Abstract
This work consisted of two studies of burn severity using infrared perimeter maps and satellite-inferred burn severity data, differenced Normalized Burn Ratio, from 42 wildland fires from central Idaho and western Montana from 2005 to 2007, and 2011.
Study 1 examined the proportion of burn severity categories for individual daily areas burned. We defined 2,697 areas, from which we calculated the proportion of three burn severity classes. The proportion of high severity was weakly correlated with size of area burned. Large areas burned do not consistently produced larger proportions of high severity.
Study 2 analyzed burn severity relative to 20 environmental variables using the Random Forest machine learning algorithm. We used ten daily weather observations, eight 34-yr climate percentiles, seven topographical index measurements, and four vegetation characteristics from 10,819 randomly located points. We found that higher percentage existing vegetation cover had larger influences on changes in burn severity.