Research ArticleCLIMATOLOGY

Changes in seasonal snow water equivalent distribution in High Mountain Asia (1987 to 2009)

See allHide authors and affiliations

Science Advances  17 Jan 2018:
Vol. 4, no. 1, e1701550
DOI: 10.1126/sciadv.1701550

Abstract

Snow meltwaters account for most of the yearly water budgets of many catchments in High Mountain Asia (HMA). We examine trends in snow water equivalent (SWE) using passive microwave data (1987 to 2009). We find an overall decrease in SWE in HMA, despite regions of increased SWE in the Pamir, Kunlun Shan, Eastern Himalaya, and Eastern Tien Shan. Although the average decline in annual SWE across HMA (contributing area, 2641 × 103 km2) is low (average, −0.3%), annual SWE losses conceal distinct seasonal and spatial heterogeneities across the study region. For example, the Tien Shan has seen both strong increases in winter SWE and sharp declines in spring and summer SWE. In the majority of catchments, the most negative SWE trends are found in mid-elevation zones, which often correspond to the regions of highest snow-water storage and are somewhat distinct from glaciated areas. Negative changes in SWE storage in these mid-elevation zones have strong implications for downstream water availability.

INTRODUCTION

The impacts of climate change on High Mountain Asia (HMA) have been the subject of intense debate over the past decade (17) and remain largely unconstrained because of the lack of in situ observational data in many areas, particularly those at high elevations (8). Large-scale satellite data sets, such as the Global Precipitation Measurement (9), Tropical Rainfall Measurement Mission (10), and the Gravity Recovery and Climate Experiment (GRACE) (11) missions, and modeling efforts, such as High Asia Refined Analysis (12), and Asian Precipitation–Highly Resolved Observational Data Integration Towards Evaluation (APHRODITE) (13), show spatially and temporally heterogeneous trends in temperature and precipitation in the region (7). Global and regional climate models also disagree on climate projections for much of HMA due to relatively poor understanding of the interactions between large-scale climate systems active in the region, such as the Indian summer monsoon (ISM), East Asian summer monsoon (EASM), and the winter westerly disturbances (WWD) (14).

Climatic shifts in HMA have spatial, altitudinal, and temporal components and are not evenly felt across the region. The total yearly precipitation in the ISM region (see Fig. 1, inset) has not changed significantly over the past decades, but extensive changes in the spatial and intensity distributions of rainfall have been observed (7, 8, 15, 16). Similarly, the timing and intensity of major precipitation events related to the WWD has shifted over the past 30 years (17). Because whether precipitation falls as rain or snow is determined by air temperature, changes in the timing of storm events can have large impacts on the amount of snow that falls in HMA (18) and downstream water availability (19). Consistent runoff is essential for year-round water provision for more than a billion people who rely on HMA as a water source (20).

Fig. 1 Study area.

(A) Topographic map of HMA with major catchment boundaries (black) derived from SRTM data (68) and names of major mountain ranges. Inset map shows political boundaries, as well as wind direction of major weather systems (WWD, ISM, and EASM). (B) Twenty-two–year average DJF daily SWE volume across the study area, as derived from SSMI data. (C) DJF SWE standard deviation. Each point represents a 0.25 × 0.25 dd grid cell.

Many studies have noted extensive glacial retreat across HMA (1, 2, 5, 21, 22), with a few notable exceptions, such as the so-called Karakoram Anomaly (6, 23). Heterogeneity in glacier response to climate change has been attributed to hypsometric and precipitation variation (6), differences in debris cover (21), snow cover shielding (14), and precipitation seasonality (24, 25). Extensive research has been focused on retreating glaciers as a symbol of regional climate change, although the majority of meltwater in much of HMA is generated by snowmelt (4, 8, 2629). Due to large spatial heterogeneity in snow cover, a regional assessment is difficult; very little in situ data are available for constraining large uncertainties in modeled trends.

