Merging computational fluid dynamics and machine learning to reveal animal migration strategies

Understanding how migratory animals interact with dynamic physical environments remains a major challenge in migration biology. Interactions between migrants and wind water currents are often poorly resolved models due to both the lack of high-resolution environmental data, understanding respond fine-scale structure environment. Here we develop generalizable, data-driven methodology study through complex environments. Our approach combines validated computational fluid (CFD) modelling animal tracking data decompose movements into two components, namely movement caused by forcing active locomotion. We then use flexible recurrent neural network model relate local conditions locomotion behaviour migrating animal, allowing us predict migrant's force production, velocity trajectory over time. apply this framework large dataset containing measured trajectories Chinook salmon section river California's Sacramento-San Joaquin Delta. show that is capable describing fish as function flow variables, it possible accurately forecast on which was not trained. After validating our model, can be used understand local-flow conditions, changes overall system change energetic cost depends space readily applied other species systems.