1. Introduction
For much of the United States, snow on the ground is the defining feature of the winter season. The local impact of snow cover ranges from traffic disruption to providing essential ecological services to a wide variety of flora and fauna (Jones et al. 2013). In many regions, snowpacks are integral to water supply and management strategies, while also presenting the potential for hazards such as infrastructure damage and flooding. Snow cover also alters surface albedo, providing an important feedback in the climate system. A detailed understanding of historical changes in snow cover is needed to inform discussions of the effects of climate variability and change.
Snow cover, defined here simply as the presence of snow on the ground at a given location, is measured in several ways. Weekly, satellite-based snow-cover extent (SCE) data began in 1966, although resolution and accuracy were improved in 1972 with deployment of a new sensor technology (e.g., Robinson et al. 1993). These data are typically aggregated to scales of tens to thousands of kilometers, and have made possible many studies addressing historical changes in the continental United States or larger-scale snow cover (Groisman et al. 1993, 1994; Karl et al. 1993; Frei and Robinson 1999; Dery and Brown 2007; Brown and Mote 2009). Satellite-derived snow products other than SCE [e.g., snowpack water equivalent (SWE) and snowpack depth (SD)] have considerable uncertainty (e.g., Hancock et al. 2013) and are currently of limited usefulness.
Snow course (www.wcc.nrcs.usda.gov/snowcourse/sc-data.html) and snow telemetry (SNOTEL; Serreze et al. 1999; www.wcc.nrcs.usda.gov/snow) data in the western United States provide less spatial coverage than satellite data but more relevant details for water supply and risk assessments. Snow course sites, each of which consists of multiple monthly winter measurements of SWE and SD to represent average snowpack over an area, had achieved relatively good coverage by the mid-twentieth century at higher elevations. Currently there are nearly 1000 snow course sites in the continental western United States, but they are gradually being replaced by SNOTEL sites, which provide collocated precipitation, air temperature, and SWE data at a daily frequency (SD and other quantities are available at a subset of SNOTEL sites). These sites are equipped with snow pillows to measure the site-specific snowpack and thus provide better temporal coverage, but more limited spatial representation, than the snow courses. There are 855 active SNOTEL stations in the western continental United States in 2015; over 80% of these are above 1500 m. Taken together, these two data sources are at the core of water supply forecasting in the western United States and have also yielded some important scientific insights into snowpack trends (e.g., Mote et al. 2005; Pierce et al. 2008).
Data from the National Weather Service Cooperative Observer (COOP) comprise mostly low- to middle-elevation sites throughout the United States with observations on a daily basis, and thus are complementary to the daily, higher-elevation SNOTEL data. COOP data currently include precipitation, air temperature, snowfall, and snow depth data (only a subset of sites measures the snow quantities). COOP SD data also provide longer temporal coverage than satellite and SNOTEL data, broad spatial coverage, and site-specific details of snow cover at lower-elevation sites across the United States, and have been applied in several studies addressing trends in snow cover (Hughes and Robinson 1996; Frei et al. 1999; Brown 2000; Groisman et al. 2001; Dyer and Mote 2006; Heim 2010). Of the COOP data types, the SD data are in some aspects the most problematic. They tend to have more missing data than other data types, and are especially vulnerable to observer subjectivity and to inhomogeneities due to changes in station location and observation practices (Karl et al. 1989; Robinson and Hughes 1991). With the exception of Heim (2010), all studies of trends in snow cover cited above attempted to address the issue of station inhomogeneity by using 1) specially selected subsets of COOP stations and/or 2) spatial and temporal averaging. In both approaches, accuracy is increased but spatial detail is reduced and, especially in areas where the stations used are sparse or the terrain complex, data representativeness can be compromised (Karl et al. 1989).
The main objective of this study is to diagnose and understand the spatial and temporal patterns of any trends in snow-cover frequency, magnitude, and persistence using COOP data. The station data are investigated directly and without spatial averaging. Although SD time series at individual stations can contain inhomogeneities, when nearly all stations in a region exhibit a coherent trend pattern, confidence in the reality of the individual trends is higher. This approach provides a direct, detailed, and comprehensive view of what the in situ observations can tell us about trends in United States snow cover.
In section 2, the data used and the methods of quality assurance and analysis are described. In section 3, the results of the annual and monthly trend analyses are presented, and all results are discussed in section 4.
