Multi-source precipitation merging (MSP) is widely used to improve the accuracy and spatial resolution of precipitation estimates for hydrological and water resources applications. However, most existing MSP methods assume fixed relationships between precipitation intensity and environmental variables, which limits their ability to represent nonlinear precipitation behavior and often results in degraded performance during heavy and extreme rainfall events. To overcome these limitations, this study proposes a three-step merging framework that integrates downscaling, precipitation event classification, and categorical merging. The framework first downscales original precipitation products to enhance spatial resolution, then classifies precipitation events according to occurrence and intensity, and finally models intensity-specific nonlinear relationships between precipitation and environmental variables using machine learning. Based on this approach, a high-resolution merged precipitation dataset, termed the Multi-Source Merging Precipitation dataset (MSMP), was developed at a 1 km spatial resolution and daily temporal resolution for the period 1981–2020. The framework was evaluated over the Pearl River Basin in South China by comparison with original precipitation products and conventional MSP methods. Results indicate that the MSMP consistently outperforms existing products in both statistical and categorical metrics, particularly for heavy and extreme precipitation. These findings demonstrate that incorporating precipitation intensity classification significantly enhances multisource precipitation merging and provides more reliable precipitation inputs for hydrological modeling and water resources management.

To mitigate flood disasters and reduce their risks, a lot of reservoirs, as a reservoir group, and some flood detention basins (FDBs) have been constructed. As the flood control targets of the reservoir group and the FDB are often the same or are interrelated, it is necessary to integrate the reservoir group and FDB operation instead of operating them independently. Hydropower benefit is one of the functions of the reservoir group besides its flood control. There should be a tradeoff between the flood risk and the hydropower benefit from the integrated flood control operation of a reservoir group and FDB. A framework is proposed to determine and understand the tradeoff for enhancing decision-making in flood control. It has been applied in a case study of the Jingjiang FDB and a mixed reservoir group (Three Gorges, Shuibuya, and Geheyan) in the Changjiang River Basin. It has been found that the integrated operation of Jingjiang FDB can enhance the exchange between extreme flood risk and hydropower benefit, while this exchange is weakened for the non-extreme flood conditions due to the operational inflexibility of Jingjiang FDB. The safety flow at the downstream protection target has a significant impact on the tradeoff under non-extreme flood conditions but only a slight impact under extreme flood scenarios. The damage coefficient exerts a significant and consistent influence on the tradeoff across both extreme and non-extreme flood conditions. It is critical to determine the damage coefficient and the safety flow for improving the performance of the flood control operation. The proposed framework not only determines the tradeoff between flood risk and hydropower benefit but also helps facilitate the integrated flood control operation of a reservoir group and FDB.

Flood retention basin (FRB) is an important measure for the flood control. As the available flood storage capacity of FRB reflects the maximum amount of flood that can be stored, it is often varied due to the changing environment, which impacts on the its flood control function. Based on the principle of allowing only specific areas to be inundated for flood control, a systematic framework is proposed to quantify the variations of the available flood storage capacity of an FRB due to the combined effects of human activities and climate change. The Variable Infiltration Capacity hydrological model is a tool of simulation of flood flow in the framework. A Geodetector model is applied to figure out the contributions of human activities and climate change. Applying the framework to 42 FRBs in the middle reaches of the Changjiang River Basin, the results show a significant 9.6% decrease in total available flood storage capacity from 2000 to 2020, with human activities contributing more than climate change. The decrease of available flood storage capacities in the FRBs bring the increases of frequencies (i.e., the decrease of the standard of flood control) of the flood inflows that represents the maximum preventing flood for FRBs from 0.111%, 0.167%, 0.100% to 0.143%, 0.250% and 0.167%, respectively. Our study will not only help assess variations of the available flood storage capacity, but also contribute to the protection of the FRBs.

