Study region The Four Lakes Basin is a typical agricultural plain river network subject to intensive regulation by artificial sluices and pumping stations. Study focus This study proposes a hybrid streamflow prediction model, SWAT-GNN-LSTM, that integrates physical processes with data-driven methods. Using SWAT-simulated hydrological states as physically informed boundary conditions, the model jointly extracts spatial topology and time-lag effects to infer and reconstruct regulated streamflow indirectly. New hydrological insights for the region (1) The model effectively mitigates peak-flow misalignment and extreme errors caused by intensive cross-boundary pumping and internal regulation, with average NSE values of 0.77 in validation and 0.66 in independent testing across six regulated nodes. (2) Ablation experiments confirm that the spatiotemporally coupled architecture successfully reconstructs the spatial transmission and routing delays of operation-related flow responses. (3) Feature sensitivity analysis shows that the contributions of wind speed, evapotranspiration, and temperature unexpectedly exceed that of precipitation, indicating a substantial shift in the dominant driving factors of the unnatural hydrological cycle under intensive agricultural water diversion. (4) Confidence intervals generated using MC Dropout quantitatively capture plausible streamflow fluctuations under typical flood-control operation periods, thereby objectively characterizing operational margins for decision-making. This study provides a reliable approach for streamflow simulation in complex regulated basins that lack fine-resolution operational data.
Study region The Ganzi–Litang fault zone (GLFZ), the eastern margin of the Tibetan Plateau. Study focus Fault segmentation can govern the circulation and geochemical evolution of geothermal fluids, but the contrasting effects of strike-slip and pull-apart structures along plateau-margin fault zones remain insufficiently quantified. We compiled 58 water and isotope samples and 28 gas samples from the GLFZ and integrated Self-Organizing Map (SOM) classification with H–O–Sr–He–C isotopic tracing, reservoir-temperature estimation, and volatile-source apportionment to evaluate segment-specific controls on recharge, water–rock interaction, mixing, and degassing. New hydrological insights for the region Geothermal fluid evolution along the GLFZ reflects a dual-mode, segment-dependent geochemical pattern. The SOM analysis categorized the geothermal waters into three hydrochemical groups, revealing both segment-related geochemical patterns and a cross-segment shallow-circulation group. In the northern Ganzi strike-slip segment, the Group 2 geochemical signature is best explained by laterally coherent deep-circulation conduits in a granitic reservoir, yielding partial-to-full water–rock equilibrium and reservoir temperatures of 200–230 °C. In the southern Xinlong–Litang pull-apart segment, fluids (Group 1 and Group 3) are more strongly diluted by shallow cold water (57–81%) and record lower deep thermal end-member temperatures, of 140–170 °C, multi-lithological water–rock interaction, and spatially localized He–CO₂ anomalies. The geochemical patterns suggest that the northern strike-slip segment favors stable, long-residence deep circulation, whereas the southern pull-apart segment is characterized by stronger shallow mixing and localized deep-volatile anomalies. The results provide a regional framework for segment-specific geothermal development and the protection of alpine water resources in composite fault zones.
Study region The Touchih right-bank slope adjacent to the Liyutan Reservoir in central Taiwan is a structurally heterogeneous fractured rock hillslope in a humid subtropical setting. Historical instability and drainage interventions indicate spatially variable groundwater responses under rainfall forcing. Study focus This study investigates event-scale infiltration, moisture redistribution, and drainage using 510 quality-controlled hourly time-lapse electrical resistivity tomography (ERT) profiles acquired during multiple rainfall events. Temporally constrained inversion results were used to derive normalized relative water saturation (RWS) indicators and observation-based metrics of response lag, spatial variability, and saturation persistence. Geological information and available hourly groundwater-level observations supported hydrological interpretation, while normalized depth-of-investigation assessment constrained the principal interpretation depth. New hydrological insights for the region Rainfall-related responses differ markedly among structural domains of the slope. Fracture-influenced zones exhibit rapid wetting followed by relatively rapid drainage and short moisture persistence, whereas relatively intact or storage-dominated zones show delayed responses and prolonged moisture retention. The continuously constrained model region extends predominantly to approximately 27–30 m depth, within which spatially coherent ERT/RWS responses provide field evidence of structurally influenced hydrological heterogeneity. These findings show that saturation persistence, together with response timing, provides useful process-oriented information for monitoring fractured hillslopes under humid subtropical rainfall conditions.
Study region: Upper and middle reaches of the Yangtze River Basin (YRB), China.Study focus: This study reconstructed daily terrestrial water storage anomalies (TWSA) from 2002 to 2023 using GRACE/GRACE-FO data combined with precipitation and temperature, and estimated daily evapotranspiration (ETWB) based on the water balance method. ETWB was compared with the Penman–Monteith–Leuning (PML) ET and Catchment Land Surface Model (CLSM) ET to evaluate its performance at different time scales and to analyze the response characteristics of water balance and energy-driven ET estimation methods.New hydrological insights for the region: ETWB inherits strong pulse signals from precipitation, introducing high-frequency noise. Low-pass filtering (Gaussian and Butterworth) significantly improves ETWB consistency with PML ET at multi-day scales, confirming its reliability for seasonal water resource assessment. In contrast, the original daily ETWB can correspond to the fast response characteristics of CLSM ET, while low-pass filtering smooths out these physically significant high-frequency signals. At the same time, there is also a phase reversal between the two. These findings indicate that ET estimation based on water balance and energy-driven have different response characteristics. The high-frequency uncertainty of the water balance method on a daily scale is a major limiting factor for accurately estimating daily evapotranspiration.
Study region The cascade reservoirs in the lower Lancang River, southwestern China. Study focus Natural river ecosystems rely on specific thermal regimes, which are frequently disrupted by deep reservoirs releasing cold hypolimnetic water. This study investigated how interannual hydrological variability controls thermal stratification, and how selective withdrawal via stoplog gates influences downstream thermal habitats and hydropower generation. By coupling a three-dimensional hydrodynamic model with a streamwise temperature model, the thermal dynamics were simulated. A systematic assessment was then conducted to compare the thermal habitat improvement and the hydropower losses between bottom-outlet operation and selective withdrawal. New hydrological insights for the region Results indicated persistent vertical stratification in the reservoir forebay across typical hydrological years, with a stable hypolimnion temperature of 15 °C and a distinct stratified regime from May to August. Utilizing multi-layer stoplog gates shifted withdrawal flows to warmer surface layers, raising downstream temperatures by a maximum of 8.34 °C. Moreover, the thermal guarantee rate for the spawning of the indicator species (Tor sinensis) increased from 14.29% to 66.67%. This thermal habitat improvement incurred negligible hydropower losses, yielding a 0.33% reduction in total generation and a 0.058% decrease in guaranteed output. These results confirmed that coupling selective withdrawal with cascade optimization effectively balances thermal habitat suitability and hydropower generation, providing information to support reservoir management in highly regulated river basins.
Study region The Heihe and Xitugou Rivers, two intermittent rivers in arid northwestern China. Study focus In the Heihe and Xitugou Rivers, we integrated shallow soil measurements, tension infiltrometer tests, and depth-resolved ground penetrating radar (GPR) attributes to compare the spatial patterns of saturated hydraulic conductivity (Ks), quantify the response of unsaturated hydraulic conductivity (Ku) to decreasing suction, and identify shallow and subsurface features associated with Ks. New hydrological insights for the region Ks ranged from 0.72 to 2.31 cm·min−1 in Heihe and from 0.73 to 1.14 cm·min−1 in Xitugou. Spatial variability was greater in Heihe, whereas Xitugou consistently had higher channel-centre than bank values. For each 1 cm WC increase in tension head toward 0 cm WC, Ku increased by 21.7% (95% CI, 9.0–35.9%) in Heihe and 21.2% (95% CI, 13.5–29.4%) in Xitugou. Structural differences were concentrated in shallow sediments. In Heihe, Ks was positively associated with the log-transformed particle-size ratio ln(P2/P3). Higher Ks also corresponded to lower log-envelope values at 5.5 and 7.5 m and more homogeneous GPR textures at 6–9 m. In Xitugou, no individual shallow-soil variable showed a stable association with Ks, whereas higher Ks corresponded to lower profile-mean relative energy across 1–9 m. Overall, Ks variability reflected river-specific combinations of shallow-soil conditions and subsurface GPR features.