Here, we leverage passive microwave (PM) data from the special sensor microwave imager (SSMI, 1987 to 2009), spatially averaged from raw swath data following the study of Smith and Bookhagen (30), to assess whether there have been any significant trends in snow water equivalent (SWE) across HMA from October 1987 to October 2009. We consider a set of 6680 grid cells with 0.25 decimal degree (dd) spatial resolution (approximately, 4175 × 103 km2), with, on average, 8340 measurements each (average of 1.03 measurements per day for 22 complete October to October years) across 10 major catchments, with a special focus on the Syr Darya, Amu Darya, Tarim, Indus, Ganges/Brahmaputra, and Tibetan Plateau catchments (Fig. 1). Together, these catchments serve more than a billion people, many of whom depend heavily on seasonal snowmelt. In some catchments, such as the Indus and Amu Darya, nearly 50% of the yearly water budget is derived from snowmelt (8, 27). The Syr Darya, Tarim, and Tibet catchments receive more than two thirds of their yearly water budget as snow (31, 32). Across the study region, the majority of the snowpack is at high elevations and is often poorly measured by sparse weather station coverage (fig. S1). Our data provide a quantification of SWE magnitudes and trends at a spatial scale that is not available to in situ studies and which can be used in future regional and global modeling studies to validate model output and constrain snow-related parameterizations.

Our results are provided with the caveats that (i) the climatically short study period limits attribution of the trends, and (ii) PM SWE estimates are affected by a wide range of uncertainties—particularly in deep snow and complex terrain—and increasingly lack physical bases as snow depths increase (30, 33, 34). Our results therefore provide an assessment of relative changes in SWE over the study period—and the spatial and temporal distribution of those changes—rather than concrete changes in water storage in HMA.

RESULTS

Seasonal patterns in regional snowpack

Average winter (December-January-February, DJF) SWE ranges from zero up to more than 140 mm in some high-elevation areas. In particular, the Tien Shan, Pamir, and Hindu Kush see significant snow buildup above 2-km elevation (Fig. 1B). These high-SWE areas generally follow the track of the WWDs as they move northeast from the Arabian Sea and deposit snow along high-elevation topographic breaks (Fig. 1A, inset). Southern HMA also receives snow from the tail end of the ISM as it moves north and west toward the Indus. Because each region of HMA is affected by different climatic systems—and the impacts of climate change are spatially diverse—trends in SWE are not uniform. Although annual trends (1987 to 2009) in SWE (see fig. S2) are generally negative across HMA, there are distinct seasonal and elevation heterogeneities.

In general, SWE in HMA is decreasing across the period of March to August (Fig. 2, B and C), but some areas, particularly in the Pamir–Tien Shan region, exhibit a significant positive DJF trend over our study period (Fig. 2A). A smaller geographic area, mainly in the Himalaya, Karakoram, and Kunlun Shan, shows increasing trends in fall (September-October-November, SON) SWE (Fig. 2D).

Fig. 2 Seasonality in SWE trends.

Significant (P < 0.05) (A) DJF, (B) MAM, (C) JJA, and (D) SON trends in SWE volume (1987 to 2009), as derived from SSMI data, with major catchments (black outlines, see Fig. 1A). We limit our analysis to regions where the seasonal average SWE is greater than 5 mm to remove spurious results in areas with shallow or infrequent snow cover. MAM and JJA trends across HMA are overwhelmingly negative, except a few isolated regions. DJF trends are more widely positive and are also present in SON in the western Himalaya, the Tien Shan, and the Kunlun Shan. Yearly aggregated SWE trends available in fig. S2.

DJF and SON SWE trends are generally negative from the Hindu Kush in the northern Indus catchment, along the Himalayan Front and through the Tibetan Plateau. The exception to this is along the border between the Indus and Ganges catchments (Garhwal), where positive SWE trends are observed. Increases in SON and DJF SWE in this region are possibly linked to changes in the WWD, which has resulted in increased snowfall in parts of the Indus (17, 35, 36), and are strong enough to show positive annual SWE trends in opposition to the general decrease in SWE in HMA (see fig. S2).

The northwestern region of HMA, following the track of the WWD (see Fig. 1A, inset) through the Amu Darya catchment toward the Tien Shan into the northwestern Tarim, has also seen increasing DJF SWE over the study period, which is in line with measured increases in precipitation in the Pamir and parts of the Tien Shan (22, 37) and increased snow cover in Western China (38). This is particularly true along the border between the Syr Darya and Tarim catchments in the Tien Shan, despite decreasing trends in mean annual precipitation as measured by sparse in situ climate stations (22). These trends align well with a proposed increase in the strength of the WWD over the past 30 years (17, 39) and observed increases in DJF precipitation in northern Pakistan (Pamir-Karakoram) (40, 41). The full extent of changes in SWE, and in particular, the positive SWE trends at high elevations, had not yet been observed outside of model data.