2. Data and methods
The main dataset used in this analysis is the Daily Global Historical Climatology Network (GHCND; Menne et al. 2012a,b), the official repository of daily U.S. climate data. GHCND draws from multiple data sources that are merged and processed through a common set of quality assurance tests. Among the data included in GHCND are daily precipitation P, maximum and minimum surface air temperatures Tmax and Tmin, SD, and incremental snowfall depth (SF). These data were extracted from GHCND for the continental United States for the period comprising snow years (SY; 1 August–31 July) 1950–2010. While data are available prior to 1950 (albeit with sparser coverage), these data were not included in the present study due in part to the practice, more common before 1950 and in the eastern half of the United States, of assuming that P on snowy days was 1/10th of the snowfall depth (rather than melting the snow and measuring this quantity). Because the ratio 1/10 is higher than the true average ratio of liquid water equivalent to snowfall, inclusion of the earlier data would result in artificially more negative P trends over the full study period (Kunkel et al. 2007). Data after 2010 were excluded from the present analysis due to a change in the National Climatic Data Center’s reporting of missing COOP data for snow quantities. The bulk of the data used here originate as COOP data [other sources are Remote Automatic Weather Stations (RAWS) and Weather Bureau Army Navy (WBAN) stations]. All GHCND data that had been flagged as having failed a quality assurance check were excluded, with the exception of failed outlier checks for SD and SF, which have a high rate of falsely identified outliers (Durre et al. 2010). Data from the Global Summary of the Day data source were also excluded because they are derived from hourly data and may not faithfully represent the daily time scale (as noted in the GHCND readme file).
In this study, the focus is on snow cover as characterized by SD within each snow year. The period studied in this paper is SY 1950–2010 (from 1 August 1949 to 31 July 2010). Because there is a great deal of variation among sites in which SY months typically have snow cover, it is also useful to define the site-specific snow-cover season (SCS) as the period within the SY during which 99% of the historical snow cover has occurred.
In a process similar to that described in Knowles et al. (2006), the station records for each of the observed quantities were culled according to the following sequential steps:
Any SY whose SCS was missing more than 20% of its daily observations was considered incomplete and was excluded from the analysis.
Any station that was missing >50% of its SCS daily observations in any given 10-yr period was excluded.
Any station that was missing >30% of its SCS daily observations in the whole study period was excluded.
For the quantities SD and SF only, any station that did not have nonzero data for an average of at least 7 days per year and 3 days per year, respectively, was discarded.
Steps 1–3 were applied to SD, SF, P, Tmax, and Tmin. These criteria ensure that the data analyzed here are sufficiently serially complete that seasonal averages or totals, trends, and other long-term patterns can be reliably calculated without significant interference from sampling errors and seasonal effects. For the quantities SD, SF, P, Tmax, and Tmin, this culling retained 766, 1478, 1126, 842, and 829 stations, respectively. For subsequent analyses involving more than one variable jointly, intersections of the respective datasets were used.
The culling parameters discussed above were arrived at after evaluation of a range of possible values. They represent a balance between the goals of spatial coverage and data completeness. When stricter values were used (e.g., 10% in step 1, 20% in step 2, 10% in step 3, and 11 days and 5 days in step 4), the results supported all major conclusions below but with considerably sparser spatial representation (for the given values, the number of retained stations was reduced by a factor of at least 5 for all quantities). When the criteria used in steps 1–4 above were relaxed (e.g., to 30% in step 1, 80% in step 2, 50% in step 3, and 3 days and 1 day in step 4), the results still supported all major conclusions in this study (and for the given values, the number of retained stations was approximately doubled for all quantities), but isolated spurious results also occurred.
To identify trends in snow cover, values for the number of days with nonzero SD (designated as Nsc) and for the dates of first and last nonzero SD in each SY were calculated. Additionally, means of SD, SF, P, Tmax, Tmin, and, to focus on trends in melt-favorable conditions, the fraction of time for which T > 0°C, designated as FT>0°C (using a sine wave to interpolate between Tmin and Tmax values), were calculated for each station’s SCS for every SY in the study period, as well as for each month in the calendar year. For each of the resulting time series, tests for completeness (less than 30% missing data), as well as for moderately significant (p < 0.10) trends in the occurrence of missing data over time, were conducted to supplement the culling steps above. Then a Kendall’s tau nonparametric trend analysis (Kendall 1938) was performed on each time series.
A least squares regression line was fitted to each time series to estimate the magnitudes of changes over the 61-yr study period, and each trend was further quantified in terms of standard deviations of the detrended yearly time series. In the rare cases when a linear fit produced (physically impossible) negative SD, SF, or P values, only the positive segment of the fit was used to determine change magnitudes.
Finally, to understand the spatial and temporal patterns of the links between quantities that underlie the long-term trends, canonical correlation analyses (CCA; Barnett and Preisendorfer 1987) were performed. CCA yields maximally correlated patterns from two sets of variables [e.g., values of FT>0°C and Nsc from all stations, or principal components (PCs) thereof]. Prior to performing the CCA, an empirical orthogonal function reconstruction (von Storch and Zwiers 1999; Beckers and Rixen 2003) was applied to fill in any missing data remaining after culling for completeness as described above. Next, all station data were standardized, then area-weighted by the normalized vector of distances from each station to its nearest station (roughly proportional to the square root of the area) (Livezey and Smith 1999) to prevent areas with denser station coverage from being overrepresented in the CCA results. A PC analysis was then performed to reduce degrees of freedom and filter noise, with only enough PCs retained to satisfy a percent variance explained (PVE) criterion of 70%. This arbitrary parameter was varied to test robustness of results. The retained PCs were standardized and the CCA was conducted. Each resulting canonical correlate’s (CC) time series was generated and low-frequency variability was explored using a low-pass filter with a 16-yr cutoff period (Mann 2004). The CC time series were also tested for trends using the Kendall’s tau test.
