[Paper Review] Adjusting Rate of Spread Factors through Derivative-Free Optimization: A New Methodology to Improve the Performance of Forest Fire Simulators

Jaime Carrasco, Cristóbal Pais+2 more|arXiv (Cornell University)|Sep 11, 2019

Fire effects on ecosystems29 references4 citations

TL;DR

This paper proposes a derivative-free optimization (DFO) framework to automatically calibrate Rate of Spread (ROS) adjustment factors in the Cell2Fire forest fire simulator, improving simulation accuracy by minimizing the difference between simulated and observed fire scars. The method efficiently identifies optimal fuel-specific ROS multipliers using black-box optimization, with BOBYQA and NEWUOA showing superior performance in convergence speed and runtime.

ABSTRACT

In practical applications, it is common that wildfire simulators do not correctly predict the evolution of the fire scar. Usually, this is caused due to multiple factors including inaccuracy in the input data such as land cover classification, moisture, improperly represented local winds, cumulative errors in the fire growth simulation model, high level of discontinuity/heterogeneity within the landscape, among many others. Therefore in practice, it is necessary to adjust the propagation of the fire to obtain better results, either to support suppression activities or to improve the performance of the simulator considering new default parameters for future events, best representing the current fire spread growth phenomenon. In this article, we address this problem through a new methodology using Derivative-Free Optimization (DFO) algorithms for adjusting the Rate of Spread (ROS) factors in a fire simulation growth model called Cell2Fire. To achieve this, we solve an error minimization optimization problem that captures the difference between the simulated and observed fire, which involves the evaluation of the simulator output in each iteration as part of a DFO framework, allowing us to find the best possible factors for each fuel present on the landscape. Numerical results for different objective functions are shown and discussed, including a performance comparison of alternative DFO algorithms.

Motivation & Objective

• To address the persistent challenge of inaccurate fire scar predictions in forest fire simulators due to input data errors and model limitations.
• To develop an automated, derivative-free optimization framework for tuning Rate of Spread (ROS) adjustment factors in the Cell2Fire simulator.
• To improve simulation fidelity by minimizing the discrepancy between simulated and observed fire perimeters using real-world fire data.
• To evaluate and compare the performance of multiple DFO algorithms in the context of wildfire simulation parameter calibration.
• To provide a generalizable methodology applicable to other fire growth models and heterogeneous forest landscapes.

Proposed method

• Formulates a black-box optimization problem to minimize the error between simulated and observed fire scars, using the ROS adjustment factor (RAF) as decision variables.
• Employs derivative-free optimization (DFO) algorithms—BOBYQA, NEWUOA, NELDER-MEAD, COBYLA—within a simulation-optimization loop to tune fuel-specific ROS multipliers.
• Uses the Cell2Fire simulator as a black-box evaluator, calling it iteratively to assess the objective function based on fire perimeter deviation.
• Applies the RAF framework to adjust head, back, flank, and rate of spread values per fuel type, enabling localized and global fire spread adjustments.
• Validates the approach on two real fire events: Dogrib and Dogrib-North, using Prometheus-simulated scars as reference.
• Implements a multi-objective error metric based on the root mean square deviation between simulated and observed fire perimeters.

Experimental results

Research questions

• RQ1 Can derivative-free optimization effectively calibrate Rate of Spread adjustment factors to improve fire simulator accuracy?
• RQ2 How do different DFO algorithms compare in terms of convergence speed, runtime, and objective function quality for fire simulation tuning?
• RQ3 Does using simulated fire scars (from Prometheus) as reference provide a reliable basis for parameter calibration despite suppression effects?
• RQ4 To what extent can the RAF framework capture local and global fire spread dynamics in heterogeneous landscapes?
• RQ5 Can the proposed methodology be generalized to other fire growth simulators and forest types?

Key findings

• BOBYQA and NEWUOA achieved the best balance between convergence speed and runtime, requiring only 96 and 104 evaluations with runtimes of 7.78 and 12.23 minutes, respectively.
• NELDER-MEAD achieved the lowest objective function value (80.54) but required 332 evaluations and 45.33 minutes, indicating high computational cost.
• The error in the real Dogrib fire instance was higher than in the Dogrib-North instance, primarily due to suppression effects in the observed scar, which biases parameter tuning.
• The RAF framework successfully captured both local discontinuities and global landscape structures, improving simulation fidelity across heterogeneous fuel types.
• The methodology is generalizable and could be extended to include additional parameters such as slope effect, curing degree, and resilience time in future work.
• The study demonstrates that using simulated scars as reference data is viable when real scars are biased by suppression actions, provided the model accounts for such effects explicitly.