Changes in snowpack accumulation and ablation in the intermountain west


Corresponding author: A. Harpold, Institute for Arctic and Alpine Research, University of Colorado Boulder, Boulder, CO 80309, USA. (


[1] Recent observations have documented declining snow water equivalent (SWE) and earlier melt in the coastal Cascade and Sierra Nevada mountain ranges, and climate models suggest that warming temperatures will decrease snowpack storage in the higher-elevation mountain ranges of interior western North America. To date, however, observations of changing SWE or snowmelt have been limited to the state of Colorado in the intermountain west (IMW), defined here as the Rio Grande, Colorado River, and Great Basins, which supply water to the driest regions of North America. We used daily SNOTEL data collected between 1984 and 2009 combined with the nonparametric regional Kendall test to demonstrate significant and widespread changes in the duration of snow cover in these river basins. Daily SNOTEL data demonstrated that basin average maximum SWE occurred as early as 7 March (Lower Colorado River Basin) and as late as 13 April (Upper Colorado, Yampa, and White River Basins). Although significant increases in winter temperature (T) were widespread, there were minimal changes in the day of maximum accumulation and no indications from SWE to winter precipitation ratios (SWE:P) and winter T observations that a transition from snow to rain had occurred. While there was little change in day of maximum accumulation, the duration of snow cover decreased in 11 of 13 drainage regions, and snowmelt center of mass (SM50) advanced 1 to 4 days per decade in 6 of 13 regions. There were significant trends toward a faster SM50 and shorter duration of snow cover in the highest-elevation regions (>2800 m) of the Colorado River Basin, suggesting that winter T and P may not be the primary driver of change. Our results show that the IMW hydroclimate is both spatially and temporally variable, with few changes in winter T and P in the Great Basin and drier and warmer winters in the Colorado River and Rio Grande Basins. The changes in snowmelt timing also were variable, with a shorter SM50 and less maximum SWE in the Colorado River and Rio Grande Basins. The variable response of snowpacks in the IMW to widespread warming highlights the need for additional research into the mass and energy balance of these continental snowpacks.

1. Introduction

[2] In most river basins of the western U.S. snow is the largest component of runoff, making this area sensitive to climatic changes [Barnett et al., 2005; Bales et al., 2006; McCabe and Wolock, 2007; Rauscher et al., 2008]. Decreases in snow accumulation and earlier melt have been observed in the Cascade and Sierra Nevada mountain ranges [Regonda et al., 2005; Mote et al., 2005; Mote, 2006; Cayan et al., 2001; Stewart et al., 2005; Kapnick and Hall, 2012; Fritze et al., 2011], and it is expected that warming temperatures will reduce the amount of precipitation that falls as snow and hasten the onset of spring throughout the western U.S. [McCabe and Clark, 2005; Cayan et al., 2001; Pierce et al., 2008; Knowles et al., 2006]. Macroscale hydrologic models forced with Global Circulation Model data sets predict that snowpacks and snowmelt runoff will respond if winter and spring temperatures continue to rise in the continental mountains [Christensen et al., 2004; Christensen and Lettenmaier, 2007; Rauscher et al., 2008]. Winter and spring temperatures in the western U.S. have increased over 1°C on average in the last 50 years to record high temperatures in the early 2000s [Intergovernmental Panel on Climate Change, 2007], yet changes in SWE and snowmelt timing have only been documented in the state of Colorado [Clow, 2010]. The effects of warming temperatures on higher-elevation snowpacks in the Rio Grande, Great Basin, and Colorado River Basins comprising the larger intermountain west (IMW) remains poorly understood.

[3] Snowpacks are the primary water resource in the semiarid IMW river basins [Bales et al., 2006] and decreases in SWE and earlier melts would be particularly troubling because water supplies are already overallocated [McCabe and Wolock, 2007; Rajagopalan et al., 2009]. Seasonal snowpacks act as “reservoirs” for water storage and shorter and earlier snowmelt seasons lead to greater downstream losses and less water available for ecological and human needs [Barnett et al., 2005; Bales et al., 2006]. Smaller SWE and shorter snow seasons also result in a longer growing season, but may lead to less carbon sequestration in subalpine forests due to the reliance on snow water late in the growing season [Hu et al., 2010]. Together with expanded water demands, potential changes to SWE and the timing and duration of snowmelt do not bode well for water resources in the IMW [McCabe and Wolock, 2007; Rajagopalan et al., 2009], yet snow course observations dating back to the 1950s have revealed inconsistent trends in 1 April snow water equivalent (SWE) across these river basins [Regonda et al., 2005; Mote et al., 2005; Mote, 2006; Kapnick and Hall, 2012], potentially due to a lack of trend detectability or possibly due to physical mechanisms that “buffer” the snowpack to changing temperatures.

