Machine learning driven multi-hazard risk framework for coastal resilience

Abstract

Coastal cities are increasingly vulnerable to climate-induced hazards, including sea-level rise, storm surges, coastal erosion, and extreme weather events. Buenos Aires, located along the Río de la Plata estuary, faces significant risks due to its dense population, vital infrastructure, and dynamic coastal environment. This study proposes a multi-hazard risk analysis framework, leveraging machine learning algorithms to model and predict risks in Buenos Aires’ coastal zones. Our approach integrates various datasets, including historical weather patterns, topographical and bathymetric data, urban development indices, and socio-economic factors, to assess hazard impacts' spatial and temporal distribution. The framework employs machine learning, specifically Random Forest (RF), to classify hazard-prone zones based on historical patterns and projected climate scenarios. The models are trained on datasets spanning the past three decades, capturing shifts in storm intensity, precipitation, and land use. Furthermore, Principal Component Analysis (PCA) is used to reduce data dimensionality and highlight key predictors of risk. By mapping hazard probabilities across different zones, our framework offers a nuanced view of hazard susceptibility, enabling local authorities to prioritize resources and adapt policies accordingly. This framework also introduces a novel adaptive learning component, where model parameters are periodically updated based on new environmental and urbanization data, ensuring that risk assessments remain relevant amidst changing climate conditions. Through this predictive model, stakeholders can anticipate hazard-prone areas, improving resilience planning and response mechanisms. This study thus provides a scalable, data-driven approach for addressing multi-hazard risk in Buenos Aires and offers insights into adaptive risk management strategies for coastal cities facing similar challenges globally.