Limited meltwater potential in the Asian Water Tower to mitigate downstream urban scarcity

Abstract

The Asian Water Tower, a vital freshwater source for billions downstream, is rapidly transforming under climate change. Accelerated glacier and snowmelt are reshaping meltwater patterns, threatening urban water security amid rapid population growth. Using a high-resolution cryosphere–hydrology model, we quantify meltwater’s role in mitigating urban water scarcity. Here we show that over the past three decades, meltwater availability has mitigated water scarcity for 22.9–28.7 million urban residents by supplementing water during dry seasons. However, future meltwater availability is projected to become increasingly uneven, rising during the monsoon when rainfall is abundant but declining substantially in the dry season. This seasonal redistribution substantially weakens the buffering capacity of meltwater during critical dry periods. As urban populations facing water scarcity are projected to increase by 31–53%, meltwater gains will remain minimal (4–10%), indicating an increasing supply–demand mismatch. Our findings highlight that despite projected increases in annual volume, the potential of meltwater to alleviate future urban water scarcity is increasingly constrained by seasonal misalignments, underscoring the urgent need to integrate meltwater variability into regional water resource planning.

Climate change is altering the timing—not just the volume—of meltwater from the Asian Water Tower. Although meltwater has historically buffered dry‑season urban water scarcity, future increases are concentrated in the monsoon season, weakening its effectiveness just as urban water demand rises.

Introduction

The Asian Water Tower (AWT)—encompassing the Tibetan Plateau and the adjacent Himalayan, Karakoram, Hindu Kush, and Tianshan mountain ranges—is the largest global reservoir of cryospheric freshwater outside the polar regions and serves as a crucial water source for downstream regions^1–3. The AWT and its downstream basins (Fig. 1) span approximately 6.5 million km², with an average elevation exceeding 4000 m a.s.l.⁴. Meltwater from glaciers and snow sustains major river systems, such as the Indus, Amu Darya, and Tarim, where rainfall-runoff contributions are limited^5–8. Functioning as a natural water regulator, the AWT retains precipitation during the monsoon season and gradually releases meltwater during the dry season, thereby stabilizing river flows and mitigating seasonal water scarcity for urban and agricultural users downstream^9,10. This buffering capacity also dampens the effects of inter-annual climate variability; in years of weak monsoon precipitation, increased meltwater compensates for rainfall deficits and reduces drought severity^5,11.

Fig. 1. The AWT and its downstream cities.

The study area encompasses nine major river basins: AMU (Amu Darya), IND (Indus), TRM (Tarim Interior), GAN (Brahmaputra–Ganges), TPL (Interior Tibetan Plateau), MEK (Mekong), SAL (Salween), YEL (Yellow River), and YAN (Yangtze). Cities are categorized by population size, as indicated by the colored circles. Hydrological basin boundaries are derived from the HydroSHEDS dataset, and glacier volume data, aggregated to a 0.05 × 0.05° grid, are sourced from Farinotti et al.⁵⁵. The background topography is derived from HydroSHEDS digital elevation model data, and river network data are obtained from Natural Earth (https://www.naturalearthdata.com).

As a global hotspot of climate change, the AWT has warmed at a rate of 0.42 °C per decade from 1980 to 2018—more than twice the global average³. This rapid warming has driven sustained glacier mass loss, earlier snowmelt, and changes in runoff components^11–13, reshaping river discharge regimes and seasonal water availability⁸. These changes threaten the water security of approximately 400 million downstream urban residents¹⁴. Meanwhile, the region has experienced rapid urbanization: between 1990 and 2020, the urban population grew from 250 million to 390 million, corresponding to an annual growth rate of 1.45%. Urbanization rates rose from 25.8% to 46.1% across the region’s countries¹⁵. With urban populations projected to expand by over 50% by 2050¹⁶, the tension between water supply and demand is expected to intensify. Simultaneously, climate change will continue to alter meltwater availability^6,17,18. Thus, accurately quantifying meltwater changes and assessing their downstream impacts on urban water security is essential for evidence-based regional water resource management and for advancing the United Nations Sustainable Development Goals (SDGs) on water security and climate action.

Recent studies have significantly advanced our understanding of cryosphere–hydrology interactions in the AWT and their implications for regional water resources^8,19,20, indicating substantial changes in meltwater availability and seasonal distribution in recent decades. Under continued warming, meltwater availability is projected to increase until it peaks, followed by a gradual decline^9,10,21. The timing and magnitude of these changes vary considerably across basins²², reflecting spatial heterogeneity in glacier cover, elevation, and climate sensitivity^13,23. Meanwhile, global-scale assessments of urban water scarcity consistently identify the downstream regions of the AWT as critical hotspots for both present and future water stress^24–26. For instance, He et al.²⁷ highlight that South Asia faces a severe urban water crisis, intensified by climate change and urbanization. Despite comprehensive mitigation efforts, water scarcity is expected to persist in megacities such as New Delhi, India, and Lahore, Pakistan^28,29. However, the extent to which climate-driven changes in mountain meltwater impact urban water security in downstream regions remains poorly understood. Previous studies have not explicitly coupled projections of high-altitude meltwater change with forward-looking, city-level socio-economic datasets required to assess water scarcity for rapidly growing urban populations. In particular, it is unclear whether the projected spatiotemporal changes in meltwater availability can effectively alleviate urban water scarcity in these increasingly vulnerable areas.

In this study, we systematically evaluate the impact of meltwater changes on urban water scarcity in the downstream regions of the AWT. First, we use a high-resolution cryosphere–hydrology model to simulate regional hydrological processes from 1990 to 2020 and quantify meltwater contributions to urban water supply. We then assess the extent to which meltwater mitigates urban water scarcity. Finally, we analyze the impacts of future meltwater changes on urban water scarcity under different climate scenarios for 2020–2050. Our findings offer valuable insights for informing basin-level water resource management and climate change adaptation.

