1106. A firebreak location model using two-stage stochastic programming
Invited abstract in session HB-7: Analytics for Prescribed Burning and Firebreaks Location in Forest Fires Prevention, cluster Use of Analytics in Forest Fires Management.
Thursday, 10:30-12:00Room: CE-210
Authors (first author is the speaker)
| 1. | Matias Vilches
|
| Industrial Engineering, Universidad de Santiago de Chile | |
| 2. | Jaime Carrasco
|
| University of Chile | |
| 3. | Sebastián Dávila
|
| Industrial Engineering, Universidad de Santiago de Chile | |
| 4. | Franco Quezada
|
| Industrial Engineering Department, University of Santiago of Chile | |
| 5. | David Palacios
|
| Departamento de Ingeniería Civil Industrial, Universidad de Chile |
Abstract
Forest fuel management constitutes a means of fire prevention through the reduction and manipulation of landscape fuels or the application of firebreaks, to decrease or prevent fire progression. The determination of the optimal location and timing of treatments within the landscape is a stochastic combinatorial optimization problem, however, involving the interaction of different management options with the realization possibilities of a random variable representing the spread of fires - not well understood at present. In this study, we propose a two-stage stochastic integer programming (SIP) model for the spatial allocation of firebreaks at the landscape scale. The model takes as input an approximation of the probability distribution of wildfires by choosing a set of weather conditions and initial ignition points, considering historical data of the landscape. The model minimizes the expected loss due to fire by determining the optimal spatial allocation of firebreaks.
We use the Cell2Fire simulator and the Canadian Fire Behavior Prediction (FBP) System to re-create fire scenarios, and we present four variations of the model to support different landscape management decisions. The first variation minimizes expected loss due to fire, while the second variation minimizes the burn probability (BP) of the most burned zone. The third variation minimizes the BP of the most burned scenario, and the fourth variation incorporates a risk measure (CVaR) over the set of possible scenarios.
Our approach solves the problem for real forests using exact methods and considering the interaction of treatments to the spread of fire; to evaluate the effectiveness of our solutions, we compare them with random solutions on various wildfires. Preliminary numerical results show an average of 7% and 8% improvement in loss due to fire and BP of the most burned zone, respectively, compared with random solutions. By considering the probability distribution of wildfires and using a stochastic optimization approach, our model provides a more effective solution than traditional methods. Results suggest a good performance of our approach to support landscape management decisions and mitigate the impact of wildfires on rural forests.
Keywords
- Stochastic Optimization
- Environmental Management
- Landscape Analysis
Status: accepted
Back to the list of papers