Study region Beijing-Tianjin-Hebei (BTH) region Study focus Understanding urban flood risk-resilience relationships facilitates the shift from defense-oriented to resilience-informed risk management, which is essential for building resilient cities. However, existing studies primarily treat risk and resilience as independent dimensions and overlook their coupling effects, limiting identifications of areas with urgent needs and greater potential for coordinated enhancement of resilience and risk mitigation under differentiated management. This often leads to an overemphasis on high-risk-low-resilience areas while neglecting other priority areas. Building on conventional analytical frameworks that examine spatial patterns, numerical values, and spatial correlations, this study advances resilience-informed risk management by explicitly incorporating risk-resilience coupling effects. The improved approach is applied to the BTH region, a flood-prone urban system facing increasing threats from extreme events, over the period 2010–2020. Priority areas are identified and implications for resilience-informed risk management are discussed. New hydrological insights for the region Results indicate that the BTH region has been characterized by low risk but low resilience. Targeted measures should be prioritized in areas with low coupling coordination but a relatively high degree of coupling. These areas are extended from being limited to high-risk-low-resilience areas to encompass those within Medium-Low, High-Very Low, High-Low, High-Medium, High-High, Very High-Very Low, Very High-Low, Very High-Medium, and Very High-High risk-resilience classes. Routine risk management considering coupling effects can reveal potential vulnerability under extreme events.
Study region The present study region exhibits six diverse physiographic divisions of Bangladesh. Study focus This study harnessed the potential of generative artificial intelligence (GenAI) to predict groundwater recharge across physiographic divisions, ranging from recent to old floodplains, terraces, and a depression area. A total of thirteen predictive variables, encompassing both surface and subsurface factors, and a recharge inventory from 227 piezometers estimated using the water table fluctuation method were used. Since classical machine-learning models often achieve limited accuracy when trained on small datasets, we deployed two GenAI models: Conditional Tabular Generative Adversarial Network (CTGAN) and Tabular Variational Autoencoder (TVAE) to augment the existing data and improve predictive performance. The generated synthetic observations reproduced the statistical structure of the original data and strengthened the downstream machine-learning predictions. New hydrological insights for the region This study provides clear insight into the applicability of machine-learning models for groundwater-recharge prediction under limited-data conditions. Following GenAI-based data augmentation (n = 10–150), relative test-R² gains ranged from 8.3% to 24.2% across the five regressors. Gradient Boosting with TVAE augmentation achieved the largest relative gain (+24.2%, n = 150), whereas XGBoost with CTGAN augmentation achieved the highest absolute test R² of 0.716 (n = 20). These model-level improvements were evaluated using the pooled dataset encompassing all physiographic divisions, with groundwater recharge expressed in millimetres (mm). Spatially, active floodplain areas exhibited the highest recharge potential, followed by terraces and older floodplains, whereas areas adjacent to the depression exhibited relatively low recharge potential. These results demonstrate the strategic value of GenAI for addressing data scarcity and advancing groundwater-recharge modelling.
Study Region: Upper Hanjiang River Basin (UHRB), headwater catchment of the Middle Route of the South-to-North Water Diversion Project, China. Study Focus: Daily flow prediction underpins water resources management, yet existing deep learning methods struggle to reproduce hydrological extremes. This study proposes Deep Process Learning-Long Short-term Kolmogorov–Arnold Network (DPL-LSTKAN). Specifically, DPL-LSTKAN captures the temporal evolution of meteorological forcings and employs Kolmogorov–Arnold Network (KAN) to represent the intrinsic dependencies between meteorological forcings and runoff. Decoupled learning scheme is adopted to separately characterize extreme and normal flow regimes. In addition, Physics-guided gating mechanism dynamically assigns predictor weights according to the dominant hydrological processes. DPL-LSTKAN is applied to flow prediction in the UHRB, with particular focus on its accuracy and physical consistency under extreme floods and droughts. New Hydrological Insights for the Region: In the UHRB, hydrological extremes are controlled by antecedent catchment wetness rather than instantaneous rainfall, most strongly by soil water storage (r=0.59). Flood and drought follow distinct mechanisms: fastflow responds near-monotonically to precipitation via saturation-excess pathways, whereas baseflow scales with subsurface storage and declines toward zero during dry spells. Because fastflow is released only as storage approaches capacity, a saturation onset precedes flood peaks, reproducing peak timing within one day (PTE=0.83day) while suppressing implausible low-flow errors. These controls support reservoir scheduling, flood warning, and ecological-flow management.