3. Results
a. Changes in SD, P, and Tmin
SY-average SD (Fig. 1a) at the sites studied here varies primarily with latitude east of the Rocky Mountains; in the western United States, topography is a major factor. Trends in SY-average SD (Fig. 1b) are expressed in terms of the standard deviation of SY-average SD at each site. The standard deviations of SY-average SD (not shown) exhibit a spatial pattern very similar to the mean SD values in Fig. 1a, with lower values at lower latitudes and elevations. Trends in SY-average SD are not in widespread agreement despite widespread increases in SCS-averaged Tmin (Fig. 1c). SCS-averaged Tmax trends (not shown) exhibit a similar, but weaker, pattern of warming (median change is +0.95°C for Tmin and +0.56°C for Tmax). Notably, trends in the SCS-averaged values of both Tmax and Tmin are stronger than trends in the corresponding annually averaged values. The primary factor countering the warming trend’s influence on SD appears to be increases in SCS-averaged precipitation (Fig. 1d), which is the prevailing trend east of the Rockies. The spatial pattern of standard deviations of P (not shown), in terms of which the P trends are expressed, is fairly uniform across the sites shown in Fig. 1d with the main exception being relatively high values in the western coastal mountains, where significant P trends are accordingly rare.
(a) Mean SD over the SY. (b) Trends in SY-averaged SD. (c) Trends in SCS-averaged Tmin. (d) Trends in SCS-averaged P. Symbol area is proportional to magnitude of trend in standard deviations for (b) and (d) and degrees Celsius for (c). Symbol color intensity is determined by trend statistical significance (e.g., darkest colors are p < 0.01 while lightest are p > 0.30). Sites with no trend significance (p > 0.50) are shown as gray dots. Smaller trends are plotted in front of larger trends to minimize obscuring of information. Trends are for SY 1950–2010.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0051.1
The main regional exception to the above pattern is in the Pacific Northwest (PNW), where P trends are muddled while Tmin trends, as elsewhere, are generally positive. These patterns concur with the general decline in SD in this region.
b. Changes in SF and FT>0°C
Trends in SF provide a more direct measure of an important driver of changes in snow cover via the process of snow accumulation. Also, SF is not affected by the “10:1” measurement issue (Kunkel et al. 2007). Trends in SF (Fig. 2a) appear to exhibit a pattern very similar to trends in SD (Fig. 1b), suggesting that changes in snow accumulation are indeed one cause of SD changes. These trends are expressed in terms of the standard deviation of the corresponding SF time series at each site, the spatial distribution of which (not shown) is again very similar to the pattern in Fig. 1a, with lower values at lower latitudes and elevations.
(a) Trends in SCS-averaged SF. (b) Trends in FT>0°C. Symbol area is proportional to magnitude of trend in standard deviations for (a) and in FT>0°C for (b). See Fig. 1 caption for trend plotting details.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0051.1
Another key process affecting SD is snowmelt. While general temperature trends (Fig. 1c) are a broad driver of this, a more focused measure is the fraction of time during the SCS for which T > 0°C (i.e., FT>0°C; Fig. 2b). As with trends in Tmin, this fraction exhibits a widespread increase over the period of study, meaning that the occurrence of melt-favorable conditions at nearly all sites has increased. The median change in this fraction is +0.028. While warming does affect precipitation form and thus trends in SF, the trends shown in Fig. 2b are not tailored to be directly relevant to SF trends; the threshold surface air temperature for precipitation form is typically above 0°C, the threshold used in that figure, due to the fact that snow forms well above the surface in a colder environment. Nonetheless, at the regional scale, trends in occurrence of above-freezing temperatures are a useful measure of how temperature trends drive both snow accumulation and melt.
c. Changes in length of snow season and number of days with snow cover
Another indicator of the effects of warming on snow cover is the length of the period between the first and last days in a snow year that have snow cover at a site, designated as Lsc. As with SD, the mean of Lsc (Fig. 3a) varies with latitude and topography. Trends in Lsc (Fig. 3b) show a widespread decline, with a median change of −8.1 days. Of the 766 sites used in this analysis, 17% experienced an Lsc decline of at least 28 days. The average period between first and last snow cover in the SY generally ranges from November–April for the northernmost and highest-altitude sites to January–February for the southernmost and lower sites. Trends in the date of first snow cover (Fig. 3c) show that many sites in the west, the Great Plains and parts of the upper Midwest have experienced trends toward earlier first snow cover, while the Ohio Valley and the Northeast experienced later first snow covers. In contrast, the date of last snow cover trends predominantly earlier (Fig. 3d).
