Assessing fire severity using imaging spectroscopy data from the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) and comparison with multispectral capabilities

Abstract

Fire severity, the degree of environmental change caused by a fire, is traditionally assessed by broadband spectral indices, such as the differenced Normalized Burn Ratio (dNBR) from Landsat imagery. Here, we used an alternative indicator, the burned fraction derived from spectral mixture analysis (SMA), to evaluate and compare the performance for assessing fire severity of broadband and narrowband imaging spectroscopy (IS) data in the visible to shortwave infrared (VSWIR, 0.35–2.5μm). We used the band specifications of the broadband Operational Land Imager (OLI) and the narrowband Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). We integrated two techniques to account for endmember variability in the unmixing process, spectral weighting and iterative unmixing, in a model referred to as weighted multiple endmember SMA (wMESMA). Based on a separability index, we evaluated the separability between the different ground components, or endmembers, that comprise post-fire environments (char, green vegetation (GV), non-photosynthetic vegetation (NPV) and substrate). We found that the near infrared region (0.7–1.3μm) had the highest discriminatory power, followed by the shortwave infrared 2 (SWIR2, 2–2.4μm), SWIR1 (1.5–1.7μm) and visible (0.35–0.7μm) regions. Individual narrowbands did not substantially outperform individual broadbands, however, the higher data dimensionality of IS resulted in significantly improved post-fire fractional cover and burned fraction estimates compared to multispectral data. Multispectral data captured a fair amount of the variability in fire severity conditions as represented by the different fractional cover estimates of the endmembers in both a multispectral narrow- and broadband scenario, however, fractional cover estimates derived from IS data using all viable bands were significantly better. This demonstrated the benefits of IS over traditional multispectral data to assess fire severity and also showed that the additional information gain was the result of higher data dimensionality and not because of certain narrowbands capturing narrow spectral features. In addition, we found that the burned fraction derived from all viable AVIRIS bands over a fire in California, USA, was highly correlated with two field measures of fire severity (Geo Composite Burn Index: R2=0.86, and the percentage black trees and shrubs: R2=0.65). Formal quantification of potential improvements of IS over multispectral methods is important with the advent of upcoming spaceborne IS missions (e.g. the Environmental Mapping and Analysis Program and Hyperspectral Infrared Imager). Our analysis showed that IS data when combined with advanced analysis techniques significantly improved fire severity assessments. The improvements of using IS data require higher computational cost and advanced processing, thus multispectral data might still suit the needs of certain applications such as rapid fire damage assessments and global analysis of spatio-temporal fire severity patterns.