Spring (March-April-May; MAM) SWE trends are distinctly negative, excepting where some of the high-elevation regions of the Pamir-Karakoram (northern Indus and Amu Darya catchments) show positive SWE trends (Fig. 2B), which may be related to the positive regional DJF trends. The dramatic DJF-MAM trend reversal in the Tien Shan could indicate that increasing spring temperatures have led to faster spring runoff despite increased winter snowpack—a trend that has already been observed in snow cover, but not SWE, changes in central Asia (22, 37, 42). Higher air temperatures across HMA could also induce a shift from snow to rain, which would also reduce SWE storage.

Trends in summer (June-July-August, JJA) SWE are negative across the study region, where enough SWE is present to allow for trend analysis (Fig. 2C). This is in line with projections of increased temperatures in HMA (16), which drive earlier melting of snow (42, 43) and increase the ratio of liquid precipitation to snowfall in much of the region (18). The decline in summer SWE in the Himalaya is consistent with observed changes in the precipitation distribution of the ISM (15, 44) and melt rate enhancement from decreases in snow albedo due to aerosol contamination (45).

When these trends are considered together, they imply an intensification of the yearly hydrological cycle, where winter storms deposit more snow, particularly at high elevations, whereas warming summer temperatures cause rapid melting. Under this scenario, late spring to early fall water storage will decrease, with potentially dire implications for year-round water provision in many communities that rely directly on the slow melting of large snowfields for dry season water provision (20).

Aggregated trends across elevations

Although seasonal trend analysis provides valuable information on potential changes in local hydrologic budgets, it does not account for whether these changes occur in low- or high-SWE areas and what the elevation distribution of these changes is. We aggregate SWE trend and distribution data on a catchment scale and segment it into a succession of five-percentile elevation slices to illustrate the variation of trends across elevation zones (Fig. 3). Catchment-aggregated trend statistics can be found in tables S1 to S5.

Fig. 3 SWE contribution and SWE trend synthesis.

(A) Elevation distribution of SWE in each catchment, where each point shows the percentage of total catchment SWE at each five-percentile elevation bin. (B) Mean SWE trend at each five-percentile elevation bin. In the majority of catchments, maximum SWE occurs below the maximum catchment elevation, despite differences in catchment hypsometry. Each catchment is characterized by a unique elevation-trend relationship. The Indus, Amu Darya, and Tibetan Plateau catchments see the most negative SWE trends at their mid-elevations. The Ganges in the central Himalaya sees the most negative trends at the highest elevations.

Across all catchments, there is a strong, nonlinear, elevation-SWE relationship; the nature and magnitude of these relationships—as well as each individual catchment’s glacier and SWE distribution—are distinct and contextualized by the unique topographic and climatic setting of each catchment. High-elevation DJF snow and glacier coverage is nearly complete, although actual catchment hypsometries vary significantly (figs. S3 to S8). The highest SWE-volume elevation slice in each catchment occurs below the maximal catchment elevation in most catchments (Fig. 3A).

In the northern, lower-elevation, regions of HMA (Amu Darya, Syr Darya, and Tarim catchments), positive SWE changes are limited to DJF (figs. S3 to S5). In particular, the Syr Darya catchment has experienced strong increases in high-elevation DJF SWE (Fig. 2A and fig. S3) while also having some of the strongest negative trends in MAM SWE (table S3). The Tarim and Amu Darya catchments also see positive DJF trends—albeit with different elevation distributions—and negative MAM-JJA trends. However, the Amu Darya and Tarim catchments see less negative MAM SWE trends, implying more persistence of SWE into the spring melt season in these catchments than in the Syr Darya. Across all three catchments, however, full-year SWE trends remain negative, except at the very lowest parts of the Amu Darya (Fig. 3 and table S1).

