Advanced techniques in Forest Management under conditions of fire uncertainty

Abstract

The general objective of this thesis is to address the problem of forest fires from a preventive perspective, mainly motivated by the increase of this phenomenon worldwide and exacerbated by climate change. The current incidents indicate that preventive policy measures must be taken to reduce the risk of fire occurrence, managing the land in an effective way to protect natural forests, agricultural areas, and human lives. However, the problem of deciding where, when, and how to perform fuel treatments is a very complex problem and remains open. To address this problem, in the first two chapters (Chapter 2 and 3), we will develop a Cell-based Fire Growth Model, suitable for decision making that includes different sources of fire uncertainty such as: (1) ignition point(s) selected via a user-defined probability distribution; (2) a coefficient of variation (cvROS) capturing the stochastic aspects of Rate of Spread(ROS); and (3) a set of user-generated weather stream files (scenarios) that can be provided to the simulator. Using the tools mentioned in the previous paragraph, in Chapter 4 a Decision Support System (DSS) is developed that integrates fire simulation and fuel treatment decision making to minimize wildfire losses. We introduce the Downstream-Protection-Value (DPV) concept, an adaptable metric that ranks units of the landscape via their impact on fire propagation by modeling a forest as a network and the fire as multiple tree graphs. Our DPV metric stands out in all experiments, rapidly detecting fire ignition areas, preventing fire occurrence and decreasing the expected area burned by more than half compared to state-of-the-art alternatives. We test our methodology using public data of real forest patches from Alberta and British Columbia provinces, Canada. In Chapter 5, we provide a device to improve the performance of our simulator by adjusting its internal input parameters in order to obtain better and more reliable fire scars, through a robust methodology and easy-and-fast implementation based in Derivative-Free Optimization Algorithms. Also, we discuss how this approach could be used to develop simulators that learn from the data provided in real-time from an ongoing fire; and also from historical fire scars. Finally, in Chapter 6, we developed a methodology based on fire risk, to delimit the Wildland Urban Interface for Chile. To meet this objective, we first built a Bagged Decision Tree (BDT) model to quantify the risk of fire occurrence from different variables, grouped into Human Activity, Geographic and Topographic, and Land Cover. Subsequently, using the mathematical elements underlying the BDT models, we determine individual relationships between the risk and the variables considered to address the problem of delineating a WUI-map.