Under the dual influence of climate change and human activities，the rainfall-runoff relationship in river basins exhibits non-stationary characteristics，and hydrological models based on the traditional steady-state assumption struggle to capture the structural transitions of the system，leading to a decline in runoff simulation performance. In this paper，hydrological states are used as the representation of non-stationarity，and Hidden Markov Models (HMMs) with 1 to 3 states are constructed，the Viterbi algorithm is adopted to identify the hydrological states of the river basin and their evolution paths，and the decoded hydrological state results are introduced into the runoff simulation process to address the impact of non-stationarity. Experiments are conducted based on 240 typical river basins in the CAMELS database，and the results show that: approximately 22.5% of the river basins have significant multi-state hydrological characteristics; the multi-state model with hydrological state constraints shows significant advantages in both probabilistic simulation and deterministic simulation，with the uncertainty interval coverage rate increased by an average of about 30% compared with the single-state model; the Nash-Sutcliffe efficiency coefficient increases from 0.45 to 0.83 (an increase of about 84%)，and the root mean square error decreases by an average of about 49%. Research shows that: the HMM-based hydrological state identification method can effectively capture hydrological state transitions，and incorporating state information into the simulation process helps improve runoff simulation performance under non-stationary conditions，providing a new path for hydrological simulation in a changing environment.

The El Nino-Southern Oscillation (ENSO) significantly influences runoff variability across river basins in China. However, most ENSO-runoff studies rely on the NINO3.4 index, which introduces subjectivity and fails to capture the different ENSO flavors. To address this limitation, in this study, Pacific Sea Surface Temperature anomalies are used as input to the Nonhomogeneous Hidden Markov Model (NHMM), which objectively classifies five distinct ENSO-like states, i.e., strong El Nino-like, weak El Nino-like, neutral-like, weak La Nina-like, and strong La Nina-like states. These hidden states are then linked to observed runoff anomalies across China to examine temporal lags, spatial responses, and predictability. Our analysis reveals distinct temporal lags in ENSO impacts across basins, with the most pronounced lag depending on basin scale and location. Under strong La Nina-like conditions, runoff anomalies in China become significantly negative with a 9-month lag, whereas strong El Nino-like events trigger a 7.3-fold increase in standardized runoff anomalies in China relative to the neutral-like state mean, with a 3-month lag. Runoff predictability is then evaluated based on these five ENSO-like states and shows statistically significant skill up to 12 months, surpassing the NINO3.4 index across 63.8 % of China at the 12-month lead. we explore potential teleconnection mechanisms, which show that strong El Nino-like events enhance water vapor transport across most of China, whereas strong La Nina-like events concentrate water vapor in the Southwest Rivers Basin and increase drought risks elsewhere.

Nonstationary flood frequency analysis (NFFA) is essential for accurately quantifying flood risks under changing environmental conditions. The core of nonstationary flood frequency analysis lies in accurately estimating the nonstationary distribution and understanding how changing environments influence flood behavior. This study proposes an eXplainable Neural Network framework for Nonstationary Flood Frequency Analysis (XNN-NFFA), which integrates feedforward neural networks (FNNs) with SHapley Additive exPlanations (SHAP). The FNN is optimized through the Generalized Extreme Value distribution likelihood, allowing it to construct the appropriate nonstationary flood frequency distribution (NFFD), while SHAP provides interpretations of nonstationary contributions. The framework is applied to four representative hydrological stations in the upper Yangtze River Basin (1951-2019). Results show that warm-season precipitation, temperature, and reservoir operation are consistently the dominant drivers of flood nonstationarity, while large-scale atmospheric circulation indices exert more complex and station-specific influences. Compared with the Generalized Additive Models for Location, Scale, and Shape (GAMLSS) models, the FNN-based NFFDs show strong performance, reducing estimation deviance by over 10% in training and with improvements of about 3% in validation, demonstrating both robustness and generalization capability. Through SHAP analysis, the framework not only quantifies the relative importance of each covariate at different stations but also characterizes the forms of their effects. In particular, large-scale circulation indices are found to influence flood distributions through three distinct patterns: consistent monotonic effects across their ranges, non-monotonic responses where impacts change direction at different index levels, and effects that become pronounced only under extreme index values. Finally, the FNNbased estimates of nonstationary design floods are generally lower than those derived from stationary assumptions, and are more reasonable than those from GAMLSS models. By offering an explainable and comprehensive approach to NFFA, this study provides methodological support and practical implications for flood risk management under changing environmental conditions.

