Regional patterns and proximal causes of the recent snowpack decline in the Rocky Mountains, U.S.


  • Gregory T. Pederson,

    Corresponding author
    1. U.S. Geological Survey, Northern Rocky Mountain Science Center, Bozeman, Montana, USA
    2. Earth Sciences Department, Montana State University, Bozeman, Montana, USA
    • Corresponding author: G. T. Pederson, U.S. Geological Survey, Northern Rocky Mountain Science Center, 2727 University Way (Suite 2), Bozeman, MT 59715, USA. (

    Search for more papers by this author
  • Julio L. Betancourt,

    1. U.S. Geological Survey, National Research Program, Tucson, Arizona, USA
    Search for more papers by this author
  • Gregory J. McCabe

    1. U.S. Geological Survey, National Research Program, Water Resources Division, Denver Federal Center, Denver, Colorado, USA
    Search for more papers by this author


[1] We used a first-order, monthly snow model and observations to disentangle seasonal influences on 20th century,regional snowpack anomalies in the Rocky Mountains of western North America, where interannual variations in cool-season (November–March) temperatures are broadly synchronous, but precipitation is typically antiphased north to south and uncorrelated with temperature. Over the previous eight centuries, regional snowpack variability exhibits strong, decadally persistent north-south (N-S) antiphasing of snowpack anomalies. Contrary to the normal regional antiphasing, two intervals of spatially synchronized snow deficits were identified. Snow deficits shown during the 1930s were synchronized north-south by low cool-season precipitation, with spring warming (February–March) since the 1980s driving the majority of the recent synchronous snow declines, especially across the low to middle elevations. Spring warming strongly influenced low snowpacks in the north after 1958, but not in the south until after 1980. The post-1980, synchronous snow decline reduced snow cover at low to middle elevations by ~20% and partly explains earlier and reduced streamflow and both longer and more active fire seasons. Climatologies of Rocky Mountain snowpack are shown to be seasonally and regionally complex, with Pacific decadal variability positively reinforcing the anthropogenic warming trend.

1 Introduction

[2] In the Rocky Mountains, U.S., regional variability in annual snowpack accumulation over the past millennium is characterized by a north-south (N-S) dipole that breaks down occasionally, most notably with the synchronous decline since ~1980. The decline was most dramatic in the north due to much lower mean and maximum elevations and, thus, greater sensitivity to temperature (Figure 1) [Pederson et al., 2011]. The N-S antiphasing evident in snowpack reconstructions from tree rings is consistent with regional precipitation anomalies associated with Pacific climate variability [e.g., El Niño Southern Oscillation (ENSO) and Pacific Decadal Oscillation (PDO)] [Pederson et al., 2011; Wise, 2010]. The unusual and synchronous decline since 1980, however, could signal a shift from precipitation to temperature as the dominant control of large-scale snowpack patterns, which has profound consequences for water resources. Storage reservoirs can effectively buffer interannual variability in snowmelt-dominated watersheds, but decadal variability and secular trends in snowpack accumulation pose difficult challenges for conventional water planning [Milly et al., 2008].

Figure 1.

The north-south winter snowpack dipole in western North America. Regional snowpack anomalies are shown for the majority of the past millennium using tree ring–based snowpack reconstructions from Pederson et al. [2011] that have been smoothed (20 year cubic smoothing spline) to highlight interdecadal variability and centered around each regions long-term mean (500 years).

[3] Based on instrumental data for the past century, several studies have identified increases in freezing elevations [Abatzoglou, 2011], the ratio of precipitation falling as rain rather than snow [Knowles et al., 2006], frequency of rain-on-snow events [McCabe et al., 2007], and a shift toward earlier and/or faster snowmelt [McCabe and Clark, 2005; Stewart et al., 2005] as contributing to a decline in accumulated winter snowpack [i.e., 1 April snow water equivalent (SWE)] in Western North America (WNA) [McCabe and Wolock, 2009; Mote, 2006]. At the subcontinental scale, most studies concur that recent snowpack losses are associated with late winter/early spring (February–March) warming that decreased snow accumulation while advancing and increasing snowmelt, particularly at elevations near the freezing level [Hamlet et al., 2005, 2007; Kapnick and Hall, 2010; McCabe and Wolock, 2009; Minder, 2010; Nolin and Daly, 2006; Regonda et al., 2005]. Integrated climate and hydrological models attribute up to 60% of these hydroclimatic trends to greenhouse gas emissions [Barnett et al., 2008; Bonfils et al., 2008; Pierce et al., 2008], whereas other studies suggest that 10%–50% of the trend could be due to decadal shifts in the Pacific North American (PNA) pattern and other large-scale modes of climate variability [Abatzoglou, 2011; Ault et al., 2011; Pederson et al., 2010; Stoelinga et al., 2009].

