Environmental Drivers of Habitat Suitability for Shortbill Spearfish (Tetrapturus angustirostris) in the Pacific Ocean: A Comparison of Single and Ensemble Models.

Understanding habitat suitability and environmental preferences of shortbill spearfish (Tetrapturus angustirostris) is essential for assessing and mitigating bycatch risk in pelagic longline fisheries. Using observer data collected by the Chinese Pacific tuna longline fishery, including 19,932 longline sets collected during 2010-2021, we compared five individual modelling algorithms and three ensemble strategies to predict habitat suitability across the Pacific Ocean and identify key environmental drivers. Model performance was evaluated using repeated 10-fold cross-validation with three repetitions. The stacking ensemble achieved the highest predictive performance (AUC = 0.872; TSS = 0.591), followed closely by random forest (RF; AUC = 0.869), although the absolute difference between these two models was small. A sensitivity analysis including log-transformed observed hooks resulted in only minor improvements in RF and stacking performance and did not materially alter the main habitat-suitability patterns. Both generally outperformed the other individual and weighted ensemble models. Permutation-based importance analyses consistently identified salinity as the strongest environmental predictor of habitat suitability, followed by chlorophyll-a concentration, distance to shore, and dissolved oxygen concentration. Partial dependence analyses showed that the stacking ensemble produced smoother and more readily interpretable responses to environmental gradients, particularly salinity and dissolved oxygen, whereas RF placed greater importance on a limited number of dominant predictors. Highly suitable habitat was primarily predicted in subtropical offshore waters of the southeastern Pacific (10° S-25° S, 100° W-130° W), which were characterized by relatively high salinity, moderate dissolved oxygen concentrations, and low chlorophyll-a concentrations. A secondary area of elevated suitability was identified in the central North Pacific (160° E-180° E, 15° N-25° N). Overall, the stacking ensemble provided robust predictions of shortbill spearfish habitat suitability, whereas RF remained useful for identifying dominant environmental correlates. The predicted habitat maps provide a spatial basis for identifying potential bycatch-risk hotspots and supporting adaptive management strategies, such as spatially targeted monitoring and mitigation measures in tuna longline fisheries.