmental policies. Information from all these sources is needed, and the richer the array of information, the better. From a scien- tific perspective, indicators are most useful if they are reliable, that is, if the measurements on which they are based are repeatable and do not vary significantly depending on who gathers them. Indicators increase in value with the time period over which their supporting data are gathered. Because the value of long-term data depends strongly on the consistency of the methods used to gather them, changes in measurement methods need to be implemented carefully. Technologi- cal advances regularly improve the speed, reliability, and accuracy with which data can be gathered. To fail to incorporate such advances is gen- erally undesirable. Therefore, good indicators should be robust to changes in measurement technology, so that long-term data sets are internally consistent even if measurement methods have changed. The calibration of methods during periods of technological change is an essential compo- nent of integrating indicators across technological boundaries. HISTORICAL AND PALEOECOLOGICAL DATA AS AIDS TO INDICATOR DEVELOPMENT Paleoecology, taken here to encompass also paleontology, paleo- limnology, paleogeochemistry, and paleoclimatology, is concerned with describing past ecological communities and their environments. Records of temporal change in the distribution of fossilized organisms, and of physical and chemical properties of the environment, can provide a very useful background to assess the influence on ecosystems of diverse natural and anthropogenic disturbances. These records are an extremely valu- able complement to environmental monitoring, and can suggest whether the causes of past changes have been natural or anthropogenic (Charles et al. 1994~. Such records, which can be gathered for a wide range of temporal and spatial scales, can help determine requirements for sampling frequency and duration.
OCR for page 32
32 ECOLOGICAL INDICATORS FOR THE NATION In conjunction with monitoring programs, paleoecological indicators can help answer five main questions: · What were communities and ecosystems like before they were subjected to natural or human disturbances? Evidence from pollen and seeds preserved in dated cores of lake sediments, along with other sources of information, helps to answer this question for regional and local vegetation (Brush 1986, Davis 1985, McAndrews and Boyko-Diakonow 1989~. Diatoms and fossilized chlorophylls and carotenoids can be used to answer this ques- tion for lakes (Engstrom et al. 1985~. · What have been the patterns of recovery from disturbances, and have the initial conditions been regained? Foster (1995) used records of fossilized pollen to show that forests in central Massachusetts did not return to their composition before the arrival of European settlers following abandon- ment of agriculture. Recovery of lakes from severe anthropogenic acidifi- cation has been assessed by analysis of past diatom and chrysophyte species in sediment cores (Dixit et al. 1992~. · What has been the nature and degree of variability in the past, especially the frequency of extreme events? Heinselman (1996), using fire scars and tree-ring analysis, reconstructed the fire history of the Boundary Waters Wilderness in northern Minnesota. His analysis showed that between 1681 and 1894, the intervals between major fires ranged from 11 to 42 years. Clark (1988a, b) showed that fire frequency in this area was greater during the dry medieval warm period than during the subsequent Little Ice Age. · Were communities and ecosystems relatively stable, were they following trajectories of gradual change, or did they exhibit sudden fluctuations or transi- tions to another state? The postglacial vegetation history of North America, established by numerous regional studies of pollen records (Wright 1971), offers extensive evidence concerning this question. In the midwestern United States, for instance, deciduous forest changed to prairie and back again during the period from 8,000 to 4,000 years ago. Renberg (1990) determined the pH history of a Swedish Lake, Lilla Oresjon, by studying its diatom stratigraphy. Many diatoms have rather narrow pH toler- ances, and their sculptured shells (frustules) can be identified to species, making them excellent indicators of lake acidification. Renberg showed that Lilla Oresjon acidified slowly from neutrality 12,000 years ago to pH 5.2 2,300 years ago. It then became much less acid owing to land clear- ances by humans and the consequent leaching of bases into the lake. It re- acidified to about pH 4.5 during the present century owing to acid deposition. · Have anthropogenic perturbations been different, in degree or kind, from natural perturbations? The history of Lilla Oresjon provides evidence that
OCR for page 33
THE EMPIRICAL AND CONCEPTUAL FOUNDATIONS OF INDICATORS 33 recent anthropogenic acidification exceeded the natural acidification that preceded it for thousands of years. A difference in kind of change is illustrated by the observed eggshell thinning in birds caused by exposure to chlorinated hydrocarbons (Ratcliffe 1967), which appears to have no natural analog. Environmental Problems for Which Paleoecological Data Have Been Useful Paleoecological data have been used to study a wide range of environ- mental problems, including climate change, acid deposition, eutrophica- tion, losses of biodiversity, fish declines and introductions, fire frequency, soil erosion, silting of lakes and estuaries, and pollution by heavy metals (e.g., lead and mercury) and trace organic molecules (e.g., DDT and toxaphene). Indeed, the understanding of most environmental problems can be improved by access to Paleoecological information. Paleoecological data have often revealed major unsuspected environ- mental changes, over time scales ranging from the mass extinctions observed in the fossil record to the sudden catastrophic loss of submerged macrophytes in Chesapeake Bay in the early 1970s (Brush and Davis 1984, Davis 1985~. Paleoecological data have been especially important in alter- ing views of the operation of the global climate system. Tree-ring records from the southwestern United States reveal numerous abrupt changes in regional precipitation between 800 and 2,000 years ago (Graumlich 1993, Hughes and Graumlich 1996~. Likewise, close-interval paleoclimatic stud- ies of the Greenland ice core have shown that mean annual temperatures there have changed by as much as 10°C in a few years (Grootes et al.1993, Alley et al. 1993), and that sudden "jumps" in regional climate may be likely as changing conditions lead to a warmer climate globally (Broecker 1987, Overpeck 1996~. To anticipate "surprise" events, possible hazards will need to be moni- tored in a variety of environments. Attention will need to be paid to outliers in the data, which may reflect not errors in measurement but unusual, rare phenomena (Kates and Clark 1996~. Thus, programs that monitor environments subject to a wide range of natural and anthropo- genic stresses can benefit greatly from complementary studies of paleo- ecological indicators. Indeed, much useful information, such as long- term baseline records, can be gathered in no other way. Retrospective Paleoecological indicators can be used prospectively to identify trends that may be useful in evaluating monitoring results.
