will be presented. For information on snowfall and its long-term variability across the North American continent, the paper later in this section by Groisman and Easterling (1995) is recommended.
In situ snow-cover data are gathered mainly over land. Only a few short-term studies have measured snow on sea ice or ice sheets (e.g., Hanson, 1980). Most observations on land are made on a once-per-day basis. The general practice is to record the average depth of snow lying on level, open ground that has a natural surface cover. At primary stations, the water equivalent of the snowpack may also be measured. In some regions, snow courses have been established where snow depth, water equivalent, and perhaps other pack properties are measured along prescribed transects across the landscape. Observations are often made only once per month, and the number of courses is extremely limited in North America. More frequent and abundant course data are gathered in the Commonwealth of Independent States, and currently this information is being recovered through a cooperative effort between the U.S. National Snow and Ice Data Center (NSIDC) at the University of Colorado and A. Krenke of the Russian Academy of Sciences.
Current station observations of snow cover are of a sufficient density for climatological study in the lower elevations of the middle latitudes. Elsewhere, while data of a high quality are gathered at a number of locations (Barry, 1983), the spatial and temporal coverage of the information is often inadequate for climate study. There is no hemispheric snow-cover product based entirely on station reports. The U.S. Air Force global snow-depth product depends heavily on surface-based observations as input into a numerical model that creates daily charts with global coverage, but it must rely on extrapolations and climatology in data-sparse regions (Hall, 1986; Armstrong and Hardman, 1991). There have been a number of regional snow-cover products over the years that are based on station data. Of greatest longevity are the Weekly Weather and Crop Bulletin charts, which have been produced since 1935. These, and the daily NOAA charts, are produced for the conterminous United States mainly from first-order station observations. Therefore, neither has a particularly high resolution, and observations may be influenced by urban heat-island effects.
In a number of countries, there are numerous stations with relatively complete records of snow extending back 50 years or more (Barry and Armstrong, 1987). Until recently, most data have remained unverified and disorganized (Robinson, 1989). As a result, few studies have dealt with long-term trends or low-frequency fluctuations of snow over even small regions (e.g., Arakawa, 1957; Manley, 1969; Jackson, 1978; Pfister, 1985). Through the cooperative efforts of a number of scientists and data centers, this situation has begun to be rectified. Examples include the exchange of data through the US/USSR Bi-Lateral Environmental Data Exchange Agreement and between the Lanzhou Institute of Glaciology and Geocryology and both Rutgers University and the NSIDC. These and other data are in the process of being quality controlled, and routines to fill in gaps in snow-cover records are being developed (Hughes and Robinson, 1993; Robinson, 1993a). Clearly, there is a need to continue efforts to identify, assimilate, in some cases digitize, and in all cases validate station and snow-course observations from around the world. These data must also must be accompanied by accurate and complete metadata.
Lengthy in situ records continue to be analyzed for individual stations, and data from networks of stations have begun to be studied on a regional level. For example, marked year-to-year variability in snow-cover duration is recognized over the course of this century at Denison, Iowa (Figure 1). Snow at least 7.5 cm deep has covered the area for as much as 80 percent of the winter, but in a number of years no or only a few days have had a cover this deep. Overall, the duration of winter and spring cover was at a maximum in the 1970s, and fall cover peaked in the 1950s. Other periods of more frequent winter cover include the 1910s and the late 1930s to early 1940s. The 1920s, middle 1940s, and late 1950s to early 1960s were periods with less abundant winter cover. Missing data around 1950 prohibit a direct assessment of winter cover at Denison, but adjacent stations suggest cover was scarce at this time. All three seasons show a greater abundance of snow-cover days in the past 40 years than in the first half of the century.
Efforts are under way to develop gridded snow files for a large portion of the central United States, using data from several hundred stations. Raw and filtered winter records from four of these stations are shown in Figure 2. They are for days with snow cover =7.5 cm, and all indicate long-term fluctuations on the order of one to several decades. The Nebraska and Kansas stations show maximum durations during the past several decades, with a similar early maximum at Oshkosh, Nebraska, in the 1910s and early 1940s. Late 1920s, early 1950s, and 1970s maxima were observed at Dupree, South Dakota, the latter two ending abruptly shortly thereafter. The North Dakota station had maximum winter snow cover in the 1930s, around 1950, and in the late 1970s. The range in filtered values over the period of record was approximately two weeks in Kansas and Nebraska and seven weeks in South Dakota and North Dakota.
Snow extent is monitored using data recorded in short-wave (visible and near-infrared) and microwave wave-