Assessing Burn Severity in the Alaskan Boreal Forest using Remote Sensing Methods

Abstract:

Over the past half-century, the circumpolar north has experienced a temperature change 1.5 to 4.5 times higher than the global average, leading to an increase in the frequency, severity, and duration of wildfires. In 2019 alone, Alaska experienced 742 fires that burned 2.6 million acres. Among these fires, an unprecedented number were near major population centers at Wildland Urban Interfaces (WUI), which caused major property damage, prompted health issues from unhealthy air quality, and placed a significant strain on local and national firefighting infrastructure already spread thin to meet increasing demands. The 2019 Shovel Creek and Nugget Creek fires, both located at a WUI on the outskirts of Fairbanks, offered a unique opportunity to test and update remote sensing methods for mapping boreal fire burn severity and fire behavior. Burn severity is a measurement of ecological changes across a landscape induced by a fire event. By assessing burn severity we can gain insight into how fires behave across a landscape (i.e. vegetation type that burned or did not burn) so that fire managers can better predict and prepare for where future fires will erupt and be the most destructive. In order to map and assess burn severity of the Shovel Creek and Nugget Creek fires we used Sentinel-2 data to test a number of machine learning classifiers including Random Forest, Support Vector Machine, and Naive Bayes. The results of these machine learning classifiers were then compared to more traditional indices (NBR,NDMI, NDVI) to assess which indices do a better job at assessing burn severity. All burn severity maps were then validated using the Composite Burn Index (a field bases assessment of burn severity). This study will benefit fire managers and researchers alike by identifying the remote sensing methods that are most effective and accurate for assessing boreal wildfire burn severity.