OCR for page 34
34 ECOLOGICAL INDICATORS FOR THE NATION SOURCES OF INFORMATION ABOUT CURRENT ECOLOGICAL PROCESSES Regularly gathered information can provide the broad spatial cover- age and substantial time series necessary to detect environmental trends. Monitoring key ecological information is important because the value of an indicator is reduced considerably if new baseline data must be gathered before it can be used. Because ecosystems are so variable in space and time, gathering enough data for national ecological indicators will be difficult. Fortunately, recent developments in remote-sensing technology offer new opportunities for measurements at extensive spatial scales. Because the committee recommends indicators that depend on remotely sensed information, we describe these developments in some detail. Remote Sensing From Satellites There are many reasons for using remotely sensed data as inputs to indicators. For many of the indicators that the committee recommends, especially for information on terrestrial ecosystem processes and some aspects of ecosystem status, remote sensing offers rapid and relatively accurate sampling. For a few remotely sensed measurements, there already is an approximately 25-year time series of reasonably well cali- brated data of nearly global coverage. In addition, for measurements over large spatial areas, remote sensing offers the only affordable means of sampling. Operational costs for satellite systems are not necessarily higher than the personnel costs of large in situ monitoring efforts, but the capital costs of developing and launching satellite missions are extremely large. The availability of time series of satellite measurements with well- established precision and accuracy is important in using remotely sensed data. Satellite-generated data are especially valuable if all instruments, both within and among satellite platforms, are fully calibrated and if the data are accessible and affordable. Fortunately, considerable attention has been paid to these issues. The two types of satellite imagery most widely used by terrestrial ecologists are Landsat TM (Thematic Mapper) and AVHRR (Advanced Very High Resolution Radiometer). Both cover most of the Earth's sur- face, TM imagery at 30 m resolution and AVHRR imagery at 1 km resolu- tion. Each records the intensity of radiation reflected from the planet in five to seven spectral bands, in the visible, near-infrared (NIR), and thermal infrared (JR) parts of the spectrum. The Landsat data record, which starts in 1972, has continued through the use of three primary instruments. The first provided video images,
OCR for page 35
THE EMPIRICAL AND CONCEPTUAL FOUNDATIONS OF INDICATORS 35 some of which are still archived. The video system was accompanied by the Multi-Spectral Scanner, which very quickly became the preferred instrument for quantitative analysis. Since 1982, the primary instrument has been the Thematic Mapper, with six spectral bands in the visible and short-wave near-infrared, all of which have 30 m spatial resolution, and one thermal band with 120 m spatial resolution. A Landsat scene is approximately 185 km on a side; the combination of swath width and orbital characteristics of the satellite means that any spot on the Earth is revisited at about 16- to 18-day intervals. Six Landsat missions were launched, of which five achieved orbit. Landsat 5 is currently in the 13th year of its planned three-year lifetime. The United States holds in its archives at the U.S. Geological Survey (USGS) Eros Data Center nearly complete global data from the Landsat system for the years 1972 to 1980, but the dates of cloud-free acquisitions vary geographically. Less comprehensive coverage is held in U.S. archives since 1980 because of the U.S. policy of commercializing the Landsat system (although the record for the United States itself remains fairly complete). However, 10 to 17 international ground stations, which were operating during the period of the Landsat missions, have maintained accessible data archives. Thus, in principle, global data sets could be derived for the 1980s and 1990s. The tape recorders on Landsat 5 failed several years ago, restricting its data downlink capabilities to direct trans- mission to ground stations (without relays to other communication satel- lites). This failure created holes in global coverage where ground stations are not operating, most notably Siberia, Alaska, and central Africa. The AVHRR a standard instrument on the National Oceanic and Atmospheric Administration (NOAA) polar-orbiting meteorological sat- ellites has several spectral bands in the visible, NIR, and IR parts of the spectrum. However, the AVHRR collects data with very different spatial characteristics from Landsat data, reflecting its origin as a meteorological instrument. Global-area coverage data, which are resampled on board the satellite before being transmitted, have a 4 km spatial resolution. Local-area coverage data, which are directly transmitted to the ground, have a 1 km spatial resolution. The combination of swath width of the sensor (about 1000 km) and orbital characteristics of the satellite platform achieve daily to twice-daily coverage. However, unlike Landsat TM, whose radiometric properties are known and monitored with a high de- gree of accuracy, the AVHRR has no provision for on-orbit calibration, and the satellite orbits have drifted substantially over the years. The data themselves are less than ideal because the instruments are poorly cali- brated, and intercalibration of instruments on different platforms is also poor. Therefore, although roughly decade-long time series of truly global
OCR for page 36
