Validation of MCD64A1 and FireCCI51 cropland burned area mapping in Ukraine

Small fires represent an important under-represented fire type within active fire and burned area datasets in addition to fire emission inventories, especially in regions with substantial agricultural areas. In order to improve regional and global burned area and fire emissions inventories, active fire and burned area algorithm developers are focusing on improving the mapping accuracy of the timing and spatial extent of these small fires. However, product developers have often relied on burned area validation methods that are designed for larger wildfires, and which are therefore not appropriate for these small fire types. Specifically, validation of crop residue burns using pre-and post-Landsat and Sentinel imagery (as recommended by the Committee on Earth Observing Satellites Working Group on Calibration and Validation (CEOS) Land Product Validation protocol) breaks down because the short duration (less than 1 day) between the burn and the subsequent field plowing removes the burned signature before the next sensor overpass. Here we describe an alternate approach that allowed us to rigorously validate two widely available, coarse-resolution global burned area products - MCD64A1 and FireCCI51 - in cropland through using exhaustively-mapped field-level burned area reference maps produced for seven reference areas in Ukraine in 2016 and 2017. Our results highlight the overall high omission errors (MCD64A1: 71-76% and FireCCI51: 63-99%) and commission errors (MCD64A1: 62-81% and FireCCI51: 49-93%) for both products within cropland, while also demonstrating the difficulty of mapping crop residue burned area within the spring, pre-planting mapping period compared to the summer, post-harvest mapping period. Product-specific artifacts and errors are also demonstrated including the confusion with the larger harvest spectral signal (MCD64A1) and the large swaths of unmapped pixels clustered in regular geometric shapes (FireCCI51). These validation results will be used to help guide the upcoming MCD64A1 Collection 7 burned area product improvements within cropland.

Automated summary

The study finds that small fires are often underestimated in agricultural areas within global burned area and fire emission inventories. Current validation methods designed for larger wildfires are not suitable for small fires. An alternative approach using detailed field-level burned area reference maps was used to validate two global burned area products, revealing high omission and commission error rates.