Integrated machine learning methods with resampling algorithms for flood susceptibility prediction.

Area of Science:

• Hydrology and Water Resources
• Geographic Information Science
• Environmental Modeling

Background:

• Standalone flood susceptibility models often yield biased results due to single train-test data splits.
• Accurate flood susceptibility mapping is crucial for mitigating risks in vulnerable regions, especially coastal areas.

Purpose of the Study:

• To develop and evaluate novel integrative flood susceptibility prediction models.
• To compare the performance of multi-time resampling approaches (random subsampling and bootstrapping) integrated with machine learning models.
• To enhance the accuracy and reliability of flood susceptibility maps (FSM).

Main Methods:

• Proposed integrative models combining random subsampling (RS) and bootstrapping (BT) with generalized additive model (GAM), boosted regression tree (BRT), and multivariate adaptive regression splines (MARS).
• Employed 10 runs of data resampling for model learning and validation, averaging predictions for FSM generation.
• Evaluated model performance using Area Under Curve (AUC) of Receiver Operating Characteristic (ROC), True Skill Statistic (TSS), and Correlation Coefficient (COR).

Main Results:

• Resampling algorithms significantly improved the performance of standalone GAM, MARS, and BRT models.
• The bootstrapping (BT) algorithm generally outperformed the random subsampling (RS) algorithm.
• The BT-GAM model achieved superior predictive accuracy (AUC = 0.98, TSS = 0.93, COR = 0.91), outperforming benchmark models like MLP and SVM.

Conclusions:

• Integrative flood susceptibility models utilizing multi-time resampling techniques offer superior predictive performance over standalone approaches.
• The proposed BT-GAM, BT-MARS, and BT-BRT models provide reliable tools for accurate flood susceptibility mapping.
• These advanced models hold promise for application in diverse geographical regions prone to flooding.