36 ECOLOGICAL INDICATORS FOR THE NATION data exist, they must be extensively processed before they can be used for quantitative analyses of ecosystem properties and processes. Ecosystem Processes. The most widely used satellite-derived indicator of processes is the normalized difference vegetation index (NDVI). NDVI is a direct measure of absorbed photosynthetically active radiation, because chlorophyll absorbs in the satellite's NIR band and reflects in the IR band, whereas rocks, stems, and soil reflect in both NIR and IR. The numerator of NDVI is IR minus NIR, which is large if abundant chloro- phyll reduces the NIR. To compensate for the variable transparency of the atmosphere, this difference is divided by the sum NIR plus IR. Global NDVI measurements at a variety of resolutions are now available through the Internet (Kidwell 1997~. As a result of an extensive effort by the National Aeronautics and Space Administration (NASA) and NOAA, a decade-long record of 4 km resolution NDVI data with consistently processed and documented tech- niques is available through the AVHRR Pathfinder Program. Many indi- vidual research groups have collected and maintained archives of AVHRR and NDVI. For example, S. Los and colleagues have compiled a one- degree aggregation of 4 km resolution data for 1981 through 1990 (Los et al. in press). A complete global record at 1 km resolution is now available for the period 1993 to 1994, and continues to be collected and processed by NASA and the USGS. Collaborating ground receiving stations around the world have recently developed a more precise measure of absorbed photosynthetically active radiation than is possible with NDVI by using multiple scenes with the sun and sensor in different positions and invert- ing a radiative transfer model. Once one knows the amount of leaf area in a location, it is a relatively easy matter to predict the region's net primary production (NPP). The simplest approach is to regress NDVI against measures of NPP. Gener- ally this is done separately for each biome (e.g., Fung et al. 1987~. More precise information can be obtained by using the vegetation index to parameterize a biogeochemical model, especially if there are data about the weather, topography, and soil type (which are widely available for the United States and to a lesser degree globally from NOAA and Depart- ment of Energy Web sites and on CD-ROM from these agencies). The best example of an ecosystem model driven by NDVI is the CASA model of Potter et al. (1993~. This model translates monthly NDVI and climate data into predictions of primary production, carbon storage, net ecosystem production, and nitrogen mineralization. It contains only a single fitted parameter (one value for the entire globe). Although the model is rela- tively simple, it has an accuracy approaching direct-measurement accu- racy of NPP and nitrogen mineralization about 25 percent.
OCR for page 37
THE EMPIRICAL AND CONCEPTUAL FOUNDATIONS OF INDICATORS 37 Several widely used land-surface models, initially developed to improve long-range weather forecasts, also rely heavily on satellite imag- ery. These models predict the transfer of matter, energy, and momentum between the land surface, vegetation, and the atmosphere at short (e.g., 20 minute) time scales. The models typically contain an enormous amount of detail at a fine scale, including the biochemistry of photosynthesis, the physiological control of stomates, the distribution of leaves in the canopy, and the vertical distribution of water in the soil. The best example of this type of model is the SiB model (SIB I, II, and III) of P. Sellers and col- leagues (Sellers et al. 1997~. Because these models contain so much detail, they rely heavily on satellite imagery to fix the properties of vegetation at each location (veg- etation type, albedo, leaf area, etc.~. Even so, a relatively large number of parameters and functions are little more than educated guesses. Despite these limitations, models such as SiB II have a remarkable capacity to predict diurnal and seasonal patterns of production, respiration, and transpiration. SiB II also improves the predictions of those weather models that incorporate it. The first international comparison of model simulations of NPP has recently been conducted under the auspices of the International Geosphere and Biosphere Program (IGBP). The study, which included models ranging from empirically fitted regressions to the most complex SiB II-type model, attempted to isolate the reasons for differences in esti- mates of ecosystem productivity and NPP. Results (Cramer et al. 1999, Cramer and Field 1999) indicate a fair degree of spread in the simulated NPP values, with some of the variation certainly attributable to differ- ences in the underlying data sets used. However, such comparison studies are likely to lead to a greater understanding of the limitations of both the underlying data and the models themselves. The result should be better and more quantitative documentation of NPP patterns. Vegetation Characterization and Classification. Classifying and mapping land cover and vegetation, the most common use of aerial photography, is served well by satellite imagery. The number of vegetation categories used is limited by the number of spectral bands, the resolution of the most commonly used imagery, and the experience of the interpreter. Although automated statistical techniques are used to cluster most data, there is no fully automated technique for land-cover classification. Even simple classifications are useful as indicators of ecosystem status. For example, the rate of loss of closed-canopy humid tropical forest in Brazil was measured from the late 1970s to the late 1980s using Landsat data (Skole and Tucker, 1993) and independently verified by scientists at the Brazilian Space Agency, INPE. INPE now uses analogous
