Optimizing burn severity assessments in Alaskan tussock tundra from optical imagery

Over the past decade Alaskan tundra has experienced an increase in fire occurrence prompting rising concerns in the scientific community. Fire occurrence in tundra ecosystems has the potential to release a large amount of organic carbon stored in the deep organic layer, modify soil moisture and respiration, and make more organic matter available for decomposition and future burning through impacts on the active depth layer. Monitoring and characterization of fire occurrence and impacts in extensive, remote, and largely inaccessible tundra regions rely on satellite observations of land surface and require robust approaches to burn severity measurements. The relatively low fire activity in tundra regions between 1950 and 2000 has resulted in overall lack of understanding of fire impacts on tundra landscapes outside the Seward Peninsula where tundra fire record is better known. Thus satellite-based mapping of burn severity is limited by the lack of quantified knowledge of fire-induced physical changes on the landscape on the one hand and the capabilities of optical remote sensing systems to capture those characteristics on the other. Here we present an analysis of satellite mapping of burn severity using multi-date Landsat imagery and two field-based measurements of burn severity - the operationally applied Composite Burn Index (CBI) and the more simplistic Burn Severity Index (BSI), also known as the Burn Severity Code Matrix. The BSI used here is a four-point scale (unburned, low, moderate, severe) assessed for the surface substrate and vegetation layers. The BSI and CBI used to compare to the remote sensing data were determined from the field data by converting the qualitative fractional assessment of burn severity within 10 x 10 m plots to a single value. Since both indices represent mostly ocular assessment of the fire-impacted surface, they can relate well to Landsat's optical sensors measurements. The analysis shows that overall satellite indices have closer relationships with CBI compared to BSI (exceeding R2 of 0.8 in 10 and 4 instances for CBI and BSI, respectively) for single-date assessments. Similarly, for multi-date differenced assessments, the R2 between CBI and various Landsat-based indices exceeded 0.8 in 76 instances compared to only 20 instances for BSI. However, there are considerable differences between the timing of image acquisition post fire when the relationships between satellite observations and CBI or BSI are the closest as well and indices which field assessments of burn severity appear to be best correlated with. CBI produces closest relationships with near infrared, short-wave infrared at 1.5μm, relative differenced Normalized Burn Ratio, and Tasseled Cap Greenness indices when the post-fire image is acquired immediately after the fire event irrespective of the pre-burn date selection. In contrast, the relationship between BSI and spectral indices peaks when pre- and post-burn indices are computed from images preceding the full green-up conditions (i.e. late May imagery), with Tasseled Cap Brightness showing the closest relationship. Although on average CBI outperforms BSI, BSI produces closer relationships to spectral indices than CBI indicating that the two field assessments provide complimentary information which can be related to biophysical properties of impacted surface.