4.2.2.2 Variability and Trends in Northern Hemisphere Snow Cover

In this subsection, following the hemispheric view provided by the large-scale analyses by Brown (2000) and Robinson et al. (1993), regional and national-scale studies are discussed. The mean annual NH SCA (1966–2004) is 23.9 × 10⁶ km², not including the Greenland Ice Sheet. Interannual variability of SCA is largest not in winter, when mean SCA is greatest, but in autumn (in absolute terms) or summer (in relative terms). Monthly standard deviations range from 1.0 × 10⁶ km² in August and September to 2.7 × 10⁶ km² in October, and are generally just below 2 × 10⁶ km² in non-summer months.

Since the early 1920s, and especially since the late 1970s, SCA has declined in spring (Figure 4.2) and summer, but not substantially in winter (Table 4.2) despite winter warming (see Section 3.2.2). Recent declines in SCA in the months of February through August have resulted in (1) a shift in the month of maximum SCA from February to January; (2) a statistically significant decline in annual mean SCA; and (3) a shift towards earlier spring melt by almost two weeks in the 1972 to 2000 period (Dye, 2002). Early in the satellite era, between 1967 and 1987, mean annual SCA was 24.4 × 10⁶ km². An abrupt transition occurred between 1986 and 1988, and since 1988 the mean annual extent has been 23.1 × 10⁶ km², a statistically significant (T test, p <0.01) reduction of approximately 5% (Robinson and Frei, 2000). Over the longer 1922 to 2005 period (updated from Brown, 2000), the linear trend in March and April NH SCA (Figure 4.2) is a statistically significant reduction of 2.7 ± 1.5 × 10⁶ km² or 7.5 ± 3.5%.

Figure 4.2. Update of NH March-April average snow-covered area (SCA) from Brown (2000). Values of SCA before 1972 are based on the station-derived snow cover index of Brown (2000); values beginning in 1972 are from the NOAA satellite data set. The smooth curve shows decadal variations (see Appendix 3.A), and the shaded area shows the 5 to 95% range of the data estimated after first subtracting the smooth curve.

Temperature variations and trends play a significant role in variability and trends of NH SCA, by determining whether precipitation falls as rain or snow, and by determining snowmelt. In almost every month, SCA is correlated with temperature in the latitude band of greatest variability in SCA, owing to the snow-albedo feedback. For example, temperature in the 40°N to 60°N band and NH SCA are highly correlated in spring (r = –0.68; updated from Brown, 2000) and the largest reductions in March-April average snow cover occurred roughly between the 0°C and 5°C isotherms (Figure 4.3). The snow-albedo feedback also helps determine the longer-term trends (for temperature see Section 3.2.2; see also M.P. Clark et al., 1999; Groisman et al., 1994).

The following paragraphs discuss regional details, including information not available or missing from the satellite data and from Brown’s (2000) hemispheric reconstruction.

Figure 4.3. Differences in the distribution of Northern Hemisphere March-April average snow cover between earlier (1967–1987) and later (1988–2004) portions of the satellite era (expressed in % coverage). Negative values indicate greater extent in the earlier portion of the record. Extents are derived from NOAA/NESDIS snow maps. Red curves show the 0°C and 5°C isotherms averaged for March and April 1967 to 2004, from the Climatic Research Unit (CRU) gridded land surface temperature version 2 (CRUTEM2v) data.

4.2.2.2.1 North America

From 1915 to 2004, North American SCA increased in November, December and January owing to increases in precipitation (Section 3.3.2; Groisman et al., 2004). Decreases in snow cover are mainly confined to the latter half of the 20th century, and are most apparent in the spring period over western North America (Groisman et al., 2004). Shifts towards earlier melt by about eight days since the mid-1960s were also observed in northern Alaska (Stone et al., 2002).

Another dimension of change in snow is provided by the annual measurements of mountain SWE near April 1 in western North America, which indicate declines since 1950 at about 75% of locations monitored (Mote et al., 2005). The date of maximum mountain SWE appears to have shifted earlier by about two weeks since 1950, as inferred from streamflow measurements (Stewart et al., 2005). That these reductions are predominantly due to warming is shown by regression analysis of streamflow (Stewart et al., 2005) and SWE (Mote, 2006) on temperature and precipitation, and by the dependence of trends in SWE (Mote et al., 2005) on elevation or equivalently mean winter temperature (Figure 4.4a), with the largest percentage changes near the 0°C level.

Figure 4.4. Dependence of trends in snow on mean winter temperature (°C) at each location. (a) Relative trends in 1 April SWE, 1950 to 2000, in the mountains of western North America (British Columbia, Washington, Oregon and California), binned by mean December to February (DJF) temperature. For each 1°C temperature bin, ‘×’ symbols indicate the mean trend, bars indicate the span of the 5 to 95% confidence interval for bins with at least 10 points, and circles indicate outliers. Total number of data points is 323 (adapted from Mote et al., 2005). (b) Relative trend in days of winter (DJF) snow cover at 109 sites in Switzerland, 1958 to 1999, binned by mean DJF temperature (adapted from Scherrer et al., 2004).

4.2.2.2.2 Europe and Eurasia

Snow cover trends in mountain regions of Europe are characterised by large regional and altitudinal variations. Recent declines in snow cover have been documented in the mountains of Switzerland (e.g., Scherrer et al., 2004) and Slovakia (Vojtek et al., 2003), but no change was observed in Bulgaria over the 1931 to 2000 period (Petkova et al., 2004). Declines, where observed, were largest at lower elevations, and Scherrer et al. (2004) statistically attributed the declines in the Swiss Alps to warming, as is clear when trends are plotted against winter temperature (Figure 4.4b).

Lowland areas of central Europe are characterised by recent reductions in annual snow cover duration by about 1 day yr^–1 (e.g., Falarz, 2002). Trends towards greater maximum snow depth but shorter snow season have been noted in Finland (Hyvärinen, 2003), the former Soviet Union from 1936 to 1995 (Ye and Ellison, 2003), and in the Tibetan Plateau (Zhang et al., 2004) since the late 1970s. Qin et al. (2006) reported no trends in snow depth or snow cover in western China since 1957.