OCR for page 40
40 ECOLOGICAL INDICATORS FOR THE NATION geographic coverage of the nation's aquatic resources cost-effectively remains largely untapped. Satellite data have been used to monitor the effects of drought on the area of shallow lakes (Brown et al. 1977a), and they could be used to monitor wetland hydrology. Most satellite applications to lake and reservoir monitoring involve use of spectral-reflectance data to estimate water clarity and related con- ditions. Reflectance of a water body is a function of optically active sub- stances, primarily algae (and their pigments), dissolved humic matter, and suspended minerals, each of which have different spectral patterns. Water clarity (and thus spectral reflectance) in most lakes depends prima- rily on phytoplankton abundance in the water column, which depends generally on the level of nutrient enrichment. Reflectance data can be used to estimate chlorophyll concentrations in water (Mittenzwey et al. 1992) and thus the trophic status of lakes. Low water clarity in reservoirs often results from nonalgal turbidity (clay-like suspended solids) caused by soil erosion, and so different relationships must be developed between reflectance and trophic status in such systems (e.g., Gallie and Murtha 1993~. Satellite data have been used to assess the trophic status and water quality of individual lakes (e.g., Sudhakar and Pal 1993, Chacon-Torres et al. 1992) and in a few cases the status of groups of lakes or all lakes in a region (Brown et al. 1977b, Lillesand et al. 1983, Lathrop 1992~. Landsat Thematic Mapper data are obtained at 16-day intervals across the United States. As many as five to six images are thus available for a given region or lake during the critical summer growth period (approxi- mately the end of tune to mid-September in northern states). Partial or complete cloud cover at the time of a Landsat overpass decreases the number of images available for processing, often to only one or two images per season. This frequency is insufficient for detailed assessment of time trends in individual lakes, but should be adequate for long-term monitoring of lake conditions on a regional basis. Other satellite plat- forms that will become available in the near future will provide more frequent coverage; the Moderate Resolution Imaging Spectroradiometer (MODIS) system, described below, will provide daily coverage of indi- vidual sites, if at lower spatial resolution (250 m, compared with 10 m resolution for Landsat). MODIS may be useful for intraseasonal monitor- ing of water quality in medium to large lakes. Future Developments. Satellite-based observation systems will soon improve dramatically. With the successful launch of Landsat 7 in April 1999, satellite-based observation systems have improved dramatically. Landsat 7 has an Enhanced Thematic Mapper (ETM) on board. The ETM maintains the same spectral bands as previous Landsats, but it also has a 15 m black-and-white band, useful both in its own right and for sharpen-
OCR for page 41
THE EMPIRICAL AND CONCEPTUAL FOUNDATIONS OF INDICATORS 41 ing the imagery from the 30 m bands. The ETM also has better calibra- tion. Most importantly, the system design for Landsat 7, which again is under government control, ensures that the U.S. archive will again be truly global. The mission is designed to update the U.S.-held global archive, and the archives held by international ground stations, on ap- proximately a seasonal basis. Data will be more affordable because the federal government will seek only to cover the costs of responding to data requests. The Earth Observing System (EOS) AM platform Terra was launched in late 1999. Terra will operate five instruments, including MODIS, which is of particular relevance for monitoring ecosystem pro- cesses. MODIS will have 32 spectral channels, with stringently defined radiometric calibration. Most of the spectral channels will have 1 km spatial resolution, although a few will have 250 to 500 m resolution. With this system, a global data set can be acquired roughly every few days. In addition, the combination of this system's spectral and spatial character- istics will enable scientists to accurately calculate a variety of vegetation indices, land-cover products, fire detection products, and land-cover change indices. In about two years, NASA will launch the Vegetation Canopy Lidar (VCL), an instrument capable of measuring foliage-height profiles every- where on the Earth. The technology has already been tested on the Space Shuttle, and the imager itself is currently on a research aircraft. The satellite-mounted VCL will allow the monitoring of harvesting and regrowth, measuring above-ground biomass, and assessing structural habitat features important to animal species. The commercial sector is increasingly active in providing remote- sensing information. Most of their proposed missions will seek very high spatial resolution data (1 to 10 m), either black-and-white or with a few spectral bands. Because these data provide little spectral information, they are of little use for assessing ecosystem processes. However, they have great utility for vegetation classification and analysis of land-cover changes. Recently declassified data provide an opportunity to analyze time series of land cover in some locations back into the 1960s. Remote Sensing from Aircraft Aerial photographs are available for most developed countries for most of the current century. Aerial photographs can provide all of the measurements available from satellites and at higher resolution, but usu- ally in a form that is more difficult to digest. Photo interpreters are routinely employed by timber-producing industries and the U. S. Forest