[4] Snow course and long-term climate data indicate that from 1950 to 2000 the changes in SWE across maritime and continental western U.S. snowpacks were related to winter temperatures [Mote et al., 2005; Kapnick and Hall, 2012] and elevation [Regonda et al., 2005], which is typically attributed to warmer temperatures causing more precipitation to fall as rain versus snow in lower elevations [Knowles et al., 2006]. It is presumed that colder winter temperatures in the higher-elevation continental mountains have “buffered” these snowpacks to regional warming over the later part of the 20th century [Regonda et al., 2005; Mote et al., 2005]. In addition to colder winter temperatures, the lack of significant changes to snowpacks in the IMW has been attributed by some studies to long-term trends in the Pacific Decadal Oscillation (PDO) and El Niño–Southern Oscillation (ENSO) that affect winter P in these basins unevenly [Mote et al., 2005; Mote, 2006; Hidalgo et al., 2009; McCabe and Dettinger, 1999]. The absence of predicted changes in snowpack in the IMW therefore may be due to the relatively short record of snowpacks and high interannual variability in winter precipitation. Alternatively, temperature typically plays a relatively minor role in the snowpack energy balance and snowmelt in these cold, continental snowpacks [Male and Granger, 1981; Marks and Dozier, 1992; Cline, 1997a], resulting in a lower sensitivity to warming temperatures.

[5] The goal of this study was to evaluate changes in the amount and timing of snowpack accumulation and ablation in the IMW using daily SNOTEL station data aggregated by water management regions in the Rio Grande, Colorado, and Great Basins and evaluated using nonparametric trend analyses. Previous studies largely have been limited by biweekly or monthly snow course measurements that provide less temporal resolution on snow accumulation and ablation. An approach developed by Clow [2010]showed significant trends in SNOTEL records for the state of Colorado, demonstrating the value of that data set and a regional statistical approach to trend detection. Similarly, we used 25 years of daily data (water year (WY) 1984 to 2009) from 202 SNOTEL stations that allowed us to quantify the maximum SWE, length of snow-covered season, and the dates of maximum accumulation and completion of melt. This study therefore quantifies both changes in SWE and snowmelt timing in a large area of the U.S. with potential water supply issues [McCabe and Wolock, 2007; Bales et al., 2006; Rajagopalan et al., 2009; Seager and Vecchi, 2010].

2. Methods

[6] Daily snow depth, SWE, precipitation, and temperature data were obtained from the NRCS SNOTEL station network ( A major advantage of the SNOTEL network is continuous observations that can detect small daily changes in snow cover not captured in biweekly or monthly surveys [Clow, 2010]. Previous work by Dressler et al. [2006]indicated that SNOTEL and snow course SWE measurements are comparable over the Colorado River Basin. A disadvantage with the SNOTEL network is the shorter data record relative to snow courses, which may not capture longer-term climate oscillations. For example, the early 1980s were characterized by strong, negative trends in both Southern Oscillation Index and Northern Oscillation Index, positive trends in PDO and Pacific–North America Index, and record or near record snowfall throughout much of our study area [Mote et al., 2005; Mote, 2006; McCabe and Dettinger, 1999; Serreze et al., 1999]. Consequently, we began our analysis in WY 1984 to avoid spurious trends that could result from beginning during the anomalously wet early 1980s.

[7] All stations within the IMW with more than 20 years of data, totaling 202, were included in the analysis. The study sites (Figure 1) cover a wide range of latitudes (32.9 to 42.3 degrees) and elevations (1780 to 3530 m above mean sea level) and the snowpacks and climate vary from the relatively warm and dry winters (1 October to 1 May) of Lower Colorado (winter temperature (T) 1.8°C, maximum SWE 22 cm) to the cold and wet Yampa/White (winter T −3.7°C, maximum SWE 62 cm). Daily precipitation and SWE data outside of three standard deviations from the annual mean value were censored (excluded from the analysis). As found by other studies, SNOTEL temperatures required the most removal of daily outliers because of poor sensor data [Serreze et al., 1999; Clow, 2010] and daily data outside two standard deviations from the mean of the daily values were removed. The poor quality of temperature records at SNOTEL stations is due to the fixed height of temperature sensors and variable boundary layer conditions, as well as cold air drainage from upland terrain. The limitations in the air temperature record were particularly variable during the ablation (melt) season due to the relatively short length of time (1 to 4 weeks) and the dependence of temperature sensor height relative to the changing snowpack depth. Additional screening was necessary capture true trends in air temperatures; winter data were censored if they were calculated with less than 50 days of record and records were excluded if the station was relocated. At most sites less than 5% of the daily data were discarded and less than 5% of the mean annual values were censored. Winter temperature trends are based on data starting in WY 1991 because temperature sensors were installed later at most sites [Serreze et al., 1999]. Trends in ten variables were determined from the daily data: mean winter temperature and precipitation (winter T and P from 1 October to 1 May), maximum SWE, SM50 (days elapsed from maximum SWE until 50% of SWE is remaining, a measure of melt duration adapted from Clow [2010]), days with snow on the ground (count of days with SWE >0), 1 April SWE, day of maximum SWE (often the beginning of snowmelt), and first and last day of snow cover (1st day without snow cover moving forward/backward from day of peak SWE).

