Decision support for wildfire asset protection: A two-stage stochastic programming approach

During uncontrollable wildfires, decision-makers dispatch vehicles for tasks aimed at reducing the hazard to key assets. The decision-making process is complicated by the need for vehicle capabilities to match asset requirements within time windows determined by the progression of the fire. This is often further complicated by a wind change that is expected but with uncertainty in the timing. This paper aims to provide a decision support approach for determining plans for the deployment of resources under various scenarios. To this end, we solve the Asset Protection Problem (APP) in the context of Australia's Black Saturday bushfires by developing a two-stage stochastic model. In this problem, we consider uncertainties in the timing of a change in wind velocity in defining various scenarios. We present a dynamic rerouting model and an Adaptive Large Neighbourhood Search (ALNS) algorithm to solve the model in a time efficient manner for decision-makers. A new set of instances were generated using realistic parameters. Thereafter, we evaluate the performance of the proposed approaches through extensive computational experiments. We solve both the two-stage stochastic program and the more limited dynamic rerouting model exactly for small instances. We observe that the restrictions of the dynamic rerouting model yields solutions within a few percent of the two-stage stochastic program. Moreover, the ALNS solution is a good approximation to its exact equivalent but with faster solution times. After further tests, it became apparent that with larger asset numbers the ALNS is the more practical method for operational purposes.