Using Landsat Imagery to Assess Burn Severity of National Forest Inventory Plots

Abstract

As the frequency and size of wildfires increase, accurate assessment of burn severity is essential for understanding fire effects and evaluating post-fire vegetation impacts. Remotely-sensed imagery allows for rapid assessment of burn severity, but it also needs to be field validated. Permanent forest inventory plots can provide burn severity information for the field validation of remotely-sensed burn severity metrics, although there is often a mismatch between the size and shape of the inventory plot and the resolution of the rasterized images. For this study, we used two distinct datasets: (1) ground-based inventory data from the United States national forest inventory to calculate ground-based burn severity; and (2) remotely-sensed data from the Monitoring Trends in Burn Severity (MTBS) database to calculate different remotely-sensed burn severity metrics based on six weighting scenarios. Our goals were to test which MTBS metric would best align with the burn severity of national inventory plots observed on the ground, and to identify the superior weighting scenarios to extract pixel values from a raster image in order to match burn severity of the national inventory plots. We fitted logistic and ordinal regression models to predict the ground-based burn severity from the remotely-sensed burn severity averaged from six weighting scenarios. Among the weighting scenarios, two scenarios assigned weights to pixels based on the area of a pixel that intersected any parts of a national inventory plot. Based on our analysis, 9-pixel weighted averages of the Relative differenced Normalized Burn Ratio (RdNBR) values best predicted the ground-based burn severity of national inventory plots. Finally, the pixel specific weights that we present can be used to link other Landsat-derived remote sensing metrics with United States forest inventory plots.