On the sampling design of spatially explicit integrated population models

Abstract

It is important to understand metapopulation dynamics and underlying demographic processes in heterogeneous landscapes. Traditionally, demographic parameters are estimated using capture–recapture data that can be difficult to collect. Spatially explicit dynamic N-mixture models allow inference for demographic parameters, including dispersal, using count data of unmarked animals, but these models have only been shown effective under constant demographic parameters and dispersal between adjacent local populations. In this study, I aimed to compensate the weakness of spatially explicit dynamic N-mixtures and multistate capture–recapture models by jointly analysing count and capture–recapture data. This spatially explicit integrated population model allows for spatiotemporal variation of demographic parameters in relation to environmental and density covariates and dispersal between any local populations. I conducted simulations to evaluate this model (a) for species with distinct life histories under different detection and capture probabilities, (b) when spatial sampling intensity varied, (c) when the length of survey period varied, (d) when the robust sampling design was adopted or not and (e) when auxiliary information is partially available. I also provided an empirical example of Gadwall Mareca strepera metapopulation dynamics in North American prairies. The results showed that the model provided unbiased parameter estimates under a variety of ecological and sampling conditions, even when the spatial sampling intensity of capture–recapture survey was low (20% of the patches) with the complement of count data (≥60% of the patches). Also, the model only required a relatively short survey period (6~8 years) to provide unbiased inferences. The robust sampling design was not necessary for the model to provide unbiased inferences when spatial counts were intense, but became critical when spatial counts were sparse. Parameter estimates remain unbiased when auxiliary information is partially available. The model showed that Gadwall had low emigration probability (11.8%) but could disperse more than 200 km. Based on the results, I provide recommendations about the trade-off between spatial sampling intensity, length of survey period and the use of the robust sampling design when applying this model in real-world studies. The model could have wide applications in the interface of metapopulation ecology and landscape ecology.