Spatiotemporal Prediction of Tuna Fishing Zones in the Arabian Sea and Western Indian Ocean: A Machine Learning Framework Integrating Remote Sensing and Oceanographic Drivers

Abstract. Tuna fisheries in the Arabian Sea and Western Indian Ocean are vital for regional economies and global food security, requiring advanced tools for sustainable management. This study introduces a novel framework for Potential Fishing Zone (PFZ) identification by integrating multi-sensor remote sensing data with machine learning. A Random Forest model was developed using eight years (2014–2021) of satellite-derived oceanographic variables—sea surface temperature, salinity, chlorophyll-a, and current velocities—alongside in-situ fisheries data from Oman's Exclusive Economic Zone. The model achieved perfect classification in cross-validation and 97% accuracy on test data. Thermohaline parameters dominated predictions, with sea surface temperature at 10m depth and surface salinity contributing >80% of explanatory power. Spatial validation showed strong agreement with observed fishing activity (sensitivity: 0.98; specificity: 0.97), capturing seasonal patterns like monsoon-driven productivity and mesoscale eddies. While 85% of predictions fell within ±0.25 error thresholds, coastal discrepancies highlighted unresolved bathymetric and fishing pressure effects. The framework effectively tracked sub-mesoscale habitat dynamics across a 1,360 km domain. Key contributions include: (1)a transferable ML architecture for PFZ forecasting, (2) evidence-based prioritization of monitoring parameters, and (3) pathways for improvement via higher-resolution coastal data. This work advances tuna resource management and demonstrates the synergy of remote sensing and machine learning in marine spatial ecology.