(a) Average Lsc. (b) Trends in Lsc. (c) Trends in date of first SY snow cover. (d) Trends in date of last SY snow cover. Symbol area is proportional to magnitude of trend in days. See Fig. 1 caption for trend plotting details.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0051.1
The last measure used here to describe snow cover is Nsc (Fig 4). Average Nsc varies from around 150 days near the Great Lakes and at some of the higher-altitude stations in the west to the culling criterion cutoff of 7 days in the PNW and at the southernmost stations east of the Rockies. Trends in Nsc were widespread and predominantly negative; 87% of sites with significant (p < 0.05) trends experienced Nsc declines. Among all sites, the median change was −4.9 days, and 22% of sites experienced Nsc declines of at least 14 days. Sites with low mean Nsc (generally at lower latitudes and in the PNW) experienced the largest Nsc changes (both increases and declines, with a tendency toward declines) when expressed as a percentage of their record-length mean Nsc.
(a) Average Nsc. (b) Trends in Nsc. Symbol area is proportional to magnitude of trend in days. See Fig. 1 caption for trend plotting details.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0051.1
d. Monthly trends
The influence of precipitation and temperature trends on SF and Nsc varies over the course of the snow season (Fig. 5). In the relatively warm months of November, March, and April, mean temperatures are higher, so temperature trends are the dominant drivers of the SF and Nsc trend patterns. Warming drove declining SF and Nsc in the eastern half of the country in November, and across the conterminous United States in March and April. This occurred despite broad patterns of increasing P in those months. In each of these three warmer months, among stations with significant (p < 0.05) trends in each quantity, fewer than 16% experienced P declines while more than 80% experienced increasing FT>0°C and decreasing SF and Nsc. The November and March patterns correspond to the trends in timing of first and last days with nonzero SD (Figs. 3c,d). By April, all but the northernmost and highest sites have, on average, stopped experiencing subzero temperatures, and their snow cover has melted. This explains the lack of Nsc trends at all but the northernmost and highest sites in April in Fig. 5.
Monthly trends. From top to bottom, rows are months from November to April. From left to right, columns are trends in SCS-averaged P (red is decreasing; see Fig. 1d), FT>0°C (red is increasing; see Fig. 2b), SCS-averaged SF (red is decreasing; see Fig. 2a), and Nsc (red is decreasing; see Fig. 4b). The reference symbols (lower left) represent trends magnitudes of one standard deviation in the P and SF columns, a change in FT>0°C of 0.10, and 7 snow-cover days in the Nsc column.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0051.1
In the coldest months of winter, December–February, SF and Nsc appear to have responded to a combination of trends in P and FT>0°C. In generally colder areas like the Great Plains, the Great Lakes region, and the Northeast, P trends were the dominant influence on SF. These regions generally experienced increasing P in these months over the study period, with the exception of the Northeast in February.
Even in the coldest months, however, many regions experienced increasing occurrence of melt conditions (increasing FT>0°C). In many cases this counteracted, or at least muddled, the effect of increasing SF on Nsc trends. That is, the December–February Nsc trend plots tend to be redder than the corresponding SF trend plots as a result of increases in FT>0°C. Specifically, of all sites with significant (p < 0.05) trends, in December–February an average of 78% of sites experienced Nsc declines while only 53% experienced SF declines, for a difference of 25% (the average difference for November, March, and April was 5%). This suggests that Nsc is more affected by midwinter warming (through melting) than SF is.
The PNW experienced declining SF and Nsc from January through April, due to a combination of declining P in January–February and increasing occurrence of above-freezing temperatures in January, March, and April. This makes the PNW the region with the most consistent decline in snow cover during the winter and spring over the study period.
e. Canonical correlation analysis
A CCA of FT>0°C and Nsc yielded modes with high canonical correlations, consistent with the strong physical link between these quantities. The leading mode (canonical correlation = 0.94) indicates a strong correspondence between widespread increases (decreases) in FT>0°C (Fig. 6a) and widespread decreases (increases) in Nsc (Fig. 6b). This mode explains an average across all stations of 34% of the variance in FT>0°C and 18% of variance in Nsc. Both CC time series for this mode (Fig. 6c) trend significantly, corresponding to an increase in FT>0°C (p = 0.003) and a decrease in Nsc (p = 0.010), commensurate with the trends shown in Figs. 2b and 4b. The low-pass-filtered versions of the corresponding CC time series (Fig. 6c) reveal the closely linked temporal variability underlying these trends, including their apparent onset in the 1970s. This leading CCA mode was robust under variations in the PVE threshold for PC truncation and under variations in the analysis period start and end dates. A CCA for SCS-averaged P and Nsc was also performed, but none of the resulting modes that exhibited significant trends was robust under variations of analysis parameters.
