Li et al. (2025) A Study on Flood Susceptibility Mapping in the Poyang Lake Basin Based on Machine Learning Model Comparison and SHapley Additive exPlanations Interpretation

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2025
• Date: 2025-10-14
• Authors: Zhuojia Li, Youchen Zhu, Danlu Chen, Qin Ji, Deliang Sun
• DOI: 10.3390/w17202955

Research Groups

The provided text does not explicitly list the research groups, labs, or departments involved in the study.

Short Summary

This study addresses bottlenecks in machine learning-based flood susceptibility mapping (FSM) in complex basins by establishing a high-precision sample database, comparing PSO-optimized hybrid models, and employing SHAP for interpretation, finding that ensemble learning models (especially RF with AUC 0.9536) exhibit superior performance and reveal complex, spatially heterogeneous driving mechanisms.

Objective

• To improve the accuracy and interpretability of machine learning-based flood susceptibility mapping (FSM) in complex basins by addressing issues of unclear model applicability, limited sample quality, and insufficient machine interpretation.

Study Configuration

• Spatial Scale: Poyang Lake basin.
• Temporal Scale: The 2020 Poyang Lake flood was used as a case study to establish the flood inundation sample database.

Methodology and Data

• Models used:
  • Hybrid models optimized by Particle Swarm Optimization (PSO): Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Convolutional Neural Network (CNN).
  • Shapley Additive exPlanations (SHAP) framework for interpreting factor contributions and interaction effects.
• Data sources:
  • High-precision flood inundation sample database established using the 2020 Poyang Lake flood.
  • Driving factors (e.g., elevation, terrain moisture index, late rainfall) used for model training and interpretation.

Main Results

• Ensemble learning models demonstrated superior performance for flood susceptibility mapping in complex basins like Poyang Lake.
• The Random Forest (RF) model achieved the best predictive performance with an Area Under the Receiver Operating Characteristic Curve (AUC) value of 0.9536.
• Elevation was identified as the most important global driving factor for flood susceptibility.
• SHAP local interpretation revealed significant spatial heterogeneity in driving mechanisms, with the susceptibility of local depressions primarily controlled by the terrain moisture index.
• A nonlinear phenomenon was observed where SHAP values were negative under extremely high late rainfall, preliminarily attributed to a "spatial transfer that is prone to occurrence" mechanism triggered by the backwater effect, highlighting complex nonlinear interactions among factors.
• The proposed "high-precision sampling, model comparison, SHAP explanation" framework effectively improved the accuracy and interpretability of FSM.

Contributions

• Established a high-precision flood inundation sample database for a complex basin, addressing the issue of limited sample quality.
• Systematically compared and validated the performance of PSO-optimized hybrid machine learning models (RF, XGBoost, CNN) for FSM, clarifying model applicability in complex environments.
• Applied the SHAP framework to provide detailed global and local interpretations of flood driving factors, including their complex nonlinear interactions, thereby enhancing machine interpretability.
• Identified key global and local driving factors (e.g., elevation, terrain moisture index) and uncovered a novel nonlinear phenomenon related to backwater effects under extreme rainfall.
• Proposed and validated a comprehensive "high-precision sampling, model comparison, SHAP explanation" framework to improve FSM accuracy and interpretability.

Funding

The provided text does not contain information regarding funding projects, programs, or reference codes.

Citation

@article{Li2025Study,
  author = {Li, Zhuojia and Tian, Jie and Zhu, Youchen and Chen, Danlu and Ji, Qin and Sun, Deliang},
  title = {A Study on Flood Susceptibility Mapping in the Poyang Lake Basin Based on Machine Learning Model Comparison and SHapley Additive exPlanations Interpretation},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17202955},
  url = {https://doi.org/10.3390/w17202955}
}