The Himalayas and Tibetan Plateau harbor hundreds of mountain lakes along with thousands of glaciers and high-elevation snowfields. This is the source of water for the upper reaches of Asia's main river systems, providing the livelihood for millions of people in the subregion. Climate change is therefore critical for the Himalaya snow and glacier hydrology, the dependent ecosystems, and the people. Whereas temperature and precipitation are common indicators for climate change, snow and glacier dynamics are reliable precursors of a warming or cooling climate. This study uses a simple empirical climate model (ECM) and the Gravity Recovery and Climate Experiment (GRACE) satellite data to analyze water storage dynamics in the 5.072 × 106 km2 Himalayas and Tibetan Plateau region. About 72 consecutive months (January 2003 through December 2008) of data are used in the study. The temperature and precipitation (snow plus rain) data are acquired from the Global Land Data Assimilation System (GLDAS) Noah land surface model and are validated with ground truth data from 205 meteorological stations. Total water storage change derived from the GRACE gravity data is fitted with a simple sinusoidal least squares regression model. A favorable agreement exists between the GRACE and sinusoidal curve (R2 = 0.81 and root-mean-square error (RMSE) = 8.73 mm), suggesting that random errors in GRACE data are small. However, the sinusoidal fit does not quantify systematic errors in GRACE data. Agreements between the GRACE- and ECM-estimated storage changes are also favorable at both monthly (R2 = 0.93, RMSE = 5.46 mm) and seasonal (R2 = 0.83, RMSE = 7.64 mm) cycles. The agreements (significant at p < 0.01) indicate not only GRACE's ability to detect storage signal but also that of the ECM model to characterize storage change in the snow and glacier hydrology. There is clear seasonality in the storage anomaly, with the highest in summer and lowest in winter. The corresponding storage change is delayed by a quarter of the year. The GRACE and ECM model indicate an overall negative storage trend of 0.36 ± 0.03 mm/month or 21.91 ± 1.95 km3/yr for the study area (significant at p < 0.1). Given that snow and glaciers are particularly sensitive to temperature change, the negative storage trend could be indicative of warming climate conditions in the region. Groundwater abstraction (mainly for irrigation) in the southern plains, coupled with dwindling snowfall in the northern massifs, is a critical storage loss factor in the region. Invariably, storage loss in the Himalayan-Tibetan Plateau region could have negative implications for the hydrology, dependent ecosystems, and livelihoods of millions of people.
 Estimates of cumulative glacier lengths show that globally, glaciers are retreating on the average of 0.25 m/yr [Hoelzle et al., 2003]. Whereas snowmelt and glacier melt could mitigate or avert catastrophic droughts during weak, delayed, or failed precipitations, enhanced glacier melt could also have devastating effects [Ren et al., 2003; Kumar et al., 2007]. Accelerated glacier melt may trigger rock-ice avalanches, high-debris flows, and floods [Ageta et al., 2000; Li et al., 2007]. On the other hand, enhanced glacier melt could threaten long-term water availability [Thayyen and Gergan, 2009]. Acute water shortages would jeopardize agroindustrial systems, the backbone of modern societies [Immerzeel et al., 2010].
 Despite its strategic importance as Asia's main source of water, the Himalayan and Tibetan Plateau snow and glaciers are fast retreating [Tang and Li, 1992; China National Committee on Climate Change, 2007]. While 82% of the Tibetan glaciers have retreated in the past half century, 10% of its permafrost has degraded in the past decade alone [Qiu, 2008]. In fact, over 64.3% of the estimated glacier retreat was in the last two decades [Liu et al., 2008]. Hydrological changes on such scales could threaten glacier-dependent lakes, rivers, and wetlands in the region and beyond.
 Permafrost loss, monsoon disruption, sea level rise caused by terrestrial ice loss, and increased earthquakes and landslides once snow and glacier weight is gone are a possible disastrous combination [Xu et al., 2007; Spencer, 2009]. To address such threats, an empirical climate model (ECM) is developed to estimate water storage change in the Himalayas and Tibetan Plateau region. The model, driven by the common climate variables of temperature and precipitation, is simple and practicable. It would also enhance our understanding about the effects of changes in climate on storage in the snow and glacier hydrology of the Himalayan and Tibetan Plateau. The results of the study will provide the basis for policy decisions and conservation strategies of the fragile, invaluable hydroecology. The logic and implementation of the ECM model are discussed in section 2.
