Prior to delineation of fire perimeters from airborne and satellite imagery, fire management agencies in Canada employed conventional methods to map area burned based on sketch mapping, digitization from a global positioning system unit, and point buffering from geographic coordinates. These techniques usually provide a less precise representation of a wildland fire’s size and shape than those derived from image data. The aim of this study is to assess the discrepancy in fire size from these techniques that contribute to uncertainty in area burned. We paired independently generated fire perimeters derived from Landsat satellite imagery with conventional perimeters (n = 2792; mean area difference per fire = 40.1%), and developed a set of prediction models to estimate a Landsat area burned from conventional perimeters by considering the mapping source, method, agency, and time period. A two-fold cross validation predicting the logarithm of area burned from the models, indicated an R ² = 0.95 (MAE = 0.10 ha; RMSE = 0.19 ha). From this, we created an adjusted area burned time series from 1950 to 2018 using the model-predicted estimates from conventional perimeters (75% of agency-reported area) in combination with unchanged estimates from agency perimeters derived from airborne and satellite imagery (13% of fires). The predicted estimates reduced the size of individual fires over 2000 ha on average in some years, contributing to an annual average reduction of approximately 11% of the area burned reported in the national agency fire database. By retrospectively applying a robust statistical adjustment to the fire size data, the historical overestimation in annual area burned—up to 1.4 Mha in a single year—could be substantially minimized.

在根据机载和卫星图像划定火场边界之前，加拿大的消防管理机构采用传统方法根据草图、全球定位系统单元的数字化以及地理坐标的点缓冲来绘制燃烧区域的地图。与从图像数据中获得的技术相比，这些技术通常提供的对野火大小和形状的表示不太精确。本研究的目的是评估这些导致燃烧面积不确定性的技术导致的火灾规模差异。我们将源自 Landsat 卫星图像的独立生成的火灾周长与常规周长 ( n= 2792; 每次火灾的平均面积差异 = 40.1%），并开发了一组预测模型，通过考虑制图源、方法、机构和时间段来估计从常规周界烧毁的 Landsat 面积。预测模型燃烧面积的对数的双重交叉验证表明R ²= 0.95（MAE = 0.10 公顷；RMSE = 0.19 公顷）。由此，我们使用来自常规周长（机构报告面积的 75%）的模型预测估计值以及从机载和卫星图像得出的机构周长的未更改估计值（13%）创建了一个从 1950 年到 2018 年的调整区域燃烧时间序列。火灾）。预测估计数在某些年份中平均减少了超过 2000 公顷的单次火灾规模，从而使国家机构火灾数据库中报告的燃烧面积每年平均减少约 11%。通过对火灾规模数据进行回顾性的统计调整，历史上对年度燃烧面积的高估——单年高达 1.4 Mha——可以大大减少。