Results

Meltwater is an important and stable source of urban water supply

Meltwater availability plays an important role in sustaining urban water supplies across the downstream regions of the AWT. A total of 782 cities—including 20 large cities with populations exceeding one million—rely on meltwater for more than 50% of their seasonal water supply (Fig. 2a). At the basin scale, meltwater contributions exhibit significant spatial heterogeneity. Highly meltwater-dependent cities (at least one month when meltwater contribution exceeds 50%) are primarily concentrated in the Indus, Yellow River, Amu Darya, and Tarim River basins. The proportion of cities exceeding this threshold ranges from 80% in the Indus basin to 100% in the Tarim River basin (Fig. 2c), reflecting a spatial transition in meltwater composition—from glacier-dominated supply in the westerly basins (e.g., the Indus, Amu Darya) to snow-dominated supply in the eastern basins (e.g., the Yellow River) (Supplementary Fig. S1). In contrast, cities in the southeastern basins of the Yangtze, Mekong, Salween, and Brahmaputra–Ganges generally exhibit low meltwater dependency, with meltwater contributing less than 5% of the seasonal water supply in most urban areas.

Fig. 2. Meltwater contributions to urban water supply in the downstream regions of the AWT.

Relative contribution of meltwater to urban water availability in the 1990s (1990–1999) (a); changes in relative meltwater contributions in the 2010s (2010–2019) relative to the 1990s (b); number and proportion of highly meltwater-dependent cities (defined as having meltwater contributions >50% in at least one month of the year) across major river basins (c); and intra-annual variability of meltwater availability for all highly meltwater-dependent catchments, as well as those in the Indus, Amu Darya, Brahmaputra–Ganges, Tarim, and Yellow River basins (d–i). Error bars represent ±1 standard deviation of interannual variability for each month. The base map is derived from Natural Earth (https://www.naturalearthdata.com).

Meltwater has provided a relatively stable source of freshwater for downstream cities. Between 1990 and 2020, variability in meltwater contributions remained within ±5% for most cities (Fig. 2b). However, cities with high meltwater dependency have shown a slight decline in meltwater contribution, with nearly 80% (624 cities) exhibiting reduced dependence. These cities are primarily located in the Indus and Yellow River basins. This apparent decline reflects a faster increase in other surface water sources, particularly rainfall-driven runoff, which expanded total water availability and reduced the proportional share of meltwater. Despite this decline, average annual meltwater availability increased by 13% across the 184 highly dependent catchments (Supplementary Fig. S2). This net increase was mainly driven by enhanced glacier melt (~10.6 percentage points), with a smaller contribution from snowmelt (~2.4 percentage points) and substantial inter-catchment variability. At the seasonal scale, meltwater availability increased in most months, with the largest absolute increases occurring during the late ablation season (September and October), while slight decreases were simulated during the dry winter period (December and January), when meltwater generation is limited by low temperatures. The largest gains in annual meltwater availability were observed in the Tarim River Basin (+22%), followed by the Indus (+15%), Brahmaputra–Ganges (+12%), Amu Darya (+11%), and Yellow River (+11%) basins.

Meltwater mitigates urban water scarcity

Meltwater has been instrumental in mitigating urban water scarcity across the study region. In the 1990s, it alleviated water scarcity (defined as Water Scarcity Index > 1.0; see Eq. (8)) for 12.3 ± 3.3 million urban residents, increasing to 18.8 ± 4.5 million by the 2010s (Fig. 3a). The mitigation effect is particularly pronounced in catchments highly dependent on meltwater (Fig. 3b). In these areas, meltwater reduced water scarcity for 9.9 ± 3.0 million people in the 1990s, constituting 14.6 ± 5.3% of the total population facing water scarcity. By the 2010s, this number had risen to 17 ± 4.9 million, representing 21.7 ± 7.4% of residents experiencing water scarcity. At the basin scale, the Indus River Basin exhibited the strongest mitigation effect (Supplementary Fig. S3). Among its 28 highly meltwater-dependent catchments, the number of urban residents whose water scarcity was alleviated by meltwater increased from 4.2 ± 0.5 million in the 1990s to 5.2 ± 2.0 million in the 2010s.

Fig. 3. The role of meltwater in mitigating urban water scarcity.

Total urban population affected by water scarcity from 1990 to 2020 (a, b), comparing scenarios with (blue lines) and without (red lines) meltwater inclusion for the entire study region (a) and highly meltwater-dependent cities (b). The gray bars represent the difference between the two scenarios (i.e., the population buffered by meltwater). Seasonal variation in the water-scarce urban population (c, d), with and without meltwater inclusion, for the entire study region (c) and highly meltwater-dependent cities (d). Shaded areas indicate ±1 standard deviation, representing the inter-annual variability within each decadal period. See Supplementary Figs. S3, S4 for detailed results at the basin scale.

The mitigation effect of meltwater demonstrates pronounced seasonality across the entire study region (Fig. 3c). During the pre-monsoon period (February–June), the effect was strongest. On average, across the three-decade historical period, 31.1 ± 2.8 million residents experienced relief from water scarcity, corresponding to a 19.6 ± 1.7% reduction in the water-scarce urban population. During the monsoon period (July–October), only 9.0 ± 5.1 million people benefited from meltwater. This seasonal discrepancy is more pronounced in catchments highly dependent on meltwater (Fig. 3d). During the pre-monsoon months, meltwater alleviated water scarcity for an average of 26.5 ± 3.8 million people (43.5 ± 6.3% of the water-scarce population). This number dropped to 8.7 ± 4.4 million during the monsoon season. Substantial spatial variation was observed across river basins (Supplementary Fig. S4). In the Indus River Basin, meltwater alleviated water scarcity for an average of 12.9 ± 2.9 million people during the pre-monsoon period, comprising 62.1 ± 13.3% of the total water-scarce population. This number dropped to 5.4 ± 3.8 million during the monsoon period. By contrast, in the Yellow River and Brahmaputra–Ganges basins, while meltwater alleviated scarcity for a maximum of 11.9 ± 1.0 million and 3.2 ± 0.4 million people in a single month, respectively, these figures represent only a minor fraction of the substantially larger urban populations facing water scarcity in these regions. Overall, meltwater played a limited role in alleviating urban water scarcity in these basins.

Meltwater has enabled a transition from perennial to seasonal urban water scarcity in several cities by supplementing water supplies during dry periods (Supplementary Fig. S5). Perennial water scarcity is defined as occurring when annual water withdrawals exceed annual water availability, whereas seasonal water scarcity occurs when this condition is met for at least one month within a year (see “Methods”). Across the study region, meltwater reduced the number of people affected by perennial water scarcity by 22.9–28.7 million. In highly meltwater-dependent catchments, this reduction ranged from 15.3 to 26.5 million people, representing 22.1–33.2% of the region’s water-scarce urban population. The strongest mitigation effects were observed in the Indus Basin (8.3–14.2 million), with notable reductions also evident in the Yellow River (3.1–7.9 million) and Amu Darya (1.4–2.0 million) basins. In the Tarim Basin, the affected population declined sharply from 2.0–2.8 million to 0.5–0.6 million. By contrast, the Brahmaputra–Ganges Basin exhibited a comparatively limited mitigation effect.

