Coupling Machine Learning with the SWAT Model to Predict Future Runoff and Extreme Hydrological Events Under the Effects of Climate Change and Human Activities: A Socio-Hydrological Attribution Study

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Projecting runoff and extreme hydrological events in human-modified basins requires climate, land-use and reported water-consumption forcings to be represented within a single, traceable scenario framework. Here, the Dawen River, the largest tributary of the lower reaches of the Yellow River, was selected as the study area. We coupled a physically based model (SWAT model) with the CEEMDAN–Attention–LSTM model to forecast future social water consumption and runoff and then diagnosed extreme hydrological events and decoupled the main driving factors driving runoff dynamics using multiple remote-sensing and hydrometeorological datasets. Five global climate models (GCMs) of the Coupled Model Intercomparison Project Phase 6 (CMIP6) were selected, and the Delta statistical downscaling method was adopted to enhance the future climate data. The GeoSOS-FLUS model was applied to simulate the future LULC distribution, and then all the climate data, the LULC data and other remote-sensing data were input into the CEEMDAN–Attention–LSTM–SWAT model. The results confirmed that the coupled model had reasonable performance in social water consumption and runoff simulation; the Nash–Sutcliffe efficiency (NSE) values for the measured and simulated runoff ranged from 0.80 to 0.84. Future runoff in all scenarios showed varying degrees of increasing trends. Future droughts display longer durations with lower severity compared to floods, while pluvial flood events are characterized by shorter durations with higher severity. The frequency of the drought–pluvial seesaw increased from upstream to downstream in the SSP126, SSP245, and SSP585 scenarios. Finally, the results of decoupling runoff dynamics revealed that the contribution of human activities showed a slight overall increasing tendency from upstream to downstream, accounting for 70% to 82% of runoff alterations, and social water consumption and the LULC factors were the dominant anthropogenic factors.
Funding Information
  • Shaanxi Province Natural Science Basic Research Program (2024JC‑YBQN‑0274)
  • Shaanxi Provincial Department of Education Service Local Special Project (23JE003)
  • Key Water Science and Technology Project of Shandong Province (SDSLKJ2024003)
  • Gansu Provincial Water Resources Scientific Experimental Research and Technology Promotion Project (21GSLK058)

This publication has 82 references indexed in Scilit: