Introducing spatially distributed Fire Danger from Earth Observations (FDEO) Using satellite-based data in the Contiguous United States

Wildfire danger assessment is essential for operational allocation of fire management resources; with longer lead prediction, the more efficiently resources can be allocated regionally. Traditional studies focus on meteorological forecasts and fire danger index models (e.g., National Fire Danger Rating System – NFDRS) for predicting fire danger. Meteorological forecasts, however, lose accuracy beyond ~10 days, as such there is no quantifiable method for predicting fire danger beyond 10 days. While some recent studies have statistically related hydrologic parameters and past wildfire area burned or occurrence to fire, no study has used these parameters to develop a monthly spatially distributed predictive model in the contiguous United States. Thus, the objective of this study is to introduce Fire Danger from Earth Observations (FDEO), which uses satellite data over the contiguous United States (CONUS) to enable two-month lead time prediction of wildfire danger, a sufficient lead time for planning purposes and relocating resources. In this study, we use satellite observations of landcover type, vapor pressure deficit, surface soil moisture and the enhanced vegetation index, together with the United States Forest Service (USFS) verified and validated fire database (FPA) to develop spatially gridded probabilistic predictions of fire danger, defined as expected area burned as a deviation from “normal”. The results show that the model predicts spatial patterns of fire danger with 52% overall accuracy over the 2004-2013 record, and up to 75% overall accuracy during the fire season. Overall accuracy is defined as number of pixels with correctly predicted fire probability classes divided by the total number of the studied pixels. This overall accuracy is the first quantified result of two-month lead prediction of fire danger and demonstrates the potential utility of using diverse observational data sets for use in operational fire management resource allocation in the CONUS.

野火危险评估对于消防管理资源的业务调配至关重要；预测提前期越长，区域资源分配的效率就越高。传统研究侧重于利用气象预报和火灾危险指数模型（如国家火灾危险评级系统——NFDRS）预测火灾危险。然而，气象预报在超过约10天后精度会下降，因此目前尚无超过10天的火灾危险量化预测方法。尽管近期一些研究已通过统计方法将水文参数与历史野火燃烧面积或发生情况关联起来，但尚无研究利用这些参数在美国本土（contiguous United States）开发月度空间分布预测模型。因此，本研究的目标是引入基于地球观测的火灾危险模型（Fire Danger from Earth Observations，FDEO），该模型利用美国本土（CONUS）的卫星数据实现两个月提前期的野火危险预测——这一提前期足以满足规划和资源调配的需求。本研究利用土地覆盖类型、水汽压亏缺（vapor pressure deficit）、表层土壤湿度和增强植被指数（enhanced vegetation index）的卫星观测数据，结合美国林务局（United States Forest Service，USFS）验证的火灾数据库（FPA），开发了空间网格化的火灾危险概率预测模型——该模型将火灾危险定义为预期燃烧面积与"正常"情况的偏差。结果表明，在2004-2013年的记录期内，该模型对火灾危险空间格局的预测总体准确率为52%，而在火灾季节期间准确率可达75%。总体准确率定义为火灾概率类别预测正确的像素数量与研究像素总数的比值。这一总体准确率是两个月提前期火灾危险预测的首个量化结果，证明了利用多样化观测数据集在CONUS地区业务消防管理资源调配中的潜在价值。