Optimal control of an invasive species with imperfect information about the level of infestation

The presence of invasive species is often not realized until well after the species becomes established. Discovering the location and extent of infestation before the invasive species causes widespread damage typically requires intensive monitoring efforts. In this paper, we analyze the problem of controlling an invasive species when there is imperfect information about the degree of infestation. We model the problem as a partially observable Markov decision process in which the decision-maker receives an imperfect signal about the level of infestation. The decision-maker then chooses a management action to minimize expected costs based on beliefs about the level of infestation. We apply this model to a simple application with three possible levels of infestation where the decision-maker can choose to take no action, only monitor, only treat, or do both monitoring and treatment jointly. We solve for optimal management as a function of beliefs about the level of infestation. For a case with positive monitoring and treatment costs, we find that the optimal policy involves choosing no action when there is a sufficiently large probability of no infestation, monitoring alone with intermediate probability values and treatment alone when the probability of moderate or high infestation is large. We also show how optimal management and expected costs change as the cost or quality of information from monitoring changes. With costless and perfect monitoring, expected costs are 20-30% lower across the range of belief states relative to the expected costs without monitoring.Invasive species Monitoring Optimal management Dynamic programming Partially observable Markov decision process