The central and southern HMA catchments (Tibet, Indus, and Ganges/Brahmaputra) store much less low-elevation SWE than the northern catchments. To simplify the discussion of SWE trends for these catchments, we have focused on elevations above 1000 m, where the vast majority of snowfall occurs. All three catchments exhibit a mid-elevation decrease in their SWE trends, where mid-elevation trends are more negative than those at higher or lower elevations (Fig. 3B). The only catchment out of the six examined in this study where the most negative SWE trends occur at the highest elevations is the Ganges/Brahmaputra (Fig. 3B, yellow line). This is in line with increased temperatures in low-precipitation, high-elevation zones of the Himalaya (1, 46). It is likely that the decreasing SWE trends in this area at high elevations have also influenced glacial declines, which are some of the fastest in HMA (1).

DISCUSSION

Implications for glaciers

Many recent studies (1, 2, 5, 14, 16, 21, 23, 47, 48) have examined glaciers in HMA across several spatial scales. Glacier retreat rates are not uniform across HMA; some distinct regions of rapid decline, stability, and even growth exist. Two regions where studies have shown stability or glacier growth are the Karakoram (6, 23, 49) and the Kunlun Shan (5, 16, 47). Our data show positive trends in DJF, MAM, and SON SWE for parts or all of both regions, which indicate that increases in winter SWE storage could be partially responsible for glacier stability and growth in these regions. These positive SWE regions generally match the increased precipitation regions found by Yao et al. (16) and are consistent with the proposed increase in the strength of the WWD (17). Previous works have argued that HMA’s glaciers are more strongly affected by changes in precipitation seasonality than by changes in annual precipitation (14, 24, 25). In the cold and dry Kunlun Shan, glaciers are less sensitive to rising regional temperatures, and thus the trend toward wintertime SWE increases (Fig. 2A) could help explain some of the observed glacial thickening. Because many glaciers are fed by both avalanching and direct snowfall (21), changes in high-elevation precipitation are an important factor to account for when estimating future glacier mass budgets.

Strongly negative summer SWE trends are found in the central Indus catchment (Jammu-Kashmir region) and correspond with some of the most rapidly shrinking glaciers in the region (2). These wasting rates are confirmed in further studies that found that the western and central Himalaya have some of the highest glacier wasting rates in HMA (5, 16). This implies a more intense melt period in the summer potentially due to increased temperatures in the region (20). Monsoon-fed glaciers along the Himalayan front could also be affected by rising summer temperatures, which would favor rain over snowfall and reduce glacier albedo by limiting the persistence of fresh snow.

Spatially and temporally variable glacier retreat rates were found for the central Tien Shan using a suite of orthorectified satellite images (50). Further work noted that although there was an overall decrease of glacial mass in the Tien Shan, some high-elevation glaciers, particularly those on the eastern inner edge of the Tien Shan, have stable or slightly positive mass balances (51). This area agrees well with both positive DJF SWE anomalies (see Fig. 2A) and overall positive SWE trends (see fig. S2). However, Gardner et al. (5) found overall negative trends for glaciers in the Tien Shan using Ice, Cloud, and land Elevation Satellite (ICESat) data. It is possible that the poor sampling frequency of ICESat led to an overestimation of glacier wasting rates in the region or that the increases in SWE observed in this study are not translated into glacier ice in the predominantly summer-accumulation Tien Shan glaciers, and thus do not have a strong impact upon glacier mass balances in this region (25).

Spatially diverse water storage trends across HMA have also been noted in GRACE data. For example, negative gravity anomaly trends are smaller in the Tien Shan than in the Karakoram or Himalaya (52). Positive storage anomalies were also found in the Pamir, Tien Shan, eastern Himalaya (53), and eastern Kunlun Shan (54). These gravity anomalies, which were attributed to changes in glaciers, may also represent a change in seasonal snow-water storage patterns. Changes in seasonal water distribution have also been observed in GRACE data; positive and negative seasonal water storage anomalies correlate well with the positive and negative anomalies present in the SWE data shown in this study. Although the overall trend in the GRACE data is negative, some areas have seasonally positive mass signatures.

Regional impact

Climate models have trouble in correctly estimating SWE buildup in high-elevation areas due to a dearth of calibration data, complex topography, and poor measurement of snowfall and SWE with current-generation weather satellites, which negatively affects model parameterization and calibration (14). For example, in the Tien Shan, only three stations exist above 3000 m, and these stations fail to show significant correlations with reanalysis data (fig. S1) (22). The data presented here indicate that there have been unmeasured changes in HMA’s SWE distribution; these changes have important implications for both glaciers and downstream water provision.

