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.
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.
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.
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