Groundwater potential mapping using a frequency ratio-LSTM hybrid model in the Andijan Region, Uzbekistan

초록

This study develops a high-resolution groundwater potential map for the Andijan Region in the eastern Fergana Valley, Uzbekistan, by integrating the probabilistic frequency ratio (FR) method with a long short-term memory (LSTM) neural network. Eleven hydrogeomorphic, lithological, and land-surface variables-Channel Network Base Level, Closed Depressions, Cross-Sectional Curvature, DEM, Downslope Curvature, LS factor, Topographic Wetness Index, Valley Depth, Geology, Soil, and Land Cover-were derived from 12.5 m ALOS PALSAR topographic data, Sentinel imagery, and national geodatabases. FR analysis quantified the contribution of each factor to observed well yields, generating factor-specific weights and using gridded predictors as inputs to the LSTM to capture nonlinear spatial interactions. Validation against an independent dataset comprising 61 surveyed production wells showed an improvement in the area under the receiver operating characteristic curve from 75.60% for the FR model to 77.54% for the FR-LSTM ensemble, demonstrating an incremental but consistent benefit of the hybrid modeling approach. High-potential zones (> 0.8 probability) are spatially associated with Quaternary alluvial fans and the Kara Darya floodplain, where thick and highly permeable sediments, together with gentle slope gradients, favor groundwater recharge, whereas uplifted Paleozoic terrains typically exhibit low groundwater potential (< 0.2). Cross-validation indicates that 74% of high-yield wells lie within the two highest groundwater potential classes. A reproducible FR-LSTM workflow provides actionable guidance for well siting, groundwater abstraction management, and integrated water resources planning in semi-arid and data-limited basins and is readily transferable to analogous hydrogeological settings.

키워드