Simulating AVIRIS-NG Hyperspectral Image from Sentinel-2 Multispectral Image for Improved Wildfire Fuel Mapping, Boreal Alaska

Abstract:

In recent decades, Alaska has experienced a significant increase in wildfire events that have been linked to drier and warmer summers with increased lightning. In 2019 alone, Alaska had 742 wildfires that burned ~2.6 million acres. Forest fuel maps derived from image data play a vital role in wildfire risk assessment and management of active fires. Freely available satellite-borne multispectral image data are widely used for land use and land cover mapping, but, due to the coarser spectral resolution, has limited ability to map forest vegetation at fuel class level. In contrast, hyperspectral image data have very high spectral resolution but limited availability. Hyperspectral data provides detailed spectral information which make them ideal for detailed land cover mapping including vegetation mapping at fuel class level. In this study, we have focused on simulating the Airborne Visual Imaging Infrared Spectrometer - Next Generation (AVIRIS-NG) hyperspectral image from a preexisting multispectral, i.e. Copernicus Sentinel-2 image so that we can map boreal vegetation at fuel level.

We used atmospherically corrected Sentinel-2 data, the spectral response function of the AVIRIS-NG sensor, and the normalized ground spectra of major land features to simulate the AVIRIS-NG hyperspectral data. Uniform Pattern Decomposition Method (UPDM) was used for spectral simulation. The simulated data have the spectral characteristics of AVIRIS-NG while the reflectance properties come from Sentinel data. We validated the simulated data by comparing it to the original AVIRIS-NG data over a test site. The validation was based on the comparison of spectral signatures, statistical correlation, and the assessment of classification results using field data. We used advanced machine learning image classifiers for improved fuel mapping from simulated AVIRIS-NG hyperspectral image data.