Temperatures across the Western United States have seen a pronounced increase over the past 50 years (e.g. Hamlet and Lettenmaier, 2007). Observed changes in hydroclimatic indicators across watersheds of the West consistent with a warming climate have led to pressing concerns over future water supplies and a host of societal and ecological ramifications. Observational findings include a reduction in the proportion of precipitation falling as snow in sub-montane regions (Knowles et al., 2006), widespread decreases in snow water equivalent (SWE) (Mote et al., 2005), an advance in the timing of snowmelt-driven streamflow (Stewart et al., 2005) and a decrease in streamflow during the driest quartile of years (Luce and Holden, 2009). Collectively, the hydroclimatological response to the observed warming has heightened flood risks during winter for transition basins (e.g. Hamlet and Lettenmaier, 2007) and resulted in seasonal shifts in streamflow (e.g. Regonda et al., 2005), both of which have implications for water management in the West.
Water storage in snowpack (SWE) at the end of the cool season is a function of both temperature and precipitation. While precipitation provides the dominant control of SWE at elevations situated above the freezing level, or 0 °C isotherm (Bradley et al., 2009), the influence of temperature is increasingly important towards the lower limits of the perennial snowpack near the 0 °C isotherm (Mote et al., 2005). Interannual to decadal variability in precipitation complicates the analysis of trends in hydrologic variables such as SWE. The influence and noise associated with interannual variations in precipitation can be removed, therein allowing for the assessment of the storage efficiency of the snowpack defined by the fraction of accumulated cold season precipitation that remains in the snowpack (SWE/P) (e.g. Pierce et al., 2008). The storage efficiency of the snowpack reflects a temperature-based signal that accounts for both the accumulation efficiency, or percent of precipitation falling as snow, and melt rate.
Attribution studies by Barnett et al. (2008) and Pierce et al. (2008) show that decreases in the SWE/P across the Western US are consistent with anthropogenic warming. However, they could only attribute approximately half of the observed decrease in SWE/P across the West to anthropogenic warming. Observed regional-scale changes in climate can arise both through anthropogenic forcing and natural variability. While observed changes at global scales are consistent with those predicted by global climate models (GCMs) run with observed 20th century anthropogenic forcing, observed changes at regional scales may often be outside the envelope of outcomes predicted by GCM ensembles (e.g. IPCC, 2007). Prior work has shown that discrepancies in observed versus predicted change can be accounted for by quantifying the influence of preferred modes of atmospheric circulation (e.g. Wu and Straus, 2004; Abatzoglou and Redmond, 2007). As snowpack losses across the West have thus far exceeded GCM predictions, the question posed here is whether differences between observations and predictions can be reconciled by accounting for observed changes in climate variability.
Climate variability across the Western US is dominated by a set of spatially fixed, recurrent perturbations to the large-scale flow field spanning decadal (Pacific Decadal Oscillation, PDO), interannual (El Niño-Southern Oscillation, ENSO) and intramonthly (Pacific-North American pattern, PNA) timescales. Modes of climate variability impose a regional-scale footprint, affecting weather regimes and subsequently surface meteorological elements. While much prior focus has been devoted to the influence of longer-lived modes such as PDO and ENSO, relatively less attention has been devoted to the influence of the PNA. The PNA is representative of the amplitude of the planetary wave field over the Pacific and North American sector, with the positive phase characterized by positive height anomalies over the western tier of North America, and negative height anomalies over both the central North Pacific and southeastern US. While prior studies have noted the seasonal influence of the PNA on hydrologic indicators across the West (e.g. Dettinger and Cayan, 1995; Cayan, 1996; Stewart et al., 2005), the PNA intrinsically operates on intramonthly timescales, with a 15-day period encapsulating the growth and decay of its life cycle (Feldstein, 2002).
Despite the intrinsic timescale of the PNA and its ability to be generated through internal mechanisms, external forcing by sea surface temperature (SST) may contribute to lower-frequency behavior in the PNA (Trenberth et al., 1998). Significant changes in boreal winter atmospheric circulation in the Pacific-North American sector have been observed over the latter half of the 20th century (e.g. Gillett et al., 2003; Deser et al., 2004). Deser and Phillips (2009) suggest that these changes are a consequence of the observed trends in tropical Indo-Pacific SST. Although these changes have not modified the inherent time scale of the PNA, the PNA has undergone a phase shift towards positive values over the period (Figure 1).
