Quantitative evaluation of the influences of multiple factors on the differenced Normalized Burned Ratio in North American ecosystems

The differenced Normalized Burned Ratio (dNBR) is one of the most commonly used indices to represent burn severity. The calculation of dNBR involves the comparison of changes in near-infrared and short wave infrared ( 2.0 μm) surface signal between two images captured before and after burning. Since its introduction almost three decades ago, it has become well known that dNBR is strongly influenced by the timing of image acquisitions, including seasonal mismatch and extended gaps between pre- or post-fire image acquisitions resulting from limited cloud-free image availability; however, this relationship has not been well-described. New complexity is added by the fundamental differences in Thematic Mapper/ Enhanced Thematic Mapper Plus (TM/ETM+) and Operational Land Imager (OLI) sensors between Landsat generations which are used for dNBR calculations within the Monitoring Trends in Burn Severity (MTBS) project. In this study, we assessed the variation in surface reflectance-based dNBR values within 20 randomly selected fire events that burned in 2015 in the North American boreal forests and tundra. By calculating dNBR for these burn scars based on all existing Landsat images acquired during the growing seasons between 2013 and 2016, we systematically evaluated how dNBR was influenced by a set of factors. Our results show that despite the differences in specific bandwidths between Landsat 8 and Landsat 7 images the impact on dNBR values is very low. The analysis reveals significant linear relationships (boreal forests: R²=61%, tundra: R²=46%) between dNBR and the difference in phenology between the two Landsat acquisitions. In addition, among the image pairs which have little difference in phenology, those established by early-growing season acquisitions are preferable over the rest. Finally, we show that it is justifiable to use images that were captured multiple years before fire to calculate dNBR but restrict the post-fire image selection to one year after fire. Overall, our study serves to provide guidelines regarding the image selection process for future research utilizing dNBR.