Meltwater’s buffering role is projected to weaken

By the 2040 s, all climate scenarios project an increase in meltwater availability in catchments with high meltwater dependency (Fig. 4; see Supplementary Fig. S6 for relative changes). Despite these gains, multi-model ensemble mean projections indicate no significant long-term trend in meltwater’s annual relative contribution (Supplementary Fig. S7). This stability is consistent with projected increases in regional precipitation, which drive concurrent increases in runoff from rainfall, thereby offsetting changes in the relative share of meltwater. In terms of magnitude, the most pronounced increase is expected under the SSP5-8.5 scenario, reaching 10.0% (~1.49 billion m³), with ~0.44 billion m³ derived from glacier melt and ~1.05 billion m³ from snowmelt, while the smallest increase of 4.1% (~613 million m³) occurs under SSP3-7.0 (Supplementary Fig. S8). Seasonally, meltwater availability is projected to rise primarily during the monsoon months (July–September) and in April, with increases ranging from 10% to 30%. In contrast, consistent declines are projected during the dry season from October to March, particularly between November and February, when relative reductions exceed 50% in most catchments. January—typically the month with the lowest absolute meltwater availability—shows the largest relative decline, which reflects reductions from an already low baseline rather than large absolute losses. At the basin scale, meltwater availability is projected to increase in the Indus (5.9–12.0%), Amu Darya (10.7–17.6%), Tarim (7.7–26.4%), and Brahmaputra–Ganges (12.2–19.4%) basins, mainly from April to October. However, all four basins face sharp declines—often exceeding 80%—from November to February. The Yellow River Basin shows overall losses.

Fig. 4. Projected absolute changes in meltwater availability under different climate scenarios.

Projected absolute changes in meltwater availability for the 2040 s compared to the 2010s in highly meltwater-dependent catchments are shown for all catchments (a) and for individual river basins (b–f). Projections are based on 18 CMIP6 models, and error bars represent one standard deviation across model simulations. Corresponding relative changes in meltwater volume are shown in Supplementary Fig. S6.

Although meltwater availability is expected to increase by the 2040 s, its capacity to mitigate urban water scarcity will remain limited due to a much faster rise in water demand. Urban populations in water-scarce regions are projected to grow substantially, increasing by 75.8–142 million people (28.5–53.4%) across different SSP scenarios. The largest population increase is anticipated under SSP3-7.0 (53.4%), and the smallest under SSP5-8.5 (28.5%) (Fig. 5a). This contrast primarily reflects differences in socioeconomic pathways, with SSP3 characterized by higher population growth and lower adaptive capacity, whereas SSP5 assumes lower population growth and more advanced water management, partially offsetting the impacts of stronger climate change. Notably, 77–98% (74.4–109.9 million) of this growth will occur in highly meltwater-dependent catchments (Fig. 5b). At the basin scale, the largest absolute increases in water-scarce urban populations are expected in the Brahmaputra–Ganges (43.3–72.3 million) and Indus basins (29.8–53.7 million) (Supplementary Fig. S9). Seasonal variation further complicates the situation. Driven by the widening gap between water supply and demand, the increase in the population exposed to water scarcity is projected to be substantially higher during the non-monsoon season (October–June), rising by 71.2–119.6 million, compared to an increase of 15.6–40.5 million during the monsoon season. In highly meltwater-dependent catchments, water scarcity will intensify perennially, with increases often exceeding 50%, particularly between November and March, when an additional 50 million people may face water scarcity (Fig. 5c, d).

Fig. 5. Urban populations facing water scarcity under future population growth.

Changes in the urban population exposed to water scarcity from 2020 to 2050 under different SSP scenarios for the entire study region (a) and for highly meltwater-dependent cities (b). The colored vertical lines for the 2040 s indicate the range of projected population changes associated with water scarcity, representing ±1 standard deviation from the multi-model ensemble mean meltwater availability based on 18 CMIP6 models. Seasonal variation in the water-scarce urban population (c, d). See Supplementary Fig. S9 for detailed results at the basin scale.

While the urban population in the study region experiencing water scarcity is projected to grow by 28.5–53.4% by the 2040 s, meltwater availability is expected to increase by only 12.2% (~0.82 billion m³) under the most extreme emissions scenario (SSP5-8.5) (Fig. 4). In 782 meltwater-dependent cities, the water-scarce population is expected to rise by 76–142 million, whereas meltwater supply will grow by less than 10% (~1.49 billion m³). For instance, Lahore’s population in the Indus Basin could increase by 47.3–112.5% (3.6–8.5 million), while meltwater availability is projected to increase by less than 1% (~1.04 million m³). Similarly, in the Ganges Basin, New Delhi’s population may increase by 36.4–69.7% (9.8–18.8 million), yet meltwater availability will increase by only 24.7% (~0.30 billion m³) under SSP5-8.5 (Supplementary Fig. S9). Furthermore, the seasonal mismatches between supply and demand severely undermine meltwater’s buffering capacity. Our analysis indicates that meltwater increases are predominantly concentrated in the monsoon months when local precipitation is already abundant, and demand is met. Conversely, during the critical dry season—which coincides with peak urban water scarcity—meltwater availability is projected to decline by more than 50% relative to the 2010s baseline (Supplementary Fig. S6). This temporal decoupling effectively converts a potential annual surplus into a functional seasonal deficit, thereby directly intensifying shortage risks during the months of highest vulnerability.

Water competition between urban and agricultural sectors arises from conflicting demands on limited resources, a tension exacerbated by climate change and rapid urbanization^30,31. In South Asia, the Rabi season (November to March) is vital for major crops such as wheat, chickpeas, and rapeseed, contributing 40% of total food production in India and Pakistan. Unlike the monsoon-dependent Kharif season, Rabi crops rely predominantly on irrigation, with peak demand from November to February coinciding with the lowest meltwater availability^8,11. Our results identify the Rabi season as the most severe urban water scarcity period. Continued urbanization will increase domestic and industrial water demand, intensifying seasonal competition with agricultural irrigation³². Declining meltwater availability will further intensify this competition, likely pushing both urban and agricultural systems toward unsustainable groundwater extraction and amplifying systemic risks to regional water security.