Reservoir surface water area (SWA) serves as a pivotal indicator for characterizing storage changes and water balance, with important implications for assessing the effects of anthropogenic regulation and protecting aquatic ecosystems. Although satellite observations offer indispensable data support for reservoir SWA monitoring, the trade-off between spatial and temporal resolutions in single-source satellite datasets severely impedes the comprehensive quantification of rapid SWA dynamics. To address this challenge, we develop a systematic deep learning framework for daily SWA inference that synergistically combines cloud-free optical image reconstruction and surface water extraction. The reconstruction module generates daily cloud-free Landsat-like optical images by fusing heterogeneous data from Sentinel-1, MODIS, and Landsat-derived NDVI via a multi-branch feature fusion mechanism. Using the reconstructed imagery as input, the surface water extraction module accurately delineates reservoir SWA. Comparative experiments demonstrate that the proposed framework achieves robust performance in cloud-free image reconstruction (RMSE = 0.021, SSIM = 0.737) and high accuracy in water extraction, with both user's and producer's accuracies exceeding 0.970. The framework is applied to 288 large reservoirs across the Yangtze River Basin, and its scalability for large-scale inference is further validated using in situ water elevation measurements from 77 reservoirs. Overall, this framework provides a robust and scalable solution for daily large-scale reservoir SWA inference, laying a solid foundation for hydrological analysis, optimal reservoir operation, and sustainable water resource management.

Flood frequency analysis (FFA) is critical for hydraulic design and flood risk management, yet the traditional stationarity assumption is increasingly challenged by climate change and intensive human activities. In large regulated basins, upstream reservoir operations substantially reshape downstream flood regimes. The controlled discharges propagate along the river network and interact with tributary inflows, thereby heightening the complexity of flood dynamics in downstream reaches. Most existing nonstationary flood frequency analysis (NFFA) methods are limited to single-site modeling and rarely provide a unified framework that simultaneously accounts for river network dependence and reservoir flood routing processes. To address this gap, this study proposes a spatial recursive framework for NFFA (SRF-NFFA) that explicitly incorporates river network dependence and reservoir flood routing within a probabilistic structure. The framework integrates GAMLSSbased marginal and conditional models, copula-based dependence structures, and physically based reservoir routing functions, recursively deriving nonstationary flood peak distributions from upstream source nodes to downstream recursive nodes. Nonstationary design floods are further estimated using the equivalent reliability method based on bias-corrected CMIP6 multi-model ensemble precipitation projections. The framework is applied to seven nodes in the Hanjiang River Basin (1964-2020), including three mega reservoirs and four gauging stations. Annual precipitation is introduced as a climatic covariate to characterize the impacts of climate change on flood peaks. Results show that, except for two upstream stations, flood peaks at all other regulated nodes exhibit significant nonstationary behavior. The application of SRF-NFFA in the Hanjiang River Basin confirms its effectiveness in flood frequency estimation across all nodes. In addition, nonstationary design floods are substantially reduced relative to stationary estimates, especially in downstream reaches. Compared with single-site covariate-based NFFA (SS-C-NFFA), although SS-C-NFFA achieves slightly lower fitting errors in conventional goodness-of-fit metrics, SRF-NFFA provides a more physically consistent representation of spatial flood dependence and reduces uncertainty in flood quantile estimates by leveraging reservoir routing processes and river network dependence for regional information sharing. Overall, SRF-NFFA provides a reliable analytical tool for basin-scale flood risk management in regulated river systems.

