INDIVIDUAL-BASED MODELING OF POPULATIONS WITH HIGH MORTALITY - A NEW METHOD BASED ON FOLLOWING A FIXED NUMBER OF MODEL INDIVIDUALS

Individual-based modeling of populations that undergo high mortality can be problematic. Large numbers of model individuals must be followed to ensure adequate numbers of survivors at the end of the simulation, but following large numbers of individuals can require excessive computer memory and computational time. Following a sample of individuals from the population partially addresses these problems for short-term simulations but not for model applications requiring long-term predictions. In this paper, we describe a resampling algorithm that permits the long-term simulation of populations undergoing high mortality. A fixed number of model individuals are followed, with each representing some number of identical population individuals. As each model individual dies, a donor individual is randomly selected from the survivors. The number of population individuals represented by the donor individual is adjusted to represent the loss of individuals due to mortality. The dead individual's attributes are then replaced with those of the donor individual. The high accuracy, reduced memory requirements, and comparable computational costs of the resampling algorithm are demonstrated using an individual-based population model of young-of-the-year striped bass. Differences between predictions without and with resampling were < 1.5% of the mean values for a suite of variables. Executable files for versions of the model using resampling were an order of magnitude smaller, and simulations required similar computational costs as versions of the model without resampling. Potential variations of the resampling algorithm to increase the accuracy of predictions of specific variables and to simulate spatially-explicit models are discussed.