PARENT SESSION
Oral Session #67: Metapopulations, fragmented landscapes, and patch dynamics. Presiding: C. Kraft.
Thursday, August 9, 2001. 1:00 PM to 4:30 PM. Hall of Ideas F.
Fitting stochastic metapopulation models to patch occupancy data in the presence of sampling error.
TYRE, ANDREW1, POSSINGHAM, HUGH1, FIELD, SCOTT1,2, TENHUMBERG, BRIGITTE1, 1 2
ABSTRACT- We construct Markov chain and simulation models of a spatially implicit stochastic metapopulation. The model has two parameters, a local extinction rate and a local per patch colonisation rate. An observer samples the model metapopulation for 25 years with and without "sampling error". The observer uses observed transitions in the number of occupied patches to estimate the local extinction and colonisation rates that are, in turn, used to estimate the probability of metapopulation extinction. Even with sampling error an observer has a good chance of estimating the ratio of the local extinction rate to the per patch colonisation rate but not either parameter in isolation. However, the observer's ability to construct a model that accurately estimates extinction probability is very poor. Decreasing the observation series to 15 years of data severely degrades all parameter estimates. Increasing the observation series to 50 years of data improves estimates of the colonization rate, but does not significantly improve estimates of the local extinction rate. Furthermore, if the observer knows their sampling error and uses that knowledge, the accuracy of the colonization rate and local extinction rate estimates are improved, but only at intermediate time series lengths (ie. 25 years). For shorter time series the error correction generates worse estimates than ignoring the error, and for longer time series there is no improvement.
KEY WORDS: metapopulation, maximum likelihood estimation, sampling error