Temporal dependence of burn severity assessment in Siberian larch (Larix sibirica) forest of northern Mongolia using remotely sensed data

Abstract

Assessing burn severity is critical for understanding both the short- and long-term effects of fire disturbance on forest ecosystems. This study proposed a methodology to reconstruct burn severity from the Landsat imagery at different time lags after a fire (≤18 years) in Siberian larch (Larix sibirica) forest. The estimated accuracy of the burn severity models we developed indicated strong effects of forest recovery, image acquisition date and remote sensing predictors on the burn severity assessment. In the first several years after the fire, the dNBR (differenced Normalized Burn Ratio) was the most important remotely sensed index for assessing burn severity, followed by the dNDMI (differenced Normalized Difference Moisture Index) and dNDVI (differenced Normalized Difference Vegetation Index). However, the dNDMI was more important than the dNBR and dNDVI in explaining burn severity when larch forest regrowth dominated. The overall accuracy of the classification and regression tree models showed a decrease in accuracy from 83% to 62% depending on the lag times of burn severity assessment. The high severity class had the lowest omission and commission errors, followed by the low and moderate classes among lag times. Our evaluation of model transferability and thresholds of burn severity index demonstrates the advantage of the proposed methodology for rapid assessment of fire effects in boreal larch forest that will assist in understanding the complex relationships among forest fires and ecological processes in Eurasian boreal ecosystems.