Study region Jialing River Basin of southwest China. Study focus Climate change exacerbates the imbalance in water-resource allocation, while land-surface conditions strongly regulate water resources across different spatial and temporal scales. In this study, the spatiotemporal characteristics of blue and green water resources (BW and GW) were evaluated based on the Soil and Water Assessment Tool (SWAT), and the impacts of climate change (CC) and land-surface change (LSC) on blue-green water resources in the basin were investigated. New hydrological insights for the region First, from 2000 to 2020, both BW and GW showed an increasing trend with 3.61 mm/year and 4.63 mm/year, respectively. The spatial distribution of BW and GW changes is uneven. Second, CC drives the variations in blue-green water within the basin. LSC can locally reverse the dominant influence of climate change. Third, CC not only directly influences the spatiotemporal characteristics of blue-green water, but also alters their dynamic partitioning by modifying the land-surface conditions, notably through vegetation greening. Our results highlight that CC impacts water resources not only directly, but also indirectly by altering vegetation. This implies that water management policies account for these complex climate-vegetation-water feedbacks to ensure sustainability and resilience in the upper Yangtze River.

Accurate rainfall data is essential for understanding the water cycle and monitoring climate change. However, precipitation estimates from satellite-only retrievals such as the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG) Early Run (IMERG-E), often contain large errors. These errors limit their usefulness for scientific studies and practical applications. To address this issue, we develop a new algorithm called Dual-Timescale Minimum Error (DTME), which combines advanced statistical techniques including the (categorical) double instrumental variable method, inverse errors-based combination, and linear scaling. The DTME algorithm is applied to improve the accuracy of daily satellite-only precipitation data by integrating it with daily and monthly global gauge-based datasets. Taking IMERG-E as an example, this study provides a new precipitation product IMERG-D (0.1 degrees /1 d, 2001-2019) over the Tibetan Plateau, an area where accurate precipitation measurements are especially challenging due to its high altitude, complex climate, and sparse station network. Using 116 independent station observations for performance evaluation, statistical scores show that the algorithm significantly enhances the performance of IMERG-E across multiple timescales, achieving higher correlation (increasing from 0.63 to 0.76), lower errors (decreasing from 3.59 to 2.63 mm/day), and better detection rates (rising from 0.49 to 0.58). Importantly, IMERG-D performs better than the bias-corrected IMERG Final Run and gauge-adjusted Global Satellite Mapping of Precipitation (GSMaP) products. These results demonstrate the effectiveness of the DTME algorithm in improving the reliability of satellite-only precipitation estimates in the alpine region with rare stations. Therefore, the algorithm shows promise for broader application of satellite-only precipitation product in hydrological and meteorological fields, such as hydrological modelling and climate assessments.

The dynamic changes in vegetation significantly affect carbon balance and water cycle. However, understanding the changes in vegetation conditions and water stress remains limited. This study proposes a novel methodology for assessing vegetation eco-regime changes, incorporating 18 indicators that comprehensively characterize vegetation conditions. By developing an ecological restoration index and employing interpretable machine learning (IML) approaches, this study systematically evaluated the impacts of both water stress and ecological restoration on vegetation eco-regimes across China. The results showed that the proportion of mutation years in China's kernel NDVI peaked in 2002. The period from 1982 to 2001 was defined as the baseline period (BP), while the period from 2002 to 2022 was designated as the change period (CP). Compared to BP, the degree of change in vegetation eco-regimes across China during CP ranged from 21.5 % to 89.4 %, with a median value of 64.9 %. The proportions of low, medium, and high change categories were 40.7 %, 59.1 %, and 0.2 %, respectively. The IML approach identified precipitation, surface solar radiation, and ecological restoration as the three dominant factors governing vegetation eco-regime dynamics, with a combined mean contribution proportion of 74.8 % during BP and CP. Furthermore, the interpretable results revealed that the transition threshold of vegetation eco-regimes increased by 40 mm for precipitation and by 0.3 hPa for vapor pressure deficit. The methodology provides a transferable approach for global assessments of vegetation-climate-restoration interactions, informing targeted ecological strategies.

Crop yield and food security worldwide are increasingly threatened by flash droughts given their rapid intensification. It is vital to understand how flash droughts affect cropland and their future projections in a warming climate. Meanwhile, the associated impacts could be strongly different depending on the hit timing of flash droughts during different growth phases of crops, an aspect that remains underexplored in previous studies. This study provides a novel perspective by integrating crop phenology within the evaluation framework for revisiting historical flash droughts in five major croplands in China, and also their projections based on the climate model simulations from CMIP6 under four future emission scenarios. Results show that for the historical period (2000-2020), nearly 40-85 % of flash droughts occurred during crop phenological phases, with the highest exposure in Northeast China and Northwest China, which were most vulnerable to flash droughts. Moreover, the vegetative phase exhibited the highest ratio of flash droughts among the three phenological phases. For CMIP6based projections (present-2100), all five regions showed the duration of drought onset would largely be reduced under the crop phenological phase than the annual averages, with more prominent changes under SSP5-8.5 (reduce by 40 %). Flash drought impacts would be substantially underestimated if overlooking crop phenology, particularly in northern regions given the dominant proportion of flash droughts during the vegetative and reproductive phases. These findings underscore the necessity of incorporating crop phenology into flash drought risk assessment and can inform early warning systems, adaptive management, and food security strategies.

This study assessed risk of major flooding across the globe based on data in the Emergency Events Database spanning 1980 to 2023 and two machine learning methods, extreme gradient boosting (XGBoost) and random forest (RF). A flood disaster index was calculated for politically defined provinces around the world using a combination of analytic hierarchy processing (AHP) and entropy weighting (EW). The resulting indices, together with hydro-meteorological, topographic, vegetation and economic variables, were used to train two machine learning algorithms, which ranked 20 variables according to their relative contribution to flood risk in areas differing in climate zones or levels of socio-economic development. The two algorithms did not substantially differ from each other in their rankings. The modeling identified the following areas as particularly vulnerable: China, South Asia, western Arabian Peninsula, western Germany, Java (Indonesia), Zulia (Venezuela), and eastern Australia. The major determinants of major flood risk depend on the climate zone: in the tropics, economy and precipitation are major determinants; in arid regions, vegetation cover; in temperate regions, population and prolonged heavy rainfall; in cold regions, precipitation and surface soil moisture; and in polar regions, topographic factors. In the socio-economically defined "Global North", precipitation may be the primary determinant, while in the "Global South", economic factors may be more crucial. This study enhances global flood risk assessment through the integration of multi-source data and machine learning techniques, providing novel insights into regional heterogeneity in flood risk.

Study region: The Wei River Basin (WRB), China. Study focus: This study employs a post-processing hybrid framework integrating the Soil and Water Assessment Tool (SWAT) with a Long Short-Term Memory (LSTM) network for streamflow simulation. The SWAT component leverages multi-source meteorological and geospatial datasets to simulate physically constrained hydrological processes, generating intermediate state variables and final discharge. The LSTM component then captures intricate nonlinear spatiotemporal dependencies between these hydrologically relevant features and observed streamflow. A comparative framework with four distinct input schemes dissects this post-processing paradigm. Furthermore, Shapley Additive Explanations (SHAP) method is adopted to interpret the hybrid model and trace its predictions back to physically meaningful hydrological processes. New hydrological insights for the region: The hybrid model achieved Kling-Gupta Efficiency (KGE) of 0.92 and Nash–Sutcliffe Efficiency (NSE) of 0.88, outperforming standalone SWAT and purely data-driven LSTM. SHAP revealed SWAT-derived hydrological features dominate predictions over raw meteorological inputs, with lateral flow identified as the most significant contributor. The synergistic mechanism involving intermediate hydrological variables and SWAT-simulated discharge enhances accuracy. This work demonstrates systematic scheme comparison coupled with interpretability methods can unlock mechanistic insights, providing a case-specific understanding of feature dominance under relatively natural conditions, and advances hybrid models from ''black-box'' tools towards interpretable systems, balancing accuracy with physical insight.

Root-zone water storage, the subsurface reservoir supplying water to plant transpiration, is essential for resilience against prolonged droughts. Previous studies have placed excessive emphasis on static root zone water storage capacity, the maximum water storage within the subsurface root zone available for plant transpiration, while neglecting the dynamics of water use (WU, the volume of water depleted from this reservoir for plant transpiration) and their associated environmental effects. This study employs an improved deficit-based approach incorporating groundwater contribution to elucidate seasonal-to-decadal WU dynamics worldwide. Our analysis reveals a widespread increase of WU across diverse biomes, land use/land cover types, and hydroclimatic gradients. Seasonally, WU increases are observed during 52% of drought months. Annually, approximately 56% of the vegetated areas exhibit significantly increasing trends in annual WU, with a trend 0.43 mm year-1 over vegetated areas. Decadal WU significantly increases across 48% of global vegetated areas, with a mean trend of 0.39 mm yr-1. Climates, particularly temperature dominated these WU dynamics, exhibiting a positive correlation, whereas rising water supply and elevation generally reduced trends. Croplands with sustainable blue and green water availability showed prevalent WU changes. However, cropland areas exhibiting unsustainable trends in increased WU account for 34% of global croplands. In addition to water resource, heightened WU trends correlated significantly with increased belowground biomass carbon sequestration. These findings highlight the critical trade-offs between increasing plants water use and water resource availability, carbon storage and agricultural productivity, necessitating integrated management strategies.

Quantifying ecosystem services (ESs) and understanding their responses to vegetation change are essential for effective ecological management and policy-making. The Yellow River Basin is an ecologically fragile yet strategically important region in China. Previous studies have provided valuable insights into ES dynamics in the basin, but continuous basin-scale assessments spanning more than four decades and explicitly comparing pre- and post-restoration periods remain limited. Accordingly, we developed an integrated framework to examine long-term ES dynamics, trade-offs and synergies, relationships with climate and vegetation, and threshold responses to vegetation change from 1982 to 2022. The results showed that three key ESs (water yield, soil conservation, and net primary productivity) generally increased across the basin over the study period. Water yield and soil conservation both decreased before 2000 and increased afterward, whereas net primary productivity increased in both sub-periods. A composite index representing total ecosystem services (TES) showed spatial heterogeneity, with generally higher values in the source region and the southern part of the middle reaches. Synergistic relationships predominated among ES pairs, although trade-offs were observed in localized areas. TES showed a stronger association with precipitation than with temperature, and its positive association with the non-climatic component of vegetation change extended across 96.3% of the basin. Elasticity analysis identified an NDVI threshold of 0.33 for the basin-scale response of TES, with spatial differences among subregions. These findings provide a long-term and integrated perspective on ES dynamics and support region-specific ecological management rather than a uniform greening strategy.

Effective water resource management requires a comprehensive understanding of runoff processes across spatial scales and their interconnections. However, scale effects pose major challenges to deriving a universal spatial scaling law when extrapolating runoff research from fine to broad scales. Previous studies have mainly focused on relatively small catchments (100 km²). Our results indicate that the runoff coefficient of most sub-basins in the Yellow River Basin exhibits multi-scaling behavior, with spatial patterns varying across scales. Rather than following a single trend, the scale effects of the runoff coefficient are complex and non-monotonic. In smaller sub-basins, the spatial distribution of precipitation primarily controls runoff scale effects, whereas in larger sub-basins, land-use patterns become the dominant governing factor.

