Evaluating the performance of individual-based animal movement models in novel environments

Simulating animal movement in spatially explicit individual-based models (IBMs) is both challenging and critically important to accurately estimating population dynamics. A number of different approaches have been developed that make different assumptions about how individuals move in their environment and use different mathematics to translate movement cues into a behavioral response. Properly calibrated movement models should produce realistic movement in both conditions encountered during calibration and in novel conditions; however, most studies to date have not tested movement models in novel conditions. We compared four distinct movement approaches or sub-models (restricted-area search, kinesis, event-based, and run and tumble) using an IBM loosely based on a small pelagic fish (e.g. Engraulidae) that simulated growth, mortality, and movement of a cohort on a 2-dimensional grid. We trained the sub-models with a genetic algorithm in one set of environmental conditions and then tested them in other three environments. The sub-models generally performed well in novel environments, except restricted-area search and event-based that needed to be trained in environments with gradients similar to the test environment. Also, run and tumble produced near-random distributions in all training environments except the one with the steepest habitat quality gradient, and it produced random distributions in all novel test environments. In selecting a movement sub-model, researchers should consider the assumptions of potential sub-models, the observed movement patterns of the species of interest, and the shape and steepness of the underlying habitat quality gradient.