Study region Yellow River Delta (YRD), north China. Study focus Based on event-scale precipitation samples collected over three complete hydrological years at the Gudao site in the YRD, this study systematically analyzed the temporal variation characteristics of stable hydrogen and oxygen isotopes (δD, δ18O) in precipitation. We evaluated the influences of local meteorological factors and large-scale atmospheric circulation on isotopic variability, and explored the lagged regulatory effect of equatorial Pacific sea surface temperature anomalies (SSTA, indicated by the Niño 3 index) on precipitation isotopes between the monsoon and non-monsoon seasons. New hydrological insights The controls on precipitation isotopes at the Gudao site in the YRD present significant seasonal differences. Local meteorological factors play a major role in the non-monsoon season, explaining approximately 41% of the δ18O variation. In the monsoon season, isotopic characteristics are largely associated with the upstream rainout effect along moisture transport pathways from the Indian Ocean and western Pacific. Monsoon-season δ18O has a significant positive correlation with the Niño 3 index at a 2-month lag (r = 0.60, p < 0.05), suggesting a potential lagged modulation of equatorial Pacific SSTA on regional precipitation isotopes. These results provide new observational evidence for understanding seasonal isotopic variations and a scientific basis for interpreting isotopic records and paleoclimate reconstruction in this coastal monsoonal setting.
Study region The Xinfengjiang Reservoir in the Dongjiang River Basin, China, is a large multi-year regulating reservoir whose outflows are strongly influenced by human operation. Study focus To represent the impact of reservoir operation on streamflow dynamics, this study develops a knowledge-guided Long Short-Term Memory (KG-LSTM) model to simulation reservoir outflow. Empirical rules derived from reservoir operation chart are extracted using Classification and Regression Trees and embedded into the loss function of the standard LSTM. To further extract reservoir operation insights, KG-LSTM is further combined with an interpretability framework integrating Shapley Additive exPlanations, t-Distributed Stochastic Neighbor Embedding, and k-means clustering. New hydrological insights for the region The KG-LSTM achieves superior performance compared with the standard LSTM, indicating that incorporating empirical reservoir operation knowledge improves the reservoir outflow simulation. More importantly, the interpretability analysis identifies previous-month outflow as the most influential factor in current outflow determination. In particular, current inflow tends to contribute negatively to outflow during the flood season, supporting impoundment, but positively during the dry season, supporting water replenishment. Furthermore, the interpretability framework identifies six representative and situation-dependent operation schemes, which correspond to three major reservoir functions: impoundment, power generation, and water replenishment. Overall, this study provides a framework that enhances both accuracy and interpretability of data-driven reservoir modeling, offering a new avenue for operation knowledge extraction and validation.
Study Region The study was conducted in Baksa District, located in the Eastern Himalayan foothills of Assam, India, where intense monsoonal rainfall, steep terrain, heterogeneous land cover, and dynamic river systems pose significant challenges to sustainable water-resource management. Study Focus This study presents a multi-scale hydrological assessment to support sustainable water management in Baksa District. Event-based runoff generation in the ungauged Pagladiya watershed was estimated using the Soil Conservation Service Curve Number (SCS-CN) method. Groundwater potential was assessed across Baksa District using a Geographic Information System-based Analytical Hierarchy Process (GIS-AHP). Basin-scale hydrological processes were evaluated in the neighbouring gauged Beki River Basin using the Soil and Water Assessment Tool (SWAT), owing to the absence of streamflow observations in the Pagladiya watershed. Riverbank migration along the Pagladiya River was quantified using multi-temporal satellite imagery and the Digital Shoreline Analysis System (DSAS). The complementary findings from these analyses were synthesized to support region-specific water-management planning. New Hydrological Insights for the Region Mean runoff across the Pagladiya sub-watersheds ranged from 71.73 to 114.07 mm, with greater runoff generation observed in downstream sub-watersheds. Groundwater assessment indicated that approximately 56.4% of Baksa District possesses moderate-to-high groundwater potential, predominantly within the alluvial plains. The calibrated SWAT model satisfactorily reproduced observed streamflow dynamics (NSE = 0.82 and 0.78; R² = 0.86 and 0.81 for calibration and validation, respectively). DSAS analysis revealed persistent riverbank instability and pronounced lateral channel migration, particularly along downstream reaches, with lateral displacement exceeding 39 m in the most unstable downstream zone. Collectively, these complementary assessments provide a multi-scale decision-support framework for identifying runoff-prone areas, groundwater-potential zones, and erosion-sensitive river reaches, thereby supporting sustainable water-resource management in the Eastern Himalayan foothills.