Figure 1.

The 202 SNOTEL stations used in the analysis. Color shades refer to the drainage regions. Note that the Great Basin is green, Colorado is blue, and Rio Grande is red in the subsequent figures.

[8] A regional approach was used to evaluate monotonic trends by aggregating data sets from 13 drainage regions based on their locations. The RKT is able to increase the statistical power by increasing the number of independent observations and thus trend detectability in very “noisy” hydrological data sets [Helsel and Frans, 2006]. The 13 drainage regions (Figure 1 and Table 1) are key water management areas in the Colorado, Rio Grande, and Great Basin, and represent major mountain ranges and regional climate signals. The use of river drainages provided regions with similar numbers of stations, limited the amount of subjective grouping, and provided a clearly communicable result for water managers and stakeholders.

Table 1. Properties for the 13 Drainage Regions Used in the RKT Analyses, Including the Number of Stations and Their Average Latitude and Elevationa
Drainage AreaMajor River SystemsNumber of StationsAverage Latitude (deg)Average Elevation (m)
  • a

    Average latitude and elevation are given as mean value ± coefficient of variation.

Central NevadaHumboldt and Reese1440.8 ± 0.022361 ± 0.11
Eastern SierrasTruckee, Carson, and Walker2238.8 ± 0.012419 ± 0.14
Green above FlamingGreen1442.8 ± 0.012588 ± 0.07
Green below FlamingGreen1840.6 ± 0.012864 ± 0.11
GunnisonGunnison1038.8 ± 0.063160 ± 0.12
Lower ColoradoSalt, Verde, Virgin, and Little Colorado2235.1 ± 0.052465 ± 0.11
Rio GrandeRio Grande1536.9 ± 0.023065 ± 0.09
SE UtahDirty Devil, Escalante, and San Rafael1038.7 ± 0.022843 ± 0.05
San JuanAnimas, Piedra, Conejos, and Dolores1337.8 ± 0.063092 ± 0.12
SevierSevier, San Pitch, and Beaver1438.5 ± 0.022777 ± 0.09
Upper ColoradoColorado1739.7 ± 0.013016 ± 0.09
WasatchJordan, Toole, Weber, and Bear2141.0 ± 0.022473 ± 0.11
Yampa/WhiteWhite, Yampa, and Little Snake1240.5 ± 0.012810 ± 0.11

[9] Trends within these regions were analyzed using the Kendall tau nonparametric correlation coefficient (Mann-Kendall) for individual stations [Yue et al., 2002] and the regional Kendall test (RKT) for groups of stations [Helsel and Frans, 2006], similar to other trend analysis on SNOTEL records [Clow, 2010]. Most previous studies have used parametric linear regression, but these analyses are limited by the requirement for normality in the residuals and/or short data records and the effects of outliers, which are common in hydrologic studies. The RKT is a nonparametric test that determines if there is a monotonic trend over time and assumes similar number of stations in each region and no serial correlation (seasonal snowpacks that return to zero each year) [Helsel and Frans, 2006]. The RKT therefore looks for consistency in the direction of trend at each location and tests whether there is evidence for a general trend across the region. Another advantage of the RKT is that it can be used with censored or missing data and different record lengths using Sen's method for estimating slope [Sen, 1968].

3. Results

[10] The Kendall tau results indicated that winter temperatures (T) increased significantly (between 0.5 and 2.5°C) at 153 of 202 SNOTEL stations (Figure 2a). In contrast, only one station had a significant change in winter precipitation (P), a decrease of approximately 5 cm per decade (Figure 2b). Similarly, there were few significant trends in site-specific maximum SWE, SWE:P, melt rate (SM50), or days with snow cover. Six stations exhibited a change in maximum SWE, with five decreases and one increase (Figure 2c), 18 stations exhibited significant changes in SWE:P, split between increases and decreases (Figure 2d), four stations exhibited faster melt while two showed slower melt (Figure 2e), and six stations exhibited significant changes in the number of snow-covered days, with five decreases and one increase (Figure 2f). The absence of significant changes is not surprising given the high interannual variability in amount and timing of snow at these sites (Table 2). For example, the coefficient of variation (CV) in annual maximum SWE had a mean of 0.35 (range of 0.19 to 0.78) across the 202 stations. A total of 19 stations had significant trends in either maximum SWE, SM50, or snow-covered days, with 15 of those trends toward reduced maximum SWE and melt duration. The stations that showed statistically significant trends came from all three river basins and encompassed wide ranges of elevation and latitude.

