Xie et al. (2025) Interpretable deep learning for dynamic rainfall-runoff prediction: Integrating adaptive signal decomposition and spatiotemporal feature extraction

Identification

• Journal: Journal of Environmental Management
• Year: 2025
• Date: 2025-12-30
• Authors: Xuan Xie, Guohe Huang, Shuguang Wang, Feng Wang, Shuai Xie
• DOI: 10.1016/j.jenvman.2025.128444

Research Groups

• School of Environmental Science and Engineering, Shandong University, Qingdao 266237, China
• Institute for Energy, Environment and Sustainable Communities, University of Regina, Regina, Saskatchewan S4S0A2, Canada
• Shandong Key Laboratory of Water Pollution Control and Resource Reuse, School of Environmental Science and Engineering, Shandong University, Qingdao 266237, China
• State Key Laboratory of Earth Surface Processes and Resource Ecology, Faculty of Geographical Science, Beijing Normal University, Beijing 100875, China
• Center for Energy, Environment and Ecology Research, UR-BNU, Beijing Normal University, Beijing, China
• Changjiang River Scientific Research Institute, Wuhan 430010, China

Short Summary

This study proposes an interpretable deep learning model for dynamic rainfall-runoff prediction, integrating adaptive signal decomposition and spatiotemporal feature extraction to enhance accuracy and provide insights into complex hydrological processes. The model significantly outperforms traditional methods, especially for short-term predictions, with data decomposition being the strongest contributing module.

Objective

• To develop an interpretable deep learning model that integrates adaptive signal decomposition and spatiotemporal feature extraction to improve the accuracy of dynamic rainfall-runoff prediction, addressing the nonlinear and non-stationary characteristics of runoff data.

Study Configuration

• Spatial Scale: Not explicitly stated, but implied for rainfall-runoff processes (e.g., catchment or watershed scale).
• Temporal Scale: Multi-step prediction tasks, with a focus on short-term predictions.

Methodology and Data

• Models used: Adaptive particle swarm optimization variational mode decomposition (APSO-VMD), spatiotemporal attention mechanisms, Gated Recurrent Units (GRUs).
• Data sources: Hydrological time series data (rainfall, runoff).

Main Results

• The proposed model consistently outperforms traditional baseline models and other combined models across different prediction horizons.
• For short-term predictions, the model achieved a Nash-Sutcliffe Efficiency (NSE) of 0.9977 and a Root Mean Square Error (RMSE) of 1.1793.
• The data decomposition module has the strongest main effect, contributing up to 62.1 %.
• The interaction between the data decomposition module and the spatiotemporal feature extraction module plays a crucial role, with average contributions of 31.74 % to RMSE, 32.04 % to NSE, 28.47 % to Mean Absolute Error (MAE), and 32.98 % to correlation coefficient (R).

Contributions

• Proposes a novel interpretable deep learning framework for rainfall-runoff prediction by integrating adaptive signal decomposition and spatiotemporal attention mechanisms.
• Enhances prediction accuracy, particularly for short-term forecasts, by effectively handling nonlinear and non-stationary hydrological data.
• Provides insights into model decision-making through the visualization of attention weights and quantifies the impact of individual components and their interactions on prediction performance.

Funding

• Not specified in the provided text.

Citation

@article{Xie2025Interpretable,
  author = {Xie, Xuan and Huang, Guohe and Wang, Shuguang and Wang, Feng and Xie, Shuai},
  title = {Interpretable deep learning for dynamic rainfall-runoff prediction: Integrating adaptive signal decomposition and spatiotemporal feature extraction},
  journal = {Journal of Environmental Management},
  year = {2025},
  doi = {10.1016/j.jenvman.2025.128444},
  url = {https://doi.org/10.1016/j.jenvman.2025.128444}
}