2.1. Snow-course data

This analysis uses SWE measurements from snow courses maintained by the California Cooperative Snow Survey of the California Department of Water Resources. A review of the data collection and verification program is provided by M. Roos (http://cdec.water.ca.gov/snow/meeting/CCSS_2003_Meeting/Calif_Coop_Snow_Program.pdf). We excluded those time series that were missing more than a total of five years of observations or more than two consecutive years. These requirements yielded a total of 201 station time series beginning in the year 1950. Throughout this paper, SWE refers to 1 April SWE measurements.

2.2. Climate data

We use monthly mean temperature and total monthly precipitation datasets for the period 1950–2002 from the US National Climate Data Center (NCDC). Where available, we utilize climate data that have undergone extensive quality control as part of the United States Historical Climate Network (USHCN) (Reference Karl, Williams, Quinlan and BodenKarl and others, 1990). However, due to the sparse distribution of USHCN corrected stations in mountainous parts of California, this dataset is supplemented with all other available NCDC distributed data. For further quality control, the analysis described in this paper was performed once using only the USHCN corrected dataset, and a second time with the full NCDC dataset. There was no difference between the general results of the two analyses, apart from a greater correlation between SWE and climate variables when all the NCDC data were used. As in the case of the SWE data, we retained only those time series which were not missing two consecutive years of measurements or five years’ total. The precipitation and temperature anomalies were averaged for December–April of each year. In California, as in other areas of the western United States, winter temperature and precipitation are largely uncorrelated (average absolute correlation is 0.12). Following an analysis of Pacific Northwest SWE by Reference MoteMote (2003a), SWE time series were correlated to the mean climate anomalies of the five nearest available climate stations, with separation distance averaging 31 km and no greater than 72 km. These distances are comparable to the differences reported in the Reference MoteMote (2003a) study. This averaging scheme is adopted to limit the sensitivity of our analysis to factors local to individual climate stations (e.g. wind patterns, changes in foliage, measurement error). Snow-course stations averaged 520m higher in elevation than the nearest climate station. A description of the limitations and uncertainties in correlating lower-elevation climate measurements to higher-elevation SWE is given by Reference MoteMote (2003a). Due to these limitations, all interpretations of results are restricted to regional-scale patterns of correlations and trends.

For correlation between temporal patterns of SWE and Pacific climate modes, time series for winter means of the Niño-3 and Pacific Decadal Oscillation (PDO) indices were calculated from monthly values obtained from the Joint Institute for the Study of Atmosphere and Ocean/University of Washington website (http://tao.atmos.washington.edu/data_sets/).

2.3. Data analysis

Trends in each SWE and climate time series are determined from the slope of the linear regression line through the data points over the 52 year period. Trends given in per cent are the total change relative to the y-intercept of the regression line. Following the methodology of Reference MoteMote (2003a), we use multiple linear regression to determine the dependence of the variation and trend of each SWE time series, S(t), on corresponding reference climate time series of temperature T(t) and precipitation P(t). The resulting regression coefficients, a _t and a _p, for temperature and precipitation respectively, yield the component of the trends in SWE 〈S〉 that can be explained by trends in temperature 〈T〉 and precipitation 〈P〉 through the relation:

3.1. Trends in SWE

California’s SWE has decreased by a station-wide average of 1.2% over the past 52 years. However, this value is the mean of more significant decreases and increases over the region. The spatial distribution of SWE trend, in both cma^–1 and per cent, is shown in Figure 2. The pattern of SWE trends shows a dependency on both latitude and elevation. Selecting 2300m as the approximate elevation of zero trend in Figure 2, we find that stations below 2300 m elevation have lost, on average, 13% SWE while higher-altitude stations have gained 12%. This spatial distinction can also be made between the northern and southern halves of the study area, averaging –15% and 10% respectively.

Fig. 2.

(a) Map of linear trends in 1 April SWE for the period 1950–2002. (b) Relative change in SWE over 52 years as a percentage of the starting value (y-intercept of the regression line). (c, d) SWE trend (c) and relative trend (d) plotted against station elevation.

Since station elevation varies strongly with latitude, reflecting the altitude distribution of California, we cannot uniquely conclude whether the observed north–south contrast in SWE trend is dependent on elevation, latitude or both. The correlation coefficients between relative SWE trend and station latitude and elevation are –0.51 and 0.56 respectively. To further constrain the cause(s) of this spatial dependency, we turn to correlations with climate-station time series.

3.2. Correlation with trends in regional climate

Numerous studies have cited increases in winter precipitation and temperatures in California over the past half-century (Reference Groisman, Knight and KarlGroisman and others, 2001; Reference Lund, Seymour and KafadarLund and others, 2001). Change has been greatest in the southeastern section of the study area, where temperature has increased >3˚C and total winter precipitation has doubled at multiple stations. In agreement with observed trends in the Pacific Northwest, winter precipitation has decreased in extreme northwestern California (Reference MoteMote, 2003b).