The rapid development of altimeter satellite has opened new pathways for monitoring inland water bodies and offers promising prospects for discharge estimation in ungauged basins. Previous studies have used altimetry data from a single virtual station (VS) to replace observed discharge data for calibrating lumped hydrological models, enabling discharge estimation in ungauged basins. However, the large spacing of satellite orbits often results in a spatial mismatch between VS locations and the natural outlet of the watershed. In this study, we apply a DEM-based Distributed Rainfall-runoff Model (DDRM) to couple with Sentinel-3 data for distributed discharge estimation in an ungauged basin and compares it with a lumped model, the Genie Rural a 4 parametres Journalier (GR4J) model. Leveraging the distributed modeling capabilities of the DDRM, we achieved spatial matching between four VSs and the DDRM. This allowed us to successfully integrate altimetry data from multiple VSs for distributed discharge estimation in ungauged basins. In addition, this study introduces ERA5-land soil moisture data for multi-objective calibration to explore its impact on discharge estimation. This study demonstrates that the discharge estimation of the GR4J model coupled with altimetry data is unstable, and due to its structural limitations, it cannot incorporate additional effective altimetry data, nor can it achieve distributed discharge estimation across the basin. While a distributed hydrological model can effectively achieve spatial matching and coupling with multiple VSs. Moreover, the model performs best when more altimetry data are incorporated alongside soil moisture data for multi-objective calibration. Under this configuration, the mean NSE across 17 hydrological stations in the basin reaches 0.73 during the calibration period and 0.62 during the validation period.

Streamflow seasonality, characterized by its unequal intra-annual distribution, is predominantly shaped by precipitation dynamics in non-snowmelt-dominated regions. However, climate change and human activities can alter the hydrological cycle, weaken the consistency in precipitation-streamflow seasonality, and further modify streamflow seasonality. Despite its importance, the mechanisms modulating precipitation-streamflow seasonality consistency remain underexplored. This study employs time-varying copulas and partial correlation analysis to identify and quantify the evolving seasonality inconsistency between precipitation and streamflow under the influence of multiple anthropogenic and climatic factors. In the Yangtze River Basin (YRB), recent environmental changes, including climate warming, vegetation greening, and reservoir regulation, have increasingly influenced the hydrological cycle. In this study, a slight intensification of precipitation seasonality is observed in half of the YRB watersheds, whereas streamflow seasonality is moderated in 90 % of the watersheds. The climatic and human-induced changes consistently weaken the dependency between precipitation and streamflow across seasons. Comparing the dependency strength between the periods 2006-2015 and 1982-1991, a decline of 17 % was observed in spring, 1 % in summer, 42 % in autumn, and 75 % in winter. Partial correlation analysis highlights three key factors driving the observed decrease in summer streamflow and increase in winter streamflow, with reservoir regulation exerting the most pronounced influence in winter. This study sheds new light on investigating streamflow seasonality in a changing environment.

Frequency analysis is crucial in low flow statistics, helping estimate the probability of water availability during low flow seasons and droughts. Low flow frequency analysis typically assumes stationarity, which has been challenged by climate change and variability. Therefore, non-stationary frequency analysis is essential when trends and non-stationarity exist in low streamflow data. This study developed a methodology that includes trend, change point, non-stationarity detection, and stationary and non-stationary low flow frequency analysis for annual minimum streamflow series of 7-day (Q7), 14-day (Q14), 30-day (Q30) and 90-day (Q90) periods, applied to selected river basins in Victoria, Australia. Significant decreasing trends were detected in several basins, with the strongest trends observed in the East Gippsland Basin, where the trend slopes were − 1.02, − 0.989, − 1.035 and − 1.534 for Q7, Q14, Q30, and Q90, respectively. Similarly, significant change points were found with year 2002 being the most common change point year, followed by year 1996, 2000 and 2001. Non-stationary frequency analysis proved superior in capturing the changing characteristics of low flow series. Moreover, the non-stationary models that included physical covariates outperformed those with only time covariates, highlighting the benefit of using covariates related to the physical mechanisms of low flow events. This study emphasizes the importance of non-stationary frequency analysis to prevent misleading conclusions in low flow-based water management, thereby enhancing the reliability and effectiveness of water management strategies.

