SwimmingIndividuals: A high-performance agent-based model for marine ecosystems

Because they are based on established theory, mechanistic models are essential for projecting ecological responses to novel conditions. Yet mechanistic agent-based model (ABM) applications have often been limited by computational constraints. This paper introduces SwimmingIndividuals, an open-source ABM framework that advances our ability to simulate physiological processes, animal movement, higher trophic level marine ecosystem dynamics, and fisheries removals at large ecological scales. The software's key contributions are a suite of mechanistic sub-models governing the full life cycle of each agent, including visual prey detection, adaptive behaviors such as diel vertical migration (DVM), a flexible bioenergetics engine with three taxon-specific metabolic architectures, and a spatial pathfinding algorithm for habitat-seeking movement. These processes are made computationally tractable through a hybrid CPU/GPU architecture written in Julia. Unlike prior ABMs that use fixed perception radii or single-equation bioenergetics, SwimmingIndividuals enables emergent ecological dynamics across populations of superindividual agents. We demonstrate the framework through three Gulf of America simulations. The model reproduced DVM, cetacean diving patterns, size-based trophic shifts, and von Bertalanffy growth curves consistent with published King Mackerel stock assessments. Long-term simulations generated realistic population dynamics and quantified impacts of fishery harvest control rules across five fleets. SwimmingIndividuals provides a powerful, flexible, open-source platform for mechanistic marine ecosystem research and ecosystem-based fisheries management.