SwimmingIndividuals: A High-Performance Agent-Based Model for Marine Ecosystems

Mechanistic models are essential for projecting ecosystem responses to novel conditions, yet the application and biological realism of agent-based models (ABMs) has often been limited by computational constraints. This paper introduces SwimmingIndividuals, an open-source agent-based model (ABM) framework that advances our ability to simulate complex biological processes at large ecological scales. The software's main contribution is its suite of mechanistic sub-models that govern the full life cycle of each agent, including physics-based visual predation, adaptive behaviors such as diel vertical migration, and a flexible bioenergetics engine with multiple taxa-specific equations. These detailed biological processes are made computationally tractable for large populations through a hybrid CPU/GPU architecture written in the Julia language. We demonstrate the framework’s capabilities through three targeted simulations. The model successfully reproduced complex ecological phenomena as emergent properties, including diel vertical migration, cetacean diving patterns, size-based trophic dynamics, population-level growth curves, and stock-recruitment relationships. Furthermore, long-term simulations generated realistic population dynamics and quantified the impacts of different fishery harvest control rules. By coupling high-resolution biological realism with technical scalability, SwimmingIndividuals provides a powerful and flexible tool for a wide range of ecological inquiries. It can be used to conduct in-silico experiments into the ecosystem-scale impacts of environmental disturbances, test fundamental ecological theory, and evaluate the efficacy of complex, ecosystem-based management strategies. This framework advances our ability to build a more mechanistic understanding of marine ecosystem resilience in a changing world.