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.
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 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