Assessment of burn severity in Middle Povozhje with Landsat multitemporal data

Abstract

Forests of Middle Povolzhje in the Russian Federation were seriously affected by severe wildfires in 2010. The importance of accurate estimation of burn severity levels is because fire-affected areas may have important societal, ecological and economic consequences to the region. The aim of the study was to explore the performance of the differenced Normalized Burn Ratio (dNBR) and Composite Burn Index (CBI) to evaluate the burnt forest areas of the 2010 wildfires in the Republics of Mari El and Chuvashia of the Russian Federation with the use of Landsat-5 Thematic Mapper (TM) time series images. In total, 625 forest sites were sampled for ground-based burn severity data following the CBI protocol. Both datasets for Mari El and Chuvashia were statistically similar with correlation coefficients of 0.87, which confirms pooling them into a single dataset for the forests of Middle Povozhje. A non-linear third-degree polynomial model form of third order best represented the relationship (R2=0.79) between the dNBR and CBI data. Our model estimates that the total forest burnt area on the study area was 113000ha in 2010, mostly in coniferous stands. Almost half (45.9% or 51900ha) of the burnt forest areas were classified in the high-severity category. The accuracy assessment shows that severity classification results were accurate for estimating burn severity levels, with both producer’s and user’s accuracies higher than 80% for the unburned, low and high classes. Lower user and producer accuracies were noticed for the moderate class of burn severity.