Spatio‐temporal model and machine learning method reveal patterns and processes of migration under climate change.

Aim: Despite extensive studies of phenological shifts in migration by climate change and driving factors of migration, a few issues remain unresolved. In particular, little is known about the complex effects of driving factors on migration with interactions and nonlinearity, and partitioning of the effects of factors into spatial, temporal, and spatio‐temporal effects. Here, we aim to elucidate migration pattern as well as its driving factors under climate change. Location: Western North Pacific. Taxon: North Pacific spiny dogfish Squalus suckleyi. Methods: We first examined long‐term changes in the timing and geographic location of migration by applying the Barrier model, a spatio‐temporal model, to c. 5‐decade time series data (1972–2019) for the presence/absence of spiny dogfish in the western North Pacific. We then evaluated the spatial, temporal and spatio‐temporal effects of driving factors (fish productivity, sea surface temperature [SST], depth and magnetic field) on seasonal occurrence patterns using a machine learning model and an interpretable machine learning technique. Results: The migration area did not change over c. 5‐decades, whereas the migration timing advanced by a month after 2000. The spatial effects of magnetic field and depth were consistently large and the spatial and spatio‐temporal effects of SST increased in the migration season, even though the temporal effect of SST was consistently weak. Main Conclusions: The migration area of spiny dogfish was stable over time because of the effect of magnetic field and a strong preference for submarine topography, whereas the migration timing advanced as a result of tracking a suitable location based on SST, which increased sharply after 2000. Therefore, temperature and other factors simultaneously influence migration under climate change, highlighting the importance of considering both biotic and abiotic factors and understanding the underlying processes in predicting future impacts of climate change on species distribution. [ABSTRACT FROM AUTHOR]