Figure 2.

(a) The change in winter temperature, (b) winter precipitation, (c) maximum SWE, (d) SWE:P ratio, (e) SM50, and (f) snow-covered days per decade from 1984 to 2009 at 202 SNOTEL stations. Filled symbols represent significant trends (p < 0.05) using Kendall tau tests. Symbol type and color refer to the three river basins inFigure 1.

Table 2. Mean and Variance of Winter Temperature and Precipitation and Eight Snowpack Properties (Maximum SWE, SM50, Snow-Covered Days, SWE:P, 1 April SWE, Day of Maximum SWE, and First and Last Days of Snow Season) for the 13 Drainage Regions From WY1984 to WY2009a
Drainage AreaAverage Winter T (°C)Average Winter P (cm)Average Maximum SWE (cm)Average SM50 (days)Average Snow-Covered Days (days)Average SWE:P (m/m)1 April SWE (cm)Day of Maximum SWEbFirst DaycLast Dayc
  • a

    Winter T is WY1991 to WY2009. Values are given as mean value ± coefficient of variation.

  • b

    Day from 1 April.

  • c

    Day from 1 October.

Central Nevada−0.3 ± 4.1750.8 ± 0.2541.45 ± 0.3227.42 ± 0.17182.45 ± 0.130.81 ± 0.1523.25 ± 0.76−4.08 ± 0.1944.32 ± 0.2218.87 ± 0.08
Eastern Sierras0.2 ± 8.2677.5 ± 0.3565.00 ± 0.5433.89 ± 0.28185.88 ± 0.200.82 ± 0.2838.99 ± 0.92−6.06 ± 0.1644.44 ± 0.38221.79 ± 0.16
Green above Flaming−4.4 ± 0.1950.3 ± 0.3344.90 ± 0.3730.90 ± 0.13213.82 ± 0.080.89 ± 0.1033.93 ± 0.579.56 ± 0.1024.77 ± 0.22235.30 ± 0.05
Green below Flaming−4.1 ± 0.4644.0 ± 0.2837.51 ± 0.3025.26 ± 0.15201.84 ± 0.120.86 ± 0.0522.42 ± 0.667.39 ± 0.1430.13 ± 0.30228.08 ± 0.07
Gunnison−5.2 ± 0.5457.4 ± 0.4451.31 ± 0.5328.43 ± 0.25219.68 ± 0.190.88 ± 0.1739.94 ± 0.6914.61 ± 0.1723.34 ± 0.86240.12 ± 0.12
Lower Colorado1.8 ± 1.0533.4 ± 0.3722.59 ± 0.5321.28 ± 0.28126.94 ± 0.250.74 ± 0.213.53 ± 2.55−24.26 ± 0.2169.84 ± 0.25182.89 ± 0.11
Rio Grande−2.9 ± 0.5549.0 ± 0.4541.37 ± 0.5327.41 ± 0.17192.32 ± 0.140.83 ± 0.1524.51 ± 1.04−1.41 ± 0.1636.16 ± 0.29221.74 ± 0.09
SE Utah−2.2 ± 0.5340.9 ± 0.2534.68 ± 0.2626.13 ± 0.16187.75 ± 0.080.85 ± 0.1017.79 ± 0.61−0.40 ± 0.0938.08 ± 0.15220.25 ± 0.04
San Juan−3.5 ± 0.7360.8 ± 0.3751.92 ± 0.4528.21 ± 0.22201.45 ± 0.180.85 ± 0.1532.01 ± 0.773.32 ± 0.1733.63 ± 0.50229.38 ± 0.11
Sevier−1.9 ± 0.5249.1 ± 0.2742.89 ± 0.3328.02 ± 0.20193.51 ± 0.150.88 ± 0.1826.33 ± 0.673.65 ± 0.1236.38 ± 0.31224.31 ± 0.09
Upper Colorado−4.5 ± 0.3451.4 ± 0.2842.85 ± 0.3126.95 ± 0.12211.75 ± 0.100.83 ± 0.0932.44 ± 0.5511.59 ± 0.1326.21 ± 0.30234.06 ± 0.06
Wasatch−1.5 ± 1.1168.9 ± 0.3559.60 ± 0.4130.80 ± 0.16200.33 ± 0.110.87 ± 0.1336.44 ± 0.732.69 ± 0.1232.01 ± 0.26225.11 ± 0.11
Yampa/White−3.7 ± 0.3972.0 ± 0.3462.39 ± 0.4229.51 ± 0.17217.50 ± 0.110.85 ± 0.0946.29 ± 0.6711.60 ± 0.1226.20 ± 0.29239.18 ± 0.08

