Assessment and characterization of sources of error impacting the accuracy of global burned area products

Understanding the sources of error that affect the accuracy of current coarse burned area products is crucial to identifying weaknesses and limitations of burned area (BA) algorithms. In this paper, we propose methods to analyse sources of errors of global BA products, which are applied to assess a widely used product, (FireCCI51), created under the ESA climate change initiative programme. The effect of several variables related to product accuracy are quantified by comparing the FireCCI51 BA product with a large global reference BA dataset for the years 2017–2019. These reference data were obtained from multitemporal analysis of Landsat-8 data. Error maps (commission and omission error layers) were generated for a sample of 316 units (100 km × 100 km) and associations between errors and potential explanatory sources of error were spatially analysed to provide a comprehensive error assessment of the FireCCI51 product. Active fire density, burned severity, burned patch size and land cover class were selected as explanatory variables for omission and commission errors. The results indicated that the largest contribution to the commission errors (70–80%) was related to the pixel size, particularly when pixels are only partially burned. Active fire densities were also associated to error distribution, with higher densities from products derived from the VIIRS sensors versus those generated from MODIS, indicating the potential benefit of the former as an alternative source in upcoming BA algorithms. Burn severity was also associated with error distribution, but this relation varied substantially across biomes. Land cover was also related to errors, with lowest accuracies for croplands with omission error percentages >85%. Fire size and, to a lesser extent, the shape of burned areas also played a crucial role in determining the accuracy of the BA products. BA accuracy increased for larger and more compact burned patches. Although our results suggest that further improvement of global BA products is possible, they also indicate that the ability of coarse spatial resolution BA sensors to meet the accuracy recommendations of the Global Climate Observing System (GCOS) programme is limited.