Flood risk prediction using synthetic finite difference data and machine learning

Abstract

The present study proposes a classifier model to assess flood risk in areas with irregular topography by integrating numerical methods and machine learning. Initially, a finite difference model is implemented to simulate water propagation using a simplified version of the Saint-Venant equations. Subsequently, a Random Forest Classifier is used to classify the risk as a function of the simulated water level. The model was evaluated using F1-score, confusion matrix, accuracy, and precision metrics. The study aims to contribute to the development of predictive systems for hydrological risk. This is achieved through hybrid models that facilitate early identification of vulnerabilities before hydrometeorological events.