The importance of patch dynamics as an extrinsic driver of metapopulation persistence has only been appreciated from a mechanistic perspective over the last several decades. The integration of metapopulation and patch dynamic theories can enhance management effectiveness through greater understanding of the spatial and temporal processes that influence species dynamics. Optimal management strategies for species dependent on mid-successional habitat induce particular challenges. Absent natural disturbance regimes, managers are confronted with considerable uncertainty regarding the ideal placement and timing of habitat treatments to create appropriate patch-state configurations that will support a metapopulation over time.
The central Atlantic coast population of Florida scrub-jays is one of a handful of remnant metapopulations of this endangered species. Threatened by fragmentation and repression of natural fire disturbance regimes, the management of rare Florida scrub habitat using prescribed burning and mechanical cutting is the primary mechanism to benefit the persistence of jay populations.
In this paper, we describe the integration of an action-dependent habitat dynamics model with a spatially explicit occupancy model that, together, inform a sequential decision policy to maximize the long-term persistence of scrub-jays across a series of independently-managed habitat reserves.
We elicited reserve managers’ expert judgment to account for the lack of empirical observations for certain patch state-action combinations. These were used to develop semi-informative Dirichlet priors on habitat transitions using a Bayesian state-space formulation. An autologistic model accounted for the occupancy of neighboring sites, in addition to current habitat state, to explain extinction/colonization processes. The optimization approach applied a category count model to define transition probabilities for our set of categorical variables (defining habitat states, soil substrate type, occupancy state and management actions) and combine them into a joint distribution.
Posterior distributions returned on habitat transitions appropriately balanced expert knowledge, manager uncertainty and the observations of habitat transitions under alternate management actions. Occupancy model results supported mid-successional habitat producing highest occupancy probabilities, with earlier mid-succession vegetation more beneficial that later mid-succession states. Underlying vital rates were strongly conditioned on the occupancy status of surrounding territories, with spatial dependency estimated to 500m from the edge of a territory. Dynamic programming provided optimal state-dependent recommendations for habitat treatments. Somewhat unexpectedly, decision ‘rules’ were fairly simple, with actions influenced by substrate and habitat, less-so on site occupancy and least on neighborhood status, suggesting possible spatial independence of management. Most reserves predicted increased occupancy and reduced management input/cost over time.