Predicting Barents Sea Cod Stock Dynamics Using Oceanographic Data and Neural Network Analysis

ABSTRACT Fluctuations in fish populations are driven by recruitment, growth, and mortality—processes heavily influenced by environmental variability, particularly in highly dynamic marine ecosystems such as the Barents Sea. The complex, nonlinear relationships between various environmental drivers and fish stock dynamics remain challenging to capture. Building upon earlier works that predict Barents Sea cod total stock biomass using lagged hydrographic variables, we combine a wide range of updated data and machine learning techniques to improve predictions. Specifically, we employ neural networks, which excel at modeling intricate, nonlinear patterns, using hydrographic variables and fishing mortality as inputs. Our results highlight the potential of machine learning to complement conventional methods, such as linear regression models, in fisheries science, providing more accurate predictions of stock biomass in response to environmental and anthropogenic pressures.