2. Data and Method
2.1. Study Area
 The Himalayas and Tibetan Plateau cover, in various proportions, the countries of Pakistan, Afghanistan, Tajikistan, China, India, Nepal, Myanmar, Bhutan, and Bangladesh (Figure 1). The study area is located between 25.59°N–40.27°N and 69.12°E–105.54°E (Figure 1) and has an estimated area of 5.072 × 106 km2. It extends (in an arc shape) for 2500 km from west to east and 100–400 km from south to north [Muskett, 2008; Mats et al., 2009; Immerzeel et al., 2010]. Although over 80% of the area is in the Tibetan Plateau, it also includes Karakoram, Hindu-Kush, Pamir, Tien Shan, Inner and Outer Himalayas, and several other mountain ranges.
 The geophysiographic features in this region are complex and rugged [Bolch et al., 2008], with over 30 of the 110 massif peak elevations above 6000 m [Kaul, 1999; Macintosh, 2005]. The 70 km long Siachen Glacier (on the India-Pakistan border) is the longest of the over 50,000 glaciers in the region [Kaul, 1999; Qiu, 2008]. The mountains are characterized by deep snowcaps, glacial valleys and gorges, and rich ecobiodiversity [Saxena et al. 2005; Harris, 2010]. The mountain valleys, depressions, and canyons provide abundant storage for summer rainwater and meltwater [Ageta et al., 2000]. The tectonic glacial lakes occur at altitudes below 5000 m and generally diminish in size at higher elevations [Thayyen and Gergan, 2009]. Pangong Tso, the largest endorheic lake in the region is ≈134 km long. It spreads across the India-China border at an average altitude of 4350 m [Winiger et al., 2005].
 Of the 519 mm average annual precipitation, 72% falls as summer monsoon rain and 28% as winter snow [Dulal et al., 2006]. Variation in diurnal temperature is large, with a long-term average of 7°С. The coldest (in January) and hottest (in July) average monthly temperatures are 12°С and 16°С, respectively [Shrestha et al., 2000]. Above-16°С summer temperatures are not uncommon on the valley slopes [Dulal et al., 2006]. At elevations over 5000 m, above-zero temperatures only occur during the daytime [Thayyen and Gergan, 2009]. Because of rising temperatures, snow precipitation is increasingly shifting toward rain. Winds generally strengthen at higher altitudes, reaching 120 km/h at 6000 m [Burbank et al., 2003].
 While the western phase of the Himalayas has a generally warm and humid monsoon climate, a typical mountain desert cold climate dominates on the northern slopes and Tibetan Plateau [Goswami et al., 2003]. As Asia's main source of water, the Himalayas and Tibetan Plateau snow and glaciers greatly influence the hydrology, ecology, and livelihoods of millions of people in the subcontinent [Burbank et al., 2003; Qiu, 2008; Thayyen and Gergan, 2009; Immerzeel et al., 2010].
2.2. Hydrologic Mass Balance
 Here storage change Δω [L/T] is the difference between storage anomalies of two successive time steps. A storage anomaly is the residual storage content ω′ [L/T] at a given time with respect to the content at the reference epoch. Reference storage ω′r [L/T] is the mean storage within the time interval for which the temporal mean is computed. Hydrologic mass balance in the Himalayas and Tibetan Plateau study area is mainly driven by storage in snow, glaciers, high-altitude lakes, and permafrost. A glacier mass balance method could therefore be used to estimate the storage change. Studies show that the climate variables with the most impact on glacier dynamics are temperature and precipitation [Meier, 1984; Oerlemans and Fortuin, 1992; Dyurgerov and Meier, 1997; Raper and Braithwaite, 2006; Kaser et al., 2006; Fujita, 2008b; Qiu, 2008]. Using seasonal sensitivity characteristics of temperature and precipitation, Oerlemans and Reighert  conducted glacier mass balance for different climatic regions of the world.
