The Effects of Spatial and Temporal Resolution in Simulating Fish Movement in Individual-Based Models

Many fisheries management decisions require predictions of spatial dynamics, and simulation of realistic movement is critically important for accurately representing population-level dynamics with spatially explicit individual-based models (IBMs). Movement approaches developed to date have been applied across a wide range of spatiotemporal resolutions. We compared four movement approaches or submodels (restricted-area search, kinesis, event based, and run and tumble) using an IBM (roughly based on Bay Anchovy Anchoa mitchilli and Northern Anchovy Engraulis mordax) that simulated growth, mortality, and movement of a cohort on a two-dimensional grid. We evaluated the submodels in 2.7- × 2.7-km environments at five resolutions defined by various cell sizes and time steps. We used a genetic algorithm to calibrate each movement submodel over a 300-generation training phase and then tested the mean movement parameters for a single generation in the training environment and a novel environment. Restricted-area search, kinesis, and event-based submodels had higher egg production than a random walk model (baseline, assuming no behavioral movement) across spatiotemporal resolutions in training and novel environments. The run-and-tumble submodel also had higher egg production than the random walk model but only under certain conditions. Although restricted-area, kinesis, and event-based submodels outperformed random walk at all resolutions, the submodels did not perform equally well across resolutions in terms of egg production and aggregation of model individuals in high-quality cells (i.e., those with high growth and low mortality). The variability in performance was due to the change in habitat quality experienced by model individuals from one time step to the next. Restricted-area and event-based submodels had higher egg production when model individuals experienced small changes in habitat quality, whereas the kinesis and run-and-tumble submodels performed better when model individuals experienced larger changes in habitat quality.