Application of the machine learning method to estimate the biomass of pacific cod in the North Kuril zone

Abstract

The biomass of pacific cod (Gadus macrocephalus) in the North Kuril fishing zone is estimated using a multifactorial approach, with evaluation of uncertainty. For this purpose, the density of fish over entire zone is restored using the data on density obtained in 2022 compared with the data of previous surveys and fishery data obtained in 2021 and earlier, converted to the same scale, with application of the machine learning method, as the random forest in the multiple imputation by chained equations procedure (MICE). The coefficient of the restored data determination with out-of-bag (test set) data was > 0.8 with the data of scientific survey in 2021 and > 0.5 with the data of Danish seine observations. The cod density variance in MICE data was in 82 % lower than in the data of the scientific survey; therefore the biomass estimation with MICE data has lower uncertainty than that one calculated just from the mean density in survey. The study showed insignificant difference of the cod biomass in 2021 and 2022. Spatial segregation is revealed for fishing gears used for the pacific cod fishery. There is proposed to extend the list of fishing gears and to expand the study area for reducing possible bias in the biomass estimation due to large area of extrapolation.

Keywords

• biomass
• pacific cod
• north kuril fishing zone
• random forest
• mice