 In this study, the method of Oerlemans and Reighert  is modified to estimate water storage change in the Himalayas and Tibetan Plateau region. To achieve this, it is assumed that the time derivative of the residual storage anomaly ω′ [L/T] with respect to the initial time of the water storage ω [L/T] is a nonlinear function of temperature T (K) and precipitation P [L]; i.e., ω′ = f(T, P). A linearization of this function at the vicinity of the reference temperature Tr (K) and reference precipitation Pr [L] yields
with the residual storage change Δω′ = ω′ − ω′r [L/T], reference storage ω′r = f(Tr, Pr) [L/T], temperature coefficient CT = ∂f/∂T [L/K], and precipitation coefficient CP = ∂f/∂P (dimensionless). The integration of the linearized relationship, after discretization, gives
The index k is the time step (annual, seasonal, monthly, or other time scales) associated with the reference temperature Tr and precipitation Pr of the reference mass balance ω′r (preferably zero). The temperature coefficient CT,k [L/K] and precipitation coefficient CP,k (dimensionless) in equation (2) are the so-called climate sensitivity characteristics (CSC) of the mass balance [Fujita, 2008b]. The CT,k and CP,k of the mass balance are quantified as
This can be determined by any simple or complex degree-day or energy balance model that computes the cumulative mass balance through the year [Oerlemans and Reighert, 2000].
 As this study focuses on the entire snow and glacier system of the Himalayas and Tibetan Plateau, a bulk CSC value is calculated using the energy–mass balance model described by Fujita [2008b]. Because of the great size, apparently all glacier types exist in this region. This allows the use of average values and eliminates the need for different sets of parameters for individual glaciers. The areas of the glaciers in the region are therefore added together and treated as a seamless glacier. Most of the model input parameters (including glacier and ice extent, albedo, etc.) are derived from Moderate Resolution Imaging Spectroradiometer (MODIS) and Landsat thematic mapper data [König et al., 2001]. The energy fraction of the model is balanced on glacier surface daily heat, radiation, sensible and latent turbulent heat, and conductive heat into the glacier mass. The mass fraction of the model is balanced on accumulated snow and refrozen meltwater and on snowmelt and evaporation [Fujita, 2008a, 2008b]. The model gains mass via solid precipitation and loses it via ablation. Conductive heat into the glacier system and the amount of water at the snow-glacier interface are used to estimate meltwater refreeze [Fujita and Ageta, 2000]. Snow surface albedo (which declines with glacier depth and melt season duration) is calculated in relation to surface snow density (which changes with compaction). This preserves climate feedback effects on the snow and glacier regimes. The model has been validated on selected glaciers in the Himalayas and Tibetan Plateau region. Further details of the model, including validation analyses, are documented by Fujita and Ageta , Fujita et al. , and Fujita [2008a, 2008b].
 In calculating CSC, the calibrated model is rerun for the reference state Δω′ = 0 by adjusting annual mean air temperature within ±2 K. Next, systematic monthly perturbations in temperature (±0.5 K) and in precipitation (±10%) are induced to derive the CT,k and CP,k in equation (3), respectively. Further details, including the rationale of this procedure, are explained by Oerlemans and Reighert . The results of the CSC analysis are depicted in Figure 2a.
 One distinctive feature of glaciers is the length of period for which perturbations in temperature or precipitation affect their annual mass balances [Oerlemans and Reighert, 2000]. The stability of the average Himalaya glacier system is relatable to climate change and global warming, which are mainly driven by summer climate trends. Hence, a sensitivity index (SI) that shows the average behavior of the glaciers through the year is used to quantify perturbations in temperature (SITy) and precipitation (SIPy) as
where y is the year and 6, 7, and 8 denote June, July, and August, respectively. The SI is related to average yearly temperatures and precipitation for 2003–2008 in Figure 2. The power fits in Figures 2a and 2c show strong correlations between SI and temperature (R2 = 0.91) and precipitation (R2 = 0.86); all are significant at p < 0.01. Figure 2 suggests that snow and glacier vulnerability in the region increases with increasing temperature and precipitation. Of course, rising temperatures favor rain precipitation (Figure 3), which in turn increases glacier vulnerability [Oerlemans and Reighert, 2000].
