The transferability of a dNBR-derived model to predict burn severity across 10 wildland fires in western Canada

Abstract

Wildfire is a complex and critical ecological process that is an integral component of western Canadian terrestrial ecosystems. Therefore, Canadian land management agencies such as Parks Canada require detailed burn severity data for the monitoring and managing of both wildland and prescribed fires. In order to gather this type of information for our study, the acquisition of Landsat imagery was acquired and the differenced Normalised Burn Ratio was computed, a technique that has been shown to be sensitive to field measured burn severity. It is less known if this technique can be transferred and extrapolated over space and time so as to make routine and operational application of the approach feasible. To answer this question, we combined burn severity data from 10 fires across western Canada to test the transferability of an overall model as well as one stratified by land cover and ecozone. Finally, we tested the statistical benefit of incorporating pre- and post-fire data directly into the statistical model using transformations of the remote sensing imagery. Our results indicated that an overall differenced Normalised Burn Ratio derived model successfully estimated burn severity for the majority of fires in the study, which supports its transferability across multiple western Canadian landscapes.