The aggregated data show an HMA-wide annual net loss in SWE [−10.60 mm/year (average, −0.3%) over 2641 × 103 km2, including only trends with P < 0.05 and areas above 500 m above sea level (m asl) over the period of 1987 to 2009 (Fig. 3, fig. S2, and table S1)]. Although there have been positive changes in SWE in some areas of HMA, particularly in the winter and at the highest elevations (see Fig. 2, fig. S2, and table S2), these changes are outweighed by the net losses in the medium- and low-elevation zones of each catchment (Fig. 3B). Unfortunately, because of the poor performance of SWE algorithms in complex terrain (30, 33), converting these changes into absolute discharge or sea-level contributions is not feasible. Although SWE data have been shown to be internally consistent (30), absolute SWE volume measurement is not precise (55). However, significant observations about the areal and elevation distribution of SWE and relative SWE changes can still be made from the data.

The highest SWE totals in many catchments occur below the maximum catchment elevation (Figs. 3 and 4). This implies that although the negative SWE trends at the highest elevations are important, particularly for glaciers, downstream meltwater discharge is more strongly affected by SWE trends at the mid-elevations of each catchment. SWE trends in the mid-elevation zones of the Indus and Ganges—between 4000 and 5000 m asl—are more negative than trends at lower and higher elevations (Fig. 3). As the mid-elevations are less heavily glaciated than high-elevation zones (see Fig. 4), it is likely that changes in glaciers and snowfields will be somewhat distinct. Figure 4 indicates that the impacts of SWE changes will be unevenly felt even in neighboring catchments, because the majority of the SWE in the Indus is stored at significantly lower elevations than in the Ganges. SWE trends at the highest elevations are often less negative than lower-elevation trends, indicating that snow in colder, high-elevation zones may be partially shielded from regional climate changes.

Fig. 4 Differences between the snow distribution of the Ganges and Indus catchments.

(A) The Ganges and (B) Indus catchments showing catchment hypsometry (gray) (68), percentage glaciated area (red) (69), and SWE elevation distribution (blue). Dashed lines indicate catchment elevation percentiles. Both catchments show SWE maxima below their elevation peaks, despite differences in their SWE distributions. The altitude of SWE maxima are also minimally overlapping with glacier areas, indicating that snow and glacier meltwaters are often distinct and are affected by different climatic processes.

Although the trend values presented in this study (see Fig. 2, fig. S2, and tables S1 to S5) are individually small, the changes in SWE volume over each point represent gains and losses in SWE over a ~625–km2 area, meaning that each millimeter of SWE change is equivalent to 6.25 × 108 liters of water. When statistically significant trends are aggregated at a watershed scale, the annual changes in water resources range from −0.46 mm/year (average, −0.14%) in the Amu Darya to −2.9 mm/year (average, −0.41%) in Tibet (table S1). Although these trends are small in comparison to the total volume of snowpack in HMA, they indicate that changes in HMA’s cryosphere are not confined to glaciers. While changes in SWE are less likely to be felt in monsoon-dominated areas, regions that rely heavily on snowmelt—particularly during low-rainfall premonsoon months—will feel the effects of diminished snow-water storage.

This work provides a first step toward understanding changes in SWE in HMA and underlines the strength of seasonal and regional variations in the hydrological regime. It also presents evidence that changes in glaciers and snowfields are somewhat distinct due to differences in their elevation distributions. We find that trends in SWE are not linearly related to elevation and are highly heterogeneous between catchments. Because these changes will affect household water availability, as well as hydropower and agricultural infrastructure, understanding the interplay between snow and glaciers in regional water budgets will continue to be important for the vast downstream populations of HMA.

METHODS

SWE data processing

We generated a 0.25-dd grid from 25° to 45°N to 67.5° to 95°E, which encompasses a wide range of topographic and climatic settings, as well as several major mountain ranges, which allows us to track large-scale patterns in SWE (Fig. 1). Here, we acquired ungridded, raw, swath data for SSMI (F08, F11, and F13; 1987 to 2009) (56). We then extracted a time series at native sensor resolution for each of our sample points, as described by Smith and Bookhagen (30). In short, we aggregated all measurements within a 0.125-dd radius of each point in the 0.25-dd grid and then generated a spatially weighted mean value for each swath at that point. Our time series was made up of 1.03 measurements per day, on average, over the study period. This process did not involve regridding the raw swath data or resampling the data to daily or otherwise even time steps, and thus preserved as much of the raw, empirical signal as possible.