OCR for page 42
42 ECOLOGICAL INDICATORS FOR THE NATION Service to assess the sizes, densities, and species identities of trees in aerial photographs. Ground-Based Measurements Despite their great value, remote-sensing techniques do not eliminate the need for ground-based measurements. Such measurements record processes that are not detectable from afar and they are needed to "ground-truth" measurements from aircraft and especially from satel- lites. The most extensive and systematic system of ecological research sites is the National Science Foundation-funded network of 21 Long Term Ecological Research Sites (LTER). Several LTER sites have been in con- tinuous operation for 15 to 20 years. The LTER sites span a range of ecosystem types on U.S. territory from arctic tundra at Toolik Lake in Alaska's Brooks Range to tropical rainforest at the Luquillo Experimental Forest near San Juan, Puerto Rico (see http://lternet.edu/network/sites/~. Although each site has a different investigator-driven mission, several measurements are made at each site every year. The results, which include estimates of primary production, nitrogen mineralization rates, standing crop, abundances of most soil cations, detritus production, and censuses of dominant plant species, are available in standard form. A second source of information that has been collected systematically for more than 50 years is the U. S. Forest Service's Continuous Forest Inventory and Analysis (FIA), which is used by the Forest Service to set timber-management policy. This system represents several thousand plots on which every tree greater than 10 cm in diameter is measured every ten years. These data obviously could be used to detect trends in the abundances of species and information about primary production. However, the usefulness of the data is reduced by the large number of errors in the archived data and the archaic format of the magnetic tape on which most of the data are stored. Precise knowledge of plot locations is not widely available to avoid the possibility that someone would manipu- late the plots to affect future forest policy. With a few exceptions, formal censuses of animal species are local and short-term. The amateur (National Audubon Society) Christmas Bird Count (CBC), begun in the winter of 1900-1901, and the North American Breeding Bird Survey (BBS), launched in 1966 by the Migratory Bird Popu- lation Station in Laurel, Maryland (now the USGS Patuxent Environmen- tal Science Center), are conducted annually and cover the continental United States and parts of Canada (Root and McDaniel 1995, Peterjohn et al. 1995, Saner et al. 1997~. Data from both the CBC and BBS are now compiled and are available on the Internet at USGS and Cornell Labora- tory of Ornithology Web sites (BBS at http://www.mbr-pwrc.usgs.gov/
OCR for page 43
THE EMPIRICAL AND CONCEPTUAL FOUNDATIONS OF INDICATORS 43 bbs.html; CBC at http://birdsource.tc.cornell.edu/cbcdata/~. Root (1993) and her coworkers have shown that these data are sufficiently reliable to detect temporal and spatial trends. In 1975, the Xerxes Society started the annual Fourth of fuly Butterfly Count (FIC), now administered by the North American Butterfly Association. The FIC is modeled after the CBC and provides data that, when used carefully, are valuable for the study of status and trends of rare and widely distributed species (Swengel 1995~. The U.S. Department of Agriculture, in cooperation with other federal agencies, funds systematic studies of crop and timber pests through such programs as the Forest Health Monitoring Program (FHM) (USDA Forest Service 1994, see http: / /willow.ncfes.umn.edu/fhm/publicat.htm). Most Fish and Game departments census game fish, birds, and mammals. Fi- nally, all officially endangered species are periodically censused and their current ranges are known by county (Dobson et al. 1997~. MODELS TO ASSESS ECOSYSTEM FUNCTIONING For many years, ecosystems have been studied to determine patterns (community structure, biogeography) and processes (energy flow, nutri- ent cycling, stream flow, and oxygen content), but the measurements and accompanying models focus on relatively small scales. More recently, attention has also turned to processes operating at watershed scales that link upstream and downstream communities and terrestrial and aquatic . . communities. Conceptual Models Conceptual models have been significant in the development of eco- system ecology, especially in limnology, where they have been used extensively for more than a century. For example, the concept of a lake as a microcosm (Forbes 1887) introduced the ecosystem approach to ecology and identified the processes that are still the primary foci of ecosystem ecology: energy flow, elemental cycling, production and decomposition of organic matter, and food web interactions. Lindeman's (1942) trophic- dynamic model of energy flow through the food web of Cedar Lake intro- duced to ecology such important topics as energy transfer efficiency, relationships between production and decomposition, and physical and chemical constraints on biological production. Conceptual models help identify key links between ecosystem com- ponents and serve as the basis for developing quantitative models. Flow- diagram models are widely used to describe nutrient cycles, food webs, and energy flows in ecosystems. The river-continuum model is the most important conceptual model for river ecosystems. It is based on the fact
OCR for page 44