Lakes play a crucial role in shaping both local and regional climates through heat exchange with the atmosphere. Amid global climate change, these interactions have undergone significant shifts. However, our understanding of the global heat release from lakes to the atmosphere, and its future trajectory, remains limited. In this study, we investigate changes in global lake heat release patterns and identify an amplified increase in heat release, particularly in mid to high latitudes (>45°N). This amplification is linked with a feedback mechanism, where the reduction in lake ice cover not only reduces the insulating effect between the warmer lake water and the colder atmosphere but also leads to increased heat absorption by lakes. As a result, lakes in mid-high latitudes experience a greater relative increase in heat release, primarily through upward thermal radiation, compared to lakes at lower latitudes with comparable surface water temperature increases. Additionally, seasonal variations in latent heat flux intensify the heat release during warmer seasons compared to colder ones. Future projections suggest substantially greater heat release compared to historical trends.

Water use simulation plays a pivotal role in water resource management globally. Simulating water use at regional raster scale enables better alignment with available water resources, facilitating efficient allocation. However, there remains a deficiency in methods of spatiotemporal scale selection for ensuring the simulation accuracy while also guaranteeing the information density at each raster scale. A novel framework has been proposed to select the appropriate spatiotemporal scales for water use simulation. The framework utilizes an iterative input variables selection (IIS) algorithm to identify optimal input variables for water use simulation and an end-to-end deep learning-based spatiotemporal scale adaptive selection (SSAS) model to determine the appropriate spatiotemporal scales. Due to China's substantial population, water demand, and the growing challenges of global warming, the country is particularly susceptible to water scarcity. The proposed framework was applied to select the appropriate spatiotemporal scales for simulating irrigation, domestic, and industrial water use across 341 prefectures in China. The results indicate that the appropriate spatial scales for irrigation water use simulation range from 1 km to 5 km in most places, while they vary from 1 km to 4 km for domestic and industrial water use simulation. Furthermore, the appropriate temporal scale generally spans from 10 days to 45 days for all three types of water use simulation. It is interesting to find that the simulation accuracy is significantly impacted by the selection of appropriate temporal scales through the parameter sensitivity analysis. Our proposed framework supports water resource management and facilitates efficient water resource allocation to mitigate water scarcity.

Drought, a recurring disturbance in water cycle, significantly impacts terrestrial ecosystems. Understanding the response and recovery times of ecosystems to drought is imperative for assessing drought impacts and unraveling ecological dynamics. However, the dominant factors influencing response and recovery times of vegetation productivity to drought remain unclear. This study first evaluated the response and recovery times of vegetation productivity to soil drought using gross primary production (GPP) from an event-based perspective in China. Random forest models were used to identify the dominant factors of response and recovery times, and the relationship between response and recovery times was also investigated. Results showed that the mean GPP response time was 3.29 months, increasing from humid southeast to arid northwest China during 1989-2019. The mean recovery time was 3.46 months, longer in humid eastern regions compared to arid western ones. GPP responded most promptly in spring drought and slowly in autumn and winter droughts, whereas recovery was quick in summer and autumn droughts, and delayed in spring and winter droughts. Moreover, a notable negative correlation existed between response and recovery times with a long response time associating with a fast recovery rate and a short response time indicating a long recovery time. Drought duration was the primary factor influencing response time, while soil moisture during the recovery period was the critical for recovery time. These findings suggest resilience strategies to enhance ecosystem adaptability by strengthening adaptive capacities in regions with prolonged recovery time and safeguarding regions with long response time and rapid recovery capacity.

The river mega cascade reservoirs play a pivotal role in water resource management, fulfilling multiple objectives such as ecological protection, power generation, navigation, and flood control. Efficiently managing the drawdown water levels across these reservoirs is crucial for maximizing water utilization and optimizing basinwide benefits. The study presents an innovative approach for multi-objective reservoir operations that integrates ecological, carbon reduction, and hydropower generation goals. A comprehensive optimization model was developed, targeting the minimization of the adjusted annual proportional flow deviation, suppression of carbon emissions, and maximization of power generation. The evolutionary optimization algorithm was utilized to address this model, producing a collection of Pareto solutions. The entropy-based decision-making analysis was then utilized to select the compromised solution. The results demonstrate that the compromised solution outperforms the practical scheme, achieving reductions in amended annual proportion flow deviation by 21.3 %- 31.7 %, lowering carbon emissions by 4.6 %-5.1 %, and boosting power generation by 3.8 %-8.1 %. This solution not only enhances ecological benefits and reduces carbon emissions but also improves energy produc-tion. This study offers an integrated perspective on reservoir drawdown optimization, providing valuable insights for decision-makers to create effective operational strategies and fostering sustainable hydropower development.

