The problem of wildfires is one of the major challenges around the world, causing immense economic losses. The primary causes of wildfires worldwide vary from country to country. In the U.S. most wildfires are human caused, e.g., due to unattended campfires, equipment malfunctions; the remaining fires are a naturally occurring phenomenon. Fuel treatments form the first line of defense against large-scale wildfires. The practice of fuel treatment involves removing all or some of the vegetation (i.e., fuel) from a landscape to reduce the potential for fires and their severity. Several fuel treatment options have been applied in practice including prescribed (controlled) burning and a variety of possible mechanical manipulations (e.g., thinning, pruning), as well as their combinations. In general, fuel treatments do not completely prevent wildfires. Their main purpose is to acknowledge that wildfires cannot be fully eliminated, aiming instead for their containment and, to some extent, controlled management.

In this project, we developed several mathematical optimization models, which determine appropriate locations and types of treatments to minimize expected losses from wildfires. The resulting mathematical models were developed under different assumptions on the data availability, the types of decisions that can be made by fuel treatment planners and their possible overall objectives (e.g., risk-neutral or risk-averse). To reflect actual decisions made by the practitioners, our models involve integrality restrictions for the decision variables (e.g., "yes" or "no" decisions). Such models are known to be computationally challenging. Hence, we explored their difficulty from the theoretical complexity perspective. It turned out that our models are very difficult to solve both theoretically and practically and cannot be handled by off-the-shelf solvers. Hence, we designed specialized solution methods for general and some special cases of the developed models. We performed experiments with semi-synthetic and real-life instances to illustrate the performance of our models. We also explored numerically the fundamental structural properties of the fuel treatment solutions arising from our models. Additionally, we conducted various types of sensitivity analysis on the performance of the obtained policies and illustrated the value of the proposed solutions from the practical perspective.

Our computational studies provide several interesting practical insights. First, for the fuel treatment planner it is important to be conservative under limited treatment budgets. Second, the treatment policy structure typically depends on the performance measure used (e.g., total area burned, rate of fire spread) and the vegetation compositions in the areas of interest. More importantly, the policies that combine prescribed burning and mechanical treatment tend to be more effective than either prescribed burning or mechanical treatment alone.

Finally, several PhD students were trained in the framework on this project. One of them has already defended, while the others continue working on the related research topics inspired by this project.

Last Modified: 01/02/2025
Modified by: Lisa M Maillart