 The CSC so computed is used in the ECM model to estimate storage change in the study area. In GRACE, the steady state portion of water storage is highly correlated with static gravity fields. It is difficult to separate the steady state portion of the water storage from the static gravity estimates. Hence, what GRACE provides is not total terrestrial water content but its anomaly field. Raw GRACE monthly solutions contain information about water storage anomaly in relation to the reference storage. In other words, GRACE is indirectly sensitive to total water storage signal. In this study, water storage change estimated independently by the ECM model in equation (1) is compared with that derived from the GRACE satellite gravity data.
 Because data are missing for June 2003, about 71 months of GRACE data (January 2003 through December 2008) are used in the analysis. The 1° resolution GRACE data are truncated to the 5.072 × 106 km2 study area and smoothed with the Gaussian filter of 300 km half width. To minimize the effect of geographic truncation [Mayer-Guerr et al., 2007] on the analysis, a 1° buffer zone is maintained around the study area. The fields are spatially averaged to create time series of total water storage anomalies. Total water storage change is then calculated from the anomaly data.
2.3.2. Climate Data
 In especially complex mountain terrains, climate data are not always available in sufficient density and distribution for realistic spatial representations. Satellite, statistical, and knowledge-based numerical methods are therefore used to interpolate, extrapolate, or predict spatial distributions of climate fields in such regions [Lindsay, 2005]. These methods often take into account factors such as elevation, slope, gradient, aspect, curvature, geographic location, orographic effectiveness, and other land physiographic elements, which affect the pattern (occurrence and distribution) of climate phenomena [Daly et al., 2008].
 The Mountain Climate Simulator (MTCLIM [Hungerford et al., 1989]) and Parameter-elevation Regression on Independent Slopes Model (PRISM [Daly et al., 1994; Gibson et al., 1997]) are specifically developed for interpolating meteorological fields in complex terrains with limited observations. As these models are, however, not freely available, their applications outside the developer organizations are limited. In this study, the temperature and precipitation (snow and rain) data products of the Global Land Data Assimilation System (GLDAS) land surface model (LSM) are used [Rodell et al., 2004]. About 72 months of the GLDAS Noah climate data, spanning for January 2003 through December 2008, are used.
 GLDAS uses satellite- and ground-based data products to generate optimal fields of land surface states and fluxes such as snow, rain, temperature, evapotranspiration, and ground heat flux. A vegetation-based tiling approach (with a 1 km global vegetation data set as the basis) is used to simulate subgrid variability. Soil and elevation parameters are based on high-resolution global data sets [Rodell et al., 2007]. The baseline meteorological forcing data are produced by the NOAA Global Data Assimilation System (GDAS) atmospheric analysis system. The simulation, initialized for 1979, is performed on a 1° global grid and forced by bias correction reanalysis products of Berg et al.  prior to 2000.
 The reliability of GLDAS data products [Rodell et al., 2007] is therefore commensurate to that of the MTCLIM or PRISM model estimates. Despite this, the 72 month GLDAS Noah data are validated with in situ measurements of temperature and precipitation from 205 meteorological stations (Figure 1). Linear correlation fits in Figures 2d and 2e show good agreement between the GLDAS Noah data and temperature (R2 = 0.76, root-mean-square error (RMSE) = 3.43°C) and precipitation (R2 = 0.84, RMSE = 43.61 mm); all are significant at p < 0.01.
 The temperature and precipitation fields are afterward spatially averaged (in much the same manner as the GRACE data) and fed into the ECM model to estimate total water storage change in the study area. Next, the storage change derived from the ECM model is compared with that from the GRACE satellite gravity data. The GLDAS soil moisture data are also used to isolate the contribution of groundwater to mass loss in the southern region of study area.