[11] Consistent with individual station analyses, the regional Kendall test (RKT) analyses identified significant winter temperature increases of 0.5 to 0.7°C per decade in the nine drainage regions of the Colorado and Rio Grande Basins from WY1991 to WY2009, with no trends in regional temperature in the Great Basin (Figure 3a). Despite few individual station trends in winter P, the RKT analyses indicated significant decreases in winter P in 6 of 13 regional drainages, with 5 of those in the Colorado River Basin (Figure 3b and Table 3). Similarly, the maximum SWE decreased significantly in 6 of the 13 regional drainages, with 4 of those also having decreasing winter P (Figure 3c). A total of four regions exhibited a decrease in SWE:P ratios that may have caused three of those same regions to have reduced maximum SWE (Figure 3d).

Figure 3.

(a) The change in winter temperature, (b) winter precipitation, (c) maximum SWE, (d) SWE:P ratio, (e) SM50, and (f) snow-covered days per decade from 1984 to 2009 at 13 drainage regions (seeFigure 1). Filled symbols represent significant trends (p < 0.05) using RKT. Symbol type and color refer to the three river basins in Figure 1.

Table 3. Sen Slope Estimates From the RKT Analyses of Winter Temperature and Precipitation and Eight Snowpack Properties (Maximum SWE, SM50, Snow-Covered Days, SWE:P, 1 April SWE, Day of Maximum SWE, and First and Last Days of Snow Season) for the 13 Drainage Regionsa
 Winter T (°C/decade)Winter P (cm decade−1)Maximum SWE (cm decade−1)SM50 (d/decade)Snow-Covered Days (d/decade)SWE:P (decade−1)Apr 1 SWE (cm decade−1)Day of Maximum SWE (d/decade)First Day (d/decade)Last Day (d/decade)
  • a

    A dash denotes no significant slope, and asterisks indicate the following: *, p < 0.05; **, p < 0.01.

Great Basin
Central Nevada−0.15−0.60−0.90−1.18−4.29*−0.038*3.02*6.01**1.78
Eastern Sierras0.042.032.03−2.200.0091.712.19*2.98*
Green above Flaming0.64**−1.380.18−0.801.800.007−1.302.42*
Green below Flaming0.51**−2.06**−1.30*−1.67*−6.20**0.002−0.51*−1.441.72−2.52*
San Juan0.60**−4.31**−4.70**−3.86**−6.00*−0.027*−4.68**−4.03*−3.92*−4.68**
SE Utah0.45*−3.30**−2.09−2.86*−8.80**0.001−1.67*3.33*−3.31**
U. Colorado0.63**−1.27−0.64−1.25−3.18*0.009−1.17*−1.39−1.28−2.54**
L. Colorado0.50**−2.23**−1.85**−1.67*−4.20*−0.0062.19−1.74
Rio Grande
Rio Grande0.60**−3.30**−4.00**−2.00*−4.20*−0.020−3.31*−1.00−3.45**

[12] The timing of snow accumulation and ablation varied over the IMW, with the average date of maximum SWE spanning 37 days across our study domain, ranging from March 7th (Lower Colorado River Basin) to as late as 13 April (Upper Colorado, Yampa/White regions). Further, maximum SWE values were as much as 100% greater than 1 April SWE, highlighting the importance of daily data in identifying trends in IMW snowpacks (Table 2). Eight of thirteen regions had significant changes in the timing of the initiation of snow cover, with five of those trending toward a later start (Table 3). Three regions exhibited an earlier day of maximum SWE and one showed a later day (Table 3), while six regions shorter SM50 (Figure 3e) of up to 10 days per decade, with the largest changes in eastern Utah and southern Colorado (Gunnison, SE Utah, and San Juan drainages). Together, these changes in initiation and duration of melt resulted in eleven regions exhibiting shorter snow-covered seasons (Table 3 and Figure 3f), with decreases of 3 to 9 snow-covered days per decade.

[13] These snowpack trends are shown on a map in Figure 4 to highlight that changes to IMW snow accumulation and ablation are focused in the Colorado River Basin, but that trends occur in all three river basins. Increased temperatures were evenly distributed across the Rio Grande and Colorado River Basins, with no temperature increases in the Great Basin (Figure 4a). The absolute decreases in maximum SWE, number of snow-covered days, and SM50 tended to be largest in the highest-elevation regions (Figure 3c), but because the snowpack SWE and duration generally increases with elevation, the percentage changes were not as large. Decreases in maximum SWE had no clear geographical bias (Figure 4c), despite clear differences in the trends of winter T and P between the Colorado River and Great Basin (Figures 4a and 4b). The changes in the length of snowmelt were mainly observed in the Colorado River Basin, where 5 of 8 drainage regions showed decreasing SM50 (Figure 4e). Fewer snow-covered days occurred in 7 of 8 drainage regions in the Colorado River Basin (Figure 4f), with the largest changes occurring in eastern Utah and western Colorado (SE Utah, Lower Green, and San Juan drainage regions).