44 ECOLOGICAL INDICATORS FOR THE NATION that stream ecosystems undergo predictable physical and biological changes from headwaters to mouth (Vannote et al. 1980~. In headwaters, terres- trial ecosystems are the principal source of the organic matter that pro- vides energy for stream organisms. Further downstream, these extrinsic sources are increasingly replaced by instream primary production. Thus, headwaters are dominated by organisms capable of processing leaves and wood, whereas most of the consumers in downstream communities depend on instream photosynthetic plants. The river-continuum model provides qualitative predictions about the kinds of species expected in a particular stream reach and region. The model links the physical structure of streams and expected biota, and provides a conceptual foundation for many biologically based stream monitoring approaches. Extensive data are available for mid-reaches of wadable streams, but few data are available for rapidly changing head- waters and large rivers (but see Patrick et al.  for a valiant attempt to survey large rivers). Other conceptual models have been developed to assist the design of biotic indicators and to evaluate the effects of oxygen-demanding wastes on stream ecosystems (Metcalfe 1989, Cairns and Pratt 1993~. These wastes increase concentrations of fine organic particles and decrease oxygen con- centrations. Among such indices are the Saprobien system (Kolkwitz and Marsson 1909) and models based on the tolerances of species to changes in turbidity and oxygen concentrations (Hilsenhoff 1982, 1987~. Conceptual models that focus on specific groups of organisms (e.g., fishes) have been used to determine minimum conditions for survival of recreationally or commercially important species. Models based on the instream flow increment method (IFIM) seek to determine stream flows necessary to support species of concern (Bovee 1996~. IFIM models com- pare the proportional use of habitats by a species with the proportional availability of particular water velocities. Other habitat-suitability models have been developed by the U.S. Fish and Wildlife Service. Recently, biogeographic models, based on expectations of regional distributions of species in intact habitats (Kerr et al. 1985, see Chapter 1), have been used to assess the condition of stream biotas. The best known model of this type is the earlier-mentioned Index of Biotic Integrity (IBI, discussed further in chapters 4 and 5), which was first applied to fish communities and later extended to stream macrobenthos and diatom com- munities. IBI models have been developed and used in the upper Mid- west (Illinois and Ohio), the South (Arkansas), and the West (Oregon). Many state agencies are using IBI analyses to develop biological criteria for evaluating the status of stream ecosystems (e.g., Whittier and Rankin 1992~. IBI models and the analytical procedures that support them are robust and capable of detecting many changes in community composi-
OCR for page 45
THE EMPIRICAL AND CONCEPTUAL FOUNDATIONS OF INDICATORS 45 lion, although often they cannot distinguish among causes of degrada- tion. Individual IBI models are region-specific and depend on availability of extensive data on species distributions in the focal region. Empirical Models Empirical models are used to describe quantitative associations be- tween variables or sets of variables and to predict the values of variables from measured values of other variables. These models are capable of making useful predictions even when cause-effect relationships between predictor and predicted variables are poorly understood. Examples of useful empirically determined relationships are the posi- tive correlation between chlorophyll a levels (a measure of algal biomass) and total phosphorus concentration (the nutrient assumed to be limiting algal growth) (Sakamoto 1966, Dillon and Rigler 1974a); the negative cor- relation between Secchi-disk transparency (a simple measure of water clarity) and chlorophyll a levels (Carlson 1977~; the positive correlation between mercury concentrations in fish and their size (Lathrop 1992~; and the correlation between log KoW (the octanol-water partition coefficient) for various synthetic organic compounds and various measures of bio- accumulation and microbial degradability of such compounds (see Brezonik 1994 for an extensive review). Semi-empirical models typically employ major simplifying assump- tions to portray process mechanisms. Short-term variations and detailed spatial patterns in ecosystem conditions are not generated by the outputs of these models. Often called reactor models, they have been most exten- sively developed for lakes, which are treated as completely mixed tank reactors. Both inputs and losses of the substance being modeled are assumed to be constant or to be simple first-order processes. Reactor models were first applied by Vollenweider (1969, 1975) to analyze lake responses to phosphorus inputs. Coupled with empirical relationships between average concentrations of total phosphorus in the water column, summertime chlorophyll a levels, and Secchi-disk transparency measures, these models were successful in describing gross features of lake eutrophi- cation. They were also used to develop phosphorus-loading criteria, at values above which the water quality of a lake would be expected to degrade (Vollenweider 1975, 1976; Dillon and Rigler 1974b, Baker et al. 1981~. The model BATHTUB (Walker 1987) is a computerized version of the reactor model for lakes and reservoirs. Reactor models have also been used to describe sulfate reduction and alkalinity generation in acid-sensitive lakes (Baker and Brezonik 1988, Kelly et al. 1988), retention of humic matter in bog lakes (Engstrom et al.