Among the 184 catchments highly dependent on meltwater, 60 are projected to face water scarcity during the Rabi season (Supplementary Fig. S10), affecting 410 cities, including 13 large cities such as New Delhi and Lahore. Under the SSP2-4.5 scenario, meltwater availability in these catchments is expected to decline by 51 ± 20% (Fig. 6), while irrigation demand is projected to increase by 11 ± 14%, and the urban population exposed to water scarcity may rise by 43%. The Indus, Tarim, and Yellow River basins exhibit a consistent pattern of declining meltwater availability, rising irrigation demand, and growing urban water-stressed populations. These trends are most pronounced in the Indus Basin, where meltwater availability during the Rabi season is projected to decline by over 50%, irrigation demand to increase by 19 ± 18%, and the water-scarce urban population to increase by 58%, affecting 189 meltwater-dependent cities. As meltwater availability declines during critical dry-season months, the threat to urban water security will intensify. Prioritizing urban water use may reduce agricultural irrigation, potentially lowering crop yields and endangering regional food security. Drawing on the Aqueduct 4.0 global maps³³, we also identified 32 catchments that are heavily dependent on meltwater and classified as hotspots of unsustainable groundwater extraction, encompassing 310 cities (Supplementary Fig. S11). As water demands from urban and agricultural sectors continue to grow while meltwater availability declines, these regions will likely become increasingly reliant on groundwater, accelerating its depletion and further destabilizing already vulnerable water systems.

Fig. 6. Projected urban–agricultural water competition during the Rabi season in the 2040 s under the SSP2-4.5 scenario.

Projected changes in meltwater availability, irrigation demand, and the urban population exposed to water scarcity in the 2040 s relative to the 2010s across major river basins. In catchments where declining meltwater availability coincides with increasing irrigation demand and a growing urban population, competition between urban and agricultural water use is expected to intensify. Error bars for meltwater availability represent the standard deviation across 18 CMIP6 models, while those for irrigation demand reflect variability across simulations driven by five GCMs. White circles indicate decreasing irrigation demand. Some schematic elements were created using EdrawMax.

Discussion

This study systematically evaluates the impact of meltwater changes on urban water scarcity in the downstream regions of the AWT. Our results demonstrate that meltwater serves as a vital resource for sustaining urban water supplies. Over the past three decades, it has mitigated seasonal water scarcity, supporting 22.9–28.7 million urban residents by augmenting dry-season water supplies. While total meltwater availability is projected to increase by 4–10% until the 2040 s under varying climate scenarios, the urban population facing water scarcity is expected to rise by 31–53%. This intensification is driven primarily by a surge in demand that far outpaces the projected supply. Critically, during peak water-stress months (October–May), meltwater availability is projected to decline by more than 50%. This growing mismatch between supply and demand implies that future meltwater gains will have marginal efficacy in alleviating urban water scarcity. Instead, the growing imbalance will likely exacerbate competition between urban and agricultural sectors and drive unsustainable groundwater exploitation.

Water scarcity in downstream regions of the AWT, particularly in South Asian cities, has attracted increasing attention in recent years^34,35. Current mitigation strategies largely emphasize improving water-use efficiency, such as water-saving agricultural technologies and allocation quotas, alongside efforts to strengthen transboundary water-sharing agreements^36–38. However, the seasonal variability of meltwater and its implications for water scarcity remain insufficiently addressed. This implies that future meltwater gains will have only marginal efficacy in alleviating urban water scarcity, as these gains largely coincide with precipitation-driven increases in water availability during the wet season, rather than offsetting deficits during the dry season. For instance, the transboundary water allocation agreement for the Indus River Basin inadequately accounts for intra-annual variability in meltwater. This variability intensifies winter low-flow conditions and causes summer flood peaks to occur earlier, significantly challenging urban water security. Moreover, current policies primarily emphasize generalized water stress assessments^39,40, with limited attention to targeted regulation and risk management for highly meltwater-dependent cities. These gaps highlight the urgent need for adaptive, region-specific water management strategies that respond to the complex and evolving impacts of meltwater variability across heterogeneous basin contexts.

Our findings highlight the vital role of meltwater in sustaining urban water supply downstream of the AWT, where seasonal supply–demand mismatches threaten water security. To address these risks, we propose three recommendations: (1) establishing high-resolution cryosphere–hydrology monitoring in high-risk basins (e.g., Indus, Amu Darya), integrating satellite and ground data to track meltwater flows and issue real-time, demand-sensitive threshold warnings that adjust allocation priorities based on short-term forecasts; (2) strengthening adaptive, transboundary basin agreements that reflect spatial meltwater variability, and update water-sharing rules to match seasonal glacier melt and demand profiles; and (3) adopting phased adaptive management: short-term measures (e.g., seasonal drought contingency plans) to assess combined meltwater impacts on cities and irrigation, and long-term strategies to detect tipping points in meltwater trends from climate change and human activity. This shift towards integrated, climate-responsive water governance will enhance resilience across diverse hydrological settings.

This study has some limitations. First, the hydrological model does not incorporate groundwater extraction. Although groundwater in the study region is predominantly used for agricultural irrigation⁴¹, substantial groundwater extraction for domestic use also occurs in some large cities, where illegal pumping is reported^42,43. Therefore, ignoring groundwater extraction may introduce some uncertainty in assessing meltwater-driven urban water scarcity. In addition, the simplified representation of cryospheric–subsurface interactions in the hydrological model may introduce additional uncertainty in the partitioning between surface runoff and groundwater contributions at monthly and basin scales, particularly in snow- and glacier-dominated catchments. Recent studies have highlighted that groundwater systems can be strongly coupled with cryospheric processes and that changes in glacierization may alter the relative contribution of surface runoff versus groundwater storage^44,45. Such processes are not explicitly resolved in our modeling framework and therefore constitute an additional source of structural uncertainty at the scales considered here. In particular, systematic limitations in representing winter baseflow and groundwater discharge in high-altitude catchments (Supplementary Fig. S12) may lead to biases in the partitioning between meltwater and rain/baseflow contributions, and consequently influence the estimation of water availability and water scarcity. This structural uncertainty is reflected in the differing runoff partitioning estimates reported across basin-scale studies using similar modeling frameworks (e.g., Khanal et al.⁶). These differences highlight the sensitivity of large-scale hydrological models to methodological choices, including meteorological forcing, simulation period, and calibration constraints. Importantly, such uncertainties affect the representation of baseflow contributions and should therefore be considered when interpreting winter water scarcity estimates.