In this study, the influence of intramonthly variability in the PNA on trends in the percent of precipitation falling as snow and snowmelt across mountainous regions in the Western US during late winter (JFM) is assessed. Changes in dynamical atmospheric circulation over the last half-century associated with the PNA are hypothesized to have accelerated losses in snowpack, and the inclusion of the influence of the PNA is hypothesized to reconcile differences between predicted changes associated solely with anthropogenic forcing and those that have been observed. Section 2 outlines the datasets and methods used in the study. Results detailing the influence of the PNA on both are presented in Section 3. Section 4 provides concluding thoughts.
2. Data and methods
Most prior work to decipher climatic influences on snowpack has employed monthly datasets. Monthly climate data (e.g. precipitation and temperature) physically relate to processes involved in snow accumulation and melt; however, aggregated monthly datasets are unable to discern the phase of precipitation (rain vs snow) as well as sub-monthly periods of snowmelt as they neglect daily-or-higher frequency data that invoke contemporaneous observations of precipitation and temperature. Other studies have used daily data to assess their influence on snowpack, but have been spatially or temporally limited by the lack of a reliable long-term observational network at high altitudes where a majority of the snowpack in the Western US resides. Knowles et al. (2006) examined changes in the fraction of precipitation falling as snow across the 11 Western US states utilizing daily data from National Weather Service Cooperative Observer (COOP) stations. While they documented widespread decreases in the fraction of precipitation falling as snow, they noted that COOP stations in the West are concentrated at elevations much lower than that of the primary snowpack where routine measurements of SWE are made. To circumvent the limitation in assessing mountain snowpack trends, an alternative approach is presented that pairs daily free-air temperatures with daily precipitation.
Daily free-air temperatures from the NCEP/NCAR reanalysis are interpolated to the 0 °C surface using geopotential height and temperature fields from 1958 to 2005. Substantial differences between free-air temperatures and surface temperatures have been documented in montane regions where surface temperatures can be decoupled from free-air temperatures (e.g. Pepin and Norris, 2005). Decoupling is maximized when large-scale ridging is present as subsidence and inhibited mixing can lead to an inversion in the boundary layer. Under these conditions, the atmospheric temperature profile may allow for multiple instances of the 0 °C surface. In this study, only the uppermost elevation of the freezing level is considered. In addition, while differences between free-air temperature and surface temperatures are generally reduced when precipitation is occurring as the atmospheric lapse rate grossly approximates the moist adiabatic lapse rate of 6.5 °C/km, a few caveats exist. During precipitation events, the freezing level can fluctuate dramatically over the course of a day due to frontal passage, and can fall locally up to several hundred metres due to diabatic cooling associated with melting snow (e.g. Marwitz, 1987), due to evaporative cooling in a non-saturated atmosphere (i.e. wet-bulb freezing level), as well as due to the influence of topography on cold air damming (Steenburgh et al., 1997). Unlike sophisticated observational campaigns (e.g. Lundquist et al., 2008), the NCEP-NCAR reanalysis is not equipped to resolve such features and likely overestimates temperature and freezing level in localized situations.
The study area covers the western tier of the continental United States north of 35°N and west of 110°W, encapsulating the Cascades, Sierra Nevada, Northern Rockies and Wasatch Range. Daily gridded precipitation is obtained from Hamlet and Lettenmaier (2005) at 1/8th degree horizontal resolution from 1958 to 2005. Given the specific interest in examining mountain snowpack, the analysis is restricted to grid points (also at 1/8th degree horizontal resolution) that have a mean 1 April SWE of at least 5 cm assessed using output from the variable infiltration capacity (VIC) model (Liang et al., 1994). The VIC model incorporates daily meteorological forcing, physiographic heterogeneity and a detailed energy balance model to simulate mountain snowpack and is well accepted in providing macroscale output for use in climate studies (e.g. Mote et al., 2005) and in operational streamflow forecasting (e.g. Wood and Lettenmaier, 2006).