Figure 4.

(a) Change in winter temperature, (b) winter precipitation, (c) maximum SWE, (d) SWE:P ratio, (e) SM50, and (f) snow-covered days per decade at the 13 drainage regions used in the RKT analyses. Colored arrows refer to significant trends (p < 0.05) using RKT. Arrow size is the magnitude of change, and color refers to the three river basins inFigure 1.

[14] Decreases in SWE:P ratios occurred in four drainage regions, potentially leading to significantly less maximum SWE in the Wasatch, Yampa/White, and San Juan regions. The two regions with decreasing SWE:P and mean elevations below 2500 m (Central Nevada and Wasatch) had no change in winter T, whereas the two regions above 2500 m in the Colorado Basin with decreasing SWE:P (Yampa/White and San Juan) had a 0.5°C per decade increase in winter T but were high elevation and cold (winter T < −3°C; Tables 1 and 2 and Figure 3d). These changes in SWE:P occurred from 2500 to 3100 m (Figure 3d), suggesting that transitions from snow to rain and/or winter losses reduced SWE across the IMW.

4. Discussion

[15] Similar to previous studies, we observed widespread warming throughout the IMW, but few clear trends in maximum SWE or timing of snowpacks at individual stations [Regonda et al., 2005; Mote et al., 2005; Clow, 2010; Kapnick and Hall, 2012]. The lack of statistically significant changes in snowpacks is consistent with the high interannual variability in snow accumulation and ablation in these regions [Serreze et al., 1999; Clow, 2010]. Utilizing both daily data and a regional trend analysis to increase statistical power, we demonstrated decreases in maximum SWE and winter P, a shorter snow-covered season, and faster melts in the IMW. Our results are consistent with a recent analysis for the state of Colorado [Clow, 2010], which found similar trends in 1 April snowpacks for the Yampa/White, Upper Colorado, San Juan, and Gunnison drainages (Figure 4). The more widespread decreases in SWE previously observed in Colorado [Clow, 2010] may have partially resulted from the slightly earlier SNOTEL record employed by Clow [2010]. Patterns observed at our Colorado sites also reflect the differences between using SWE on 1 April as done by Clow [2010] and using a variable day of maximum SWE (Table 3) as in our analysis. Significant trends in 1 April snowpack properties and melt timing are generally more common than trends in maximum SWE (Table 3), particularly in the Colorado River Basin, where the day of maximum SWE occurs ±2 weeks from 1 April (Table 2), despite a high correlation between 1 April SWE and maximum SWE [Clow, 2010]. The SNOTEL records are generally consistent with snow course measurements [Dressler et al., 2006], and thus provided high temporal resolution information on snowpack SWE and climate compared to 1 April measurements [Bohr and Aguado, 2001]. Although it is well recognized that SNOTEL stations are underrepresented at high elevations [Serreze et al., 1999] and overrepresented in protected, forested environments [Molotch and Bales, 2006], the consistency of these data set with longer-term trends from snow course data sets combined with the high temporal resolution and concurrent climate data provide a valuable resource for addressing changes both in snowfall, and the ablation of snowpacks before and during melt.

[16] Decreasing trends in maximum SWE could be due to a combination of reduced winter precipitation and vapor losses due to sublimation prior to melt. Our results suggest that transitions from snow to rain are not impacting maximum SWE and melt timing in the continental IMW sites, despite extensive observations to this effect in the warmer and lower-elevation Sierra Nevada and Cascade mountain ranges [Regonda et al., 2005; Mote et al., 2005; Mote, 2006; Knowles et al., 2006; Kapnick and Hall, 2012]. The lack of temperature and precipitation trends in the lower-elevation Great Basin regions, where significant decreases in SWE:P occurred, suggests that neither changing transitions from snow to rain nor reductions in winter P can fully explain the lower maximum SWE. Instead, the reduced SWE:P could be due to increased sublimation vapor losses prior to snowmelt, which are a function of the vapor pressure deficit and the radiative and turbulent energy fluxes. Previous research has estimated 15% to 30% vapor losses via sublimation in the IMW [Male and Granger, 1981; Hood et al., 1999; Rinehart et al., 2008; Gustafson et al., 2010] associated with interception by vegetation, blowing snow, and increased turbulent and radiative energy fluxes. The limited empirical investigations of sublimation in semiarid environments have demonstrated the importance of threshold wind speeds and air temperatures above which sublimation increased rapidly [Jackson and Prowse, 2009; Zhou et al., 2012]. Because the saturation pressure of the atmosphere increases exponentially with temperature, an increased gradient in vapor pressure between the snow interstitial spaces and the atmosphere due to warming winter temperatures would increase sublimation even if temperatures remain well below 0°C [Hood et al., 1999; Jackson and Prowse, 2009; Zhou et al., 2012]. Increased solar radiation or turbulence could also act to increase sublimation and the ways that air temperature, relative humidity, solar radiation, and turbulent kinetic energy interact to control the rates and timing of sublimation are poorly understood.