Study region The Loess Plateau, China, encompassing grassland, cropland, shrubland and forest ecosystems. Study focus Ecosystem water-use efficiency (eWUE) reflects carbon–water coupling, yet how environmental effects are transmitted through the carbon (GPP) and water (ET) fluxes remains poorly resolved in water-limited regions. We addressed this across the Loess Plateau (2001–2019) at sub-kilometer resolution, combining a spatial random forest, partial-derivative decomposition and conditional probability ratio (CPR) analysis to resolve vegetation-specific controls, their GPP- and ET-mediated pathways and extreme responses. New hydrological insights for the region eWUE was highest in autumn (1.39 gC·m⁻²·mm⁻¹) and lowest and largely stable in spring (1.08 gC·m⁻²·mm⁻¹; 77.34% of pixels with non-significant trend). eWUE extremes were primarily associated with VPD and DSR in spring and summer but shrublands in summer (CPR = 2.14–3.86). Pathway attribution was driver-specific rather than a uniform seasonal switch. Environmental effects were mostly transmitted through ET (fET = 0.55–0.94) in spring. In summer, VPD shifted to GPP-mediation in all vegetation types (fGPP = 0.56–0.85), whereas AW remained ET-mediated (fET = 0.52–0.68). DSR and PRE diverged by vegetation type, GPP-mediated in herbaceous but ET-mediated in woody ecosystems. AW switched to GPP-mediation only in autumn (fGPP = 0.55–0.90). Our findings provide a mechanistic basis for predicting eWUE responses to climate variability.
Study region Kamp catchment, Austria Study focus This study developed a comprehensive framework to assess consecutive hydrological drought-to-flood (D-F) events. The framework integrates a performance-based selection of ÖKS15 climate projections for RCP2.6, RCP4.5, and RCP8.5, hydrological modelling with SWAT+, and ensemble-based assessment of D-F events. New hydrological insights Consecutive D-F events are projected to decline across all scenarios by the end of the century (EOC), with a median reduction of 22–44% relative to the reference period (1976–2005). However, only five climate projections are statistically significant. The direction of change in ensemble signal remains consistent despite uncertainties. This decline in D-F events results from reduced future drought frequency (14–42%) rather than weakened coupling. Both short- and long-duration droughts preceded floods at comparable frequencies (12 each) in the reference period, with future projections showing high inter-model variability (–77 to +47%). The proportion of D-F events among total flood events is projected to decrease from 18– 22% to 10–16% at the EOC, indicating a temporal separation. Projections reveal a decline in winter and spring D-F events and an increase in autumn events. Median transition times between drought termination and flood onset are projected to increase from 37 to 49 days in the EOC (RCP8.5). A future with a compositional shift towards more independent hydrological extremes is anticipated. These results are essential for water resources management in the Kamp catchment.