The daily mean freezing elevation is interpolated to the resolution of the VIC dataset to develop a dataset of snowfall equivalent (SFE, as per the nomenclature of Knowles et al., 2006), or the accumulated daily water equivalent that falls as snow. The procedure considers the aggregate mean elevation at each grid cell from VIC, the daily mean freezing elevation and the daily accumulated precipitation. For a given day all the precipitation at an elevation above the freezing level is considered to fall as snow, whereas all the precipitation at elevations below the freezing level falls as rain. Although precipitation often reaches the surface as snow at elevations below the freezing level (e.g. at temperatures above freezing) in the presence of locally lower wet-bulb temperatures as discussed above, for this analysis, snowfall entrained into snowpack storage is considered to occur only at elevations above the freezing level. While this represents an oversimplification of rain–snow delineation, prior studies using surface temperature observations show this metric to work well in regions of frequent snowfall (e.g. Casola et al., 2009).
Mountain snowpack at the end of winter is a function of both accumulation (e.g. SFE) and melt processes. The latter is examined using a positive degree-day (PDD) method. This method considers the potential for melt when daily mean temperature is above 0 °C, with degree days representative of the sum of values above freezing over the course of late winter (JFM). While the PDD approach is a simple, yet robust proxy for snowmelt, estimations of actual decreases in SWE require the use of degree-day factors (e.g. Braithwaite, 1995). Given the dependence on topographic setting, radiation and snow albedo in determining degree-day factors (e.g. Hock, 2003), raw PDD values computed from daily mean gridded free-air temperatures are utilized as a proxy for potential snowmelt.
The daily PNA index is calculated using the rotated empirical orthogonal function of 500-hPa height field northward of 20°N for all days in JFM following the procedures of Barnston and Livezey (1987). The JFM period is selected due to the fact that temperature and hydrologic indicators show significant changes in late winter, whereas comparably little change has been observed in early winter (OND) (e.g. Knowles et al., 2006; Abatzoglou and Redmond, 2007). Secondly, observations show a positive trend in the late winter PNA over the period of record (Figure 1(a)), and an increase (decrease) in the frequency of daily PNA values greater than (less than) one standard deviation (negative one standard deviation) (Figure 1(b)). By contrast, changes in the PNA during early winter have been insignificant (not shown). Linear regression between the daily PNA index and daily freezing heights for JFM reveals a quadrapole pattern similar to that seen in lower tropospheric height fields, with positive anomalies over Western North America and negative anomalies over the central North Pacific and southeastern tier of the US (Figure 2). While these results are largely consistent with a regression to surface temperature (not shown), contextualizing the influence of the PNA on freezing levels provides additional inference for snowpack storage in rain–snow transition zones.
The PNA influences both temperature and precipitation across the West; however, as the focus of this paper is on temperature-driven changes in snowpack storage efficiency, the influence of the PNA on daily precipitation is not addressed. To determine the temperature-driven trend independent of the PNA, the linear contribution of the PNA to daily freezing levels is removed from the observational record and residuals are examined to see if they are consistent with an anthropogenic signal. The linear contribution of the PNA is removed on daily freezing levels by multiplying regression fields (i.e. Figure 2) by the daily PNA index following the methodology of Hurrell et al. (1996). This procedure results in residual daily mean freezing levels void of the linear contribution of the PNA. Upon removal of the PNA signature in freezing height, daily SFE and PDD are recalculated, and trends are computed for both the observed and PNA-adjusted dataset. Statistical significance trends are qualified at the 95% level using the method of Santer et al. (2000).
The trend in late winter SFE over the period of record depicts the north–south dipole in precipitation trends across the Western US, with decreases across the Cascades and Northern Rockies, and increases across the southern Sierra Nevada largely reflected in changes in precipitation associated with the 1977 Pacific shift (Mote, 2006). Trends in the fraction of accumulated precipitation that falls as snow, hereafter referred to as SFE/P, show region-wide decreases with a nearly 10:1 ratio of decreasing trends to increasing trends (Figure 3(a)). The median trend across mountain regions in the study area is − 1.4% per decade, similar in magnitude to trends computed at weather stations in the Western US (Knowles et al., 2006; Feng and Hu, 2007). The spatial pattern of JFM SFE/P trend roughly approximates that of the trend in 1 April SWE/P, reinforcing the importance of changes in late winter snowpack accumulation efficiency in driving snowpack storage efficiency. Statistically significant negative trends (locally exceeding − 5% per decade) are observed across 35% of the analysed domain, concentrated primarily over the Cascades and Northern Rockies. By contrast, no grid points within the domain of interest exhibited statistically significant increases in SFE/P. On monthly timescales (not shown), large negative trends over lower-elevation watersheds in the Cascades, Northern Rockies and Sierra Nevada are observed in January, spatially incoherent and insignificant trends are observed in February, and significant negative trends across all montane watersheds save for the uppermost stretches of the southern Sierra are observed in March. These results are consistent with the findings of Knowles et al. (2006), and also reinforce that the largest trends are observed near the climatological freezing level, which rises between 300 and 500 m in elevation between January and March.
