Strategic optimization of fishing predictors enhances target species catches while minimizing impact on marine biodiversity

Abstract Small-scale fisheries are vital to coastal communities, providing food security, economic livelihoods, and cultural continuity, yet operational inefficiencies often result in elevated bycatch with negative consequences for marine biodiversity. Addressing this challenge requires understanding the conditions under which target catches are maximized while explicitly considering the trade-offs for non-target species. Here, we used a machine-learning framework to analyse the subtropical Sparisoma cretense fishery of the Canary Islands, aiming to identify combinations of fishing operations and environmental conditions that maximize catch efficiency and to evaluate their consequences for bycatch. Using an Extreme Gradient Boosting model, we examined relationships between S. cretense catch biomass and 15 fishing-related and environmental variables, explaining 65% of the observed variance in catches. The analysis highlighted average fishing depth, net height, number of nets, net length, and fishing month as the most influential predictors shaping catch outcomes. Optimization of these five key predictors could increase S. cretense catches by up to 815% while simultaneously reducing the impact on bycatch diversity. This approach decreases overall bycatch species richness by 40% and completely eliminates the impact on threatened species. By explicitly linking catch efficiency to biodiversity outcomes, our study demonstrates how machine-learning models can define operational windows that guide fishing decisions, revealing inherent trade-offs between target catch success and bycatch, and providing a transferable, evidence-based framework for supporting more balanced and sustainable management of small-scale fisheries.