Leading mode from CCA. (a) Homogeneous correlation map for FT>0°C and (b) Nsc. (c) CC time series for correlates represented in (a) and (b). The darker lines represent a low-pass-filtered version of each time series (with a cutoff of 16 yr).
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0051.1
4. Discussion
Direct analysis of all COOP sites with sufficient temporal coverage to assess trends provides a clear picture of the spatial patterns of changes in snow cover and the processes driving them. The main potential problem with this approach is the inaccurate identification of trends due to the occurrence of station inhomogeneities such as changes in station location and observation practices. These and other errors are often assumed to be random and are addressed using spatial averaging, or by only using curated subsets. These approaches are often usefully employed to improve accuracy of trend magnitudes and/or help isolate trend signals in noisy datasets, but they are not well suited for comprehensive identification of spatial patterns. By detecting or avoiding widespread data issues such as the “10:1” issue described earlier (Kunkel et al. 2007) and the large number of station relocations in the 1940s (Easterling et al. 1997), trends can be presented for all stations that have sufficient data with some confidence. Additionally, while many COOP sites are in more developed areas where trends may be affected by nonclimatic factors, the implications of this are minimized here by basing conclusions on broad patterns of change that encompass many stations.
Because trend analysis results can depend strongly on start and end dates (e.g., Abatzoglou et al. 2014), the analyses in this study were repeated with start years of SY 1960 and SY 1970, holding the end year at SY 2010, and then with end years of SY 1990 and SY 2000, with a start of SY 1950. Later start years generally resulted in more pronounced and widespread warming and snow-reduction trends. For example, the east–west opposing pattern of trends in the date of first snow cover (Fig. 3c) for the SY 1950–2010 becomes a widespread trend toward later snow-cover onset with a start year of SY 1970. Earlier end years produced the opposite effects—smaller warming and snow reduction, notably with a wide band of cooling and increasing snow cover across the southern edge of sites with snow, extending into the lower Midwest and mid-Atlantic states. These results suggest that warming and snow-reduction trends have been more pronounced and widespread in the latter decades of the study period, consistent with the CCA results.
Although this analysis does include rigorous culling of stations to support accurate trend detection, it does not use a subset of stations curated specifically to minimize artificial inhomogeneities. However, all analyses in this paper were repeated using such a dataset from Kunkel et al. (2009b). They developed a subset of all COOP stations for which station inhomogeneities in SF due to station moves, sensor changes, etc., were not evident. Although curation with regard to SF does not ensure minimization of inhomogeneities in SD observations, this is a reasonable assumption. The Kunkel et al. dataset was updated through 2010 for use in the present study. The results derived using the Kunkel et al. dataset, while considerably sparser spatially than the results generated using the entire dataset, supported the interpretation of broad trend patterns and conclusions presented here. While analysis of the entire dataset provides a much more complete spatial coverage of the study domain, it is important that results from individual stations or even clusters of a few stations not be interpreted in isolation; this has been avoided here.
The trends documented here include mixed changes in SD since 1950, with the most consistent SD trends being a regional decline in the PNW. Interpretation of SD changes is somewhat complicated by possible trends in snowpack density, and any influence of warming on SY-averaged SD tends to be swamped by the effect of high interannual precipitation variability. The more sensitive indicators Lsc and Nsc clearly depict widespread overall declines in snow-season length and number of days with nonzero SD.
Spatial trend patterns for SCS-averaged SF (Fig. 2a) were in general agreement with the SF trend patterns found using a curated subset of SF data in Kunkel et al. (2007). While SF declines are the predominant trend, the regions of increasing snowfall seen by Kunkel et al. and here in the lee of the Rockies, in the northern Great Plains, and around the Great Lakes appear to be related to a broad pattern of increasing SCS-averaged P across the middle of the country (Fig. 1d). Localized causes of SF increases have also been proposed in the cases of the lee of the southern Rockies (Kunkel et al. 2009b) and the Great Lakes (Kunkel et al. 2009a). Also, Kluver and Leathers (2015) found that in the upper Midwest/northern Great Plains region, snowfall frequency increased for all percentiles from 1930 to 2007, while in the PNW, above-median snowfall-frequency percentiles decreased.
