Commercial catch prediction of chub mackerel in the Hyuga-nada, southern Japan, using random forest regression

Abstract

This study predicted the catch of spawning spring chub mackerel that migrates to spawn in the Hyuga-nada Sea, which substantially impacts the business of the medium-sized purse-seine fishery in Miyazaki Prefecture. After selecting the explanatory variables that are important for the conditions of migration to the Hyuga-nada Sea using the Boruta feature selection algorithm, a catch forecasting model (R²=0.79) was constructed using random forest regression. The explanatory variables selected for the model were the sea surface temperature (SST) of the Kumano-nada Sea and the distance of the Kuroshio current from Cape Ashizuri and Cape Toi. The influence of these explanatory variables on the model is examined in conjunction with partial dependent plots, regional SSTs, and previous findings. The results indicate that the optimum conditions for spring catches of chub mackerel in Miyazaki Prefecture’s medium-sized purse-seine fishery during high abundance period of the Pacific stock of chub mackerel are the typical Kuroshio large meander when warm water spills over into the Kumano-nada Sea and the Kuroshio comes in contact with the waters off Cape Ashizuri and Cape Toi.