3. Results and Analysis
3.1. Uncertainty Analysis
 The GLDAS Noah monthly temperature and precipitation data are validated with ground truth data from 205 climate monitoring stations. The method described by Strassberg et al.  is used to gauge uncertainties in the GLDAS Noah data. On the basis of the analysis, the RMSE difference between the GLDAS Noah and ground truth data is 3.43°C (R2 = 76) for temperature and 43.61 mm (R2 = 0.84) for precipitation. Artificial disturbances (at the magnitude of the estimated uncertainties) are added to the ECM model to analyze their propagation in the estimated water storage. In principle, a disturbance at the beginning of the integration affects further evolutions of the balance. The model is initiated at the start of 1979 (the time GLDAS Noah data logging started) and is run continuously for 30 years (1979–2008) with monthly temperature and precipitation forcing. Comparison at seasonal cycle shows a good agreement, with R2 = 0.81 and RMSE = 7.82 mm.
 As an important condition, GRACE data application requires validation with in situ hydrological measurement data. This set of data is currently not in our repository for the Himalayas and Tibetan Plateau region. It is, however, important to note that the GRACE monthly solutions are differentiated prior to application in this study. In other words, the GRACE-based water mass changes are considered, not the monthly solutions (see sections 3.2–3.6). Furthermore, studies [e.g., Rodell et al., 2009] show that GRACE satellite gravity data reliably signal storage anomaly in the region. Muskett  and Immerzeel et al.  have also shown that GRACE sufficiently detects storage change in the Himalayas and Tibetan Plateau region.
3.2. Monthly and Seasonal Storage Anomaly
 Time series of the monthly and seasonal anomalies of snow, rain, precipitation, temperature, and GRACE total water storage are depicted in Figure 3. The negative (for snow) and positive (for rain) linear trends (at the seasonal cycle) in Figures 3a and 3b suggest increasing rainfall at the expense of snowfall in the study area. Overall, precipitation (snow plus rain) is increasing in the region, indicated by a positive linear trend for the seasonal cycle in Figure 3c. Temperature too is increasing (Figure 3d), which indicates a warming condition [see Fujita, 2008b; Thayyen and Gergan, 2009]. The corresponding GRACE total water storage anomaly shows an overall decline. The anomaly trends suggest that the increase in (rain) precipitation could be in response to the rising temperatures.
 The phases of the monthly and, more so, the seasonal time series of precipitation, temperature, and GRACE track one another. There is a near-synchronized seasonality, with the highest amplitudes in summer (June to August) and the lowest in winter (December to February). Whereas the linear trends in the seasonal anomalies for GRACE and snowfall are negative, those for rainfall and precipitation are positive. This shows that rain is increasingly becoming a dominant form of precipitation in the region.
 Air temperature generally dictates the fraction of precipitation that falls as either low-albedo rain or high-albedo snow on glacier surfaces [Dulal et al., 2006]. Studies show that under warming climate conditions, precipitation favors low-albedo rain, not high-albedo snow [Thapa, 1993; Fujita, 2008a]. Because of accelerated absorption of solar radiation, glacier melt increases in the absence of high-albedo snow [Ren et al., 2003; Fujita, 2008a]. In fact, Higuchi et al.  and Fujita [2008b] noted increasing snowmelt and glacier melt under decreasing snow precipitation. Low albedos increase glacier surface exposure, which in turn accelerates glacier melt even under stable temperature conditions [Fujita and Ageta, 2000]. Warming climate conditions generally trigger temperature rise, leading to less snow precipitation and high snow and glacier ablation and storage loss. Such conditions could negatively impact the hydrology, ecology, and livelihoods in the Himalayas and Tibetan Plateau region [Muskett, 2008; Qiu, 2008; Immerzeel et al., 2010].
3.3. Average Monthly Storage Anomaly
 Along with GRACE total water storage, average monthly snow, rain, precipitation, and temperature anomalies for 2003–2008 are plotted in Figure 4. On the basis of Figure 4a, snowfall increases from January through March, then decreases through August, before increasing again through December. On the other hand, rainfall, precipitation, temperature, and GRACE-based storage anomalies have positive trends for January through July and negative trends through December [see also Fujita, 2008a, 2008b]. Because temperatures are highest in summer, snow and ice ablation likely occurs during this period. This, coupled with the declining (high-albedo) snowfall, induces storage loss in the region.
