Machine Learning Enhances Flood Resilience Measurement in a Coastal Area – Case Study of Morocco

Abstract

Understanding the characteristics contributing to enhancing flood resilience is a matter of urgency in managing urban areas, especially for developing countries, given the challenges imposed by climate change, social growth and urbanization. Identifying resilience metrics remains challenging, mainly because the concept is relatively new, methodological approaches are almost absent, and many types of resilience-related data are still unavailable. A number of indices for flood resilience have been introduced in the literature, typically based on clustering algorithms that allow complex behaviors to be mapped to specific levels of resilience. Consequently, the qualitative aspects of such indices are highly sensitive to the availability, quality and heterogeneity of data. Historically, this assessment has often been performed using rather simple algorithms such as Principal Components Analysis (PCA). Whilst they allow reliable resilience metrics in some areas, their use in a complex urban system such as the northern coastal area in Morocco is arguable. In the present study, we introduce an advanced Machine Learning (ML) method, namely the Self-Organizing Map (SOM), to build a Flood Resilience Index (FRI). Compared to classical methodologies, this present technique allows an improved assimilation of the complex relationship between data representing the social, economic and physical status of the area and resilience level. The success of this approach is mainly due to the ability of SOM to deal with complex, heterogeneous and sparse datasets. The results demonstrate great potential for such algorithms to shed light on systems that are too complex for classical techniques.

Keywords: resilience, flooding, composite indicators, machine learning, flood resilience index