Optimization with fire spread simulation for forest management

Abstract

We propose a method for forest management in which wildfire is modeled explicitly through the integration of optimisation and simulation. Given a forest, the decision problem is to select a plan (i.e. a prescription and a periodicity for brush cleaning) for each of its stands. Each plan is associated with values for a set of criteria for each period of the temporal horizon. Considered criteria are net present value, biodiversity, carbon stock, and erosion. The problem is modelled by a mixed integer programming (MIP) with the objective of maximizing the net present value and imposing limits for the remaining criteria. A fire spread simulator, based on shortest path algorithms following the minimum travel time principle, is responsible to identify sets of plans that are not acceptable together as they result in a high rate of fire spread. That information is included in the MIP as constraints. This cycle optimization-simulation is repeated until the plans provided by the MIP are acceptable in all scenarios. Data from a real landscape case-study has been collected and processed to obtain management and fire parameters required to validate the proposed method, which is being implemented in Python (with Gurobi as a MIP solver, GeoPandas for managing and processing geospatial data, and NetworkX implementation of graph algorithms).