Kanani-Sadat, Yousef; Safari, Abdolreza; Nasseri, Mohsen et Homayouni, Saeid 
ORCID: https://orcid.org/0000-0002-0214-5356
(2025).
A novel explainable stacking ensemble model for estimating design floods: A data-driven approach for ungauged regions.
Advanced Engineering Informatics
, vol. 66
.
p. 103429.
DOI: 10.1016/j.aei.2025.103429.
Résumé
Accurate flood estimation is crucial for the sustainable design of resilient urban infrastructures, especially in ungauged regions where data scarcity limits traditional hydrological analyses. This study introduces an innovative explainable stacking ensemble learning framework to predict design floods with return periods of 5, 10, 25, 50, 100 and 500 years. The model integrates diverse machine learning techniques to capture the nonlinear and complex relationships between flood magnitudes and catchment characteristics. In the first level of the ensemble, four base models, Multilayer Perceptron (MLP), Support Vector Regression (SVR), Random Forest (RF), and Extremely Randomized Trees (ERT), are trained. A linear Elastic-Net regression model at the second level combines the results of these base models for the final flood estimates. Performance analysis shows that MLP performs best for lower return periods, while ERT excels for higher ones. Overall, the stacking model outperforms individual models across all return periods, demonstrating its robustness. To enhance model interpretability, Shapley Additive Explanations (SHAP) values are used to reveal the contribution of catchment characteristics to flood behavior, identifying key flood drivers. The results show how feature importance changes with return periods, supporting context-specific decision-making. This research pioneers the use of explainable AI (XAI) in stacking ensemble models for flood estimation, offering a transparent and actionable methodology for urban flood risk management. By integrating advanced modeling, explainability, and remote sensing data through Google Earth Engine, the proposed framework provides a robust decision-support tool for practitioners and policymakers, enhancing the resilience of urban infrastructure to flood hazards.
| Type de document: | Article |
|---|---|
| Mots-clés libres: | explainable artificial intelligence; stacking ensemble learning; urban resilience; flood risk management; regional flood frequency analysis; Google earth engine |
| Centre: | Centre Eau Terre Environnement |
| Date de dépôt: | 26 août 2025 14:46 |
| Dernière modification: | 26 août 2025 14:46 |
| URI: | https://espace.inrs.ca/id/eprint/16509 |
Gestion Actions (Identification requise)
 |
Modifier la notice |