The broad declines in number of days with nonzero SD in November and March are especially striking. These are also the months during which most stations experience their first and last snow cover of the SY, respectively. This explains the general resemblance of the Nsc trend patterns for these months to Figs. 3c and 3d. The widespread reduction of snow cover in March, combined with the particularly widespread and strong warming during that month, may reflect a positive feedback between changing albedo and air temperatures. Northern Hemisphere analyses of satellite data by Groisman et al. (1994) and Dery and Brown (2007) and of a subset of available in situ data by Brown (2000) suggest that the snow–albedo feedback has contributed to spring warming at continental scales. Knowles et al. (2006) and Feng and Hu (2007) found strong, widespread shifts in precipitation form from snowfall to rain for the month of March as well.
It is instructive to place the findings presented here in the context of findings from previous studies using satellite data and from previous studies of the United States in situ snow-cover data. The declines in various measures of snow cover presented here are consistent with broader (continental and hemispheric) declines in satellite-derived SCE found by Groisman et al. (1994) and Dery and Brown (2007). Also, the in situ data support the finding of Brown and Mote (2009) based on satellite data that Nsc is a sensitive responder to changes in climate. Hughes and Robinson (1996), studying in situ data in the central and northern Great Plains, found increases in Nsc for the period 1910–93, while the present study found a weak decline in that region, likely a result of the different study periods. Frei et al. (1999), aggregating North American in situ data to the continental scale in conjunction with satellite data, diagnosed a possible shift to an earlier snow season over the period 1900–94, in general agreement with findings of this study, although Fig. 3c herein indicates that the eastern part of the United States experienced a later snow-season start over the different period studied here.
Groisman et al. (2001) used 1950–98 COOP data to calculate United States regional averages of March SCE and mean date of last snow on the ground, finding declines across most of the country, with statistically significant trends in the west. Brown (2000) found that aggregate North American indices of SCE and SWE based on COOP data exhibited long-term (1915–97) increases in the late fall and winter months, while spring months showed clear decreases in spring since the mid-1970s. The findings presented here are generally consistent with these results, and complement these studies by providing spatial and temporal details of the trends underlying the aggregate results.
It is important when considering the results presented here to remember that the COOP dataset on which they are based severely undersamples higher elevations, which is of particular concern in the west. However, studies of higher-altitude snow-course data have documented warming-induced declines in snow-related quantities across the west that are consistent, in particular, with the findings presented here for the PNW. Mote et al. (2005) found declining SWE in western North America from 1950 to 1997 using snow-course data, and Pierce et al. (2008) attributed about half of a 1950–99 decline in the ratio of 1 April SWE to October–March total precipitation at snow courses in the western United States to forcing by anthropogenic greenhouse gases.
The most comparable analysis to the one presented here is by Heim (2010), who, in a direct analysis of COOP data covering the continental United States, found regionally varying trends in snowfall and number of days with snow cover for the period 1948–2008. Seasonal SF trend patterns presented here (Fig. 5) agree well with those in Heim (2010). However, Nsc trends presented here are generally larger and more widespread than those in Heim (2010), with an especially clear difference in the PNW, where the present analysis shows strong Nsc declines in the winter and spring, while the results of Heim (2010) show relatively small downward trends in both seasons. The longer averaging periods in Heim (2010) may contribute to these differences.
These results raise interesting questions. How much of the March warming is due to the snow–albedo feedback? Will snow-cover days begin to decline more in February, such that a snow–albedo feedback begins to affect February temperatures? More research is needed to better understand such feedbacks from the land surface to the climate system. A better understanding is also needed of the climatic and synoptic-scale phenomena underlying distinct broad regional trend patterns such as the increase in snow-cover-season P over the center of the United States (Fig. 1d) that appears to partially drive SF changes across the region (Fig. 2a).
The trends presented here exhibited an onset (within the study period) in the 1970s, commensurate with the observed temporal pattern of global warming. Under a “business-as-usual” emissions scenario, global mean temperature is projected to rise by 3.7°C by the end of the century, over 5 times the increase that has occurred over the period examined here (Hartmann et al. 2013; Collins et al. 2013). It is important to consider implications of the findings presented above in this context. If winter precipitation and/or temperature continue to increase over the interior of the country and the northeast, flooding is likely to increase in frequency and magnitude. With continued warming across the country, the number of days with snow cover would continue to decline. At the southern edge of seasonal snow cover east of the Rockies, and at low to middle elevations in the west, snow cover would become an increasingly rare occurrence. If continued warming were to become large enough, these outcomes would begin to manifest even in the currently colder regions. In addition to the questions raised above, the socioeconomic, hydrologic, and ecological implications of such changes in snow cover are important areas of continued research.
Acknowledgments
Thanks to Mary Tyree, Alexander Gershunov, and Emelia Bainto of Scripps Institution of Oceanography for sharing GHCND extraction code, to Ken Kunkel for providing the list of homogeneous stations, to Dan Cayan and the reviewers for their helpful comments, and to the multitude of COOP observers and other volunteers who make science like this possible. This work was funded by the USGS National Research Program and by USGS Priority Ecosystems Science.
REFERENCES
Abatzoglou, J. T., D. E. Rupp, and P. W. Mote, 2014: Seasonal climate variability and change in the Pacific Northwest of the United States. J. Climate, 27, 2125–2142, doi:10.1175/JCLI-D-13-00218.1.
Barnett, T. P., and R. Preisendorfer, 1987: Origins and levels of monthly and seasonal forecast skill for United States surface air temperatures determined by canonical correlation analysis. Mon. Wea. Rev., 115, 1825–1850, doi:10.1175/1520-0493(1987)115<1825:OALOMA>2.0.CO;2.
Beckers, J. M., and M. Rixen, 2003: EOF calculations and data filling from incomplete oceanographic datasets. J. Atmos. Oceanic Technol., 20, 1839–1856, doi:10.1175/1520-0426(2003)020<1839:ECADFF>2.0.CO;2.
Brown, R. D., 2000: Northern Hemisphere snow cover variability and change. 1915–97. J. Climate, 13, 2339–2355, doi:10.1175/1520-0442(2000)013<2339:NHSCVA>2.0.CO;2.
Brown, R. D., and P. W. Mote, 2009: The response of Northern Hemisphere snow cover to a changing climate. J. Climate, 22, 2124–2145, doi:10.1175/2008JCLI2665.1.
Collins, M., and Coauthors, 2013: Long-term climate change: Projections, commitments and irreversibility. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 1029–1136.
Dery, S. J., and R. D. Brown, 2007: Recent Northern Hemisphere snow cover extent trends and implications for the snow–albedo feedback. Geophys. Res. Lett., 34, L22504, doi:10.1029/2007GL031474.
Durre, I., M. J. Menne, B. E. Gleason, T. G. Houston, and R. S. Vose, 2010: Comprehensive automated quality assurance of daily surface observations. J. Appl. Meteor. Climatol., 49, 1615–1633, doi:10.1175/2010JAMC2375.1.
Dyer, J. L., and T. L. Mote, 2006: Spatial variability and trends in observed snow depth over North America. Geophys. Res. Lett., 33, L16503, doi:10.1029/2006GL027258.
Easterling, D. R., P. Jamason, D. Bowman, P. Y. Hughes, and E. H. Mason, 1997: Daily snow depth measurements from 195 stations in the United States. Dataset Documentation ORNL/CDIAC-95, NDP-059, 23 pp., doi:10.3334/CDIAC/cli.ndp059.
Feng, S., and Q. Hu, 2007: Changes in winter snowfall/precipitation ratio in the contiguous United States. J. Geophys. Res., 112, D15109, doi:10.1029/2007JD008397.
Frei, A., and D. A. Robinson, 1999: Northern Hemisphere snow extent: Regional variability 1972–1994. Int. J. Climatol., 19, 1535–1560, doi:10.1002/(SICI)1097-0088(19991130)19:14<1535::AID-JOC438>3.0.CO;2-J.
Frei, A., D. A. Robinson, and M. G. Hughes, 1999: North American snow extent: 1910–1994. Int. J. Climatol., 19, 1517–1534, doi:10.1002/(SICI)1097-0088(19991130)19:14<1517::AID-JOC437>3.0.CO;2-I.
Groisman, P. Ya., T. R. Karl, and R. R. Heim Jr., 1993: Inferences of the North American snowfall and snow cover with recent global temperature changes. Glaciological Data Rep. GD-25, World Data Center for Glaciology, 44–51.
Groisman, P. Ya., T. R. Karl, R. W. Knight, and G. L. Stenchikov, 1994: Changes of snow cover, temperature, and radiative heat balance over the Northern Hemisphere. J. Climate, 7, 1633–1656, doi:10.1175/1520-0442(1994)007<1633:COSCTA>2.0.CO;2.
Groisman, P. Ya., R. W. Knight, and T. R. Karl, 2001: Heavy precipitation and high streamflow in the contiguous United States: Trends in the twentieth century. Bull. Amer. Meteor. Soc., 82, 219–246, doi:10.1175/1520-0477(2001)082<0219:HPAHSI>2.3.CO;2.
Hancock, S., R. Baxter, J. Evans, and B. Huntley, 2013: Evaluating global snow water equivalent products for testing land surface models. Remote Sens. Environ., 128, 107–117, doi:10.1016/j.rse.2012.10.004.
Hartmann, D. L., and Coauthors, 2013: Observations: Atmosphere and surface. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 159–254.
Heim, R. R., 2010: Trends of U.S. snowfall and snow cover in a warming world, 1948–2008. 18th Conf. on Applied Climatology, Atlanta, GA, Amer. Meteor. Soc., J8.4. [Available online at https://ams.confex.com/ams/90annual/techprogram/paper_159730.htm.]
