Malaise traps in biodiversity assessments - II. Evaluating changes in community structure

We will in this talk present applications of the bivariate Poisson lognormal distribution to evaluate the spatio-temporal dynamics of species abundances. To illustrate this approach we use samples from chironomid communities, caught by Malaise traps, along an alpine-boreal watershed in Central Norway. In the univariate case, a community of S species fits the lognormal distribution if the log of the abundances can be regarded as a sample from the normal distribution. By assuming a Poisson sampling distribution, the species counts can then be described by the Poisson lognormal distribution. When we extend this approach to consider samples from two communities jointly, either from two different locations or sampled at different times from the same location, we can assume that the log abundances in the pair of communities have the binormal distribution. The correlation of this binormal distribution will then serve as a robust measure of the similarity between communities, and can be estimated without any knowledge of the two sampling intensities. With correlation close to 1, e.g. for two parallel samples from the same community, the relative abundances of any species should be approximately the same in both samples although they may differ due to random sampling effects. Sampling methods that yield correlations close to 1 for parallel samples can thereby be expected to give good estimates also for the actual spatial or temporal change in community structure between samples. These correlation structures can then be modeled by e.g. distance in space or time, or by environmental covariates. Using samples from chironomid communities we found correlations between parallel samples ranging from 0.94 to 0.99 for a given week and location, indicating a high consistency in relative species abundances between samples. We will further present models for how the correlations decrease with time between samples from the same location.