Study regionYellow River Basin (YRB)Study focusHydrological drought characteristics are increasingly altered by climate change and human activities. However, the identification of drought development stages and their response to human regulation remains limited. This study developed an integrated non-stationary framework to quantify human impacts on drought development. A non-stationary standardized streamflow index was constructed using the GAMLSS framework from observed and VIC-simulated naturalized streamflow at seven hydrological stations during 1961-2020. Drought events were identified using a three-threshold run theory, and development-stage characteristics, including duration, severity, rate, and seasonal timing, were extracted at station and event scales.New hydrological insights for the regionAt upper-reach stations, human activities increased drought development duration by 69-94% and severity by 42-55%, while slowing development rate by 24-25%. In contrast, at middle- and lower-reach stations, development duration and severity decreased by 13-53%, whereas development rate increased by 35-63%. At the basin scale, human activities exerted a more consistent influence on development rate (~21%) than on duration or cumulative severity. Human regulation altered the seasonal timing of drought development by increasing the frequency of winter-related onsets and peaks while suppressing warm-season development. Overall, human influences were more pronounced for drought intensification rate than for duration or severity, demonstrating that the proposed framework provides a transferable approach for identifying and attributing human influences on hydrological drought development under non-stationary conditions.

Precipitation has obvious spatial and temporal variability, making it difficult to accurately capture precipitation patterns using only a single instrument. To address this issue, we proposed a Time-varying Weighted (TVW) merging method that considers error variations, aiming to minimize the difference between the merged product and the station measurements for optimizing weights in every day, and then through weight interpolation to achieve time-variant fusion of multi-source precipitation products in ungauged areas. The TVW method was applied in mainland China between 2010 and 2015 using daily observations from 2278 stations, with 70% of the data used to determine the weights and 30% used for validation. Two merged products were generated by integrating precipitation data from the fifth-generation of the European Centre for Medium Range Weather Forecasts atmospheric reanalysis (ERA5), the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement Final Run (IMERG-F), and the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks- Climate Data Record (PERSIANN-CDR) or the Climate Prediction Center (CPC) global unified gauge-based analysis, and they called MEIP (Merged ERA5, IMERG-F, and PERSIANN-CDR) and MEIC (Merged ERA5, IMERG-F, and CPC), respectively. The results showed that both MEIP and MEIC performed significantly better than their parent datasets, with improvements in all the metrics considered, including correlation, relative bias, absolute difference, and detection probability (i.e., hit and false alarm rates). Notably, even when the reference station density was reduced to 10% of the total stations, the TVW method maintained a robust performance, with MEIC generally outperforming state-of-the-art precipitation products such as the Asian Precipitation-Highly-Resolved Observational Data Integration Towards Evaluation (APHRODITE), the China Meteorological Forcing Dataset (CMFD), and the Multi-Source Weighted-Ensemble Precipitation (MSWEP). These findings suggest that TVW is an effective and robust merging method for producing high-quality precipitation estimates and it will provide great benefits for hydrological and meteorological applications over the ungauged and large-scale areas.

Study region 531 ungauged basins in the contiguous United States (CONUS). Study focus: We propose XNLSTM, an interpretive framework combining regionalized Long Short-Term Memory (LSTM) modeling, 10-fold basin cross-validation, and explainable artificial intelligence (XAI) methods, including Shapley Additive Explanations (SHAP), the Morris method, and permutation feature importance (PFI), to examine how catchment attributes affect runoff prediction in ungauged basins. The framework jointly uses multiple XAI methods to reveal interpretational variability and identify catchment attributes with relatively stable importance across methods. It then trains models under different catchment attribute configurations derived from the XAI-based interpretation results and interprets fold-specific models to assess the stability of catchment attribute importance across basin splits and configurations. New hydrological insights for the region: SHAP, Morris, and PFI produce different rankings for the same LSTM model, but p_mean, aridity, and other catchment attributes remain consistently important. Across the three XAI rankings, climatic and topographic attributes are consistently more important than soil, geological, and vegetation attributes for overall runoff prediction. Using more catchment attributes, and using attributes with higher XAI-derived importance, improves cross-basin generalization. However, lower ranked attribute combinations show slight tendencies to reproduce low-flow and high-flow metrics closer to observations. Importance rankings are relatively stable for all-attribute models but change more clearly when models use attribute subsets. Attributes such as elev_mean, pet_mean, and soil_conductivity remain important across basin splits and configurations.