Second, there is an inherent inconsistency between the Spatial Processes in Hydrology (SPHY)-simulated water availability and the GCM-based withdrawal projections, as the latter may partly depend on simulated availability rather than solely on demand. Third, reservoir and lake processes are not explicitly represented. Although the SPHY model includes modules to simulate these dynamics, their implementation requires detailed observational data (e.g., operational rules, storage capacity), which are scarce across this transboundary region. As a result, these processes were excluded, representing a notable limitation. Reservoir operations are known to strongly influence downstream hydrology⁴⁶; studies in heavily engineered basins have shown that reservoir management can substantially alter seasonal hydrographs by storing peak meltwater and releasing water during low-flow periods^47,48. Consequently, our unmanaged (naturalized) flow simulations may overestimate pre-monsoon water scarcity and underestimate late- and post-monsoon water availability in these regulated basins. Fourth, while the 1990–2020 study period captures the most rapid phase of recent urbanization and water demand growth, it is relatively short for characterizing low-frequency, multi-decadal climate variability.

Finally, our estimates of the urban population exposed to water scarcity should be regarded as conservative. We adopted the Water Scarcity Index (WSI) threshold of 1.0 to define water scarcity, while previous studies²⁷ adopt a broader range from 0.4 to 1.0. Although a WSI threshold of 1.0 has a clear physical meaning—indicating that water demand equals available supply—it does not account for environmental flow requirements⁴⁹. Thus, it may underestimate the severity of water scarcity. Sensitivity analysis indicates that under the standard WSI = 1.0 threshold, an average of 204.4 million urban residents were affected annually between 1990 and 2020. When the threshold is lowered to 0.4, the affected urban population increases by 19%, while the population experiencing seasonal water scarcity rises by 37%. In highly meltwater-dependent catchments, the corresponding increases are 14% and 28%, respectively (Supplementary Fig. S13). These findings suggest that incorporating environmental flow requirements would substantially increase estimates of urban populations exposed to water scarcity. Future research should integrate groundwater use, reservoir operations, and lake storage into high-resolution, multi-process hydrological models to more comprehensively evaluate the long-term impacts of meltwater changes on regional water security under climate change.

Methods

First, we used the SPHY model to simulate hydrological processes across the AWT and its downstream regions from 1980 to 2020 at a spatial resolution of 5 × 5 km, and to evaluate the importance of meltwater in urban water supply. Second, we quantified the actual role of meltwater in alleviating water scarcity in downstream cities. Finally, by integrating projected hydrological changes with gridded population projections under Shared Socioeconomic Pathways (SSPs)⁵⁰, we projected the potential impacts of meltwater changes on urban water scarcity under different climate change scenarios for the period 2020–2050 (Supplementary Fig. S14).

Cryosphere–hydrology modeling

We employed the SPHY model, a spatially distributed, raster-based “leaky-bucket” hydrological model⁵¹. The model integrates various input variables, including topography, daily mean, maximum and minimum temperatures, precipitation, soil properties, land use, and glacier cover. It has been widely applied in cryosphere–hydrology studies across high-mountain Asia, spanning various spatial and temporal scales^6,9. To account for complex topography, the coarser WFDE5 (WATCH Forcing Data methodology applied to ERA5) temperature data⁵² were downscaled to the 5 km model resolution using the SPHY model’s preprocessing module. This process applies a vertical temperature lapse rate (–0.0065 °C/m) to correct for elevation differences between the forcing data and the high-resolution DEM. SPHY comprises six interconnected modules: glacier, snow, evapotranspiration, groundwater, routing, and reservoir/lake. The study area is divided into uniform grid cells. The model dynamically simulates key hydrological processes within each grid cell based on meteorological forcings and surface characteristics. These processes include snow and glacier melt, precipitation, evapotranspiration, and infiltration. It then calculates total runoff ($R_{Tot}$), glacier runoff ($R_{glacier}$), snowmelt runoff ($R_{snow}$), and rainfall runoff ($R_{rain}$). These runoff components are routed through the landscape according to flow accumulation paths, ultimately determining streamflow and its components at the catchment outlet.

$$R_{Tot}=R_{glacier}+R_{snow}+R_{rain}+R_{basic}$$ 1

In this study, meltwater is a composite term defined as the sum of glacier runoff and snowmelt runoff. Glacier runoff represents water generated solely from the melting of glacier ice (both debris-covered and debris-free). Snowmelt runoff represents water released from the snowpack and includes both snowmelt and rainfall on existing snowpack, adjusted for refreezing or temporary storage within the pack⁵³. Rainfall runoff is treated separately and refers to runoff from liquid precipitation over snow-free, non-glacierized land. This definition of meltwater is consistent with previous hydrological studies in high-mountain regions^8,11.

$$R_{melt}=R_{glacier}+R_{snow}$$ 2

The glacier and snow modules of the SPHY model simulate cryospheric hydrological processes at a daily time step using the degree-day approach⁵¹. The snow module determines the phase of precipitation (rain or snow) by comparing the daily mean air temperature ($T_{avg}$) with a calibrated critical temperature threshold ($T_{crit}$), and subsequently estimates potential melt using degree-day factors (DDFs). $T_{crit}$ is a calibrated model parameter rather than a quantity prescribed by the forcing data, ranging from −2.0 °C to 0 °C across the study basins (see calibrated values in Supplementary Table S1). It also accounts for the snowpack’s water-holding capacity, enabling the simulation of delayed meltwater release and capturing the spatiotemporal dynamics of snow accumulation and ablation. Glacier melt is similarly calculated using the degree–day approach. Glacier outlines were derived from the Randolph Glacier Inventory (RGI 6.0). To reflect the differential melt behavior of clean and debris-covered glaciers, glacier type was identified following the method proposed by Paul et al.⁵⁴, using elevation and slope thresholds derived from the DEM. Distinct DDFs were applied to each glacier type, with the DDF for debris-covered ice constrained to be lower than that for clean ice to account for the insulating effect of supraglacial debris. These DDFs were treated as independent parameters and calibrated separately against observed glacier mass balance data.

