Modelling Potential Fishing Zones in Kenya's Exclusive Economic Zones Using Machine Learning and Satellite-Derived Oceanographic Variables

Kenya's Exclusive Economic Zone (EEZ) in the Indian Ocean spans approximately 142,400 km², of which around 100,204 km² constitutes the industrial fishing zone beyond the 12 nautical mile territorial sea. Despite supporting commercially significant pelagic species — including skipjack tuna (Katsuwonus pelamis), yellowfin tuna (Thunnus albacares), bigeye tuna (Thunnus obesus), and swordfish (Xiphias gladius) — the spatial distribution of industrial fishing effort and environmentally suitable fishing zones within this zone has not been systematically characterized using quantitative, evidence-based approaches. This study predicts areas of high fishing suitability based on vessel presence records — a proxy for environmental quality — rather than direct measures of fish abundance. This study develops a validated spatial modelling framework to map potential fishing zones (PFZs) in Kenya's EEZ by integrating Global Fishing Watch (GFW) vessel tracking data with four satellite-derived oceanographic variables: sea surface temperature (SST), chlorophyll-a concentration (Chl-a), sea surface height (SSH), and sea surface salinity (SSS). A Random Forest (RF) classifier was trained on GFW fishing presence and pseudo-absence data for January 2019 and January 2021 (1,105 training points), with SST from NASA MODIS-Aqua, Chl-a from Sentinel-3 OLCI, SSH from the Copernicus DUACS multi-satellite altimetry product, and SSS from the Copernicus SMOS/BEC product — all harmonized to a 4 km spatial resolution across the industrial EEZ. The model was independently validated against a January 2022 dataset of 220 points withheld entirely from training. The Random Forest model achieved strong predictive performance on the independent validation dataset, recording an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.885, an accuracy of 79.5%, precision of 80%, recall of 78%, and an F1 score of 0.789. Sea Surface Height was identified as the most important predictor (variable importance = 28.1%), followed by Chl-a (24.8%), SST (23.7%), and SSS (23.5%) — reflecting the dominant role of mesoscale eddy dynamics in structuring the spatial distribution of industrial fishing effort in the Kenyan EEZ. Application of the validated model to the full EEZ generated a spatially continuous fishing suitability probability surface for January 2022. At a probability threshold of 0.3, approximately 22.8% of the Kenyan EEZ was classified as a potential fishing zone, with high-suitability areas concentrated between 40.5–41.5°E and 4–5°S, coinciding with a persistent SSH anomaly hotspot. The study demonstrates that the integration of open-access GFW vessel tracking data and multi-satellite oceanographic products within a Random Forest framework provides an effective, reproducible, and temporally transferable approach to PFZ prediction in Kenya's EEZ. The methodology and outputs are directly applicable to operational fisheries management by the Kenya Fisheries Service (KeFS), and the workflow provides a replicable blueprint for spatial modelling of industrial fishing effort distribution in other data-limited EEZ systems across the Western Indian Ocean. This upload includes the full dissertation PDF, reproducible analysis code (Jupyter Notebook and Python script), and final output maps (PFZ.jpeg). The code is available as a ZIP file and requires Python 3.x with the packages listed in requirements.txt.