Figure 4f illustrates a comparison of the average monthly total water storage change derived separately from GRACE data and the ECM model. In the study area, storage change generally increases from January through June, with a slight dip in March. It decreases through September before gradually increasing through December [see Rodell et al., 2009]. The winter-to-spring rise in storage could be driven by snow accumulation during this period [Ren et al., 2003; Siebert et al., 2005; Qiu, 2008; Thayyen and Gergan, 2009]. Then high summer storage could be due to water storage in the vast crayons, reservoirs, and (endorheic) lakes and tarns in the region. Storage loss is noticeable either directly from Figure 4e or as a negative storage change in Figure 4f. In both cases, storage loss occurs between August and December. This could be mainly driven by groundwater depletion for farm irrigation, especially in the southern half of the study area. [Sarwar and Bastiaanssen, 2001; Singh and Bengtsson, 2004; Mall et al., 2006; Kumar et al., 2007; Muskett, 2008; Rodell et al., 2009; Immerzeel et al., 2010; Wada et al., 2010]. Dwindling snowfall, particularly in the Tibetan Plateau region, could be another important driver of storage loss [Qiu, 2008]. There is strong correlation (R2 = 0.93, RMSE = 5.46 mm, significant at p < 0.01) between average monthly storage change from GRACE data and the ECM model (Figure 4f). This suggests that the temperature- and precipitation-driven ECM model sufficiently characterizes storage dynamics in the study area.
3.4. Seasonal Storage Change
 This section illustrates the dependence of water storage in the Himalayas and Tibetan Plateau region on temperature and precipitation as a time derivative of storage, not its function. Such analysis is not only informative but also efficient in terms of data and computational requirements. Furthermore, most studies involving GRACE hydrology are presented as time derivative of storage, not its function. This approach therefore allows direct comparisons of the results here with those of other studies.
Figures 5a–5c depict seasonal storage change in the study area, separately computed from the GRACE, temperature (at CP,k = 0), and precipitation (at CT,k = 0) data. Figure 5 shows the relative importance and suitability of the climate variables of temperature and precipitation as indicators for storage change in the region. The storage change (Figure 5) is delayed by a quarter of the year with respect to the storage anomaly (Figure 3). The minima and maxima of the GRACE-only, temperature-only, and precipitation-only estimated storage changes are in spring and autumn, respectively. However, the amplitudes (separate for the GRACE, temperature, and precipitation) are totally different. This indicates that temperature or precipitation variations alone definitely cannot explain water storage dynamics in the study area. However, the sum of the temperature and precipitation components almost completely explains the variations in GRACE-estimated storage change. The trends in the three storage terms are negative, suggesting storage loss acceleration in the region [see Tang and Li, 1992; Hou et al., 2000; Muskett, 2008; Qiu, 2008; Rodell et al., 2009; Immerzeel et al., 2010].
3.5. Storage Change Comparison
Figure 6a depicts a simple least squares sinusoidal fit for the GRACE-derived total water storage change. A good agreement exists between the GRACE and sinusoidal curve, with R2 = 0.81 and RMSE = 8.73 mm (significant at p < 0.01). The favorable agreement shows that random errors in GRACE data are small. However, the test does not allow us to quantify systematic errors in GRACE data. This second error type can be quantified using ground-based measurements of storage change, data which are currently not in our database. The sinusoidal fit is therefore assumed to give a measure of GRACE's ability to detect storage signal in the study area. In fact, GRACE is shown to reliably detect storage change in the Himalayas and the surrounding regions [Muskett, 2008; Rodell et al., 2009; Immerzeel et al., 2010].
