To What Extent Can Simulation Optimisation be Used in Wildlife Reserve Design?

Establishing nature reserves is a key method for preventing biodiversity loss. This thesis addresses the reserve site selection (RSS) problem, which aims to select sites for nature reserves to ensure species survival. Specifically, it examines the extent to which simulation optimisation (SO) can be used in the RSS problem. The applicability and effectiveness of SO are evaluated by applying an SO method and its adaptations across three scenarios of a Grey Wolf (Canis lupus) RSS problem. The problem is formulated as a chance-constrained SO problem, with a deterministic objective that minimises conservation costs, subject to a probabilistic species survival constraint. This probability is estimated using a grey wolf simulation model that simulates the wolves’ birth, growth, dispersal and death in discrete time steps. The problem is solved using the sequential feasibility test procedure from Hong, Luo, and Nelson (2015), hereafter CCSB-F. Three scenarios of the RSS problem, each with different characteristics, are investigated in this research. Scenario 1 demonstrates how CCSB-F can tackle a basic problem. Several observations are made: first, since solution costs are trivial to obtain, computational effort (measured by the number of simulation runs) is required solely for establishing solution feasibility. Second, due to sampling error in simulation results, solution feasibility can only be assured subject to a ‘statistical guarantee’. Lastly, the computational effort required depends on how close the solutions are to the feasibility boundary and the required level of statistical guarantee. To address likely computational hurdles in more difficult versions of this problem, Scenarios 2 and 3 are designed to demonstrate and evaluate two solution space filtering approaches. The first approach ‘temporarily removes’ solutions with equivalent alternatives, identified based on the simulation model, without affecting CCSB-F’s statistical guarantee. Applying this approach to Scenario 2 (28 solutions) reduces computational effort by approximately 26% compared to using CCSB-F alone. When applied to Scenario 1 (28 solutions), it achieves an estimated savings of 40% while maintaining the same level of statistical guarantee. The second approach is a heuristic that uses expert knowledge to create solution dominance rules and then removes dominated solutions before applying CCSB-F. It reduces computational effort by approximately 80% in Scenario 3 (210 solutions) compared to using CCSB-F alone. When applied to Scenario 1 (28 solutions), the estimated computational savings is 60%. Even though it cannot guarantee to find the best solution in the entire solution space because it removes solutions, it still provides a statistical guarantee on the filtered solution space and the feasibility of the selected solutions. Although these estimates represent conservative lower bounds (as they do not fully account for the additional reduction in the number of replications per solution), they clearly demonstrate the potential of the proposed approaches to significantly reduce computational effort.