Advancing hydrological and cryospheric simulations in Indus River Basin: First comparative evaluation of GESSI-3/ESSI-3 and a CNN–LSTM deep learning framework

Hydrological models of glacier-fed river systems are critical for assessing climate impacts, yet conventional single-variable calibration, often against discharge alone, may obscure key climate sensitivities and internal dynamics. To address this, we conduct the first multi-variable, basin-wide evaluation of process-based and deep learning hydrological models across the climatically diverse Indus River Basin (IRB). Using identical meteorological forcing, we compare the performance of the GESSI-3 and ESSI-3 models against a hybrid CNN–LSTM framework and validate the outputs against independent observations of runoff, evapotranspiration (ET), terrestrial water storage (TWS) anomalies, and glacier mass balance (GMB) over the period 2001–2024. Results indicate that the CNN–LSTM model more consistently reproduces observed variability, particularly during seasonal extremes, high-flow events, and rapid changes in hydrological storage. Multivariable analysis reveals that runoff variability is predominantly temperature-driven, even in the monsoon-affected lower basin, and that persistent declines in TWS are consistent with sustained warming and long-term storage depletion. This study demonstrates that reliance on discharge-only evaluation can mask critical climate-driven transitions, and that incorporating cryospheric and hydrological constraints significantly strengthens the assessment of climate change impacts in large, glacier-fed, transboundary basins.