Estimating the Spatial Dynamics of Plant Recruitment using Approximate Bayesian Computation

Abstract

We present a general statistical framework to infer processes (underlying ecological mechanisms) from patterns (observed arrangements in nature) in spatial ecology. We demonstrate our method by investigating the process of plant recruitment, how new individuals join a plant community, combining seed dispersal and environmental factors that determine the success of seeds germinating and surviving to juvenile maturity (establishment). Observed data includes seed and seedling counts at discrete spatial locations for the tree species slash pine (Pinus elliottii). The patterns in the data are described by their spatial correlation and we incorporate these correlations into historically used spatial models. We use a Bayesian simulation-based inference algorithm (Approximate Bayesian Computation, ABC) to estimate model parameters. Interest in ABC and Bayesian inference methods is growing in ecology, but they still remain behind classic approaches. Our results highlight techniques to validate the method to ensure accuracy and detect issues. Simulation tools are discussed to improve computational efficiency. We conclude with ABC parameter estimates that capture valuable spatial information ecologists can interpret. A small comparison study with classic likelihood-based parameter estimates is performed to illustrate the flexibility and informativeness of ABC. Our method is purposefully kept general to make it applicable to many spatial ecological problems.