Using history matching to speed up management strategy evaluation grid searches

Management strategy evaluations (MSEs) are a valuable tool for assessing the performance of different management strategies under varying ecological and economic conditions. They can be used to optimise the management procedure. However, MSEs are computationally demanding, especially as the complexity of operating models increases. I propose a scheme based on history matching to speed up grid searches when looking to optimise control parameters in harvest control rules. The approach uses an emulator, a fast statistical model that mimics the MSE, to exclude points that have a high probability of not being the optimal. The emulator is updated until only one point remains. The methods are introduced and demonstrated using an MSE on North Sea cod. Typically, the method found the optimal solution with only 9% of the grid evaluated in 12% of the clock time. I compared history matching with alternative optimisation algorithms, including hill climbing, simulated annealing and the genetic algorithm, and found that history matching consistently outperformed the alternatives.