Populations divided across multiple sites, or subpopulations, presents a challenge to risk assessment in conservation biology because species of conservation concern are often data-limited and critical movement and spatial-structure data are usually missing. This hampers the use of detailed spatial models that require knowledge of movement and dispersal throughout a landscape. Current diffusion-approximation methods for estimating population viability parameters require only time-series data but cannot incorporate data from multiple subpopulations where those subpopulations are interdependent. In this talk, we extend univariate state-space theory and present an analytical framework based on multivariate state-space models for the analysis of time-series data of abundance from multiple sites within a population. State-space models incorporate a stochastic model of the population growth process with a separate stochastic model of the observation process. State-space models allow ecologists to partition the variance in population time series into process error which is driving the population dynamics and observation error which is corrupting the observations of the population process. A multivariate state-space approach allows a stochastic metapopulation process to be modeled as multiple population processes that are inter-dependent.
Results/Conclusions
We demonstrate how the multivariate state-space framework can be used to model a wide-variety of alternative structures within the metapopulation. For example, a metapopulation that is a collection of independent populations with different subpopulation growth rates versus a metapopulation that is a collection of dependent subpopulations with the same population growth rates. We demonstrate how a Kalman smoother can be used to obtain optimal estimates of the true, unseen, population abundance within each subpopulation—given a particular assumed structure within the metapopulation. We use the Kalman smoother within an Estimation-Maximation algorithm to determine maximum-likelihood estimates of the stochastic growth rates and process variances within the subpopulations. We apply the multivariate state-space approach to data from an endangered metapopulation of salmon in the