[4] Although snowpack generally has declined across WNA, not all stations experienced the same magnitude, or even sign, in the long-term trend. For example, recent studies [Kapnick and Hall, 2012; Mote, 2006; Regonda et al., 2005] showed that statistically significant positive trends (1950 up to 2010) in monthly SWE are generally confined to stations in the Sierra Nevada and southern Rockies (south of 45°N), where pronounced increases have happened mostly in early winter (before February). Negative trends are more pronounced in March, when maximum temperatures rise above the freezing level, across nearly all stations in the western U.S. [Kapnick and Hall, 2012]. Here we use a first-order, monthly snow model [McCabe and Wolock, 2009] and observed climatology to further disentangle and explain historic temperature and precipitation influences on snowpack in the Rocky Mountains, where regional physiography is inherently complex, cool-season temperature anomalies are broadly synchronous, and winter/spring precipitation is normally antiphased from north to south. Additionally, cool-season temperature and precipitation are uncorrelated within and between regions, making their individual effects on snowpack separable (see Table A1 in the auxiliary material) [Cayan, 1996]. Our analysis extends previous studies [Cayan 1996; Hamlet et al., 2005; Kapnick and Hall, 2012; Mote, 2006] by deconvolving contributions of seasonal temperature and precipitation to 1 April snowpack anomalies from 1900 to 2010, specifically comparing the northern versus southern Rockies. We also place recent snowpack variations and trends in the context of tree ring reconstructed snowpack variability for the past 800 years across the two regions [Pederson et al., 2011] and show the strong coherence between results obtained from observed snow course records, proxy-based reconstructions, and snow model estimates of 1 April SWE. Finally, the snow model allows regional estimation of trends in the extent of snow area with accelerated springtime warming since 1980.

2 Data and Methods

[5] Monthly temperature and precipitation data for the period January 1895 through June 2011 were obtained on a 4 × 4 km grid from the Parameter-Elevation Regressions on Independent Slopes Model (PRISM) data set [Daly et al., 2008] ( Temperature and precipitation data for grid cells in the western U.S. (west of −102.0° longitude) serve as input to a monthly time step snow accumulation and melt model (snow model) to estimate monthly SWE values for each grid cell. Model-estimated SWE for the years 1900 through 2010 were used for analysis. SWE estimates for 1895 through 1899 were discarded to avoid effects of initial model conditions on SWE estimates.

[6] The snow model used in this study is based on concepts previously used in monthly water balance models [e.g., McCabe and Wolock, 1999, 2009, 2010; Wolock and McCabe, 1999]. Inputs to the model are monthly temperature (T) and precipitation (P); the occurrence of snow is computed as

display math

where S is monthly snowfall in millimeters (mm), P is monthly precipitation in mm, Ta is monthly air temperature in degrees Celsius (°C), Train is a threshold above which all monthly precipitation is rain, and Tsnow is a threshold below which all monthly precipitation is snow. When the monthly air temperature is between Train and Tsnow, the proportion of precipitation that is snow or rain changes linearly.

[7] In the snow model, 1 April SWE is calculated as the total accumulated snowpack after the month of March is processed. In each month, snowfall is added to the snowpack and subject to melt if the air temperature is warm enough for snowmelt. Thus, for some cases, snow, rain, and snowmelt can occur in the same month. Snowmelt is computed using a degree-day method of the following form:

display math

where M is the amount of snow storage that can be melted in a month, α is a melt rate coefficient, and d is the number of days in a month.

[8] Details of the calibration and verification of the snow model can be found in [McCabe and Wolock, 2009]. The snow model parameter set determined by McCabe and Wolock [2009] was Train = 3.0°C, Tsnow = −1.0°C, and α = 0.5. Limitations of this model are fixed through time and include the following: (1) a temperature-driven energy balance process (e.g., long-wave radiation) based on empirical relationships between snowmelt and temperature and (2) a monthly time step operation when snow accumulation and melt occur on a daily basis. These issues aside, temperature index models such as this only require widely measured air temperature data (unlike more complex models) and provide generally good model performance despite their simplicity.

