Abstract: Species distribution models are commonly used to predict krill distribution but often suffer from data heterogeneity, which compromises the accuracy of modeling relationships with environmental factors. To address these challenges, our study employs geostatistical models with spatiotemporal effects using Integrated Nested Laplace Approximation and Stochastic Partial Differential Equations to analyze the distribution of Antarctic krill in the northern Antarctic Peninsula region. We utilize krill density data from scientific acoustic surveys conducted between 2013 and 2020 in summer. Our findings demonstrate that incorporating spatiotemporal effects in generalized linear mixed-effects models enhances model performance and accuracy, as indicated by lower Relative Root Mean Squared Error and higher cross-validation R-squared, thereby yielding more reliable predictions for krill density. Furthermore, our study reveals inter-annual variations in the spatial distribution of Antarctic krill, with sea salinity and flow velocity emerging as the most influential environmental variables. These results underscore the critical importance of considering spatiotemporal effects when predicting krill density, offering valuable insights for krill fisheries management.
Key words: Antarctic krill, distribution, spatiotemporal effects, Integrated Nested Laplace Approximation