[17] The length of the snow-covered season is changing throughout the IMW due to both later snowpack initiation and earlier snow disappearance, with the earlier disappearance driven largely by faster melt (shorter SM50) and not by an earlier initiation of the melt season. In the Great Basin drainage regions the occurrence of a seasonal snowpack was later by 1 to 6 days per decade, corresponding with 4 to 6 fewer snow-covered days per decade. In contrast, the Colorado River Basin had a mixture of changes to the first day of the snow season, but consistent earlier disappearance of snowpacks and 3 to 8 fewer snow-covered days per decade. Warmer fall temperatures can delay the snow season by melting small initial snowfalls before changes in albedo are sufficient to buffer energy inputs, whereas warmer springs would increase the sensible and latent heat available for melt. Although the initiation of snowmelt did not change appreciably, the timing of the center of mass of the snowmelt (SM50) was shorter by 1 to 4 days per decade in most of the Colorado and Rio Grande drainage regions. The faster snowmelt (shorter SM50) was comparable to the values found byClow [2010] using a similar analysis in the state of Colorado; however, our analysis showed the changes in SM50 were focused to the southern Colorado River and Rio Grande Basins where droughts and reduced winter P have occurred during the last several decades. The differences in regional climate trends (winter T and P) between the Great Basin and the Colorado and Rio Grande River Basins may be driven by ENSO and/or PDO, which operate on this time scale (∼25 years) [Mote, 2006; McCabe and Dettinger, 1999; Seager and Vecchi, 2010]. In contrast, the decreases in SWE:P and shorter snow-covered seasons observed in both the Great Basin (where snow comes later and winter P did not change) and the Colorado River Basin (where winter P decreased and snow melted faster), indicated that the mechanisms causing differences in snowpack mass and energy balance across the IMW are variable and highlight the need to understand regional differences in snow processes to predicting possible impacts on water resources.

[18] The largest decreases in SM50 occurred in regions with mean elevations over 2800 m and the largest decreases in snow-covered days were over 2700 m (Figures 3e and 3f), even when precipitation and SWE:P changes were mixed (Figure 3d). These results are contrary to the expectation that colder, high-elevation snowpacks will be more resistant to changes in air temperature [Regonda et al., 2005; Mote et al., 2005; Rauscher et al., 2008], but consistent with observations from PRISM data sets [Diaz and Eischeid, 2007] and recent SNOTEL data sets [Clow, 2010]. These observations highlight the importance of quantifying sublimation and vapor fluxes that affect snowpack mass and energy balance in addition to the current foci on rain to snow transitions and timing of melt. Both modeling and empirical evidence suggests that the greatest increases in temperatures in the Colorado River Basin are likely to occur at the highest elevations due to a feedback between snow cover and albedo [Diaz and Eischeid, 2007; Rauscher et al., 2008]. As stated previously, warmer temperatures would increase the vapor pressure deficit in the atmosphere and enhance sublimation. Once the cold content of the snowpack has been eliminated and melt begins, greater sensible and latent turbulent fluxes can be expected to increase melt rates. Disproportionate increases in spring temperatures, relative to the winter temperatures, could therefore increase turbulent energy fluxes and alter the duration of melt [Cayan et al., 2001], which was not explored in this work due to limitations in the SNOTEL air temperature records (see Methods). In a similar finding to this study, the sublimation and melt rates were shown to increase significantly with elevation in a semiarid region of western Canada [Jackson and Prowse, 2009], despite temperatures below 0°C and low atmospheric vapor pressure. In the present study, the minimal changes in the date of maximum accumulation (4 of 13 regions) suggested that the timing of snowmelt onset is rather stable despite a trend toward decreasing SM50 values, which is consistent with the major role that solar radiation and sun angle play in initiating melt in these regions [Male and Granger, 1981; Marks and Dozier, 1992; Cline, 1997a, 1997b] or alternatively the data is too noisy for trend detection.