[9] Using the calibrated parameters, the snow model was used to estimate 1 April SWE for the years 1900 through 2010 for each of the PRISM grid cells in the western U.S. Grid cells with large numbers of years with zero SWE cause numerical problems for some statistical analyses; therefore, only grid cells with 1 April SWE values greater than zero for at least 50% of the years during 1900 through 2010 were retained for analysis. Additionally, to ensure model results were comparable with observed and proxy reconstructed 1 April SWE, grid cells below 1 standard deviation (σ) of the mean regional snow course site elevation (ele) (northern Rockies ele ≤ 1120 m, southern Rockies ele ≤ 2496 m) were omitted (Figure 2a).

Figure 2.

Regional patterns of snowpack variability from observational snow course records, tree ring–based reconstructions, and snow model estimates of 1 April snow water equivalent (SWE). (a) Study area map showing the (black) northern and (red) southern Rockies regions for which 1 April SWE has been modeled. Delineated watersheds (thick black and red lines) show regions with snowpack records constructed from snow course observations and tree ring–based reconstructions [Pederson et al. 2011]. Colored boxes (thick dashed lines) show regions with 1 April SWE estimates from the PRISM-based snow model and include contours (thin dashed lines) showing the lowest elevation from which SWE estimates were extracted. Smoothed time series (11 year moving average) of (b) northern and (c) southern Rockies historic snowpack anomalies plotted as anomalies from their twentieth century (100 years) and long-term means (500 years).

[10] To explore proximal causes of regional 1 April SWE variations over the past century, snow model results were compared for the northern and southern Rockies, excluding areas between 40°N and 42°N that commonly exhibit mixed precipitation responses to Pacific variability [Cayan, 1996; Pederson et al., 2011; Wise, 2010]. Contributions of seasonal temperature and precipitation to SWE variability were computed annually to show the changing influence of winter (November–January) and spring (February–March) season climate on the underlying regional snowpack variability. Spring was defined as the months of February and March due to the rapid increase in Northern Hemisphere insolation and the demonstrated sensitivity of snowpack to temperature over these months. The contributions of temperature and precipitation to 1 April SWE were computed by regressing the annual time series of 1 April SWE z scores against time series of z scores of seasonal (November–January and February–March) temperature and precipitation. The resulting regression coefficients were then multiplied by the time series of temperature and precipitation z scores to derive contributions of temperature and precipitation to the SWE z scores. Regional snowpack sensitivity to temperature and precipitation with elevation was estimated by comparing areas with climate normals (1971–2000) ≥1°C and ≤−1°C in mean spring (February–March) temperature. This temperature interval served to bracket the 0°C melt/freeze isotherm previously investigated by Nolin and Daly, 2006 and buffered analyses from overlapping data. Analyses were also run for areas ≥2°C and ≤−2°C to assess temperature sensitivity differences on relatively warmer and cooler snowpack, respectively. Total and percent area covered by snow during 1 April in each of the two regions and isotherm brackets were estimated by summing up the area for all PRISM grid cells with any snow in March, and general trends were described by least squares linear regression.

[11] To contextualize recent observation and modeled snow records, we compared them to tree ring reconstructions (800+ years) of 1 April SWE for watersheds in the northern and southern Rockies, generated from a previous study [Pederson et al., 2011]. More information on the specific methods and data quality can be found in the Pederson et al. [2011] Supporting Online Material.

3 Results

[12] Comparison of the observed, tree ring reconstructed, and snow model estimated 1 April SWE records shows that the snow model skillfully captures both interannual (Figure A1) and decadal variability unique to each region (Figures 2b and 2c). To avoid the influence of extreme events (e.g., 1977) at the end of a selected interval, we used departures from the 11 year running means to identify decadal periods of anomalous high and low snowpacks. Relative to the 100 year mean, the northern Rockies show three intervals of anomalously high snowpack spanning 1900–1928, 1942–1957, and 1964–1980. In the context of previous centuries (500 year mean), however, these high snowpack events rank as average or slightly below average (Figures 1 and 2a). Conversely, significantly lower snowpack persisted from 1929 to 1941 and from 1958 to 1963, with losses intensifying from 1981 to 2010. For the southern Rockies, the twentieth century and long-term means were similar. Periods of anomalously high snowpack prevailed from 1900 to 1928, from 1937 to 1954, and from 1974 to 1990. Below average snowpack conditions in the southern Rockies persisted from 1929 to 1936, from 1955 to 1973, and from 1991 to 2010.