Large‐scale inter‐basin water transfer projects have been implemented to mitigate the uneven spatial distribution of water resources at regional and national scales. Assessing the effectiveness and impacts of these projects remains a substantial challenge. The Middle Route South‐to‐North Water Diversion Project (MR‐SNWDP) in China has reversed the declining trend of groundwater storage (GWS) in parts of the water‐receiving area. However, the effects of the project on surface water quantity and quality, particularly in lakes and reservoirs, remain unclear. Here, a novel framework was developed to quantify the impact of the MR‐SNWDP on surface water by integrating altimetry, gravity, and optical remote sensing satellite data. Following the implementation of the MR‐SNWDP in December 2014, surface water volume increased significantly. In the project‐affected zone, the increase averaged over 0.9 km 3 per year, equivalent to approximately 13.5% of the annual water diversion and 19 times that in the unaffected zone. Surface water transparency, a proxy for surface water quality, also improved in most lakes, with a significant positive correlation between transparency and volume increase in the affected zone. Water balance analysis indicated that the declining trend of GWS reversed in 2020, five years after the recovery of surface water. Our findings suggest that surface water recovered more rapidly and effectively than groundwater in the water‐receiving area of the MR‐SNWDP, highlighting the positive effects of transboundary water diversion projects on the quantity and quality of surface water.

Study region: Xiangjiang and Hanjiang River basins in the Yangtze River basin, a humid subtropical inland region of central-southern China. Study focus: The general precipitation predictive skills of the mainstay numerical models are still rather limited beyond 10 days, further deteriorating the performance of sub-monthly streamflow prediction. This study proposes a sub-monthly streamflow prediction framework for organically combining stochastic weather generator (SWG) and monthly precipitation prediction to generate ensemble sub-monthly precipitation for streamflow prediction. The SWG-based schemes are then compared with the common numerical hydrometeorology ensemble streamflow prediction over two river basins in China. New hydrological insights for the region: Results show that the numerical streamflow predictions exhibit a less accurate deterministic performance than the SWG-based framework with the climatology scheme over sub-monthly horizon, with leadtime-averaged mean absolute relative error dropping from 19.9 % to 8.3 %, and 21.8-11.1 % for Xiangjiang and Hanjiang river basins, respectively. In addition, the more restrictive parametric adjustment procedure with adjusted precipitation amounts can bring added values and further improve the accuracy of SWG-based daily streamflow prediction for sub-monthly leadtimes. In terms of the probabilistic prediction performance, the SWG-based methods yield approximately equivalent results with the numerical hydrometeorology streamflow prediction, with leadtime-averaged Continuous Ranked Probability Skill Score of the scheme with modified parameters of precipitation amounts being 0.51 and 0.56 for Xiangjiang and Hanjiang River basins, respectively. Furthermore, the SWG-based schemes are more suitable for predicting high-flow events during the flood season, with a significantly smaller Brier Score. However, there are little improvements in probabilistic prediction performance when transition probabilities of precipitation occurrence are progressively modified.

