Assessment of coastal vulnerability using AHP and machine learning techniques

Abstract

Vila Belmiro is a significant development in the coastal urban scenario and is nestled along the rugged cliffs of the southern coast of Brazil. Cyclones caused tremendous damage to the shore, destroying infrastructure, eroding the coast, flooding, and displacing people. Determining such low-lying regions' coastal vulnerability index is crucial for developing efficient mitigation strategies and enhancing readiness for calamities. In light of this viewpoint, the current study attempted to determine the Coastal Vulnerability Index (CVI) using the analytical hierarchical process (AHP). The study combined the Physical Vulnerability Index (PVI) and the Social Vulnerability Index (SVI) to create the Coastal Vulnerability Index. This study generated the LULC layer for Vila Belmiro using machine learning techniques and Random Forest based model. A component of the process was evaluating the parameters pairwise based on their importance and relevance to the study's objective. The CVI assessment classified 19.28 Sq.km of the coastline, including Solemar, São Vicente, Santo, and Ponta Da Praia, as very high risk. High-risk areas, such as São Vicente, Vila Caicara, and Guaruja, span 25.26 Sq.km. Medium risk was observed across 16.85 Sq.km, covering Praia Grande, Nova Mirim, and Balneario Praia. Low-risk regions, including Japui and Jarfim Guaiuba, extend over 13.54 Sq.km. Lastly, 10.78 Sq.km of the coastline, including Canto Do Forte, were classified as very low risk. These measurements highlight the varying levels of vulnerability along the coastline. The study concluded that coastal vulnerability zones were effectively identified and assessed, enhancing understanding and management of coastal risks. Planning and mitigating future cyclones can benefit from the identified vulnerable zones. The current study included machine learning, remote sensing, Geographic Information System (GIS), and advanced facilities that were spatially integrated with AHP approaches.