Regression models forecasting seasonal variations in Korean Delphinus delphis bycatch in its key coastal habitat

In Korean waters, common dolphins Delphinus delphis are mostly bycaught by coastal set net fisheries. Given that these fisheries operate nearly year-round in dolphin key coastal habitats, implementing flexible temporal conservation strategies based on reliable predictive model assessments of bycatch levels is essential to reduce bycatch, while minimizing associated social costs. We compared autoregressive integrated moving average (ARIMA) models with generalized linear mixed models (GLMMs) to develop robust regression models for forecasting seasonal variations in the bycatch of common dolphins in set nets around their key habitat in Korean waters from 2011 to 2022. A seasonal ARIMA model was identified as a significantly fitting model for seasonal common dolphin bycatch due to strong seasonality observed in the training and validation sets, which were created by splitting the entire data set for cross-validation. A negative binomial GLMM with 2 significant fixed effects (Pacific herring catch and sea surface temperature) and a random effect (year) was selected based on the lowest AIC C . Comparing the models based on RMSE values obtained from rolling cross-validations, the GLMM demonstrated greater accuracy in forecasting seasonal dolphin bycatch than the ARIMA model. During the testing period, the actual bycatch showed a significant and strong positive correlation with the bycatch forecasted by the GLMM. A more feasible conservation strategy for common dolphins, such as seasonal fishery restrictions in their key habitats during high-risk periods identified by the predictive model, may be implemented as a socially cost-effective measure to protect the species.