We simulated hydrological processes across nine major river basins within the AWT and its downstream regions from 1980 to 2050 at a spatial resolution of 5 × 5 km. To improve the accuracy of glacier melt estimates, we incorporated sub-grid variability. Specifically, glacier fractional cover and elevation were derived from 100 m resolution data to define unique sub-grid simulation units (U_IDs) within each coarse 5 km grid cell (MOD_ID). Melt rates were computed for each specific U_ID to capture elevation-dependent temperature gradients, and the generated meltwater volumes were subsequently aggregated to the 5 km grid scale for hydrological routing. Additionally, we integrated glacier-related parameters into the model, including mean glacier elevation and initial ice thickness. Initial ice thickness was obtained from the global ice-thickness dataset of Farinotti et al.⁵⁵ and was used to initialize the dynamic glacier evolution component of the model. Unlike elevation and glacier type, initial ice thickness does not enter the degree–day melt calculation, but instead defines the initial ice volume available for long-term glacier retreat and area change. The model accounts for future glacier changes using a mass-conserving glacier evolution algorithm. While explicit ice flow dynamics are not simulated, the model incorporates a feedback mechanism between mass balance and glacier geometry via an ice redistribution scheme. At the end of each hydrological year, the model calculates the net mass imbalance and redistributes ice from the accumulation zone to the ablation zone. This updates the glacier volume and area based on the simulated accumulation and ablation, allowing for the transient simulation of glacier retreat.

Given the complex composition of runoff sources in the AWT, conventional discharge observations alone are inadequate for accurately capturing the spatiotemporal variations in glacier and snowmelt processes. To validate the performance of the SPHY model, we implemented a stepwise calibration strategy⁸ designed to minimize equifinality (i.e., achieving accurate discharge simulations through incorrect internal process representations). First, parameters related to glacier and snow processes were calibrated and constrained for the period 2002–2020 using MODIS snow cover data and regional glacier mass balance observations, with a split-sample strategy applied to both glacier and snow components to distinguish calibration and evaluation periods. This ensures that the simulated cryospheric contributions are physically realistic. Second, the accuracy of discharge simulations was assessed using discharge records from 13 gauging stations spanning the period 1981–2019 (Supplementary Fig. S15), with the Nash–Sutcliffe Efficiency (NSE) and Pearson Correlation Coefficient (CC) used as performance metrics. This two-step calibration procedure reduces compensating errors among glacier, snow, and runoff components, thereby improving the internal consistency of the simulated hydrological processes.

The model demonstrated satisfactory performance, with simulated discharge closely matching observed values across all gauging stations. NSE values exceeded 0.6 at all sites and surpassed 0.8 at four of them, while CC values remained consistently above 0.8 (Supplementary Fig. S12). However, it is important to acknowledge that in specific western high-altitude catchments, the model exhibits limitations in accurately capturing winter recession behaviors and low-flow periods, which is likely linked to structural challenges in representing slow groundwater release from frozen subsurface reservoirs. Beyond discharge, the simulation also showed strong consistency with glacier mass balance observations (Supplementary Fig. S16). However, it is important to note that global geodetic datasets⁵⁶ inherently carry uncertainties and should be treated with caution at the site-specific scale, as highlighted by discrepancies when compared directly against available in-situ field measurements (Supplementary Fig. S17). To further contextualize glacier runoff projections, we compared SPHY-derived glacier runoff with outputs from established glacier evolution models (Supplementary Note 1, Fig. S18–S21). Regarding snow dynamics, the simulated snow cover strongly agreed with MODIS observations, yielding an overall CC of 0.76, with basin-specific values ranging from 0.75 to 0.96 (Supplementary Figs. S22, S23).

Due to the limited availability of systematic observations on snowmelt, glacier melt, and baseflow in most AWT basins, we further evaluated the model performance using the upper Brahmaputra as a representative catchment⁵⁷. In this case, simulated runoff components were compared against alternative hydrological models. The simulated runoff partitioning patterns generally align with those from these established models (Supplementary Fig. S24). Regarding baseflow (Supplementary Figs. S25, S26), however, since neither SPHY nor GLDAS has been thoroughly validated against in-situ groundwater observations in these data-scarce regions, this alignment highlights shared modeling behaviors rather than absolute accuracy. The systematically low winter flow across these models likely reflects common structural limitations in representing slow subsurface drainage in frozen environments.

Quantifying the contribution of meltwater to urban water supply

The reliable assessment of urban water scarcity requires high-resolution and temporally continuous data on urban extents and population dynamics. However, existing datasets in the AWT and its downstream regions present two major limitations: (1) limited coverage of small and medium-sized cities, and (2) the absence of consistent long-term time series data. To address these gaps, we developed a spatiotemporal dataset of urban population dynamics from 1990 to 2050 for the AWT and its downstream regions using a multi-source data integration approach.

The methodology comprised the following steps: First, we identified 2092 urban locations in the region based on data from OpenStreetMap and the National Fundamental Geographic Information Database of China. Second, urban extents for the historical period (1990–2020) were refined using a combination of impervious surface maps and urban boundary products^58–60. Visual inspection was conducted to manually exclude misclassified areas, such as industrial zones and dispersed rural settlements. Third, the refined urban boundaries were overlaid with annual gridded population data¹⁴ to extract city-level population time series. Based on this reconstruction, the dataset at the 2020 baseline comprises 10 megacities (>5 million residents), 47 large cities (1–5 million), 55 medium-sized cities (0.5–1 million), and 1980 small cities (<0.5 million). For future projections, we used the 2020 urban boundaries as a baseline and incorporated urban expansion data⁶¹ along with population projections⁵⁰ under the SSPs to develop a spatiotemporal population dataset for 2020–2050. This dynamic approach explicitly accounts for the spatial expansion of existing urban areas and the transition of smaller towns into larger cities.