OCR for page 46
46 ECOLOGICAL INDICATORS FOR THE NATION 1988), and concentrations of various synthetic organic contaminants in lakes and reservoirs (Schnoor 1981~. The advantages of reactor models include their simplicity and low data requirements. They require few coefficients and small amounts of physical information on the system being modeled. Their disadvantages, which also stem from their simplicity, are that their coefficients lack simple physical meaning and must be determined empirically, that they are unable to capture short-term dynamics, and that they lack explicit ties between modeled output of substances and biotic responses. Compartment models are similar to reactor models and use many of the same mathematical formalisms (Brezonik 1994~. The compartments in such models generally represent mass quantities (or reservoirs) of sub- stances or elements within discrete biotic and abiotic components. Flows of substances between compartments are expressed as simple first- or second-order differential equations. Such models have been used to describe in-lake dynamics of phosphorus cycling, including regeneration rates of inorganic P from organic P by zooplankton and microbial decom- posers (Lyche et al.1996~. On a much broader scale, compartment models have been used to describe global cycling of carbon and phosphorus among major biotic and abiotic reservoirs (Lasaga 1985~. The forerunner of all deterministic water quality models is the Streeter-Phelps model for dissolved oxygen in rivers (Streeter and Phelps 1925~. Developed in the mid-1920s, long before the advent of computers, the model expressed changes in oxygen concentration in a river as the difference between a loss term representing biochemical oxygen demand, caused by microbial degradation of organic matter, and a source term representing atmospheric re-aeration. Other source and sink terms, such as planktonic primary production and sediment oxygen demand, later were added to the model; it was computerized in the 1960s. Currently it can be applied to complicated river-estuarine systems and can produce time-varying output at any desired distance along the stream. Simulation Models The most advanced simulation models for water quality (e.g., HydroQual, Inc. 1991, 1998; fin et al. 1998) are capable of modeling river- ine-lake and riverine-estuarine ecosystems in three spatial dimensions at integration times on the order of minutes. They can produce output as daily averages over a year or several-year period for a wide range of physical, chemical, and biological variables, including water elevation (lakes) or flow (rivers), temperature, concentrations of inorganic nitrogen forms, dissolved organic phosphate, dissolved organic N and P. particu-
OCR for page 47
THE EMPIRICAL AND CONCEPTUAL FOUNDATIONS OF INDICATORS 47 late N and P. silica, inorganic carbon, pH, dissolved and particulate or- ganic carbon, and phyto- and zooplankton biomass. Although some early simulation models included fishes, these models' outputs failed to accurately portray fish population dynamics. Also, current simulation models of aquatic ecosystems typically exclude benthic invertebrate and macrophyte populations. These components of aquatic ecosystems are difficult to model for several reasons: these popu- lations are influenced by many factors other than food availability (the major driver in the models); the models simplify life cycles of organisms; and the rates at which the ecosystem populations fluctuate are often much slower than the rates of changes in other processes in the models. Models that do not attempt to simulate variations in all three spatial dimensions have much simpler data requirements and are more practical for long-term assessments of biological processes. The one-dimensional lake simulation model MINLAKE (Riley and Stefan 1987) ignores areal variability, but treats important vertical variations, and is suitable for modeling relatively small lakes. It accurately simulates seasonal patterns of thermal stratification in temperate lakes and is fairly successful in simu- lating temporal trends in vertical profiles for a variety of chemicals, including dissolved oxygen. The model has been used to predict the effects of climate warming on cold-water and warm-water fish communi- ties in Midwest lakes (Stefan et al. 1995, 1996~. A number of process models provide well-tested empirical estimates of changes based on point or diffuse source inputs of oxygen-demanding wastes (Beck 1987, Thomann and Mueller 1987~. They require only mea- sures of stream discharge and estimates of degradation and improvement parameters; no biological data are used. Landscape Models Spatial models of landscapes using geographic information systems (GIS) are used to make a variety of predictions about landscape changes and to indicate how those changes may affect stream ecosystems. For example, soil erosion from landscapes is predicted by the Universal Soil Loss Equation (USLE, Wischmeier and Smith 1978~. The Area Nonpoint Source Watershed Environmental Response Simulation (ANSWERS; Beasley and Huggins 1982) simulates surface runoff and erosion in agri- cultural watersheds. This model incorporates data on overland flows that are not included in USLE. The Agricultural Nonpoint Source (AGNPS) model simulates runoff and sediment and nutrient transport across a range of watershed sizes (Young et al. 1989~. This model uses grid cells to evaluate hydrology and material transport, and incorporates data on streambanks, eroded gullies, and nutrient sources. Flowpath models,
OCR for page 48
48 ECOLOGICAL INDICATORS FOR THE NATION such as TOPMODEL (Beven and Kirkby 1979, Beven 1997), make pixel- by-pixel estimates of conditions using land-surface data inputs. The Regional HydroEcological Simulation System (RHESSys; Band et al. 1991, 1993) and CENTURY model (Parson et al. 1992, 1994) are also widely used ecosystem models, the former including hydrology and the latter with a focus on soil organic matter (SOM). Despite the extensive development of these and many other ecological simulation and process models, the linkages between landscape processes and stream biota remain poorly understood. Understanding these connections is difficult because time lags between landscape changes and instream responses are highly variable. THE COMMITTEE'S CONCEPTUAL MODEL FOR CHOOSING INDICATORS To guide its selection of national ecological indicators, the committee assessed the current status of empirical and conceptual knowledge of the factors that most strongly influence ecosystem functioning. With a few local exceptions, terrestrial and freshwater ecosystems are open systems powered by sunlight. Solar energy is incorporated into ecosystems by photosynthesis, which is carried out by green plants, protists, and photo- synthetic bacteria. The goods and services that ecosystems provide to humans depend directly or indirectly on ecosystem productivity, i.e., their ability to capture solar energy and store it as carbon-based molecules. Therefore, the committee recommends several indicators of ecosystem productivity. The rate of capture of solar energy by