Evaluating the Flexibility of Fine-Scale Individual-Based Movement Models Calibrated with a Genetic Algorithm

Population and community models are becoming increasingly spatially-explicit requiring the development of fine-scale movement sub-models. Calibrating the movement sub-models is challenging because movement data are often unavailable at a fine enough scale. We developed a simple individual-based model of a hypothetical fish population to explore the utility of genetic algorithms (GAs) in calibrating three types of movement sub-models (kinesis, neighborhood search, and game theory). The model simulated growth, mortality, and movement of 1000 super-individuals for 30 days with a 5-minute time step on a 540 x 540 cell grid with 5 m cells. We assigned each cell a predator and prey multiplier ranging from 0 to 1. The multipliers were then used to adjust maximum mortality and growth rates to generate realized mortality and growth rates in each cell. At the end of 30 days, we calculated the number of eggs produced by each super-individual based on their weight and worth. We selected and mutated the movement parameters from individuals with high egg production to repopulate the next generation. We repeated this procedure for 500 generations. We trained each movement sub-model on four predator and prey density grids with different combinations of patchy versus smooth gradients.  The grids included two with optimal habitat (predator density=0, prey density=1) and two with only sub-optimal habitat (forces growth-mortality trade-off). We tested each trained sub-model on the other three grids not involved in the training. Kinesis and neighborhood search sub-models, in both training and testing simulations, moved individuals away from areas of high predator density and toward areas of high prey density. In most cases, the GA converged on consistent movement parameter values regardless of the training grid, indicating a model calibrated on one grid can be used on a novel grid. One caveat to this finding is that training on grids that forced individuals to balance growth and mortality trade-offs produced different values for some of the kinesis movement parameters than grids that allowed individuals to maximize growth and survival. However, the differences in movement parameters did not produce substantially different spatial distributions of individuals on the grid. We are still in the process of developing and testing the game theory movement sub-model.   We plan to further test the sub-models with dynamics grids, and to evaluate the successful sub-models when imbedded in population and community models.