Machine learning approach for estimating length-weight relationships in key shark species to support fisheries management

Abstract

Sharks are critical to marine ecosystems; however, they have become endangered because of overfishing over the years. In Taiwan, sharks are frequently caught as bycatch in tuna longline operations. Accurate data regarding length-weight relationships (LWRs) are essential to stock assessments and fisheries management. Nevertheless, most data regarding sharks in the Northwest Pacific Ocean are limited and outdated, which has introduced uncertainty into population assessments. In this study, we addressed these gaps by analysing the growth characteristics, specifically the relationships between body length and weight, of 10 key shark species in the Northwest Pacific Ocean. We applied both conventional statistical approaches (linear and nonlinear regression) and a more advanced method based on artificial neural networks (ANNs) in the analysis. The ANN-based method was significantly more accurate than the linear and nonlinear regression methods, as assessed using the mean squared error, mean absolute error, mean absolute percentage error, and root-mean-squared error. The ANN-based method was more effective in capturing complex, nonlinear growth patterns. For species exhibiting marked sexual dimorphism, i.e. distinct growth trajectories between male and female individuals, we derived sex-specific LWRs. The study findings highlight the utility of ANNs in fisheries management. ANNs can also be used by those performing stock assessments to refine biomass estimates and calculate catch limits. The insights provided by this study support enhanced conservation efforts and sustainable management of shark populations in the Northwest Pacific Ocean, facilitating the implementation of adaptive strategies that account for species-specific growth characteristics and population dynamics.