Modeling Metacommunities: A comparison of matrix and agent-based methods with empirical data

Background/Question/Methods The importance of space has long been recognized by ecologists in the form of metapopulation theory, and more recently in the theory of metacommunities. These models assume that stable local coexistence is rare, and instead long-term coexistence is achieved through trade-offs such as competition/colonization ability. Using two different modeling approaches, we model a simple three species metacommunity of two interacting competitors and one predator. In this community species 1 is a poor competitor but good colonizer and species 2 is a good competitor but poor colonizer. The predator species over exploits species 2 and can coexist with species 1. In our first approach we use community assembly rules to create a matrix of transition probabilities. This matrix is similar to a Leslie matrix but instead of transition probabilities between age classes. The transition probabilities are between one of eight possible community states consisting of all possible combinations of the three species. In a second approach we apply those same assembly rules to an agent based model, using them to define the behavior of three types of agents. We then compared our results to a bi-weekly sampled 25-year three species tree-hole mosquito metacommunity data set.

Results/Conclusions We calculated the average of the proportion of time over the last 25 years our tree-hole mosquitoes were in any one of the possible eight states. Our matrix model accurately predicted the proportion of time a patch was empty, had only species 2, and had the community of species 2 and a predator. It over predicted the proportion of time only species 2 was present. Our agent based model was also accurate, but for different categories. It predicted the proportion of time for a community of only species 1, of species 1 and species 2 and of species 2 and a predator. Our agent based model also over predicted the proportion of time the three species community was present. Our results demonstrate that different modeling approaches built with the same species interaction rules can lead to different results. These differences do not come from the biological assumptions, but instead from the different assumptions made by the modeling methods about time and space.