We quantified the relative contribution of meltwater to urban water supply by calculating the ratio of available meltwater (${AM}_i$) to total available water (${AW}_i$) at the catchment level (Supplementary Fig. S27). Hydrological processes were simulated at a daily time step, and all runoff components were aggregated (summed) to monthly volumetric totals at the catchment scale prior to analysis, ensuring consistency with monthly water withdrawal data and preservation of water mass balance. This ratio, referred to as the meltwater contribution index (${MC}_i$), is defined as:

$${MC}_i=\frac{{AM}_i}{{AW}_i}\times 100\%$$ 3

where ${AM}_i$ represents the monthly aggregated amount of meltwater available in catchment $i$, which includes both locally generated meltwater ($R_{melt,i}$), and upstream meltwater inflow into catchment $i$ (${UM}_i$):

$${AM}_i=R_{melt,i}+{UM}_i$$ 4

The upstream meltwater inflow (${UM}_i$) is estimated based on the difference between total available water (${AW}_{i-1}$) and water withdrawal (${WW}_{i-1}$) in the upstream catchment $i-1$, multiplied by the ratio of accumulated meltwater in the upstream inflow:

$${UM}_i=\left({AW}_{i-1}-{WW}_{i-1}\right)\times \frac{\sum R_{Melt}}{R_{Tot}}$$ 5

where the fraction represents the upstream-aggregated meltwater mixing ratio of the incoming river flow. Specifically, $\sum R_{Melt}$ denotes the total meltwater runoff generated from all upstream catchments draining into catchment $i$, and $R_{Tot}$ represents the corresponding total river discharge accumulated over the same upstream area and time period. This ensures that meltwater generated in headwater regions is correctly tracked as it travels downstream. $WW$ includes agricultural irrigation, livestock, industrial, and domestic uses (Supplementary Note 2, Supplementary Fig. S28).

A catchment was classified as ‘highly meltwater-dependent’ if the relative contribution of meltwater exceeded 50% in at least one month of the year. This criterion was adopted to identify regions where meltwater functions as the dominant water source during critical periods, consistent with standard classifications of dominant hydrological regimes and critical resource dependency^62,63. Cities located within these catchments were therefore defined as highly meltwater-dependent cities. Total available water in catchment $i$ (${AW}_i$) is defined as the sum of local runoff ($R_{Tot,i}$), upstream inflow (${UR}_i$), and inter-basin water transfers (${WT}_i$).

$${AW}_i=R_{Tot,i}+{UR}_i+{WT}_i$$ 6

where $R_{Tot,i}$ represents the total naturalized runoff simulated by the SPHY model, including glacier melt, snowmelt, rainfall runoff, and baseflow. Although the SPHY model includes a reservoir module, dynamic reservoir regulation was not activated in this study due to the lack of consistent operational data at the regional scale. Consequently, $R_{Tot,i}$ represents naturalized flow conditions. The influence of major water infrastructure is partially reflected through prescribed inter-basin water transfers (${WT}_i$). ${UR}_i$ is defined as:

$${UR}_i=\max \left(0,{AW}_{i-1}-{WW}_{i-1}\right)$$ 7

Inter-basin transfer volumes were derived from the City Water Map dataset⁶⁴. These inputs were incorporated as fixed annual contributions to total water availability, serving as a proxy for structural redistribution capacity, as specific operational schedules for seasonal transfer regulation are not publicly available for this region.

Evaluating the mitigating role of meltwater in urban water scarcity

We adopted the WSI to evaluate urban water scarcity³³. WSI was calculated at the catchment level on a monthly basis as the ratio of monthly water withdrawals to monthly water availability, thereby explicitly capturing seasonal variations in both water supply (e.g., meltwater timing) and water demand. For catchment $i$, the WSI is defined as:

$${WSI}_i=\frac{{WW}_i}{{AW}_i}$$ 8

Following previous studies^25,27, we defined water scarcity using a WSI threshold of 1.0. Catchments with an annual WSI (aggregated over the standard calendar year) greater than 1.0 were classified as perennial water-scarce catchments. This classification indicates a structural annual deficit, where total annual withdrawals exceed annual availability, distinguishing long-term resource depletion from temporary seasonal mismatches^65,66. Catchments with an annual WSI of 1.0 or less but with at least one month exceeding 1.0 were considered seasonal water-scarce catchments. To estimate the urban population affected by water scarcity, we performed a spatial overlay analysis by integrating these classifications with gridded urban population datasets. This approach assumes that water scarcity at the catchment level reflects scarcity for all sectors, including urban areas, even though urban water use often receives higher allocation priority than agriculture or industry. While this simplification may overestimate the urban population actually experiencing shortages, identifying urban populations located within water-scarce catchments remains a widely adopted practice in large-scale water scarcity assessments^27,30,64, ensuring consistency and comparability with previous studies.

To evaluate the role of meltwater in alleviating urban water scarcity, we compared the size of the water-scarce urban population under two scenarios: with and without meltwater contribution⁶⁷. In the latter, only runoff from rainfall and baseflow were included in total water availability, while water withdrawal was held constant. This design isolates the supply-side contribution of meltwater by ensuring that changes in the WSI arise solely from altered water availability, rather than from demand-side adjustments. The resulting difference in WSI therefore directly quantifies the magnitude of water demand that is currently buffered by meltwater under otherwise identical conditions. It is important to note that this ‘no-melt’ scenario is designed as a counterfactual experiment to isolate the current meltwater contribution, not to simulate a warmer climate state. While a physical loss of melt would likely be accompanied by a shift from snow to rain (thereby increasing direct runoff), we intentionally held the precipitation phase constant to strictly isolate the meltwater component. The population exposed to meltwater-buffered scarcity was then derived from the difference between the two scenarios. The calculation is expressed as follows:

$${\mathrm{\Delta }P}_{melt}=P_{urban}\times \left[\left(H\left({WSI}_{no-melt}-1\right)-H\left({WSI}_{with-melt}-1\right)\right)\right]$$ 9

where ${\mathrm{\Delta }P}_{melt}$ represents the urban population for whom meltwater mitigates water scarcity. $P_{urban}$ denotes the urban population within each catchment, ${WSI}_{no-melt}$ refers to the WSI calculated without meltwater contribution, and ${WSI}_{with-melt}$ refers to the WSI calculated with meltwater contribution. $H\left(x\right)$ denotes the Heaviside step function, defined as:

$$H\left(x\right)=\left\{\begin{array}{c}\hfill 0,x\le 0\hfill \\ \hfill 1,x>0\hfill \end{array}\right)$$ 10

Future projections

To project future changes in meltwater availability, we adopted the ScenarioMIP (Scenario Model Intercomparison Project) framework, a component of CMIP6, which provides a widely accepted scenario framework for linking climate and socio-economic pathways⁶⁸. ScenarioMIP was specifically developed to integrate SSPs with RCPs, facilitating comprehensive, interdisciplinary research and improving our understanding of global climatic and socio-economic interactions^69,70. We employed the four Tier 1 scenarios (i.e., SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5). Detailed descriptions of these scenarios are available in O’Neill et al.⁶⁸.

