A two-stage stochastic approach for the asset protection problem during escaped wildfires with uncertain timing of a wind change

TL;DR

This paper introduces a novel two-stage stochastic model for asset protection during wildfires, explicitly accounting for uncertain timing of wind changes, and demonstrates its effectiveness over dynamic rerouting in case studies.

Contribution

It is the first mathematical model to incorporate uncertainty in the timing of wildfire scenario changes for asset protection planning.

Findings

The model outperforms dynamic rerouting in generating better deployment plans.

Solutions are achievable within operational time for realistic problem sizes.

The model's computational performance is validated with realistic wildfire data.

Abstract

Wildfires are natural disasters capable of damaging economies and communities. When wildfires become uncontrollable, Incident Manager Teams (IMT's) dispatch response vehicles to key assets to undertake protective tasks and so mitigate the risk to these assets. In developing a deployment plan under severe time pressure, IMT's need to consider the special requirements of each asset, the resources (vehicles and their teams), as well as uncertainties associated with the wildfire. A common situation that arises in southern Australian wildfires is a wind change. There is a reliable forecast of a wind change, but some uncertainty around the timing of that change. To assist IMT's to deal with this situation we develop a two-stage stochastic model to integrate such an uncertainty with the complexities of asset protection operations. This is the first time a mathematical model is proposed which considers uncertainty in the timing of a scenario change. The model is implemented for a case study that uses the context of the 2009 Black Saturday bushfires in Victoria. A new set of benchmark instances is generated using realistic wildfire attributes to test the computational tractability of our model and the results compared to a dynamic rerouting approach. The computations reveal that, compared with dynamic rerouting, the new model can generate better deployment plans. The model can achieve solutions in operational time for realistic-sized problems, although for larger problems the sub-optimal rerouting algorithm would still need to be deployed.