[19] Lower maximum SWE, shorter snow seasons, and faster melts have previously been observed in warmer maritime mountain ranges [Regonda et al., 2005; Mote et al., 2005; Mote, 2006; Cayan et al., 2001], where sensible and latent heat inputs are a greater proportion of the snowpack energy budget [Male and Granger, 1981]. In contrast, latent heat is a sizable energy loss from the snowpack in drier conditions more similar to the IMW [Male and Granger, 1981; Marks and Dozier, 1992; Cline, 1997], and net radiation typically provides 60% to 90% of the total energy measured during snowmelt [Male and Granger, 1981; Marks and Dozier, 1992; Cline, 1997a; Marks and Winstral, 2001]. Changes in air temperatures of 1.3°C between two melt seasons were shown to increase the sensible and latent heat flux from 25% to 54% of the total snowmelt energy budget at an alpine site in Colorado [Cline, 1997b]. Although increased sensible heat input to the snowpack would be expected from increasing temperatures, latent heat fluxes may represent either an energy input or a mass and energy loss from the snowpack, depending on atmospheric vapor pressure. Effects on turbulent sensible and latent heat fluxes would be more sensitive in exposed, alpine areas, whereas most SNOTEL stations are in protected, forested areas where turbulent fluxes are reduced and shortwave radiation generally controls the energy budget during melt [Cline, 1997a; Marks and Winstral, 2001; Rinehart et al., 2008].

[20] The importance of solar radiation on IMW snowpack energy balance is highlighted by recent work on dust deposition to snow in the Colorado River Basin [Painter et al., 2007; Neff et al., 2008; Painter et al., 2010], where dust deposition can increase net radiation 10% to 20% by reducing albedo from 0.7 to 0.4 and cause earlier and faster snowmelts. Furthermore, incoming all-sky shortwave radiation has increased by 2 to 3 W m−2 per decade from 1960 to 1990 [Wild et al., 2004] and about 8 W m−2 per decade from 1995 to 2007 over North America due to both a decrease in dry aerosols and cloudiness [Long et al., 2009]. Increased dust deposition has been documented in the San Juan and Rio Grande [Neff et al., 2008; Painter et al., 2007], where we also observed an earlier day of maximum SWE (Table 3), which combined with both increasing solar radiation and temperatures may also be contributing to shorter melt durations. Widespread vegetation die-off throughout the IMW [Breshears et al., 2005; Raffa et al., 2008] is removing vegetation shading of the snowpack and increasing solar radiation and turbulent fluxes, potentially leading to greater winter sublimation losses and reduced runoff over this time period [Bewley et al., 2010]. Several recent studies have documented that decreased canopy shading of the snowpack results in increased sublimation fluxes before melt [Veatch et al., 2009; Gustafson et al., 2010], decreasing the water potentially available for ecological services and downstream water resources.

[21] The results of this study suggest that the response of IMW snowpacks to climate change is both qualitatively and quantitatively different in the Great Basin than it was in the Colorado River and Rio Grande Basins. Both spatial and temporal heterogeneity in the relative importance of different energy balance terms (e.g., radiation versus sensible and latent heat) has likely contributed to the absence of clear, consistent trends in the IMW. Regional variability in the energy balance controls the sensitivity of snowpack SWE and timing to changes in climate [Reba et al., 2011], specifically winter T and P. Temperature and turbulent energy fluxes typically play a smaller role in this colder and drier region than in the maritime snowpacks of more coastal ranges. Thus there are few changes to the timing of maximum SWE in the Colorado River Basin, despite a shorter SM50 and a faster melt. Increasing regional air temperatures or net radiation however, would be expected to increase sublimation water losses during winter and continue to speed melt, with both of these potentially contributing to increased downstream water stress. Water scarcity is of particular concern in the Colorado River Basin [McCabe and Wolock, 2007; Rajagopalan et al., 2009] where the largest changes in snowpack SWE, snowmelt duration, and snow-covered season were observed.

5. Conclusions

[22] Despite relatively few trends in the amount and timing of maximum SWE at the individual SNOTEL stations, a regional statistical analysis indicated widespread decreases in the duration of snow cover throughout the intermountain west and reduced maximum SWE and faster melts in the Colorado River and Rio Grande Basins. These observations are consistent with similar studies from the state of Colorado [Clow, 2010], but this study extended them to the larger region and highlighted the importance of using maximum SWE versus 1 April SWE. The large interannual variability in snow accumulation and ablation required daily data and regional (RKT) analysis for robust trend detection. Reduced SWE in parts of Utah and Nevada mountains were partially due to less winter P, but potentially also due to increased sublimation water losses during winter that are consistent with previous observations over semiarid snowpacks [Hood et al., 1999; Jackson and Prowse, 2009; Zhou et al., 2012]. The most widespread changes to IMW snowpacks were delayed snow cover in the fall and shorter and faster melts in the spring. However the snowpack response was spatially variable and difficult to observe, which highlights a need for additional research into snowpack mass and energy balance at distributed sites across the IMW. Our results suggest that sublimation and the snowpack partitioning prior to melt may be critical issues to consider when evaluating the sensitivity of water resources to climate change in the IMW and improved water resource predictions will require a better understanding of how radiative and turbulent energy covaries in space and time.