Vector Autoregressive Spatio-Temporal (VAST) models for biomass distribution of pacific cod Gadus macrocephalus (Gadidae) considering water temperature at the sea bottom in the West Bering Sea zone

Abstract

High statistical significance of water temperature at the sea bottom T and depth D for distribution of pacific cod in the West Bering Sea fishing zone is found in several tested models tuned on the data of bottom trawl surveys conducted in the period between 1977 and 2021 not deeper than 400 m. The vector autoregressive spatio-temporal (VAST) models which included nonlinear dependencies of cod catches from T and D have the best generalization ability. Correlation between predicted by VAST models and observed distribution density of cod in the test data set are higher than that in simpler models trained using the full set of data. The VAST models produce continuous time series of cod biomass with estimates of their uncertainty and statistical weights of the model configurations relative to the test data. After stacking with statistical weights and previously published estimates of biomass, the obtained time series allow to estimate dynamics of biological processes deviations from stationary assumptions and to estimate approximately the volume of “extra” cod not considered by these processes in the Bayesian State-Space Surplus Production Model. The portion of “extra” cod increased sharply above 40 % in 2016 and reached the maximum of 49 % by 2018, then began to decrease. Sharp changes in the main EOF modes for T are revealed in these years. Thus, the hypothesis of cod redistribution in the Bering Sea due to changes of the cold pool area at the bottom was tested for the first time by statistical methods in space. Due to high errors of forecasts based on analysis of biological processes only, there is impossible to predict accurately dynamics of the cod biomass without predicting the water temperature distribution at the bottom of shelf.