Study Region The European Alps feature steep elevation gradients, snow-influenced hydrology, and strong spatial variability in water and energy fluxes. Hydrological modeling is limited by sparse high-elevation observations, while growing multi-source data availability introduces inter-product uncertainty. The Adige River Basin, Italy, was selected to evaluate calibration under these conditions. Study Focus We developed interval-based robust calibration (IROC), which uses multiple evapotranspiration (ET) products as observational bounds rather than selecting a single reference. IROC combines interval consistency, controlling average and upper-tail departures, with interval plausibility, identifying parameter sets supported across subbasins and calibration–validation periods. Implemented with SWAT using MOD16, GLEAM, ERA5-Land, and PML, IROC was compared with four single-reference calibrations. Temporal validation used the four-product ET envelope, while hydrological performance was independently evaluated against observed discharge at Trento. New Hydrological Insights for the Region IROC identified consensus parameter sets that maintained monthly ET within multi-product bounds across subbasins and uncertainty levels. Single-reference calibrations favored different parameter combinations and produced distinct discharge responses, demonstrating reference-dependent parameter inference and prediction. Although IROC did not minimize every discharge metric, it balanced multi-product ET consistency, temporal robustness, central discharge accuracy, and retained ensemble compactness. Overall, IROC reduces reliance on an unverifiable single reference and offers a transferable calibration strategy for integrating multiple uncertain datasets in data-scarce or ungauged basins.
Study region The middle–lower Yangtze River Basin, China, represented by four hydrological stations: Yichang, Luoshan, Hankou, and Datong. Study focus This study investigates nonstationary flood frequency and climatic and anthropogenic drivers under changing environmental conditions. Generalized Additive Models for Location, Scale, and Shape (GAMLSS) were used to model annual flood peaks to assess the effects of atmospheric and terrestrial covariates, by incorporating antecedent precipitation, flood-season temperature, large-scale climate indices, and reservoir regulation as physically interpretable covariates. New hydrological insights Flood-frequency nonstationarity exhibits pronounced spatial heterogeneity along the middle–lower Yangtze River. Statistically significant nonstationarity is identified at Yichang, Luoshan, and Hankou Stations, whereas Datong Station shows neither a significant monotonic trend nor a significant change point during the study period. An empirically selected 27–28-day antecedent precipitation window, flood-season maximum temperature (TMAX), the modified reservoir index (MRI), and El Niño–Southern Oscillation (ENSO) improve the representation of nonstationary flood-frequency distributions. Reservoir regulation is more strongly associated with flood-frequency variation at Yichang Station, whereas antecedent precipitation and ENSO become increasingly prominent at the lower part of Luoshan and Hankou Stations. Nonstationary models indicate that flood quantiles generally increase with antecedent precipitation and with El Niño conditions in the preceding winter–spring but decrease with stronger reservoir regulation. Seasonally accumulated TMAX shows heterogeneous quantile effects, suggesting possibly different thermal associations between ordinary and extreme flood conditions.
Study Region This study examines the Western United States including five HUC2 watersheds (Upper Colorado, Lower Colorado, Great Basin, California, Pacific Northwest) and 44 HUC4 sub-basins. This region supports over 60 million people and extensive agriculture with water stress from climate change and multi-year droughts. Study Focus Hydrologic synchronization is quantified using satellite datasets (GLDAS runoff, PRISM precipitation, SMAP soil moisture) and analytical methods including cross-correlation, Principal Component Analysis, co-occurrence analysis, lagged correlation, and whiplash analysis to identify spatial patterns, synchronized extremes, and lead-lag relationships. New Hydrological Insights for the Region The California-Great Basin pair shows 65% runoff synchronization and 67% soil moisture synchronization, with six major basin pairs exceeding 50% synchronization, demonstrating that hydrologic variability is organized into co‑varying regional clusters rather than independent basins. Synchronized extremes occur every 3–4 years (~25% of months), suggesting that multi basin droughts are an intrinsic feature of western hydroclimate. PCA identifies four spatially coherent regions (coastal Pacific Northwest, arid Southwest, coastal California, northern Rockies) explaining 75% of variance. Lagged correlations show that Pacific Northwest anomalies precede those in the Lower Colorado River basin by about five months and that Upper Colorado anomalies lead California, Lower Colorado, and Great Basin by approximately one month. Hydrologic whiplash analysis identifies coastal Pacific Northwest and arid Southwest as hotspots of rapid wet–dry transitions (whiplash in 20% of years), distinguishing these as abrupt shifts and providing a new regional perspective on hydroclimatic volatility.
