Enhanced Hazard Mapping for Flood and Landslide Risks Using Memetic Programming and Machine Learning Techniques to Support Sustainable Development Goals

Abstract

Identifying regions susceptible to co-occurring flash floods and landslides is essential for devising effective mitigation strategies and minimizing future impacts, particularly in urbanized river basins within tropical and subtropical climates. This study introduces Memetic Programming (MP), an innovative data-driven technique, alongside Ridge Regression (RR) and Support Vector Regression (SVR), to produce highly accurate flood and landslide susceptibility maps for the Nishikigawa and Takatsu River basins in the western Chugoku region of Japan. MP, which is rarely applied in combined flood and landslide susceptibility mapping, demonstrated superior predictive performance and greater spatial consistency than conventional models, enabling more precise identification of high- and low-risk zones. MP performed risk assessments, identifying 75% of the basins as low-risk for flooding and only 12% as high-risk, thereby providing clearer guidance for targeted interventions in the future. In contrast, SVR tended to spatially overestimate moderate- and high-hazard zones, likely because of suboptimal parameter tuning and its sensitivity to outliers. Beyond its technical accuracy, MP offers a robust and flexible approach that enhances hazard mapping and supports informed decision making. The findings highlight that implementing mitigation measures based on MP results can directly contribute to achieving four primary SDGs and support three additional SDGs. Overall, this study underscores the potential of MP to advance sustainable risk management in urban river basins globally.