From this data set, we removed points adjacent to major water bodies, because water is known to interfere with PM SWE estimation (34). We also removed those points that do not see frequent or extensive snow accumulation, because the seasonal decomposition methods used in this study are not well-suited to sporadic and irregular time series. We chose to examine only the SSMI data set, as previous studies have noted differences in PM SWE retrievals between instruments (30). Although this time series was climatically short, it was internally consistent and was thus suitable for trend analysis.

We note that PM-derived SWE estimates have large uncertainties, especially in mountainous environments (30, 57, 58). These uncertainties are mainly due to forest cover, topographic complexity, and signal saturation in deep snowpack. The majority of our study area is above the tree line, meaning we did not expect significant vegetation-induced uncertainties. Terrain complexity can modify the assumed path between the PM sensor and the ground surface, and thus affect measured temperature brightness (Tb) values (30, 59). However, terrain complexity, as modeled by slope and relief, was shown to have a small impact on SWE uncertainty in the study region relative to the impacts of forest density (30). We examined here changes internal to single point-locations, which will be affected by the same set of terrain-related SWE uncertainties throughout the study period.

Signal saturation in deep snow remains a primary source of uncertainty across SWE algorithms (30, 33, 34, 55, 60). Here, we relied on a well-established and computationally efficient SWE algorithm (61), rather than more complex modern methods, which have primarily been developed to address issues related to vegetation cover. We used a constant snow density of 0.24 g/cm3, which was shown to be a reasonable global average (33).

PM signal saturation occurs even in pixels where only some of the pixel area exceeds the signal saturation depth, as the estimated SWE for a PM pixel is sensitive to total snow depth within that pixel (62). However, when glacier areas were removed from our analysis (defined here as any pixel with more than 25% glacier cover), the large-scale spatial patterns and elevation relationships were maintained (see fig. S9). As a further test, we calculated the percentage of time spent above a nominal saturation threshold for each pixel (see fig. S11) [120 mm, after the study of Takala et al. (33)]. Although some areas—particularly in the Tien Shan—see frequent deep snow in DJF and MAM, most snow depths in HMA do not exceed the saturation depth, and thus the large-scale spatial and seasonal patterns in SWE trends remain robust (see fig. S12).

Here, we did not attempt to create a well-calibrated SWE product for all of HMA because the SWE magnitudes were likely influenced by signal saturation, and we lacked convincing control data for the study area. We instead focused on examining changes internal to the SWE time series over our study period to examine the spatial patterns of positive and negative relative SWE changes.

Because the SWE data were drawn from a single continuous measurement record, each year of data is likely to be affected by similar measurement errors (that is, saturation depths at a given point will be similar between years; terrain-related SWE error will be consistent). Given this, analysis of SWE data can still yield valuable information on changes in SWE internal to each location’s SWE distribution. Our trends thus provide valuable information on the direction, rather than the magnitude of SWE trends, and the spatial pattern of those trends. We thus present our results with the caveat that our results are representative of relative, rather than absolute, differences in SWE trends between regions and across large spatial scales.

Trend analysis and significance testing

First, we removed shallow and infrequent snow-covered areas by applying a long-term average SWE threshold of 5 mm to each annual and seasonal analysis. The 5-mm limit was derived from previous studies, which had noted that the detection of shallow snow below 5-cm depths is unreliable (34, 60). We tested using additional metrics to remove shallow SWE areas or misclassified snow cover areas, such as the cross-polarized gradient ratio (42, 63), but found minimal changes in our results.

We then removed the seasonal signal of snow accumulation and melt from each individual point in this subset data set using Seasonal-Trend Decomposition by Loess (STL) (64, 65). This procedure effectively removes the seasonal oscillations from our data, leaving only the “residual,” multiyear, SWE signal at a given point location. To decompose the full-year signal, we resampled our data to a daily timestep and applied a 365-day decomposition window. To decompose the seasonal signals, we broke our time series into seasonal components and decomposed each season individually on a 90- to 92-day window depending on the length of the seasonal period. The Loess filter we used was adaptive, in that the full-year signal removed from the data varies from year to year. A full description of the parameters used for the STL filter used in this study is available in the Supplementary Materials.