Linear trends of SFE/P with the influence of the PNA on daily freezing levels removed, hereafter referred to as the residual, exhibit significant changes across the major mountain ranges in the West (Figure 3(b)). Overall, there is a 2:1 ratio of decreasing trends to increasing trends, with statistically significant decreases observed across 15% of the domain. While the spatial pattern of residual trend remains relatively unchanged compared to the observed trend, decreases in SFE/P are substantially reduced in magnitude. The residual SFE/P trend for transition basins in the low-to-mid elevations of the Cascades and Northern Rockies is about half that of the observed trend, suggesting that changes in the PNA have strongly contributed to decreases in snow accumulation efficiency. The residual SFE/P trends upon removing the linear influence of the PNA on both the freezing level and precipitation (effectively modifying both the numerator and denominator) are statistically indistinguishable from those shown in Figure 3(b), suggesting that the influence of the PNA can be assessed by incorporating just the influence of temperature.
The residual trend in the percent of precipitation that falls as snow reveals a region of positive values in the Sierra Nevada. An examination of daily freezing levels and precipitation exhibits a trend, albeit not statistically significant, for a greater fraction of JFM precipitation to fall when temperature and freezing levels are climatologically lower (i.e. during January and early February). These findings are broadly consistent with regional climate change projections for the Sierra Nevada that show an increase in precipitation during mid-winter, with compensating decreases during the shoulder seasons (e.g. Cayan et al., 2008).
The influence of the PNA on SFE/P represents only a single means through which it influences the mass balance of mountain snowpack. The strong influence on the daily freezing levels and temperatures across the West (i.e. Figure 2) is also reflected in snowmelt. Observations show positive trends in PDD across lower-elevation mountains (Figure 4(a)). Statistically significant increases are present across only about 9% of the domain, mainly at elevations near the climatological freezing level in the Sierra Nevada, southern Cascades and Olympics. The much larger area with significant decreases in SFE/P (35%) is suggestive of the accumulation efficiency playing a larger role in the widespread decreases in SWE/P. The residual PDD exhibits positive trends; however, residual trends across much of the Cascades and Northern Rockies are smaller in magnitude, again suggesting the influence of the PNA on trends in snowmelt (Figure 4(b)).
Detailed analysis of observed versus residual trends in SFE/P across an elevation and temperature transect are examined for the Cascades in Washington state (46°–49°N, 120°–122.3°W). Grids are binned by elevation and by climatologic JFM mean temperature, and the median SFE/P trend is taken for each bin. The largest negative trends are observed near the climatological 0 °C line, or near 1000 m, with values exceeding − 3% per decade (Figure 5(a) and (b)). Trends calculated for the residual SFE/P show similar characteristics, but are in general about half the magnitude of observed trends in transition zones (defined here by JFM temperatures between − 3 °C and 0 °C). A time series of both observed and residual SFE/P across these transition zones demonstrate the influence of the PNA in reducing the percent of precipitation falling over the period of record (Figure 5(c)). These results suggest that the PNA has contributed up to 50% of the regional decrease in SFE/P in transition basins of the Washington Cascades, complimenting the work of Mote (2006), who found that the winter mean North Pacific Index (NPI), a close cousin of the PNA, could account for half of the trend in 1 April SWE across the Pacific Northwest over the latter half of the 20th century. Stoelinga et al. (2010) attributed a majority of the decline in snowpack in the Cascades between 1930 and 2007 to natural variability, leaving a residual trend of 1 April SWE of about − 2% per decade, similar to results shown in Figure 5 for SFE/P. While these studies considered the influence of the North Pacific atmospheric circulation on snowpack through its combined influence of precipitation and temperature, this study highlights the explanatory power of the temperature-only based PNA signal in regional decreases in SFE/P.
