where is the projected latitude. A relatively dense station network exists in the industrialized countries of North America, Europe, and East Asia. In arid, mountainous, and polar regions, however, low station densities are apparent.

2.2 Oceanic Estimates

Oceanic measurements of surface air temperature were taken from the Comprehensive Ocean-Atmospheric Data Set (COADS) for the period 1950-1979 (Fletcher et al., 1983; Slutz et al., 1985; Woodruff, 1985; Woodruff et al., 1987). Within COADS, nearly fifty-million shipboard reports were condensed into 6955 median monthly estimates for 2° of latitude by 2° of longitude boxes (Fig. 2). After a compatibility evaluation (discussed below), these gridded median records were combined with the terrestrial station records. Even coverage is apparent for most of the world's oceans except for much of the Southern and Arctic Oceans.

Median air temperature, such as contained in COADS, can be used as an unbiased estimate of the mean air temperature. Legates (1987), for instance, has demonstrated that, for a global network of stations, median monthly air temperature differed from the mean by less than 0.1°C fifty percent of the time and by less than 0.5 °C ninetyfive percent of the time. Over the oceans, median air temperature, therefore, was assumed to be compatible with the terrestrial mean data and was used as a surrogate for that field.

2.3 Reliability Problems

Calculation of means from the daily minimum and maximum measurements often introduces a bias because of asymmetry in the diurnal variation of air temperature. Schaal and Dale (1977), for example, have demonstrated that mean daily air temperature computed from a single maximum and a single minimum (common practice for much of the world) may produce estimates that are different from the true (time-integrated) daily mean by as much as 1°C. This error also can be accentuated by variations in the time of observation from place to place (Mitchell, 1958: Baker, 1975).

Jones et al. (1986) indicate that many non-firstorder weather stations in the United States take more morning observations than evening obser

vations. This translates into a decrease in the mean daily air temperature below the true value. Mean monthly estimates (averages of the daily means), therefore, will not adequately represent the true values. Jones et al. (1986) suggest transforming monthly averages "to anomaly values [calculated] from a common reference period” (p. 162). While this does not address the problem of removing bias from the reference period, it underscores this inadequacy in the mean field.

Changes in instrumentation, exposure, and station location are additional problems associated with long-term air temperature records (Jones et al., 1986). Many station moves, for instance, were documented in the NCAR data (Spangler and Jenne, 1984)-stations which moved more than 0.1° of latitude or longitude were in fact treated as different stations. Mitchell (1953) determined, however, that instrument and exposure changes have only a small effect on decadal averages at least within the United States. Environmental effects such as urbanization may have a more significant impact although their effects may be considered representative of actual changes in the ambient temperature field.

at

Variable thermometer heights are a potential source of discrepancy between the land and ocean measurements. On land, surface air temperature is usually measured shelterheight-approximately four feet (1.22 meters) above the ground. Oceanic measurements of surface air temperature, however, are taken at a shipboard height of twelve meters (Woodruff, 1987). Since the lapse rate usually is small over the oceans (cf., Fleagle et al., 1958), differences between measurements taken at twelve meters and 1.22 meters should be rather small.

No viable means of correcting these biases on a global scale was apparent and, therefore, no correction was made. The existence of such biases in this and other large-scale data sets should be recognized, however, and this presentation of these data should be interpreted accordingly.

3. Grid-Point Interpolation

Station data and oceanic box averages then were interpolated to the nodes of a 0.5° of latitude by 0.5° of longitude lattice. Many procedures have been developed for interpolating grid-point values from irregularly-spaced data (cf., Lam, 1983; Ben

16

Andes, Himalayas, Alps, and Rocky Mountains as well as over the high plateaus of Tibet, Iran, Ethiopia, and Brazil. Owing to the high-resolution of these data, even relatively small mountain ranges including the Pyrenees and the mountains of eastern Equatorial Africa exhibit their effects on surface air temperature. Mountainous islands such as New Zealand, Tasmania, New Guinea, Madagascar, Sri Lanka, and Sumatra also exhibit nontrivial decreases in average air temperature. Variations in mean annual air temperature are apparent even over small islands including St. Helena and the Cape Verde, Caroline, Fiji, Mariana, Marshall, and Solomon Island chains.

4.2 Average Intra-Annual Variation

Intra-annual variation in monthly surface air temperature is described by the temporal standard deviation of the twelve mean monthly values. The standard deviation field is inversely related to the annual mean field; that is, the largest values usually appear in high latitudes and the minimum is along the equator (Fig. 4). Within much of the intertropical convergence zone, deviations over the oceans are less than 0.5°C while they exceed 12.0°C poleward of the continental interiors of North America, Asia, and Antarctica. Intra-annual variations in surface air temperature generally are larger in the northern hemisphere owing to the greater land area.

The well-known moderating influence of the oceans on surface air temperature also is evident in the standard deviations. This effect is primarily a result of the oceans' 1) larger heat capacity, 2) lower albedo (except at high solar zenith angles), 3) increased latent heat exchange, 4) deeper penetration of sunlight, and 5) vertical mixing of heat. In addition to the large-scale manifestations of this effect (ocean-continent differences), it also is apparent at the small scale. Many small islands such as the Falkland, Canary and Mascarene Islands and New Caledonia, for example. exhibit greater temperature variations than the surrounding ocean.

In mid-latitudes, intra-annual variations in surface air temperature are smaller on the western side of each continent (e.g., Mediterranean Climate) than on the eastern side (e.g., Subtropical Humid Climate). Conversely, tropical climates exhibit smaller intra-annual variations on the eastern

side (e.g., Tropical Wet-and-Dry Climate) than on the western side (e.g., Tropical Desert Climate). Atmospheric circulation patterns and the timing of precipitation (winter maxima for Mediterranean Climate; high-sun maxima for Tropical Wetand-Dry) regulate the release of latent heat and establish these patterns.

While elevation plays an important role in decreasing the surface air temperature, its influence on the intra-annual variation appears to be minimal. The Alps, Atlas, Himalayas, Pyrenees, and the Great Dividing Range (Australia), for example, do not significantly alter the general poleward increase in intra-annual variation. A slight decrease is observed, however, along the frontrange of the Rocky Mountains and along the Sierra Nevada.

Anomalous regions of high intra-annual variation are observed in the southern portions of the Pacific, Atlantic, and Indian Oceans. These areas occur near the center of the counter-clockwise, oceanic gyres that are prevalent at approximately 30° S. Although these anomalies are intriguing, we have found no physical basis for the enhancement of the seasonal variance in these regions. A possible explanation may be that the mean air temperatures averages for these areas are based on a relatively small number of ship traverses and may not adequately represent the true mean field. Anomalously low intra-annual variations that appear farther south are probably attributable to interpolation since almost no data exist in these areas (Fig. 2).

4.3 Seasonality

Using a lower-resolution subset of the 0.5° by 0.5° field, monthly mean surface air temperature was decomposed into the annual mean (Fig. 3) and the first two annual harmonics (Figs. 5 and 6). This subset is associated with (approximately) a 4° of latitude by 5° of longitude grid in the projected (8) coordinates. All grid-point records were not evaluated because the spatial density of nodes is too high to effectively present in the form of a vector map (cf., Fig. 5).

Sabbagh and Bryson (1962) were among the first to use harmonic decomposition to evaluate a large-scale climate field (i.e., precipitation in Canada). Hsu and Wallace (1976) later used harmonic analysis to evaluate terrestrial precipita