Advancing flood susceptibility modeling using stacking ensemble machine learning: A multi-model approach

Flood susceptibility modeling is crucial for rapid flood forecasting, disaster reduction strategies, evacuation planning, and decision-making. Machine learning (ML) models have proven to be effective tools for assessing flood susceptibility. However, most previous studies have focused on individual models or comparative performance, underscoring the unique strengths and weaknesses of each model. In this study, we propose a stacking ensemble learning algorithm that harnesses the strengths of a diverse range of machine learning models. The findings reveal the following: (1) The stacking ensemble learning, using RF-XGB-CB-LR model, significantly enhances flood susceptibility simulation. (2) In addition to rainfall, key flood drivers in the study area include NDVI, and impervious surfaces. Over 40% of the study area, primarily in the northeast and southeast, exhibits high flood susceptibility, with higher risks for populations compared to cropland. (3) In the northeast of the study area, heavy precipitation, low terrain, and NDVI values are key indicators contributing to high flood susceptibility, while long-duration precipitation, mountainous topography, and upper reach vegetation are the main drivers in the southeast. This study underscores the effectiveness of ML, particularly ensemble learning, in flood modeling. It identifies vulnerable areas and contributes to improved flood risk management.