We first tested the deseasoned data using a Mann-Kendall test (66, 67)—a nonparametric test often used for testing for trends in time series data—and only considered those areas that exhibit a statistically significant (P < 0.05) monotonically increasing or decreasing trend. We then performed a linear regression on the deseasoned data to examine changes in SWE over the study period (Fig. 2 and fig. S2). When these trends were compared to straightforward linear regressions performed on the original data set without detrending, the direction of the slope was the same, but the magnitude of the slope varies slightly. The main difference was in the significance of the results, where the seasonally detrended data were more statistically reliable. Similarly, when the linear regression was replaced with a weighted regression where the weights are related to the inverse of the SWE amount, the large-scale trend patterns were similar, albeit with different trend magnitudes (see figs. S13 to S15).

Elevation analysis

To examine elevation dependence in both our SWE trends and estimated SWE volumes, we segmented each catchment into five-percentile zones and used those elevation zones to subset our data. The total sum of each five-percentile slice was compared to the catchment-wide total SWE in Fig. 3A, and the average of all statistically significant SWE trends within each elevation slice was plotted in Fig. 3B. Catchment-wide trends thus reflect only statistically significant data and should be properly contextualized by examining the pixel-level data presented in Fig. 2 and fig. S2.

The hypsometries presented in Fig. 4 (gray) were generated from the SRTM V4.1 (68) data set aggregated to 50-m elevation bins. Glacier outlines from the Randolph Glacier Inventory V5.0 (69) were used to extract glacier elevation distributions from the SRTM V4.1. These elevations were then aggregated into 50-m elevation bins and compared to the size of the same SRTM elevation bin to derive glacier coverage at each elevation slice (Fig. 4, red polygon). SWE volumes, derived as in Fig. 3, were normalized to the maximum single-slice SWE volume to provide a relative measure of SWE volume across each catchment’s elevation range (Fig. 4, blue line).

SUPPLEMENTARY MATERIALS

Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/4/1/e1701550/DC1

Catchment-averaged SWE trend characteristics

By-catchment seasonal trends

Seasonal trend decomposition coefficients

fig. S1. APHRODITE station density (13).

fig. S2. Annual trends in SWE volume (1987 to 2009), as derived from SSMI data, with major regional watersheds (black outlines).

fig. S3. Syr Darya seasonal trends.

fig. S4. Amu Darya seasonal trends.

fig. S5. Tarim seasonal trends.

fig. S6. Tibetan Plateau seasonal trends.

fig. S7. Ganges seasonal trends.

fig. S8. Indus seasonal trends.

fig. S9. Impact of removal of glacier areas on trend patterns.

fig. S10. Percentage of time by season above 150-mm SWE.

fig. S11. Percentage of time by season above 120-mm SWE.

fig. S12. DJF and MAM SWE trends with areas impacted by PM signal saturation removed.

fig. S13. Impact of weighted regression on full-year SWE trends.

fig. S14. Impact of weighted regression on DJF SWE trends.

fig. S15. Changes in SWE trends when an uncertainty margin is introduced.

table S1. Full-year catchment-aggregated SWE trends above 500 m asl.

table S2. DJF catchment-aggregated SWE trends above 500 m asl.

table S3. MAM catchment-aggregated SWE trends above 500 m asl.

table S4. JJA catchment-aggregated SWE trends above 500 m asl.

table S5. SON catchment-aggregated SWE trends above 500 m asl.

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.

REFERENCES AND NOTES

Acknowledgments: Funding: The State of Brandenburg (Germany) through the Ministry of Science and Education supported T.S. for part of this study (grant to. B.B.) Author contributions: T.S. and B.B. designed the study. T.S. prepared and analyzed the SSMI data. B.B. contributed to the development of the methodology. Both authors wrote the manuscript led by T.S. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary materials and publicly available from the National Snow and Ice Data Center. Additional data related to this paper may be requested from the authors. Correspondence and requests for further information should be addressed to T.S.
View Abstract

Navigate This Article