Using NPP-Suomi VIIRS I-band data to delineate high- and low-intensity burn areas for forest fires in interior Alaska

The aim of this work is to present and evaluate an algorithm that generates near real-time fire detections suitable for use by fire and related hazard management agencies in Alaska. Our scheme offers benefits over available global products and is sensitive to low-intensity residual burns while at the same time avoiding common sources of false detections as they are observed in the Alaskan boreal forest, such as refective river banks and old fire scars. The algorithm is based on I-band brightness temperature data form the Visible Infrared Imaging Radiometer Suite (VIIRS) on the NOAA's NPP Suomi spacecraft. Using datasets covering the entire 2015 Alaska fire season, we first evaluate the performance of two global fire products: MOD14/MYD14, derived from the Moderate Resolution Imaging Spectroradiometer (MODIS), and the more recent global VIIRS I-band product. A comparison with the fire perimeter and properties data published by the Alaska Interagency Coordination Center (AICC) shows that both MODIS and VIIRS fire products successfully detect all fires larger than approx. 1000 hectares, with the VIIRS I-band product only moderately outperforming MOD14/MYD14. For smaller fires, the VIIRS I-band product offers higher detection likelihood, but still misses one fifth of the fire events overall. Furthermore, some daytime detections are missing, possibly due to processing difficulties or incomplete data transfer. Second, as an alternative, we present a simple algorithm that uses the normalized difference between the 3.74µm and 11.45 µm VIIRS-I band at-sensor brightness temperatures to map both low- and high-intensity burn areas. Such an approach has the advantage that it makes use of data that is available via the direct readout station operated by Geographic Information Network of Alaska (GINA). We apply this scheme to known Alaskan boreal forest fires and validate it using GIS data produced by fire management agencies, fire detections from near simultanous Landsat imagery, and sub-pixel analysis. We find that our VIIRS derived fire product more accurately captures the fire spread, can differentiate well between low- and high-intensity burn areas, and has fewer errors of omission compared to the MODIS and VIIRS global fire products.