The largest observed decrease in SFE/P occurs near 1000 m and 0 °C, whereas the largest decrease in the residual occurs 200 m higher in elevation. These results suggest a nonlinear influence of the PNA as a function of elevation, with the largest difference between observed and residual trends both at lower elevations (<1100 m) and higher elevations (>1700 m), with comparably less difference at mid-elevations. Large differences at lower elevations are a consequence of significant contributions to lower-elevation snowpack coincident with negative PNA episodes when the freezing level drops. Given that the frequency of strong negative PNA episodes has declined over the period of record (Figure 1(b)), removing the temperature influence of the PNA during such episodes (i.e. increasing freezing levels several hundred metres) largely accounts for the observed decline. Likewise, at higher snow-dominated elevations (>1700 m), late winter precipitation typically falls as rain only during infrequent moisture-laden warm storms with exceptionally high freezing levels. Preliminary analysis suggests that such events, known as Pineapple express events (Dettinger, 2004), preferentially make landfall across the Pacific Northwest (45°–50°N) during the positive phase of the PNA. Increases in the number of such events over the second half of the observational period have likely contributed to increases in precipitation falling as rain at higher elevations and observed decreases in SFE/P. Removing the temperature influence of the PNA during such events (i.e. decreasing freezing levels several hundred metres) results in a comparably small trend in SFE/P at snow-dominated elevations.
Pierce et al. (2008) demonstrated that the late 20th century decrease in mountain snowpack storage efficiency at lower elevations across the West could be attributed to anthropogenic forcing. They also showed that observed trends in SWE/P were nearly twice those simulated by GCMs with anthropogenic forcing applied. Overall changes in SWE/P can be attributed to changes in accumulation, melt and nonlinear processes (e.g. rain on snow events), with temperature playing a role in all three factors. In this study, the observed decrease in snowfall accumulation efficiency (SFE/P) and increase in snowmelt (PDD) is demonstrated to have been hastened by late winter changes in the PNA. The proclivity for the circulation to preferentially tend towards the positive phase of the PNA in recent decades has exacerbated any anthropogenic induced increase in the elevation of the freezing level, therein contributing to the decrease in SFE/P and to the increase in snowmelt potential. Across transition elevations of the Cascades in Washington and Oregon, the PNA has regionally contributed up to half of the observed decrease in SFE/P and snowmelt over the last half-century. Results indicate that even after removing the influence of the PNA on observed SFE/P trends, significant decreases are apparent across the West, a result consistent with anthropogenic forced change.
These results support the conclusions of Dettinger and Cayan (1995) who showed that changes in atmospheric circulation over the Pacific were a contributing factor to wintertime warming and advancement in the timing of snowmelt in the Sierra Nevada over the latter half of the 20th century. Significant changes in prominent modes of atmospheric variability have been observed in both the Atlantic and Pacific sectors over the past half-century concurrent with the acceleration of global increases in temperature. It is unclear whether observed trends in modes intrinsic to the atmosphere such as the PNA are consistent with natural climate variability, or instead consistent with dynamical component of anthropogenic forcing. A subsample of the ensemble model members used in Pierce et al. (2008) showed no apparent anthropogenic induced trend in the PNA, consistent with the findings by Gillett et al. (2003, 2005) that anthropogenically forced GCMs fail to capture the patterns and magnitude of circulation changes observed over the late 20th century. Deser and Phillips (2009) suggest that discrepancies between modelled and observed circulation trends over the Pacific-North American sector are associated with differences in modelled versus observed tropical SST trends.
The sensitivity of snow accumulation and melt to temperature makes understanding regional changes in cool season temperature a crucial component in climate change assessment studies. This is particularly true given the large degree of intermodel variability in precipitation changes across portions of the mountainous Western US. Regional climate projections for mid-latitudes are critically dependent on the ability of GCMs to resolve and depict midlatitude and tropical–extratropical teleconnection patterns. Whether or not dynamical and radiative perturbations align in the future will likely determine the rate of regional changes in water in the West.
Constructive scientific and editorial comments by Crystal A. Kolden, Laura M. Edwards as well as by anonymous reviewers helped to improve this manuscript. Gridded precipitation and SWE datasets obtained from the VIC model courtesy University of Washington and Andrew W. Wood. This research is supported by the NSF Idaho EPSCoR Program and by the National Science Foundation under award number EPS-0814387.