The correlation coefficient between winter precipitation and SWE averages 0.72 and is as high as 0.91. This positive correlation is significant at the 0.05 level for nearly every SWE time series. The correlation is strongest in the southern Sierra Nevada and in the northwestern coastal mountains. Mean winter temperature is significantly anticorrelated with low- and mid-elevation SWE time series in the central Sierra Nevada and the northwest corner, with a correlation coefficient ranging down to –0.55 and averaging –0.29 across all of the study area. Consistent with observations from the Rocky Mountains, we observe a low correlation between mean winter temperature and SWE at high elevations because these stations are above the freezing altitude for much of the winter and would be insensitive to temperature variations (Reference MoteMote, 2003a).

Employing the multiple regression analysis outlined in section 2.3 to assess the possible contributions of trends in winter precipitation and temperature to trends in SWE, we find a distinct contrast between northern and southern halves of the study area (Fig. 3). The southern half of the Sierra Nevada presents a precipitation-dominated regime where trends in precipitation positively contribute, on average, 85% to trends in SWE. The highest elevations show a positive contribution of increasing temperatures to SWE trends, likely due to an increase in warmer, precipitation-bearing storms. In the northern half of the state, increased temperatures dominate the contribution to decreasing SWE trends, accounting for nearly 100% of the total trend in the northern Sierra Nevada. The exceptions to this north–south contrast are in the corners and center of the study area. Both precipitation (positively) and temperature (negatively) contribute significantly to SWE in the northeastern high desert and mid-Sierra Nevada, the later marking the transition zone between the temperature- and precipitation-dominated regimes discussed above.

Fig. 3.

Vector map of the contributions of trends in winter temperature (vertical axis) and precipitation (horizontal axis) to the trend in SWE for the period 1950–2002, expressed as a percentage of total SWE trend. A downward- or leftward-pointing arrow indicates a negative (anticorrelated) contribution of the climate trend to SWE trend. An upward- or right-pointing arrow indicates a positive contribution. Reference arrow lengths are 100% of SWE trend.

Correlating trend in SWE with the spatial distribution of climate forcing, we find that trends in SWE have the greatest absolute magnitude where the contribution to the trend is dominated by either precipitation or temperature. Weak trends in SWE are found where increased precipitation and temperature have effectively offset each other, such as the northeastern high desert region and in the transition zone between the northern and southern Sierra Nevada.

3.3. Correlation with Pacific climate indices

In order to link the observed trends in California’s SWE to global-scale climate, we determine the extent to which the observed trends in California’s SWE can be explained by cyclical behavior of the Pacific Ocean. Previous studies have found a significant, but spatially variable, correlation between western United States spring SWE and both the Niño-3 index of the El Niño–Southern Oscillation (ENSO) and the PDO (Reference Cayan, Dettinger, Diaz and GrahamCayan and others, 1998; Reference McCabe and DettingerMcCabe and Dettinger, 1999, Reference McCabe and Dettinger2002; Reference MoteMote, 2003a).

The past half-century has been characterized by a switch from a negative to a positive phase of the PDO in the mid-1970s, corresponding to an anomalous strengthening of the Aleutian low and a resulting weakening and offshore shift of the Pacific winter storm track (Reference MinobeMinobe, 1997). This regime has its largest effect on the climate of northern Pacific latitudes. In contrast, there have been several anomalously strong ENSO events in the past 20 years, which strengthen winter storms, resulting in more precipitation in California (Reference Piechota, Dracup and FovellPiechota and others, 1997). The contrasting influence of these two indices on California’s climate can be seen in Figure 4. ENSO has a strong, positive correlation with SWE in both the north and south portions of the study area, and at mid- and high elevations. The PDO is only significantly (p<0.05) anticorrelated to SWE at ten stations, all below 2000ma.s.l. in the northern half of the study area.

Using the regression analysis outlined in section 2.3, we find that the observed trends in SWE can partially be attributed to shifts in these indices and their effects on temperature and precipitation. ENSO variability accounts for, on average, 32%, and up to 77%, of the trends in SWE while the PDO accounts for, on average, 28%, and up to 80%, of the trend. There is no clear spatial pattern in the contribution of ENSO to trends in SWE, as the contribution is equally positive at varying elevations and latitudes. The PDO, however, appears to have a stronger, negative, contribution to SWE at lower elevations and at higher latitudes (r = 0.62 and –0.56 for correlations between the contribution of the PDO to SWE trend and elevation and latitude, respectively).