Figure 6b is a comparison of total water storage change estimated from GRACE data and the ECM model. There is a near-synchronized annular seasonality in the GRACE- and ECM-estimated storage change, with R2 = 0.83 and RMSE = 7.64 mm. The phases and amplitudes of the estimated storage changes closely track each other. This further indicates that the ECM model characterizes storage dynamics in the study area. GRACE data are processed using spherical harmonic (Stokes) coefficients and, as such, have leakage problems. Like GRACE, uncertainties also exist in the ECM model. To better represent the storage dynamics, therefore, the average of the GRACE- and ECM-estimated storage changes is used. The long-term storage change is negative. It is on the order of 0.36 ± 0.03 mm/month or 21.91 ± 1.95 km3/yr for the 5.072 × 106 km2 study area (significant at p < 0.1). This finding is consistent with earlier studies in and around the Himalayas and Tibetan Plateau region [Muskett, 2008; Qiu, 2008; Rodell et al., 2009; Immerzeel et al., 2010; Cogley et al., 2010; Wada et al., 2010].
Figure 6c compares GRACE-derived total water storage change with that obtained from ECM plus GLDAS Noah soil moisture data for the region south of the study area. It is readily noticeable that the amplitudes of the GRACE storage change for this region are higher than those for the entire study area. The amplitudes of the ECM plus soil moisture storage change are also far lower than those of GRACE. This suggests that storage change in the southern region of the study area is mainly driven by groundwater depletion (see section 3.6 for details).
3.6. Annual Storage Distribution
Figures 7a–7d depict average annual spatial distributions of snowfall, rainfall, precipitation, and temperature for 2003–2008. Snowfall (Figure 7a) is highest in the west central Tibetan Plateau, northeast Pakistan and northwest India. It is lowest in the regions east of the study area. Rainfall (Figure 7b) is high along the west–east southern flank and highest in northeastern India. This is the rainward belt of the Himalayan massifs. Rainfall is lowest along the west–east northern leeward region of the Tibetan Plateau. The precipitation distribution reflects the snow and rain distributions combined. It is generally high along the west–east southern rainward regions and low in the northern rain shadow areas. The only exception is the west central Tibetan Plateau region, where snowfall is high (Figure 7c). The temperature distribution (Figure 7d) depicts a subtle replica of the elevation. To some extent, it also moderately mimics the precipitation trend. This shows the unique influence of elevation and precipitation on temperature.
 The precipitation distribution and, to a certain extent, that of temperature are influenced by the westerlies and monsoon winds [Ageta and Higuchi, 1984; Fujita, 2008a, 2008b]. Mount Everest and the sister mountains along the southern arc are the rainward and rain shadow divide in the region. The ranges of snow, rain, and precipitation are 28–997, 128–3116, and 132–3116 mm, respectively. Average annual precipitation, and especially that of rain, is low for over two thirds of the study area. Temperature, with an average annual range of 17°C to +30°C, is also negative for over three quarters of the study area. The west–east rainward belt to the south is the highest temperature region (Figure 7d).
Figure 7e illustrates a map of linear trend in total water storage derived from GRACE monthly water storage maps for 2003–2008. Implicitly, the dynamics of the climate variables of temperature and precipitation in the glacier-driven hydrology largely dictate the distribution pattern of the GRACE storage trend map. The GRACE-derived water storage trend shows a significant depletion in the south. This is especially noticeable in the border regions of Nepal, Bhutan, India, and Bangladesh; i.e., southeast of Mount Everest. Storage depletion is also visible farther in the southeast, along the India-Myanmar border region. A moderate depletion trend exists in the region northwest of Indian.
 A negative storage trend exists in the high rain precipitation region along the west–east southern arc of the study area. The negative storage trend is a signal for groundwater depletion, especially in the southern half of the study area [Muskett, 2008; Rodell et al., 2009; Wada et al., 2010]. To justify this point, the GLDAS Noah data are used to quantify soil moisture storage change in the region south of the zero-contour storage line in Figure 7e. The ECM model is also applied to simulate snow and glacier storage change in this region. The sum of the ECM and soil moisture storage changes is then subtracted from that of GRACE. This analysis isolates groundwater storage change in the region south of the zero-contour storage line. In terms of water equivalent per area, the trend in the derived groundwater storage loss is ≈70% of the estimated storage loss in the region (see Figure 6c). In terms of volume loss, however, it accounts for <20% of the estimated long-term storage loss in the Himalayas and Tibetan Plateau region. This is ascribed to irrigation farming, a critical storage depletion factor in this region [Sarwar and Bastiaanssen, 2001; Siebert et al., 2005; Mall et al., 2006; Kumar et al., 2007; Cogley et al., 2010].
