Exploring Flood Susceptibility in the Amazon River Basin Using Explainable AI

Floods, responsible for 44% of global natural disasters and impacting over 1.6 billion people between 2000 and 2019, are increasing in frequency and severity due to climate change and human activities. In the Amazon River Basin, this trend is evident with rising flood frequency and intensity since 2000, yet detailed flood susceptibility maps for the region remain scarce. To address this limitation, this study utilized Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) to develop flood susceptibility maps for the Amazon River Basin. The analysis incorporated a flood inventory dataset along with fourteen conditioning factors, encompassing meteorological, hydrological, topographical, and geological variables. The multicollinearity among the variables was addressed through Variance Inflation Factor (VIF) analysis. The models' performance was evaluated using accuracy, precision, recall, F1-score, and Kappa score. To enhance the interpretability of both models, SHAP (SHapley Additive exPlanations) was employed to identify and evaluate the key factors influencing the models' outcomes. Results confirmed the effectiveness of both models, with XGBoost delivering an accuracy of 0.91 and a Kappa score of 0.83, outperforming RF’s accuracy of 0.90 and Kappa score of 0.81. SHAP results revealed that for both models the most important factors were land use/land cover, rainfall, elevation, curve number, slope, drainage density, and soil. We assessed the robustness of the models by removing the least important features. Both models demonstrated stable performance, maintaining consistent accuracy, precision, recall, and F1-scores, with XGBoost surpassing RF. Ultimately, RF and XGBoost proved effective in generating accurate and reliable flood susceptibility maps for large regions like the Amazon River Basin, with SHAP offering significant insights into the interpretability of model outputs.

 

Funding:

This research was supported by Disaster-Safety Platform Technology Development Program of the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT. (No. 2022M3D7A1090338).