Abstract Fire severity is a function of dynamic interactions between vegetation and burning conditions. To understand the factors that control it, accurate methods for estimating prefire vegetation structure and composition as well as fire propagation conditions are required. Here we analyzed the spatial variability of fire severity in a mixed-severity fire (3217 ha) that occurred in southeast Spain (Yeste, Albacete) from 27th July to 1th August 2017, burning mostly a pine woodland, including part of an earlier fire in 1994. Fire severity was estimated using three satellite-based indices derived from the Normalized Burn Ratio (NBR) using Sentinel 2 and Landsat 8 images from the dates before and immediately after fire. The field-based Composite Burn Index (CBI) was used for validation. Prefire vegetation conditions and fuel models were derived from LiDAR metrics and other vegetation data. Fire propagation conditions were estimated based on a fire progression map provided by the Forestry Services of Castilla-La Mancha. In addition, hourly fire weather and aligned (i.e., in the sense of the propagating fire-front) slope and wind speed were calculated for each burning period. Regression models using different spectral fire severity indices and their driving factors were obtained applying Boosted Regression Trees (BRTs). Fire severity was highly predicted by both burning conditions and prefire vegetation (mean adjusted R2 [Adj.R2]: 86% ± 0.04 and 68% ± 0.05 for training and validation sets, respectively). Alone, burning conditions explained more variance than LiDAR metrics and vegetation separately. The single variables that contributed most to the models were the rate of spread of the fire-front, biomass proxies (i.e., Leaf Area Index [LAI] and fraction of Photosynthetically Active Radiation [fPAR]) and understory vegetation (i.e., density of LiDAR points 1–2 m). Higher fire severity occurred in areas burning uphill, with a high rate of spread driven by high velocity winds and under high maximum temperature. Fire severity was high in wooded stands that were heterogeneous in height, composed by scattered and small Pinus halepensis trees, with high and homogeneous understory cover. In contrast, lower fire severity occurred in mature stands dominated by tall Pinus pinaster and Pinus nigra trees. There were strong interactions between vegetation, weather, fire-aligned topography and rate of spread. Because vegetation variables were important drivers of fire severity, even under extreme fire weather conditions, fuel management treatments to limit fire severity and, potentially, fire size should be implemented.

中文翻译：

---

解开火灾前植被与燃烧条件对西班牙东南部大型森林火灾火灾严重程度的作用

摘要 火灾严重性是植被和燃烧条件之间动态相互作用的函数。为了了解控制它的因素，需要准确的方法来估计火灾前植被结构和组成以及火灾传播条件。在这里，我们分析了 2017 年 7 月 27 日至 8 月 1 日在西班牙东南部（耶斯特，阿尔巴塞特）发生的混合严重性火灾（3217 公顷）中火灾严重程度的空间变异性，主要燃烧松林，包括早期火灾的一部分1994 年。使用三个基于卫星的指数估计火灾严重性，这些指数来自标准化燃烧率 (NBR)，使用 Sentinel 2 和 Landsat 8 图像，来自火灾发生之前和之后的日期。基于现场的综合燃烧指数 (CBI) 用于验证。火灾前植被条件和燃料模型源自 LiDAR 指标和其他植被数据。火势蔓延条件是根据卡斯蒂利亚-拉曼恰林业局提供的火势发展图估算的。此外，还计算了每个燃烧期的每小时火灾天气和对齐（即，在传播火锋的意义上）坡度和风速。使用不同频谱火灾严重性指数及其驱动因素的回归模型是通过应用增强回归树 (BRT) 获得的。燃烧条件和火灾前植被都高度预测了火灾的严重性（平均调整 R2 [Adj.R2]：训练集和验证集分别为 86% ± 0.04 和 68% ± 0.05）。单独而言，燃烧条件比单独的 LiDAR 指标和植被解释了更多的差异。对模型贡献最大的单个变量是火锋的蔓延速度、生物量代理（即叶面积指数 [LAI] 和光合有效辐射的分数 [fPAR]）和林下植被（即 LiDAR 的密度）点 1–2 m）。在上坡燃烧的地区发生了更高的火灾严重性，在高速风和高最高温度的推动下蔓延速度很快。在高度不均匀的林分中，火灾严重程度很高，由零散的小松树组成，林下覆盖率高且均质。相比之下，在以高大的松树和黑松树为主的成熟林分中发生的火灾严重程度较低。植被、天气、与火灾相关的地形和传播速度之间存在很强的相互作用。