 On the basis of the GRACE map, storage trend is high in the high-snow regions northeast of Pakistan. It is also high in the headwater region of the Huang, Yangtze, and Mekong rivers. This region has relatively moderate snow, rain, and temperature (Figures 7a–7c), conditions that generally favor storage. This suggests that low-temperature zones in the Himalayas and Tibetan Plateau are hydrologically more stable. These regions are less sensitive to the effects of human activity, including climate change and global warming.
 The foregoing analysis suggests that the climate variables of temperature and precipitation are critical mass balance elements in the snow and glacier hydrology of the Himalayas and Tibetan Plateau [see Hou et al., 2000; Qiu, 2008; Cogley et al., 2010; Immerzeel et al., 2010]. Temperature and rain precipitation are increasing with decreasing snow precipitation (Figure 3). While snow and rain are negatively correlated (R = −0.57), temperature is rising on the average of 0.023°C/yr. Of course, temperature influences the fraction of precipitation that falls as either snow or rain [Dulal et al., 2006; Thapa, 1993; Fujita, 2008a]. Snowmelt and glacier melt and retreat accelerate with rising temperatures and warming climate conditions. This has an overall effect of storage loss.
 This study has used the GRACE and ECM model to estimate total water storage dynamics in the Himalayas and Tibetan Plateau snow and glacier hydrology. About 72 consecutive months of data, spanning from January 2003 through December 2008, are used in the study. The GRACE data are verified with a sinusoidal model, and the ECM forcing data are verified with in situ measurements of temperature and precipitation. The GRACE-estimated storage change is also compared with that of the ECM model for the 5.072 × 106 km2 study area. All the correlation agreements are statistically significant at p < 0.01. The favorable agreements suggest that the ECM model sufficiently characterizes storage dynamics in the Himalayas and Tibetan Plateau snow and glacier hydrology. The ECM model is simple, practicable, and readily transferable to other snow and glacier regions. The main drivers of the model are temperature and precipitation, which are routinely measured by common weather stations.
 Rising temperatures in the region favor rain precipitation, snowmelt and glacier melt, and storage depletion. Snowfall is highest in winter and lowest in summer, while rainfall, precipitation, and temperature exhibit the reverse behavior (Figure 4). Direct snow and glacier accumulation (storage gain) occurs in winter, and ablation (storage loss) occurs in summer. There is, however, a high summer storage anomaly in the study area, as over 70% of the dominant summer rain falls during this period. A considerable amount of the summer rainwater and meltwater is also stored in the vast reservoirs, endorheic lakes, and canyons, leading to high summer storage anomalies. The corresponding storage change is, however, lowest in spring and highest in autumn. This is because the storage change is delayed by a quarter of the year.
 The GRACE and ECM model show an overall negative storage change in the Himalayas and Tibetan Plateau study area. Snow and glacier ablation and the subsequent storage loss via discharge are mainly driven by rising temperatures and rain precipitation. Heavy groundwater irrigation, especially in the northwest regions of Pakistan and India and in southern Nepal and Bhutan, is another mass loss factor. Storage loss in the far north and northeast regions of the Tibetan Plateau is largely due to dwindling snowfall. Human activities (e.g., farm irrigation) and climate change (e.g., temperature rise and global warming) are therefore the possible causes of storage loss in the study area. Invariably, storage loss in the Himalayas and Tibetan Plateau region could have negative implications for the hydrology, dependent ecosystems, and livelihoods of millions of people.
 This study was funded by the International Collaborative Project of the Ministry of Science and Technology (2009DFA21690), the KZCX1-YW-08-03-04 Innovative Project, and the Hundred Talent Program of Chinese Academy of Sciences. We duly acknowledge the valuable inputs of the reviewers, especially that of Pavel G. Ditmar.