Water Mass Fingerprint Detection Using Machine Learning for the Sustainability of Capture Fisheries Economy in Southeast Sulawesi

Abstract

Capture fisheries in Southeast Sulawesi play a strategic role in supporting regional food security and economic development. However, its utilization still faces challenges due to the limited availability of accurate and sustainable spatial data. This study aims to map the spatial distribution of fisheries potential in the waters of Southeast Sulawesi using a quantitative machine learning approach. The method employed is the Multi-Layer Perceptron (MLP) algorithm, trained using key oceanographic variables including sea surface temperature, chlorophyll-a, salinity, ocean currents, sea level height, seafloor topography, and vessel activity data from the Vessel Monitoring System (VMS). All datasets were extracted through Google Earth Engine (GEE), processed in Python (scikit-learn) through normalization, outlier handling, and minority class oversampling, and visualized as regional fingerprint maps in QGIS. The results reveal a high degree of spatial heterogeneity, with the highest fisheries potential concentrated in the southern and southeastern parts of Southeast Sulawesi. The economic valuation analysis of six fisheries potential classes identified a total area of approximately 15.8 million hectares, with an estimated economic value of IDR 1.18 trillion. The highest class (Class 5) accounts for 44% of the total area and contributes about IDR 376.6. The integration of machine learning and spatial data provides a comprehensive understanding of the marine and fisheries potential distribution, serving as a strategic foundation for developing sustainable fisheries management policies in Southeast Sulawesi.