Hughes, M. G., and D. A. Robinson, 1996: Historical snow cover variability in the Great Plains region of the USA: 1910 through to 1993. Int. J. Climatol., 16, 1005–1018, doi:10.1002/(SICI)1097-0088(199609)16:9<1005::AID-JOC63>3.0.CO;2-0.
Jones, H. G., J. W. Pomeroy, D. A. Walker, and R. W. Hoham, 2013: Snow Ecology: An Interdisciplinary Examination of Snow-Covered Ecosystems. Cambridge University Press, 400 pp.
Karl, T. R., J. D. Tarpley, R. G. Quayle, H. F. Diaz, D. A. Robinson, and R. S. Bradley, 1989: The recent climate record: What it can and cannot tell us. Rev. Geophys., 27, 405–430, doi:10.1029/RG027i003p00405.
Karl, T. R., P. Ya. Groisman, R. W. Knight, and R. R. Heim, 1993: Recent variations of snow cover and snowfall in North America and their relation to precipitation and temperature variations. J. Climate, 6, 1327–1344, doi:10.1175/1520-0442(1993)006<1327:RVOSCA>2.0.CO;2.
Kendall, M. G., 1938: A new measure of rank correlation. Biometrika, 30, 81–93, doi:10.1093/biomet/30.1-2.81.
Kluver, D., and D. Leathers, 2015: Regionalization of snowfall frequency and trends over the contiguous United States. Int. J. Climatol., doi:10.1002/joc.4292, in press.
Knowles, N., M. D. Dettinger, and D. R. Cayan, 2006: Trends in snowfall versus rainfall in the western United States. J. Climate, 19, 4545–4559, doi:10.1175/JCLI3850.1.
Kunkel, K. E., M. A. Palecki, K. G. Hubbard, D. A. Robinson, K. T. Redmond, and D. R. Easterling, 2007: Trend identification in twentieth-century U.S. snowfall: The challenges. J. Atmos. Oceanic Technol., 24, 64–73, doi:10.1175/JTECH2017.1.
Kunkel, K. E., L. Ensor, M. Palecki, D. Easterling, D. Robinson, K. G. Hubbard, and K. Redmond, 2009a: A new look at lake-effect snowfall trends in the Laurentian Great Lakes using a temporally homogeneous data set. J. Great Lakes Res., 35, 23–29, doi:10.1016/j.jglr.2008.11.003.
Kunkel, K. E., M. A. Palecki, L. Ensor, K. G. Hubbard, D. Robinson, K. Redmond, and D. Easterling, 2009b: Trends in twentieth-century U.S. snowfall using a quality-controlled dataset. J. Atmos. Oceanic Technol., 26, 33–44, doi:10.1175/2008JTECHA1138.1.
Livezey, R. E., and T. M. Smith, 1999: Considerations for use of the Barnett and Preisendorfer (1987) algorithm for canonical correlation analysis of climate variations. J. Climate, 12, 303–305, doi:10.1175/1520-0442-12.1.303.
Mann, M. E., 2004: On smoothing potentially non-stationary climate time series. Geophys. Res. Lett., 31, L07214, doi:10.1029/2004GL019569.
Menne, M. J., I. Durre, R. S. Vose, B. E. Gleason, and T. G. Houston, 2012a: An overview of the Global Historical Climatology Network-Daily Database. J. Atmos. Oceanic Technol., 29, 897–910, doi:10.1175/JTECH-D-11-00103.1.
Menne, M. J., and Coauthors, 2012b: Global Historical Climatology Network-Daily (GHCN-Daily), version 3.12. NOAA National Climatic Data Center, date accessed 16 June 2014, doi:10.7289/V5D21VHZ.
Mote, P. W., A. F. Hamlet, M. P. Clark, and D. P. Lettenmaier, 2005: Declining mountain snowpack in western North America. Bull. Amer. Meteor. Soc., 86, 39–49, doi:10.1175/BAMS-86-1-39.
Pierce, D. W., and Coauthors, 2008: Attribution of declining western U.S. snowpack to human effects. J. Climate, 21, 6425–6444, doi:10.1175/2008JCLI2405.1.
Robinson, D. A., and M. G. Hughes, 1991: Snow cover variability on the northern and central Great Plains. Great Plains Res., 1 (1), 93–113.
Robinson, D. A., K. F. Dewey, and R. R. J. Heim, 1993: Global snow cover monitoring: An update. Bull. Amer. Meteor. Soc., 74, 1689–1696, doi:10.1175/1520-0477(1993)074<1689:GSCMAU>2.0.CO;2.
Serreze, M., M. Clark, R. Armstrong, D. McGinnis, and R. Pulwarty, 1999: Characteristics of the western United States snowpack from snowpack telemetry (SNOTEL) data. Water Resour. Res., 35, 2145–2160, doi:10.1029/1999WR900090.
von Storch, H., and F. W. Zwiers, 1999: Statistical Analysis in Climate Research. Cambridge University Press, 496 pp.