[13] Periods of regionally synchronous high and low snowpack anomalies, relative to 100 year means for each region, occurred before the 1980s. The proximal climatic causes for these events are apparent in Figure 3. Note that positive (negative) temperature contributions indicate cool (warm) temperatures (Figures 3, A2, and A3), and the associated regression diagnostics (e.g., R2 and RMSE) and predictor coefficients are provided in Table A2. The 1900–1928 interval, for example, shows high snowpack across the Rockies (Figures 3a and 3b). High snowpack across the northern Rockies resulted from cool winter and spring temperatures that allowed more snow to be retained despite below average precipitation (Figure 3a). Across the south, cool winter and spring temperatures combined with above average precipitation to generate one of the longest-duration high snowpack anomalies of the last millennium (Figures 1 and 3b). Compared to the 500 year mean, however, high snowpack in the early twentieth century was a strong dipole event with the northern Rockies experiencing below average conditions (Figure 1). Within this longer-term (500 year mean) context, only two events classify as synchronous negative anomalies, the “Dustbowl Drought” (1929–1936), which was caused mostly by deficits in winter and spring precipitation, and the post-1980s decline, which was driven by springtime warming (Figures 3a and 3b). Particularly since the 1990s, unusually warm springs have overwhelmed the influence of N-S precipitation dipole in Rocky Mountain snowpack.

Figure 3.

Regional estimates of the seasonal temperature and precipitation contributions to 1 April SWE anomalies and percent snow cover declines. Smoothed time series (11 year moving average) of the snow model–based 1 April SWE estimates (thick black lines) are shown for the (a) northern and (b) southern Rockies. Precipitation and temperature contributions (stacked bars) resulting in a given years snowpack anomaly are shown for the winter (November–January) (precipitation = dark blue, temperature = yellow) and spring (February–March) (precipitation = light blue, temperature = red). In these plots, cooling (warming) results in positive (negative) values for temperature contributions. Percent of snow-covered area through time estimated for the (c) northern and (d) southern Rockies landmasses lying (gray line) above the 1°C and (black line) below the −1°C isotherm in spring (February–March).

[14] The influence of spring and winter temperatures on regional snowpack anomalies shows distinct differences in timing and magnitude. Across the north, spring and winter temperatures generally exert a positive forcing that enhances snow accumulation until 1957 (Figure 3a). From 1958 to present, warming spring temperatures (especially in the month of March) contribute a strong negative influence on the regions snowpack that intensified in the mid-1980s. The southern Rockies show spring and winter temperatures that generally favored snow accumulation up until the early 1980s (Figure 3b). The negative impact of spring temperatures on 1 April SWE intensifies substantially in the early 1990s, delayed perhaps by wet winters and springs between 1974 and 1990, and enhanced by the drying trend since then.

[15] Examination of temperature relationships by elevation using the spring temperature isotherms (i.e., ±1°C and ±2°C) shows that lower, warmer elevations have been more susceptible to declines due to springtime warming (Figures A2a–A2d). The higher colder elevations exhibit snowpack variability primarily controlled by the delivery of winter and spring precipitation (Figures A2a and A2b). Recent decades, however, indicate that spring and winter temperatures have decreased snow accumulation at high elevations, with spring temperature-driven SWE declines beginning earlier (~1958) across the northern Rockies. The same relationships hold true using the ±2°C isotherm, with the exception of the lower, warmer elevations also exhibiting greater sensitivity to winter temperatures (Figures A3a–A3d). Consistent with temperature-driven declines in SWE, snow cover is also disproportionately impacted at lower elevations and across the northern Rockies (Figures 3c, 3d, A4, and A5). The low-elevation areas of both regions exhibit 20% declines in snow cover over the past century, with the northern Rockies exhibiting the greatest absolute snow cover area loss.

