Incorporating a Machine Learning Approach into an Established Decision Support System for Coastal Vulnerability in the Eastern Baltic Sea

Barzehkar, M.; Parnell, K., and Soomere, T., 2024. Incorporating a machine learning approach into an established decision support system for coastal vulnerability in the Eastern Baltic Sea. In: Phillips, M.R.; Al-Naemi, S., and Duarte, C.M. (eds.), Coastlines under Global Change: Proceedings from the International Coastal Symposium (ICS) 2024 (Doha, Qatar). Journal of Coastal Research, Special Issue No. 113, pp. 58-62. Charlotte (North Carolina), ISSN 0749-0208. Assessing coastal susceptibility to hazards is a critical input for coastal management for countries with long and greatly varying shorelines, such as Estonia. The use of different decision support systems eventually contributes to a better understanding of the functionalities and limitations of tools. The conventional multi-criteria decision analysis (MCDA) techniques using fuzzy logic, analytical hierarchy process, and weighted linear combination were applied to map the coastal vulnerability index (CVI) for the Estonian coast. As a promising alternative, we used machine learning based on the Random Forest (RF) algorithm based on the same set of 16 vulnerability parameters. The results of the MCDA methodology showed that about 47% and 42% of the Estonian coast were in the classes of low and moderate vulnerability, respectively. The estimate based on the RF technique indicates a much larger proportion of the coast (65.6%) in the class of low vulnerability and 24.9% in the class of moderate vulnerability. Almost all vulnerable locations identified by machine learning were in the south of the country. The differences may reflect the larger role of the opinions of experts in the MCDA analysis, providing a more detailed consideration of the local situation and the integration of qualitative data.