Evaluation of Multi-Spectral Band Efficacy for Mapping Wildland Fire Burn Severity from PlanetScope Imagery

Abstract

Increased spatial resolution has been shown to be an important factor in enabling machine learning to map burn extent and severity with extremely high accuracy. Unfortunately, the acquisition of drone imagery is a labor-intensive endeavor, making the capture of drone imagery impractical for large catastrophic fires, which account for the majority of the area burned each year in the western US. To overcome this difficulty, satellites, such as PlanetScope, are now available which can produce imagery with remarkably high spatial resolution (approximately three meters). In addition to having higher spatial resolution, PlanetScope imagery contains up to eight bands in the visible and near-infrared spectra. This study examines the efficacy of each of the eight bands observed in PlanetScope imagery using a variety of feature selection methods, then uses these bands to map the burn extent and biomass consumption of three wildland fires. Several classifications are produced and compared based on the available bands, resulting in highly accurate maps with slight improvements as additional bands are utilized. The near-infrared band proved contribute most to increased mapping accuracy, while the green 1 and yellow bands contributed the least.