Abstract As bias and uncertainties inevitably exist on both wildland fire model states and parameters, fire simulations do not always accurately forecast the temporal and spatial progression of wildfires. In this paper, a novel approach is proposed to estimate fire perimeters and fuel adjustment factors simultaneously for FARSITE tool. Fire perimeters estimation is the key to reduce model state bias, while fuel adjustment factors estimation is an essential component to reduce model parameter uncertainties. For those purposes, ensemble transform Kalman filter algorithm with adaptive proposal is adopted for correcting the fire perimeters, Monte Carlo based radial basis function neural network (RBFNN) is used to estimate fuel adjustment factors. The proposed method is first evaluated on homogeneous fuels distributed over flat ground for an experimental grassfire, then an intensive validation study is done on heterogeneous fuels distributed over complex terrain for a fire accident corresponding to the 2018 California Camp Fire. Results show that combined estimation of fire perimeters and fuel adjustment factors provides an interesting framework to produce accurate forecast of the fire propagation.

中文翻译：

---

FARSITE 中火周长和燃料调整因子的组合估计用于预测野火蔓延

摘要 由于野火模型状态和参数不可避免地存在偏差和不确定性，火灾模拟并不总是准确地预测野火的时间和空间进展。在本文中，提出了一种新方法来同时估计 FARSITE 工具的火灾周长和燃料调整因子。火灾周长估计是减少模型状态偏差的关键，而燃料调整因子估计是减少模型参数不确定性的重要组成部分。为此，采用具有自适应提议的集成变换卡尔曼滤波器算法来校正火灾周长，并使用基于蒙特卡罗的径向基函数神经网络 (RBFNN) 来估计燃料调整因子。所提出的方法首先在试验性草地火灾中分布在平坦地面上的均质燃料上进行评估，然后对分布在复杂地形上的异质燃料进行了密集的验证研究，以应对与 2018 年加州营火相对应的火灾事故。结果表明，火灾周长和燃料调整因素的组合估计提供了一个有趣的框架来产生火灾蔓延的准确预测。