Water-use efficiency (WUE) and carbon-use efficiency (CUE) are critical indicators of ecosystem function and hydrologic processes, reflecting the water-carbon flux exchange rate. Climatic variables, land use and land cover change (LUCC) and water diversion project (WDP) have altered water-carbon cycle; however, their roles in modulating WUE and CUE remain uncertain. To explore these effects, a framework is proposed and Han River basin (HRB) in China is selected as a case study including the data sets from both remote sensing and in situ observations during 2000-2020. The process-based Regional Hydro-Ecological Simulation System model and a supervised machine learning model are applied to simulate the impacts of climatic variables, LUCC and WDP on WUE and CUE, which are conducted by designing four experiments. We find that no significant WUE and CUE trends attributed to contrasting trends in the dry (October to March) and wet (April to September) seasons. Temperature variations greatly affect WUE and CUE, with WUE decreasing in the wet season and increasing in the dry season due to minimum temperature changes. LUCC has litter impacts on WUE and CUE changes. From 2014 to 2020, the middle route of the South-to-North WDP decreased WUE by 0.22 gCkg-1H2O in the middle-low HRB's wet season, slightly affecting CUE. Seasonal CUE was stable, with the largest decrease of 0.04 in the upper HRB during the wet season. The WDP also increased WUE sensitivities to minimum and maximum temperatures, while CUE sensitivities remained constant. Our case study has proven that the proposed framework is an effective way to understand the roles of climate change and WDP in modulating WUE and CUE.

The water level-area-storage volume (Z-A-V) relationship serves as the cornerstone of reservoir operations, governing water allocation, flood mitigation, and power generation. Sedimentation-induced capacity alterations can progressively degrade Z-A-V accuracy, yet systematic curve updates remain inadequately implemented across developing nations. Traditional reconstruction approaches face inherent limitations due to resource-intensive requirements including costly field surveys and data scarcity. Emerging satellite remote sensing technologies show transformative potential for dynamic reservoir monitoring, though their application in Z-A-V curve updating still requires substantive exploration. This study evaluates the capability of multi-source satellite imagery, including optical data from Landsat 8 and Sentinel-2, and synthetic aperture radar (SAR) data from Sentinel-1, for accurately reconstructing the Z-A-V curve. To extract high-accuracy reservoir surface extents, two advanced algorithms–Random Forest Classification (RFC) and Otsu thresholding–are applied to the optimal and SAR imagery, respectively, to delineate water and non-water pixels. The suitability of the satellite-derived Z-A-V curve is further assessed by estimating the storage capacity loss due to sedimentation accumulation and comparing these estimates with that derived from design curve. Using the Hongjiadu Reservoir in the upper reach of the Wujiang River, China, as a case study, the results show that: (1) all three satellite datasets accurately extract the reservoir surface areas, achieving average accuracies of 96 Sentinel-1 SAR achieves 96

Developing accurate large-scale drought prediction models is challenging due to the complex temporal and spatial correlation patterns that govern drought dynamics, as well as the compounding effects of anthropogenic activities and global climate change. Although recent advances in deep learning have yielded effective drought prediction models, many struggle to fully capture the heterogeneous spatial linkages over large-scale regions. In this study, we proposed a novel large-scale drought prediction framework that considers spatial heterogeneity and leverages a causal network connecting regions delineated by drought centroids of severe agricultural events identified through dynamic drought analysis. Using the predefined causal network, we employed the state-of-theart deep learning algorithm, the Spatio-Temporal Graph Convolutional Networks (STGCN) model, with a recursive multi-step forecasting strategy to predict root-zone soil moisture (RZSM) -based drought indices (DIs) up to four weeks in advance for the Yangtze River Basin (YRB). The results show that compared to the meteorological drought events, the corresponding agricultural drought has a later onset and smaller affected areas, yet greater intensity. The proposed model demonstrated robust predictive performance in drought predictions with an average root mean square error (RMSE) of 0.45 and an R2 value of 0.66 across the YRB for the spatial weekly agricultural DIs on the test dataset. Applying the STGCN with the recursive multi-step forecasting strategy can significantly improve the prediction performance, improving R2 values by 0.15 and reducing RMSE by 0.1 on average, with the most substantial improvements observed during the first three weeks (R2 increases of 0.32, 0.24 and 0.09, respectively). These findings underscore the importance of incorporating spatial correlations and demonstrate the advantages of the STGCN approach for large-scale agricultural drought prediction and inform water resource management at large-scale watersheds.