Zhou et al. (2026) Enhancing mid- to long-term runoff simulation in human-impacted basins through coupled SWAT-LUT and LSTM modelin

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-01-06
• Authors: Wenqi Zhou, Gaoxu Wang, Wei Wu, Yongxiang Wu, Xingnan Zhang, Siyang Feng, Xingchi Zhou
• DOI: 10.1016/j.ejrh.2025.103050

Research Groups

• The National Key Laboratory of Water Disaster Prevention, Nanjing Hydraulic Research Institute, Nanjing 210029, China
• The National Key Laboratory of Water Disaster Prevention, College of Hydrology and Water Resources, Hohai University, Nanjing 210098, China
• State Key Laboratory of Hydraulics and Mountain River Engineering, College of Water Resources and Hydropower, Sichuan University, Chengdu 610065, China

Short Summary

This study developed an integrated SWAT-LUT-LSTM model to improve mid- to long-term runoff simulations in human-impacted basins by dynamically incorporating land-use changes and correcting for reservoir regulation effects, demonstrating superior accuracy compared to conventional models. It also introduced novel indices to quantify the spatiotemporal impacts of land-use change and elucidated the role of human activities in altering runoff dynamics.

Objective

• Enhance mid- to long-term runoff simulation accuracy in human-impacted basins.
• Elucidate the spatiotemporal effects of land-use changes and reservoir regulation on runoff dynamics.

Study Configuration

• Spatial Scale: Dawen River Basin (DRB) in China, approximately 8726 square kilometers (km²). The basin is divided into upstream, midstream (above Dawenkou), and downstream (below Daicunba) sections. It includes two major reservoirs (Xueye Reservoir and Guangming Reservoir) with a combined storage capacity exceeding 100 million cubic meters (m³), and Dongping Lake.
• Temporal Scale:
  • Study period: 1961–2020 (meteorological data), 1965–2020 (runoff data).
  • Calibration period: 1965–1988.
  • Validation period: 2008–2020.
  • Land-use data intervals: 1980–2020 at 5-year intervals.
  • Simulation scale: Monthly.

Methodology and Data

• Models used:
  • SWAT (Soil and Water Assessment Tool) as the baseline physical model.
  • SWAT-LUT (SWAT Land-use Update Tool) for dynamic land-use integration and HRU reclassification.
  • LSTM (Long Short-Term Memory) network for error correction of SWAT-LUT outputs.
  • Coupled model: SWAT-LUT-LSTM.
  • Calibration algorithm: Sequential Uncertainty Fitting algorithm (SUFI-2) in SWAT-CUP.
  • LSTM optimization: Whale Optimization Algorithm (WOA).
  • Land-use change assessment: Inconsistency index (IC) and Permeability Change Coefficient (PCC).
  • Error analysis: Morlet wavelet transform, Welch power spectral density analysis, and autocorrelation functions.
• Data sources:
  • Runoff data: Daily average discharge records (1965–2020) from Dawenkou and Daicunba hydrological stations (Hydrological Yearbook, People’s Republic of China).
  • Meteorological data: Daily precipitation, temperature, air pressure, wind speed, relative humidity, evaporation, and sunshine duration (1961–2020) from 13 stations (China Surface Climate Daily Dataset (V3.0), National Meteorological Information Center).
  • River network: 2018 China Hydrography Dataset (OpenStreetMap).
  • DEM: 30 meter (m) resolution GDEMV2 data (Japan’s METI and NASA, via Geospatial Data Cloud of the Chinese Academy of Sciences).
  • Land use: 30 m resolution raster data from China Land Use/Cover Change Dataset (CNLUCC) (Institute of Geographic Sciences and Natural Resources Research, CAS; via Resource and Environment Science and Data Center).
  • Soil data: 1 kilometer (km) resolution raster data from Harmonized World Soil Database (HWSD) (FAO and IIASA, via FAO Soils Portal; China-specific data from Nanjing Institute of Soil Science).

Main Results

• The coupled SWAT-LUT-LSTM model achieved superior runoff simulation accuracy, with R² ≥ 0.92, NSE ≥ 0.91, and RSR ≤ 0.28 during both calibration and validation periods. This represents an improvement of 17.83% in R² and 32.68% in NSE compared to the baseline SWAT model.
• Urbanization in the Dawen River Basin (1980–2020) significantly reduced surface permeability (mean PCC of 0.08, reaching 0.30 in urban clusters), increasing the proportion of surface runoff from 46% to 64% and altering the basin's water balance.
• The Inconsistency Coefficient (IC) identified a critical threshold of 0.25, beyond which dynamic land-use inputs are essential for accurate simulation. The basin's current IC is approximately 0.30.
• SWAT-LUT model errors exhibited strong annual periodicity (12-month lag peaks) and a high correlation with reservoir water levels (R² > 0.63), indicating that reservoir regulation significantly reshapes intra-annual runoff distribution. The LSTM component effectively learned and corrected these human-induced periodic errors.
• The SWAT-LUT model, by incorporating dynamic land-use, effectively corrected the systematic overestimation of low flows observed in the conventional SWAT model and improved medium-flow simulation.

Contributions

• Development of a novel SWAT-LUT-LSTM coupling framework that significantly enhances mid- to long-term runoff simulation accuracy in human-impacted basins by integrating dynamic land-use changes and machine learning-based error correction.
• Introduction of two new indices, the Inconsistency Coefficient (IC) and Permeability Change Coefficient (PCC), to quantify the spatiotemporal impacts of land-use change on runoff and provide transferable thresholds for guiding land-use input selection in hydrological models.
• Provision of a mechanistic explanation for periodic model residuals, attributing them to reservoir regulation effects, and demonstrating how LSTM can effectively capture and correct these unmodeled anthropogenic influences while preserving the physical interpretability of the base model.
• Demonstration of the model's robust generalization ability and strong temporal and spatial extrapolation performance in a representative human-impacted basin (Dawen River Basin).

Funding

• National Key Research and Development Program of China (Grant NO. 2022YFC3204603)
• Province-Municipality Integrated Innovation Scientific Research Program of Yunnan Province (Grant NO. 202202AH210007)
• Central Public-interest Scientific Institution Basal Research Fund for Nanjing Hydraulic Research Institute (Grant NO. Y522021)

Citation

@article{Zhou2026Enhancing,
  author = {Zhou, Wenqi and Wang, Gaoxu and Wu, Wei and Wu, Yongxiang and Zhang, Xingnan and Feng, Siyang and Zhou, Xingchi},
  title = {Enhancing mid- to long-term runoff simulation in human-impacted basins through coupled SWAT-LUT and LSTM modelin},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2025.103050},
  url = {https://doi.org/10.1016/j.ejrh.2025.103050}
}