Logbook fish yield data improving species distribution model forecasting for purse seine fishery in the Western and Central Pacific Ocean

Species distribution models (SDMs) are widely used to examine the shifts in fish school distribution under environmental changes and can assist in fisheries management and conservation. Although SDMs commonly incorporate multiple marine environmental variables, their forecasting performance can be influenced by spatial variability and the type of fishing data. This study therefore compared the forecasting performance of SDMs, specifically the maximum entropy (MaxEnt) algorithm, based on logbook fish yield data and fishing effort data estimated from Automatic Identification System (AIS), each aggregated at spatial resolutions of 0.01∘,0.1∘,and0.5∘. The matched dataset of fishing location and satellite-derived factors in the Western and Central Pacific Ocean during 2012 to 2019 was analysed. The results demonstrate that the MaxEnt model based on fish yield data achieved a prediction accuracy exceeding 88%, and consistently outperformed the models trained with AIS-based fishing effort data, particularly during periods of fish aggregating device closures. The catch-based model also exhibited the highest spatial agreement with actual fishing positions, more than half of the fishing events occurring within 5 km of the predicted fishing grounds when using fine-resolution (0.01°) free-swimming (FS-only) catch records. Additionally, the contribution of environmental variables in the MaxEnt model based on fishing effort data varied with spatial scale, whereas the sea surface temperature consistently remained the most influential predictor in the fish yield-based model. These findings suggest that fine-scale fish yield data, especially FS-only logbook records, can significantly enhance prediction accuracy and provide more reliable habitat forecasts.