photosynthesis is called primary productivity. Primary productivity is strongly influenced by temperature, moisture, soil fertility, and the structure and composition of ecological communities. Information on these factors can be used as the basis for accurate estimates of the primary productivity of most ecosystems. There- fore, useful indicators of ecological conditions and the productivity of ecosystems are based on data about these factors. Extensive climatic data are already being gathered and are available to be incorporated into models of ecosystem performance. Rates of photo- synthesis are measured locally for different types of ecosystems. To cal- culate the overall status and productivity of the nation's ecosystems, information is needed on the extent of each of the major types of eco- system used to determine photosynthetic rates. The condition for maintenance of soil fertility, a key determinant of productivity, is that inputs and losses of nutrients must be balanced. In natural ecosystems, new nutrients are made available by weathering of rocks and soils and by atmospheric deposition. In most agroecosystems, natural nutrient inputs are supplemented, sometimes massively, by
OCR for page 49
THE EMPIRICAL AND CONCEPTUAL FOUNDATIONS OF INDICATORS 49 application of fertilizers. Indicators of fertility can use data from direct measurements of nutrient concentrations in soils and of nutrient exports to other ecosystems. Rates of photosynthesis are strongly influenced by the structure and composition of the species in the ecosystem. Structure is important because more complex vegetation may be able to intercept more sunlight than structurally simpler vegetation. The species composition of biologi- cal communities influences primary production in part because species differ in their abilities to photosynthesize under different weather condi- tions. Although relationships between ecosystem productivity and the number of species in the system are as yet poorly understood, it is clear that without some minimal number of species, ecosystems would func- tion poorly (Grime 1997, Tilman 1996~. Therefore, although relationships between species richness and ecosystem functioning cannot yet be quan- tified, the loss of species is a cause for concern. If one discovers that a species had great ecological or economic importance after it has dis- appeared, it is too late to do something about it. In addition, species are valued by societies for moral, aesthetic, and cultural reasons (Sagoff 1996), as expressed in international treaties and national laws (NRC l999b). Species composition also influences ecosystem performance by influ- encing the frequency and severity of diseases and pest outbreaks (Gunderson et al. 1995, Mooney et al. 1996~. In addition, exotic species, many of which have escaped from their natural enemies, often achieve higher abundances than in their native lands and hence cause ecological problems (Drake et al. 1989~. Therefore, measures of the presence of native and exotic species are important inputs to national ecological indicators. How the committee used this conceptual model is described in Chapter 4, where we recommend a set of indicators that use data on the key factors that influence ecosystem functioning. These indicators are intended to provide the basis for a comprehensive national assessment of the current state and trends in the nation's ecosystems. Policy Perspectives on Indicators Indicators are most likely to be useful if they are understandable, quantifiable, and broadly applicable. They are likely to command atten- tion if they capture changes of significance to many people in many places. Although indicators of local effects are not without value, they must be aggregated into some composite indicator if they are to serve broad policy purposes. Indicators are most policy-relevant if they are easily inter- preted in terms of environmental trends or progress toward clearly articulated policy goals, and if their relevance is made clear (Landres 1992~. In other words, indicators that convey information meaningful to
OCR for page 50
50 ECOLOGICAL INDICATORS FOR THE NATION decision makers and in a form these decision makers and the public can understand are more likely to be observed and acted on. Indicators are also more likely to be influential if they are few in number and capture key features of environmental systems in a highly condensed but under- standable way. The manner in which data are aggregated to yield a small number of general indicators should be clear, especially to those who wish to understand how the indicators were developed. The reasons for choosing indicators, and the selection criteria, should also be clear (Landres 1992~. Any objective ecological indicators should be expressed numerically, so that results can be compared with those of indicators in other places and times. For the indicators to command attention and be used, the data and calculations they are based on must be credible. The choice of what indicators to use and how to define them is necessarily somewhat subjec- tive, but the procedures for measurement and calculations associated with a particular indicator, once defined, must be clearly specified, repeatable, and as free of subjective judgments as possible. Where they are unavoid- able, the sources of subjectivity should be defined and identified (Landres 1992, Susskind and Dunlap 1981~. For example, the Consumer Price Index and the percent of people unemployed are calculated by well-defined rules that have been agreed on, regardless of a person's view about the value of full employment or low inflation or even the validity of these indices. Debates about these numbers do not involve who calculated them. Similarly, ecological indicators need to be based on calculations that are well defined and agreed on. In addition to being based on credible measurements and calcula- tions, the choice, motivation, and interpretation of indicators should be publicly trusted for them to be of greatest use. That means that the people and organizations who produce the indicators should be generally trusted (Greenwalt 1992~. The committee cannot specify the best methods for achieving this goal, but notes that in at least some cases separating the responsibility for preparing indicators from responsibility for carrying out policies based on them seems to enhance trust in the indicators. For example, the Bureau of the Census has no policy-making responsibility; so, despite recent political arguments about the validity of sampling as opposed to counting everyone, the population estimates produced by the Bureau are usually trusted. Similarly, the National Weather Service has no responsibility for environmental policies, and so, beyond some scien- tific questions about the nature and placement of its instruments, its statistics are generally widely respected and trusted. The importance of public trust in the indicators is even more critical if ecological indicators are to be used as input for a national assessment of the state of the nation's ecosystems, as we hope they will be.
Representative terms from entire chapter: