Reliable aquifer recharge prediction is essential for climate-resilient and sustainable groundwater management, yet uncertainty persists due to subsurface heterogeneity and the lack of direct basin-scale recharge measurements. We present a serial hybrid eXplainable Artificial Intelligence (XAI) framework that leverages hydrological model-derived recharge estimates to train AI models, improving prediction accuracy, transparency, and interpretability. The framework was applied to two basins within the karstic Edwards aquifer system in Texas, USA. The XAI models identified recharge events in the test dataset that were missed by the hydrological model, with findings corroborated by in-situ hydroclimatic records, HSPF recharge estimates, GRACE-derived groundwater storage anomalies, and bootstrap analyses. The results demonstrated the XAI model's superior learning capability beyond emulators to identify limitations in the training model and test data while robustly predicting high and low aquifer recharge events. Using long-term (1946-2023) hydroclimatic records and SHapley Additive exPlanations (SHAP), the best-performing AI model (Extremely Randomized Trees) identified basin-specific recharge drivers: current-month precipitation dominated in the larger, warmer, and drier basin with perennial streams, while lagged recharge, a proxy for antecedent soil moisture, was the primary driver in the smaller urbanizing basin characterized by small ephemeral streams and highly fractured zones. E ach driver explained similar to 32% of the variability in recharge estimates, underscoring the model's generalizability. SHAP-based analysis further enabled probabilistic identification of hydroclimatic conditions conducive to enhanced recharge. Projections based on downscaled CMIP6 climate data under intermediate- and high-emission scenarios indicate a decline in large recharge events in both basins through 2100, highlighting potential risks to groundwater sustainability. (c) 2026 China University of Geosciences (Beijing) and Peking University. Published by Elsevier B.V. on behalf of China University of Geosciences (Beijing). Thi s is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

The chronic water crisis in Iran stems from decades of water-intensive development, fragmented governance and national priorities that sidelined environmental protection. Without major governance reform and reprioritization, technical solutions alone cannot stop worsening water and environmental degradation.

Study Region Upper Red River Basin, Texas-Oklahoma, U.S. Study Focus This paper presents a topology-aware transfer learning (TL) approach using machine learning models that leverage stream network connectivity and publicly available stream salinity data with varying temporal frequencies within a basin to generate robust daily stream salinity predictions. To ensure robust predictions from models with stable parameters and performance across potential uncertainties, the lower upper bound estimation (LUBE) method was used to quantify the uncertainty in the generated stream salinity data. A rich original record (∼2 years) of continuous sub-daily in-situ salinity measurements was used to verify estimated prediction intervals. New hydrological insights for the region Stream salinity data availability varies across time and space within various tributaries of the Upper Red River Basin, USA, complicating water supply management. To address this challenge, this study generated robust stream salinity data for all monitoring sites within the basin. The topology-aware TL framework helped improve prediction accuracy by 6% lower RMSE for best fitted model and an average of 0.25 higher NSE for all 600 models fitted for a site. This improved accuracy of TL models is more pronounced at downstream sites compared to local models, while enhancing robustness and decreasing the uncertainty of predictions for all sites. Additionally, TL enabled fitting an accurate model with NSE higher than 0.8 for a site with only 26 pairs of observation, which was infeasible via local models. Comparison with the recent sub-daily in-situ data indicated the framework’s spatiotemporal generalizability, and confirmed the reliability of both the PIs and the generated synthetic salinity data.

Abstract The Caspian Sea, the Earth's largest inland water body, faces water level decline, drawing comparisons to the collapse of the Aral Sea. Unlike the Aral Sea, the relative roles of climatic variability, hydrological changes, and anthropogenic pressures on the Caspian Sea remain poorly understood. Here, we integrate satellite observations, in situ hydrological records and reanalysis data to examine recent drivers of the Caspian water loss. We show that total river inflow to the Caspian Sea has declined significantly, primarily due to reduced discharge from the Volga River. At the same time, precipitation over the basin has remained broadly stable, while evaporation over the sea has shown a modest upward trend. These findings point to compound anthropogenic and climatic influences on the regional water balance. We also detect a long‐term increase in chlorophyll‐a concentrations in the shallow Northern Caspian, signaling growing ecological stress associated with ongoing hydrological change. Avoiding further ecological disruption requires coordinated international action and policies to mitigate shrinkage by optimizing water allocation and environmental releases, as well as prioritizing long‐term ecosystem resilience. Without urgent intervention, the Caspian Sea risks following the trajectory of other desiccating inland water bodies, with long‐lasting ecological and socioeconomic consequences.

The Gravity Recovery and Climate Experiment (GRACE) and its Follow-On mission (G-FO) are promising tools as satellite-derived monitoring of terrestrial water storage (TWS) and groundwater storage (GWS) changes globally. However, the coarse spatial resolution of these datasets (ranging from 0.25° to 1°) limits their utility decision-relevant characterization of water resource changes. Herein, we used geospatial modeling approaches including machine learning (ML) (random forest (RF) and hybrid geographically weighted random forest (hybrid RFgw)) and spatial regression (geographically weighted regression (GWR)) in geospatial downscaling of three different GRACE datasets (JPL-SH, JPL-M, and CSR-M), to generate 1 km2 downscaled-GRACE data for purpose of quantifying water resources changes. Specifically, we quantified TWS anomalies (TWSA) and GWS anomalies (GWSA) across natural and managed landscapes in different catchments of the Indus Basin from 2002 to 2020. We evaluated the effectiveness of downscaled results against in-situ GWSA over 1238 monitoring wells. Results indicate that: (1) the hybrid RFgw model outperformed GWR and global RF models in predicting high-resolution TWSA and GWSA; (2) downscaled-GWS anomalies from the hybrid RFgw model agreed well with in-situ observational data, with higher R2 values for the downscaled JPL-SH (0.60 - 0.86), JPL-M (0.54 - 0.75), and CSR-M (0.54 - 0.85); (3) the highest TWS (−133.49 mm/year) and GWS (−150.38 mm/year) decline rates were detected in catchments dominated by managed landscapes; and (4) croplands experienced the largest storage losses (62-85 km3) followed by urban areas (8-15 km3) despite their much smaller spatial extent. The paper demonstrates the effectiveness of 1-km2 downscaled GRACE/G-FO datasets to provide localized information for water resource management across landscapes facing data-scarcity and water-food security challenges.

Accurate simulation of irrigation is key for effective water resources planning and management across various scales. This paper presents SWAT-IRR, a new irrigation algorithm for the Soil and Water Assessment Tool (SWAT) model, designed to enhance the simulation of how different irrigation systems and schedules influence hydrologic fluxes in irrigated agricultural areas. SWAT-IRR enables explicit simulation of three irrigation systems (surface, sprinkler, and drip) parameterized using irrigation application efficiency, conveyance efficiency, surface runoff ratio, and an additional area adjustment factor parameter for drip irrigation. SWAT-IRR offers three simulation options to accommodate varying user needs. Option 0 is the original SWAT model irrigation algorithm. Option 1 addresses the original SWAT model's limitation by improving control of irrigation simulation during growing seasons. Option 2 builds on Option 1 by integrating a new irrigation algorithm to advance the representation of irrigation processes and their associated impacts on hydrologic fluxes. Option 3 further strengthen Option 2 by adopting the U.S. Department of Agriculture (USDA) Natural Resources Conservation Service (NRCS) curve number approach to estimate irrigation surface runoff, as an alternative to the surface runoff ratio. An application of SWAT-IRR at the Fort Cobb Reservoir Experimental Watershed (FCREW) in central Oklahoma illustrates its effectiveness in enhancing irrigation simulation. Comparisons between the original SWAT model and SWAT-IRR, as well as among the various SWAT-IRR options, demonstrates its ability to improve control and representation of irrigation simulation and capture practical aspects of water allocation from various sources and water application using different irrigation systems.

High-resolution gridded data sets provide valuable opportunities to enhance drought forecasting, but applying complex machine learning algorithms across large spatial domains is computationally challenging. This study presents a novel hybrid approach for forecasting the gridded Standardized Precipitation-Evapotranspiration Index (SPEI) across the U.S. Southern Plains (SP), with lead times of 1 and 3 months. We developed a clustering-based method using 21 centroid grid cells, each representing a unique cluster of similar grid cells based on various hydrologic characteristics, to train and evaluate multilayer perceptrons (MLPs), long short-term memory (LSTM), and genetic programming (GP). Based on the superior performance of the trained MLPs in terms of Nash-Sutcliffe efficiency and root-mean-square error, they were extended to corresponding grid cells for each cluster, enabling spatially adaptive drought prediction at a high resolution. The use of discrete wavelet transform (DWT) further enhanced model accuracy by capturing key temporal patterns in the SPEI series. Notably, our results showed that physical and hydrologic attributes strongly influenced input selections. While a 12-month lag period worked well in regions with weaker seasonality, areas with strong seasonality benefited from selection of effective lags by using mutual information. For 3-month-ahead forecasts, including decomposed potential evapotranspiration in addition to precipitation as inputs improved accuracy in drier regions but decreased accuracy in humid areas. The forecast maps based on the hybrid DWT-MLP models effectively captured the spatial variability of drought, with high correlations to observed values, demonstrating their effectiveness for regional drought early warning systems to inform water resources management adaptations.

Study region: Saudi Arabia. Study focus: The major goal of this study is to downscale GRACE (Gravity Recovery and Climate Experiment) groundwater storage (GWS) anomalies to assess the local-scale vulnerabilities of groundwater changes across western regions of Saudi Arabia (Al Jumum, Makkah, Jeddah, and Bahrah). This was accomplished by using multi-model ensemble machine learning (ML) approach leveraging Random Forest, CART, and Gradient Tree Boosting algorithms within Google Earth Engine (GEE). Additionally, we used the downscaled GWS and CMIP6 climate data with the Generalized Additive Model (GAM) to project the future GWS changes under climate change. New hydrological insights for the region: The ensemble results demonstrated robust performance (R2 = 0.92 and RMSE = 20 mm) compared to the individual model (R2 = 0.84-0.88 and RMSE = 25-28 mm). The areas of higher groundwater depletion were predominantly observed in Jeddah and Makkah, with average annual rates of - 165 mm/year and - 150 mm/year, respectively, from 2002 to 2023. The total volumetric losses range from 11.38 km3 to 15.31 km3 across different sub-regions. Seasonally, the peak GWS drop (-90 to - 125 mm) was detected during the summer months (April-July), aligning with periods of maximum water demand. Several key drivers that control the GWS changes were also identified, including anthropogenic effects, local climate anomalies, and large-scale climate oscillations. Projections for GWS reveal an irreversible decline throughout the 21st Century with potential reductions surpassing - 216 mm/year in

Study region: Lower Arkansas River Basin (LARB) in Colorado, USA. Study focus: The process of implementing irrigation in large river basins often results in significant changes in hydrologic pathways and fluxes, such as canal seepage, runoff, recharge, pumping, and groundwater-river exchange. The objective of this study is to quantify the hydrologic fluxes in a highly irrigated river basin and investigate the controls on these fluxes, using the Lower Arkansas River Basin (LARB) (64,000 km2) in Colorado, USA as a demonstration case. We use the SWAT+ watershed model, enhanced with the new groundwater module gwflow, canal seepage, and irrigation application driven by daily canal diversions and groundwater pumping. The model is tested against streamflow and groundwater head, showing good performance along the Arkansas River and the alluvial corridor. New hydrological insights for the region: On average, precipitation in the basin is 380 mm/yr., of which 2 % (10 mm/yr.) becomes recharge and 2 % is irrigation (80 % surface water irrigation). Water yield is 18 mm/yr. (5 %), principally surface runoff and net groundwater discharge. Canal seepage is only 0.2 % of precipitation. Irrigation fluxes, canal and plant ET are highest in the downstream regions. Sensitivity analysis reveals the controlling watershed features on streamflow, groundwater head, and hydrologic fluxes for each region. Main parameters include streambed conductivity, plant uptake factors, snowmelt factors, aquifer properties, soil available water capacity, and soil percolation coefficient, with each parameter ranked by influence for each region within the basin. The calibrated models can be used to explore the impact of changes in climate, irrigation practices, and general water management schemes.

This study advances the DRASTIC groundwater vulnerability assessment framework by integrating a multi-hazard groundwater index (MHGI) to account for the dynamic impacts of diverse anthropogenic activities and natural factors on both groundwater quality and quantity. Incorporating factors such as population growth, agricultural practices, and groundwater extraction enhances the framework's ability to capture multi-dimensional, spatiotemporal changes in groundwater vulnerability. Additional improvements include refined weighting and rating scales for thematic layers based on available observational data, and the inclusion of distributed recharge. We demonstrate the practical utility of this dynamic DRASTIC-based framework through its application to the agro-urban regions of the Irrigated Indus Basin, a major groundwater-dependent agricultural area in South Asia. Results indicate that between 2005 and 2020, 54% of the study area became highly vulnerable to pollution. The MHGI revealed a 13% decline in potential groundwater storage and a 25% increase in groundwater-stressed zones, driven primarily by population growth and intensive agriculture. Groundwater vulnerability based on both groundwater quality and quantity dimensions showed a 19% decline in areas of low to very low vulnerability and a 6% reduction in medium vulnerability zones by 2020. Sensitivity analyses indicated that groundwater vulnerability in the region is most influenced by groundwater recharge (42%) and renewable groundwater stress (38%). Validation with in-situ data yielded area under the curve values of 0.71 for groundwater quality vulnerability and 0.63 for MHGI. The framework provides valuable insights to guide sustainable groundwater management, safeguarding both environmental integrity and human well-being.