Attributing changes in streamflow processes is crucial for water resource management as well as for understanding and mitigation of flood and drought risks. However, most existing attribution methods lack a unified approach to handle various causal variables, making them unsuitable for comprehensive attribution assessments. Therefore, this study proposed a framework to quantitatively attribute the impacts of natural and anthropogenic climate change, land use and cover change (LUCC), and human water withdrawal on streamflow and its seasonality. The framework consists of three steps: (1) bias correction of Coupled Model Intercomparison Project Phase 6 (CMIP6) data and construction of a dualistic nature-society water cycle model; (2) simulation of streamflow processes and identification of streamflow seasonality under different climate forcing and LUCC scenarios; and (3) quantitative attribution of streamflow evolution characteristics. The Weihe River Basin (WRB) in China has been selected as a case study area for the proposed attribution framework. The quantitative analysis indicates that natural and anthropogenic climate change, LUCC, and human water withdrawal account for 20.8%, 27.9%, 4.6%, and 46.7% of the decreasing trend in streamflow volume and -42.4%, -28.1%, -5.1%, and 175.6% of the weakening trend in streamflow seasonality in the WRB, respectively. These results suggest that human water withdrawal reduces streamflow and weakens its seasonality, while the other three factors contribute to streamflow reduction but enhance its seasonality. Overall, this study effectively distinguishes the impacts of anthropogenic and natural climate change on streamflow processes, thus providing a deep understanding of the influences of human-induced hydro-climate change.

Recently, differentiable modeling techniques have emerged as a promising approach to bidirectionally integrating neural networks and hydrologic models, achieving performance levels close to deep learning models while preserving the ability to output physical states and fluxes. However, there remains a lack of systematic exploration into the performance and physical interpretability of hybrid models that use neural networks to replace the runoff generation and routing processes in regionalized modeling. This research developed 12 regionalized hybrid models based on a differentiable parameter learning (DPL) framework, utilizing the Hydrologiska Byr & aring;ns Vattenbalansavdelning (HBV) model as the foundational backbone. These hybrid models incorporate neural networks to replace the various physical processes within the runoff generation and routing modules. The publicly available CAMELS dataset is employed to evaluate the performance and interpretability of these hybrid models. The results show that while the median Nash-Sutcliffe efficiency (NSE) and Kling-Gupta efficiency (KGE) coefficients for all hybrid models are lower than those of the purely data-driven regionalized long short-term memory neural network (LSTM) model (median NSE: 0.742, median KGE: 0.762), the best- performing hybrid model (median NSE: 0.731, median KGE: 0.761) approaches the LSTM model and has better physical interpretability. Embedding neural networks does not inherently guarantee improved performance and may, in some cases, even result in reduced performance. The degree of performance enhancement is not significantly correlated with the number of embedded neural networks. Compared to replacing the runoff generation process, substituting the routing process with neural networks yields more substantial performance improvements and enables the learning of different routing patterns based on the catchment's static attributes. This study underscores the importance of reasonably balancing the location, complexity, and quantity of embedded neural networks to achieve a trade-off between model performance and interpretability in hybrid modeling. These insights contribute to advancing regionalized hybrid modeling development.

