Adjusted ∆NBR Index Thresholds for Forest Fire Severity Mapping: A Study in Central Amazonia

Abstract

The 2006 study of fire severity in coniferous forest by Key and Benson derived threshold values of the Delta Normalized Burn Ratio (∆NBR) for interpreting satellite imagery of fire scars, and ∆NBR has been widely used for forest fire severity worldwide. We hypothesized that these thresholds underestimate the severity of fires of tropical forests. Our goal was to find appropriate thresholds to map fire severities in the central Amazon and evaluate trends in recent years. These forests are increasingly threatened by deforestation, severe droughts and wildfires. We adjusted the ∆NBR decision thresholds using new methods that are relatively fast and cheap, with a combination of field data, hemispherical photographs of the opening of the forest canopy with different fire severities, and the use of Landsat data for fires that occurred in 2015 images. The agreement between the classification of severity in the field and the classification by the adjusted ∆NBR thresholds was satisfactory (overall accuracy = 74.2%; Kappa coefficient = 0.635). Using the Key & Benson thresholds resulted in a Kappa value of only 0.184, and the severity classification would be underestimated. We applied the new threshold values to map forest fire severity in the central Amazon from 1995 to 2017. The year 2015 had the greatest area of fire, of which 60.7% was classified as moderate, 19% as high, and 20% as low severity. The results corroborate the importance of adjusting decision thresholds for each study area to classify fire severity using ∆NBR.