Reliable information about the extent of irrigated areas is critical for water resources management in agricultural regions facing mounting water scarcity challenges due to climate variability and change, extreme events, and increased competition over limited water resources. To this end, we introduce a multi-model ensemble mapping (MEM) approach to develop high-fidelity, high-resolution (30 m) annual maps of irrigated areas from 2007 to 2022 in the Upper Red River Basin (URRB), U.S., using remote sensing, machine learning (ML), and ground truth data. Our approach combines the outputs of different ML classifiers, including Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Gradient Tree Boost, and Classification and Regression Trees (CART) in Google Earth Engine. ML classifiers were trained using different input variables including vegetation indices acquired from high-resolution (30 m) Landsat imagery, soil, topography, and climate data. Furthermore, we developed a rich ground truth dataset of 910 irrigated fields in 2022 to enhance the predictive performance of ML classifiers and assess model accuracy. While CART and SVM classifiers outperformed other models with higher ground truth accuracies of 71 % and 73 %, respectively, the MEM approach improved the ground truth accuracy to similar to 84 %. Results indicate a notable upstream expansion of irrigation in the URRB, particularly near tributaries, where new croplands were frequently irrigated even during droughts when downstream irrigation was halted due to diminished surface water availability. The combination of the expansion of upstream irrigated areas and consistency of irrigation has critical long-term implications for downstream agricultural water availability.

The convergence of climatic and anthropogenic factors that triggered the August 2022 mega-flood in Southern Pakistan caused 1486 fatalities and approximately $30 billion in economic damages. After a multi-year drought, the pre-monsoon rainfall was 111% higher than the long-term average of 1951–2021, increasing soil moisture by 30% in the Indus Basin floodplains. Monsoon rains were 547% above average, with record-breaking cumulative weekly rainfall in July (200 mm) on already saturated soils. Upstream drainage catchments (e.g., Chenab, Jhelum, and Ravi) received 33% and 41% more rain in pre-monsoon and monsoon periods, respectively. August 2022’s streamflow at Sukur Barrage, just upstream of the floodplains, was 170% larger than the historical average, due to the compounding effects of rain-on-snow and warmer temperatures accelerating snowmelt. The magnitude of multi-day consecutive rainfall events is projected to increase in Southern Pakistan by 2099 in a high-emission scenario (SSP5-8.5), making the catastrophic 2022 flood a forewarning of elevated future flood risks.

Droughts may exhibit spatiotemporal heterogeneity at a regional scale. Effective drought assessment and management necessitates identifying homogeneous areas. However, previous studies often simplified clustering analysis by focusing only on a single variable. In this study, we present a novel drought clustering map for the southern plains (SP)region of the United States by integrating the wavelet-entropy approach with k-means clustering algorithm to capture spatiotemporal patterns of drought-related variables across various resolutions while eliminating redundant information. We considered multiple drought indicators and indices including gridded precipitation (P), potential evapotranspiration (PET), normalized difference vegetation index (NDVI), and standardized precipitation evapotranspiration index (SPEI) as well as geographical coordinates and topography map. Through evaluating five different combinations of input datasets, we selected the one demonstrating optimal results based on the Davies-Bouldin and Calinski-Harabasz criteria. In addition to P, PET, and NDVI, including the coordinates and elevation as secondary variables significantly enhanced the clustering performance. Using these variables, the region was subdivided into 21 clusters. Pearson's correlation coefficients for the SPEI between centroid members and corresponding cells within clusters averaged between 0.84 and 0.94. Comparison with an existing cluster map [Drought Risk Atlas (DRA)] for the region revealed that our proposed cluster map showed higher variability between clusters for P, PET, and NDVI, confirming the robustness of the clustering results for drought conditions in the SP. The new clustering framework is expected to provide valuable insights for understanding and addressing drought dynamics in the SP region.

As global climate change poses a challenge to crop production, it is imperative to prioritize effective adaptation of agricultural systems based on a scientific understanding of likely impacts. In this study, we applied an integrated watershed modeling framework to examine the impacts of projected climate on runoff, soil moisture, and soil erosion under different management systems in Central Oklahoma. The proposed model uses measured climate data and three downscaled ensembles from the Coupled Model Intercomparison Project Phase 6 (CMIP6) at the water resources and erosion watershed to understand the impact of climate change and various climate conditions under three management systems: (1) continuous winter wheat (Triticum aestivum) under conventional tillage (WW-CT; baseline system), (2) continuous winter wheat under no-till (WW-NT), and (3) cool and warm season forage cover crop mixes under no-till (CC-NT). The study indicates that the occurrence of agricultural drought is projected to increase while erosion rates will remain unchanged under the WW-CT. In contrast, climate simulations imposed on the WW-NT and CC-NT systems significantly reduce runoff and sediment while preserving soil moisture levels. Especially, implementing the CC-NT system can bolster food security and foster sustainable farming practices in Central Oklahoma in the face of a changing climate.