A Coupled SWAT-LSTM Approach for Climate-Driven Runoff Dynamics in a Snow- and Ice-Fed Arid Basin

Authors:
Open Access
Publisher website
Google Scholar
Abstract
As global climate change intensifies, hydrological processes in arid inland river basins are undergoing profound transformations, posing severe challenges to regional water security and ecological stability. This study aims to develop a coupled SWAT-LSTM model integrating glacier melt processes to simulate runoff dynamics in the Keria River basin under climate change, providing a basis for local water resource management. Based on natural monthly runoff observations from the Langgan hydrological station (1961–2015), glacier data extracted from Landsat 8 remote sensing imagery (2013–2019), and downscaled data from the CMIP6 Multi-Model Ensemble (MME), this study constructed a SWAT-LSTM coupled model to simulate future scenarios (2026–2100). Research indicates that this hybrid model significantly enhances the accuracy of hydrological simulations in high-altitude glacier-fed catchments. The Nash efficiency coefficient (NSE) during the validation period reached 0.847, representing a 15% improvement over the SWAT model. SSP5-8.5 is identified as a high-risk scenario, underscoring the urgency of emissions reduction; SSP1-2.6 represents the most desirable pathway, with its relatively stable pattern offering sustained advantages for long-term water resource management in the basin. The study further reveals a negative feedback mechanism between glacier ablation and runoff increase, validating the regulatory role of Jiyin Reservoir’s “store during floods to compensate for droughts” operation strategy in balancing basin water resources. This study explores the coupling path between the physical model and the deep learning model, and provides an effective integration scheme for the hydrological simulation of the global watershed with ice–snow meltwater as the main recharge runoff, especially for the adaptive management of water resources in inland river basins in arid areas.
Funding Information
  • Natural Science Foundation of Xinjiang Uygur Autonomous Region (2022D01C675)

This publication has 71 references indexed in Scilit: