FloodGenome: interpretable machine learning for decoding features shaping property flood risk predisposition in cities

The study introduces FloodGenome, an interpretable machine learning model, to assess flood risk disposition in urban areas by analyzing hydrological, topographic, and built-environment features and their interactions. Utilizing data from the U.S. National Flood Insurance Program (2003–2023) across four metropolitan areas, it employs k-means clustering and a random forest model to classify and predict property flood risk levels. The model’s effectiveness is proven across different metropolitan areas, highlighting the importance of factors like elevation, and impervious surfaces in determining flood risk. FloodGenome’s analysis aids in evaluating future urban development impacts on flood risk and refining property flood risk assessments at a detailed level. This tool offers critical insights for flood risk management, supporting the development of urban design strategies to mitigate flood risks.