Study region The Xitiaoxi River Basin, a highly regulated river network in eastern China's Taihu Lake Basin, was studied using daily hydrological and meteorological data (2012–2023) from five control stations. Study focus This study develops a dynamic framework for ecological flow monitoring and tiered early warning, addressing the limitations of static thresholds and the lack of predictive capacity. Baseflow was separated using the Lyne–Hollick digital filter, hydrological stages were delineated via Fisher's optimal segmentation, and stage-specific ecological flow thresholds were derived from baseflow frequency analysis using the Qp90 (90% exceedance probability) criterion. An ensemble deep learning model, incorporating precipitation, evaporation, baseflow index, reservoir releases, and antecedent hydrological conditions, was built to forecast ecological flow status up to 5 days ahead. Based on predicted flow deficits relative to thresholds, a four-level warning system was established. New hydrological insights for the region Results reveal pronounced seasonal variability in ecological flow requirements, with higher thresholds during the flood season, lower thresholds during dry periods, and intermediate thresholds during transitional stages. The pre-flood period, dry season, and ecologically sensitive stages are most vulnerable to deficits. The model achieved an average test accuracy of 89.69% over 1–5-day lead times and 84.51% at a 5-day lead time. For red warnings, precision and recall reached 91.55% and 89.08%, respectively, indicating high reliability in detecting severe deficits. This framework provides a sound scientific basis for adaptive water resources and ecological flow management in hydrologically complex river network regions.
Study region The study covers the Qinghai-Tibet Plateau (QTP), where climate warming is shifting precipitation from solid to liquid forms, potentially altering soil erosion risk. Study focus We estimated erosion using the Revised Universal Soil Loss Equation, quantified phase change with the rainfall ratio to precipitation (RRP), characterized event-scale intensity by the maximum 60-min intensity (I60), and identified dominant drivers using an interpretable XGBoost-SHAP framework. New hydrological insights for the region RRP increased across most of the QTP, but its linkage with rainfall erosivity was strongly season-dependent, with the closest coupling occurring in autumn (ρ = 0.683, p < 0.001). At the interannual scale, RRP was positively correlated with I60 (ρ = 0.710, p < 0.001), while I60 was also positively associated with rainfall erosivity (ρ = 0.647, p = 0.002), indicating a process linkage among precipitation-phase shifts, short-duration rainfall intensity, and erosive forcing. Rainfall (38.4%) and slope (27.2%) remained the primary controls on spatial erosion heterogeneity, whereas RRP (7.8%) acted as a secondary regulator. Erosion responses shifted markedly across 4500–5500 m, where vegetation buffering weakened and sensitivity to precipitation-phase variability increased. These findings improve understanding of how precipitation-phase shifts and event-scale intensity shape erosive forcing, while supporting elevation-specific soil conservation under continued warming.
The Upper Blue Nile Basin (UBNB) in Ethiopia is a critical hydrological region where rain-fed agriculture supports over 80% of employment, making assessment of long-term rainfall trends essential for water resources management and food security. This study integrates the Mann–Kendall (MK), Modified MK (mMK), Sen's slope, Innovative Trend Analysis (ITA), and Percent Bias to evaluate annual and seasonal rainfall variability across 41 stations in the UBNB from 1981 to 2024, using National Meteorology Agency records supplemented by CHIRPS data for gap-filling. The framework is extended with Pettitt change-point and Sequential Mann-Kendall (SQMK) tests to capture trend evolution, and Benjamini–Hochberg false discovery rate (FDR) testing to assess whether locally significant station trends constitute genuine basin-wide signals. Results show about 75% of stations exhibit increasing annual rainfall, with significant increases at Asendabo (9.22 mm/yr) and Nekemte (7.35 mm/yr) and significant decreases at Mota (−4.11 mm/yr) and Debre Tsige (−2.55 mm/yr). Summer and autumn rainfall show dominant increasing trends, while winter shows widespread declines across ~65% of stations. ITA consistently detects subtle sub-trends across low-, medium-, and high-rainfall regimes overlooked by MK/mMK, with strong inter-method correlations (R² up to 0.8774). Change-point analysis identifies significant shifts at 34% of stations, clustered around 2002–2005 and 2011–2012, while FDR testing confirms the annual and autumn increases as genuine field-significant signals, whereas the widespread spring increase does not survive correction for multiple testing and spatial dependence. These findings provide a more rigorously validated basis for climate adaptation and water resource management in the basin.
