Validation of National and Global Satellite-Derived Products for Mapping Burned Areas in Russia

The paper presents satellite-derived burned area spatial accuracy assessment for the territory of Russia. The reviewed remote sensing products include the national product SRBA (Surface Reflectance Burnt Area), developed at the Space Research Institute, Russian Academy of Sciences, as well as the global products CGLS Burned Area 300m 3.1 NTC (CGLS BA 3.1), FireCCI51, FireCCIS311, GABAM, MCD64A1 C6 and VNP64A1 C2. The reference samples used cover the entire territory of Russia and include more than 106 km2 burned area derived from remote sensing data with a spatial resolution of 10–30 m. The obtained estimates indicate a high accuracy of forest burned area mapping by all presented products in the summer period using the F1-score criterion, or the Dice coefficient (DC = 0.65–0.8), with the highest accuracy for the SRBA product (DC = 0.83). In the spring period, dominated by low-severity small-area forest fires, the accuracy of estimates decreases significantly (DC = 0.25–0.6; DC = 0.4 for the SRBA product). For grasslands, the CGLS BA3.1, FireCCI51, and FireCCIS311 products show the highest burned area mapping accuracy (DC = 0.55–0.7), while the SRBA product has low accuracy (DC = 0.33). All the considered remote sensing products are characterized by low cropland burned area delineating accuracy (DC < 0.3). When considering all burned areas in different land cover types, the SRBA product has a moderate accuracy (DC = 0.5). The highest accuracy results are observed for the FireCCI51 and FireCCIS311 products (DC ≈ 0.6). The GABAM, MCD64A1C6 and VNP64A1C2 products are characterized by a high omission error (approximately 60–70%). Most products tend to systematically underestimate the burned area by 20–50%. An exception is the CGLSBA3.1 product, which is characterized by a high commission error in the summer–autumn period in the forest zone.