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



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement