Moore, C. T., M. J. Conroy and K. Boston. 2000. Forest management decisions for wildlife objectives: system resolution and optimality. Computers and Electronics in Agriculture 27:25-39.
Managers of forest wildlife populations make recurring management decisions based on incomplete knowledge of system states. For example, animal population estimates may ignore spatial structure that may influence population viability. We built a spatially-explicit model for a population of birds in a forested landscape. Rates of bird population growth within forest compartments and rates of bird dispersal among compartments were functions of stand age and basal area, compartment population size, and inter-compartment distance. Stand characteristics were imbedded in a dynamic model and assumed perfectly observable and under the complete control of managers. We constructed a genetic algorithm to search for the schedule and spatial distribution of silviculture to maximize total bird abundance at the end of a fixed planning horizon, under combinations of initial habitat and population distribution. We also found policies for a smaller set of population distributions that a manager may only presume to occur (e.g. birds equally distributed among stands), as when managers are only able to observe abundance and not spatial distribution. We investigated the effect of this loss of system resolution on optimality by examining differences in projected population sizes under the two types of policies. That is, we used the set of ‘presumed-state’ policies to project population size from each true initial system state, then we compared these to projections under the best policy for that state. For the planning horizon that we considered, loss in optimality was highly dependent on initial habitat state and on choice of presumed population distribution. Generally, loss in optimality and species extinction rate were both greater for habitat states that were initially poor than initially favorable. For some initial habitat states, population projections based on policies for presumed states often exceeded objective function values for known-state policies, suggesting that the genetic algorithm frequently fell short of finding bona fide optima.