4 Discussion and Conclusions

[16] Regional snowpack variability in WNA historically has been defined by a quasi-stationary N-S dipole (Figure 1) [Pederson et al., 2011]. Recent decades, however, experienced regionally synchronous and persistent declines in snowpack beginning by ~1980 and driven predominantly by warmer springs (February–March) (Figures 2b, 2c, and 3) [McCabe and Wolock, 2009; Pederson et al., 2011]. Both the northern Rockies and the lower elevations north and south were most susceptible to spring temperature-driven snowpack declines, with the northern Rockies losing the greatest absolute amount of snow cover, and the lower elevations in both regions exhibiting a ~20% loss in snow-covered area (Figures 3 and A2–A5). The overall reduction in snowpack and snow cover has been linked to ecosystem and resource impacts such as earlier runoff and peak streamflow [e.g., Barnett et al., 2008], as well as extended and more active fire seasons [e.g., Westerling et al., 2006].

[17] Other empirical and modeling studies attribute post-1980 hydroclimatic trends in WNA to the superimposition of external forcing, including greenhouse gases, on inherent decadal variability that is most likely internal to the climate system [Abatzoglou, 2011; Ault et al., 2011; Barnett et al., 2008; Meehl et al., 2009; Pederson et al., 2011; Pierce et al., 2008]. Similarly, we find about 20%–50% of the post-1980s snowpack decline may be due to natural variability associated with the Pacific North American (PNA) pattern, the Northern Annular Mode (NAM), and the El Niño Southern Oscillation (ENSO) (Figure A6 and Tables A3–A5). Since the 1980s, these teleconnected ocean-atmosphere modes have trended toward more positive values in February–March, enhancing atmospheric blocking of storm systems and warming from compressional heating [Ault et al., 2011; McAfee and Russell, 2008]. Results from correlation and regression analysis (Tables A3–A5) suggest that northern Rockies 1 April SWE levels are modulated by a combination of spring (February–March) ENSO and cool-season (November–March) PNA pattern (up to 48%), with shifts in warming spring temperatures consistent with changes in the baseline PNA state ~1958 and the mid-1980s. Variability in southern Rockies SWE levels stem from a combination of winter (November–January) ENSO and spring (February–March) NAM conditions (up to 20%). On multidecadal time scales (i.e., ~20–60 years), an additional 15% of the change in northern Rockies snowpack may be tied to the PDO (Tables A3–A5), although this may be confounded by colinearity between predictors at high and low frequencies.

[18] These teleconnected modes of climate variability influence regional snowpack accumulation in the northern versus southern Rockies differentially. A positive PNA, for example, exerts the greatest influence on low snowpack conditions across the north, with a positive spring NAM contributing to low snowpack across the south (Tables A3–A5). Positive spring PNA conditions promote increased atmospheric ridging and blocking in the north, causing increased compressional heating and greater advection of warm southern air [Abatzoglou, 2011; Ault et al., 2011; McCabe and Clark, 2005]. A positive spring NAM produces similar conditions across the southern Rockies, blocking storm systems in early spring [Ault et al., 2011; McAfee and Russell, 2008]. The degree of ENSO influence is the same, though of opposite phase, due to the steering of cool-season moisture to the north during La Niña and to the south during El Niño (Figure A6 and Tables A3–A5). Decadal variability is important, with PDO conditions influencing the north and ENSO influencing the south (Figure A6). With the exception of the recent shift to a negative (cool) PDO, both PDO and ENSO appear to have reinforced the impact of anthropogenic springtime (February–March) warming on declining snowpack over the past 25 years.

[19] Although natural climate variability may explain up to half the interannual- and decadal-scale variability in snowpack, it mostly accounts for regional differences in past snowpack anomalies, and the timing and magnitude of the recent temperature-driven snowpack declines. Together with other dynamic responses of seasonal stormtracks to a warming atmosphere (e.g., Arctic Amplification) [Francis and Vavrus, 2012], decadal variability either amplifies or reduces the negative long-term trend in snowpack accumulation resulting from warming temperatures. Our results support the notion that the post-1980s era may well be the turning point for both spatial and temporal nonstationarity in snowpack accumulation. Still, decadal variability internal to the climate system will continue to drive regional anomalies in the foreseeable future and complicate efforts to improve near-term forecasting (years to decade). Fast advances in near-term (decadal) predictability of cool-season precipitation and spring temperatures could overcome the death of stationarity [Milly et al., 2008] and revolutionize water planning in the western U.S.


[20] We thank Dan Fagre and three anonymous reviewers for their helpful comments. This project was supported by the U.S. Geological Survey. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

[21] The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.