To ensure consistency between historical observations and future projections, we applied the standard Bias Correction and Spatial Disaggregation (BCSD) method⁷¹. Both CMIP6 outputs and observational datasets were first regridded to a common 1° × 1° resolution, at which bias correction was performed. CMIP6 climate model outputs were corrected via a quantile mapping algorithm⁷², using cumulative probability distributions derived from observational data. For each climate variable and grid cell, this correction was conducted separately for each calendar month, such that month-specific cumulative distribution functions (CDFs) and transfer functions were constructed and applied.

This approach has demonstrated robustness in high-altitude mountainous regions⁷³ and preserves key aspects of climatological variability. By correcting the full CDF rather than individual moments, the method maintains not only long-term means but also variance and distributional shape. Moreover, deriving transfer functions independently for each grid cell retains spatial heterogeneity, seasonal patterns, and interannual variability. These transfer functions were assumed to remain stable across historical and future periods and were applied to adjust future projections at the 1° resolution. This reliance on the stationarity assumption treats model biases as systematic and time-invariant; while this is a standard convention in statistical downscaling and allows preservation of long-term climate change signals, it may introduce uncertainty by not accounting for potential non-stationary biases arising from future shifts in atmospheric circulation or feedback mechanisms.

Following bias correction, climate change signals were spatially disaggregated from 1° to the 5 km hydrological model grid. Bias-corrected anomalies relative to the baseline period were interpolated to 5 km resolution and superimposed onto the high-resolution historical climatology, thereby preserving fine-scale spatial heterogeneity and topographic patterns while imposing large-scale climate change trends projected by the CMIP6 models. The resulting downscaled climate variables were used as inputs to the SPHY model to simulate hydrological processes from 2020 to 2050 and to estimate future changes in meltwater availability.

Validation and uncertainty

First, we validated our results against existing assessments of meltwater contributions conducted at various spatial and temporal scales. Given substantial differences in spatial coverage, time periods, and methodological definitions among these studies, this comparison is intended to provide contextual consistency rather than a direct one-to-one evaluation. Accordingly, our results fall within the range reported by previous studies (Supplementary Tables S2, S3). GCMs exhibit substantial uncertainty when projecting meltwater changes in mountainous regions, particularly in Asia’s climatically complex high-mountain areas^8,74. To address this, we applied statistical downscaling techniques to refine CMIP6 outputs and correct systematic biases in temperature and precipitation fields commonly reported in complex terrain. All hydrological simulations were forced using individual bias-corrected CMIP6 model outputs, thereby preserving inter-model variability and climate extremes relevant for melt processes. We further quantified uncertainty by calculating the ensemble standard deviation across CMIP6 models. Despite notable inter-model variability in the projected magnitude of future meltwater changes, consistent trends in overall increases and seasonal variations are evident across climate scenarios (Fig. 4).

Second, due to the lack of comprehensive spatiotemporal population data for cities of varying sizes within the study area, we constructed a dataset of urban population dynamics from 1990 to 2050 by integrating urban land data with gridded population datasets (Supplementary Fig. S29). However, uncertainties remain regarding the method’s accuracy in representing actual population sizes across cities of different scales. To evaluate the reliability of the constructed dataset, we compiled census data from 209 cities of various sizes and conducted a comparative analysis against the modeled urban population estimates. The results indicate a strong agreement between the datasets (Supplementary Fig. S30), with correlation coefficients exceeding 0.9 at all three analyzed time points. This high level of agreement supports the dataset’s reliability in capturing urban population trends and city-scale dynamics across diverse contexts.

Author contributions

L.L. and C.H. designed the study and developed the analysis plan. L.L. prepared the basic data, did the data analysis and visualization. L.L., C.H., and B.M. drafted the manuscript. T.Q., K.Z., and A.L. contributed to revising the manuscript. All authors contributed to the interpretation of findings, provided revisions to the manuscript and approved the final manuscript.

Peer review

Peer review information

Nature Communications thanks Magali Ponds, Elizabeth Ultee, and the other anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

C.H. discloses support for the research of this work from the National Natural Science Foundation of China [grant No. W2412143 and 42371296], and support for publication from the Project Supported by State Key Laboratory of Earth Surface Processes and Disaster Risk Reduction. L.L., T.Q., K.Z., A.L., and B.M. declare no relevant funding.

Data availability

Historical temperature data are available from the Copernicus Climate Data Store (https://cds.climate.copernicus.eu/datasets). Historical precipitation data are from the CHIRPS dataset (https://www.chc.ucsb.edu/data/chirps). CMIP6 climate model outputs are accessible via the ESGF portal (https://esgf-node.llnl.gov/search/cmip6). Land-use data are available from the GlobCover project (http://due.esrin.esa.int/page_globcover.php). MODIS snow cover data are available at https://nsidc.org/data/modis/data_summaries#snow. Glacier outlines are from the Randolph Glacier Inventory (https://www.glims.org/RGI/). Soil data are available from the HiHydroSoil dataset (https://www.futurewater.eu/projects/hihydrosoil/). Digital elevation model and watershed boundary data are from HydroSHEDS (https://www.hydrosheds.org). Urban boundary data are available from the iEarth DataHub (https://data-starcloud.pcl.ac.cn/iearthdata/). Artificial Impervious Area data are from Li et al. (ref. ⁵⁸) and Huang et al. (ref. ⁵⁹). Historical population data are available at https://zenodo.org/records/10088105, and future population projections at 10.6084/m9.figshare.19608594.v2. Historical water withdrawal data are available at https://zenodo.org/records/1209296, and future water withdrawal projections at 10.7910/DVN/VIQEAB. Interbasin water transfer data are from the City Water Map dataset (https://wbwaterdata.org/dataset/city-water-map). Detailed descriptions of all datasets are provided in Supplementary Table S4. Source data are provided with this paper.

Code availability

The SPHY model code is available at https://github.com/FutureWater/SPHY. The code used in this study is publicly available on Zenodo at 10.5281/zenodo.18185923.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-73245-3.