Under the global warming context, alpine plateau regions face escalating water and land resource (WLR) risks due to their unique ecological functions and accelerating climate change. This study focuses on the source regions of the Yangtze and Yellow Rivers, establishing a multidimensional WLR risk assessment framework integrating risk, exposure, vulnerability, and resilience. A random forest model was employed to identify driving factors of these risks, and future trajectories of risk evolution under projected scenarios were quantified. Key findings include: (1) During the historical period (2000–2020), medium–high risk areas expanded at 1.11% per five-year interval, with risk factors (extreme drought-flood frequency, freeze–thaw erosion sensitivity) being dominant drivers (contributions: 35.85). Resilience factors (vegetation resilience, water retention capacity) significantly regulated low-risk areas (contribution: 40.81%). (2) Under SSP scenarios, high-risk areas are projected to occupy 28.15% by 2060, concentrated in the southern Yangtze River source and western Yellow River source regions, where risk intensification (1.9–32.50%) substantially outweighs resilience enhancement (0.09–15.15%). (3) Multifactorial synergy analysis reveals habitat quality degradation (12.26% contribution to high-risk areas), livestock overloading (18.03% contribution to medium-risk areas), and extreme hydrological events (10.13% contribution to flood risks) as primary causes, while vegetation resilience reinforcement (14.65% contribution to low-risk mitigation) and WLR allocation optimization (7.89% contribution to medium–low risk reduction) emerge as critical pathways. The study proposes prioritized implementation of an integrated “grazing prohibition compensation − permafrost protection − ecological corridor restoration” strategy in risk hotspots, and establishing a resilience regulation system based on climate adaptation thresholds.

Plant-available groundwater water is well-documented. However, the impact of groundwater flux on root zone storage capacity (Sr), the maximum water storage within the subsurface root zone available for plant transpiration, remains poorly understood. In this study, we present a more conservative, lower-bound global estimate of Sr, incorporating groundwater for the first time into the deficit-based calculation of Sr using a novel conceptual method that accounts for groundwater contribution fraction. Our findings reveal widespread plant reliance on groundwater, with a mean use of at least 20 mm, equivalent to a water volume of 1700 km3. In western United States, this use can exceed 100 mm. Distinct spatial patterns in Sr emerge globally, with higher values in mountainous forests and lower values in boreal grasslands. Comparisons with observed rooting depths confirm that the deficit-based method, when incorporating groundwater, effectively predicts underground root traits. Biotic and abiotic factors critically influence Sr values, with irrigation and topographic convergence exacerbating this reduction. Groundwater-dependent ecosystems rely heavily on root-zone water storage, utilizing an average of 2151 km3 of water. Despite inherent uncertainties in input data, our study provides the first systematic evaluation of groundwater's role in shaping Sr estimates, offering key insights for water resource management and ecosystem sustainability.

Vegetation restoration is an important approach to improve ecosystems and address climate warming. However, there is significant debate regarding climate and hydrological impacts of large-scale vegetation restoration. This study proposes a framework for dynamically separating the cumulative climate and hydrological effects of vegetation restoration, offering a perspective on the seasonal and regional variations in these effects. The framework applies WRF-NoahMP land-atmosphere coupled model with various surface parameters to distinguish the seasonal climate and hydrological effects of the Grain for Green Program as well as the inconsistency in climate and hydrological effects between important grassland restoration and afforestation regions. The Middle Yellow River Basin, a region severely impacted by large-scale vegetation restoration, was selected to demonstrate the proposed approach. The results indicate that seasonal variations in albedo, fraction of vegetation cover, and leaf area index contribute to distinct seasonal patterns in the climate and hydrological effects of vegetation restoration in the study region, with maximum effects observed in summer. A significant shift in the cumulative climate and hydrological effects occurred around 2010 during 2001-2020. Compared to the afforestation region, the grassland restoration region showed significantly reduced land surface temperature and soil moisture, and enhanced evaporation and precipitation recycling (p < 0.05). Our study contributes to an efficient method for distinguishing the seasonal cumulative climate and hydrological effects of vegetation restoration, as well as the inconsistency in climate and hydrological effects resulting from afforestation and grassland restoration regions, providing insights to better implement vegetation restoration initiatives.