Machine learning-based groundwater quality classification can be affected by target leakage and inadequate assessment of temporal and geographic generalization. This study compares conventional machine learning, CNN-LSTM, and Transformer-based architectures in a leakage-aware manner, using 1,106 post-monsoon groundwater observations collected from 378 sampling locations in Telangana, India, during 2018–2020. We classified groundwater samples into three electrical conductivity (EC)-derived categories: Moderate, Poor, and Bad. Data auditing revealed an exact linear relationship between total dissolved solids and EC (TDS = 0.64 × EC), indicating deterministic target leakage. Both variables were therefore excluded from the principal predictor set. We evaluated models using chronological holdout, unseen-station testing, and geographical block validation. RF achieved 94.29
Reliable prediction of qanat flowrate is important for assessing the availability and supporting the sustainable management of these groundwater resources. In this study, seven machine-learning methods were investigated for predicting the flowrate of qanats in the Mokhtaran watershed, South Khorasan Province, Iran. Nine environmental, hydrological, hydrogeological, meteorological, and land-use variables were used as input predictors, while qanat flowrate (L/s) was defined as the target variable. Model performance was evaluated using RMSE, MAE, NSE, R², MAPE, and RMSLE. The results showed that random forest provided the lowest RMSE (2.7021 L/s), lowest MAE (1.7136 L/s), and highest R² and NSE (0.7064), indicating its comparatively stronger performance for predicting spatially qanat flowrate. The results demonstrate the potential of machine-learning-based flowrate prediction as a tool to support the assessment and sustainable management of qanat resources.
The governance of transboundary river basins sustainably in the face of uncertain future climate and socioeconomic conditions calls for an integrated assessment of policy measures. Although some water management policies seem effective in normal or near-normal climate conditions, they may not be as effective under more frequent climate extremes. Evaluating the effectiveness of policy options benefits from an interdisciplinary approach that tracks the hydrological and socioeconomic implications of projected water availability under future climates, especially in large and multi-jurisdictional river basins with interwoven natural and human systems. To address this need, we developed an interdisciplinary water management framework that simulates human and natural systems and evaluates policy performance under a range of climate and socioeconomic scenarios from water management, transboundary river governance, ecological, economic, and social perspectives. The framework consists of three main blocks: (a) Scenario Generation, (b) Modelling, and (c) Evaluation. We apply this framework to the Saskatchewan River Basin in western Canada and examine a combination of water policy and climate change scenarios driven by the outputs of General Circulation Models (GCM). Investigating over 300 scenarios, we demonstrate that whilst some policies are helpful under certain climate conditions, their mitigation potential is relatively marginal compared to the impacts of climate change. We demonstrate that policies like improving irrigation efficiency can save water under normal climate conditions, however, relying only on this may compromise long-term water security under more extreme climates.
Reliable sectoral water-demand forecasts are needed for regional water-network planning, drought preparedness, and demand-side scenario analysis. However, conventional approaches usually model domestic, agricultural, industrial, and ecological demands separately, leaving historical associations among demand categories underused. This study proposes a Cross-Sector XGBoost framework for regional multi-sector water-demand forecasting from a water–food–energy–ecology nexus perspective. Regional demand is represented by six management-relevant categories: domestic and public, grain-irrigation, other agricultural, energy-intensive and strategic industrial, general industrial, and ecological and environmental demand. The framework augments sector-specific predictors with candidate cross-sector historical features and uses training-period expanding-window validation to select the retained cross-sector representation while maintaining a region–sector–month forecasting structure. Using monthly water-demand records from Taizhou, China, the framework was compared with seasonal-average, Holt–Winters, LSTM, and Independent XGBoost benchmarks. The selected Cross-Sector XGBoost reduced RMSE from 95.57 to 87.89 × 104 m3 (8.04
Joint flood control operation of a reservoir and a flood detention basin (FDB) plays a critical role in mitigating extreme flood disasters. To improve the flood control efficiency of a coupled reservoir-FDB group, it is essential to quantify the equivalent flood control storage between the reservoir and the FDB. After defining the concepts of equivalent flood control storage and the equivalent ratio for a reservoir-FDB group, an equivalent flood control storage is derived that is independent of FDB operation rules and the inherent characteristics of flood events. The Three Gorges Reservoir- Jingjiang FDB group is selected as the study case. Driven by the four representative design flood events, the impacts of the FDB’s operation period, the flood magnitude characterized by its exceedance frequency, the upstream reservoir inflow hydrograph into the upstream reservoir and pre-utilized downstream FDB flood control storage on the equivalent ratio have been figured out. The results indicate that an earlier-started and shorter-duration FDB operation period, as well as a lower pre-utilized flood control storage in the FDB, generally tend to contribute to a larger equivalent ratio. Meanwhile, flood events with larger magnitude and a delayed-peak inflow hydrograph also tend to result into a larger equivalent ratio. The derived equivalent flood control storage can effectively support the flood control operation of the reservoir-FDB groups.
