Hyperspectral imagery provides a scene analyst with a wealth of spectral (100s of channels in VNIR/SWIR) and spatial (10s of meters) information. Various techniques for scene characterization can utilize individual or combinations of narrow spectral bands to identify specific features in an image. Processing of the entire spectral domain (e.g., through principal components analysis) can also be used to reduce the dimensionality of HSI data as well as to facilitate spectral feature extraction and material identification.

In this presentation, we will provide two approaches for identifying and characterizing features in a set of AVIRIS scenes dominated by areas of smoke, plumes, clouds and burning grassland as well as scarred (burned) areas. Both a physics-based and a semi-automated feature extraction approach are used. In the physics-based approach, natural occurring water clouds are contrasted with smoke areas, burn scarred land is extracted using an NDVI-like index formula and hot plumes and smoldering fires are identified with enhanced NIR/SWIR signatures. Thick smoke plumes are identified via contrast between VIS and SWIR while thin smoke plumes are delineated by contrasts at short (blue) wavelengths.

In contrast to the physics-based approach, a semi-automated principal components analysis technique is applied to the same images and a scene classification is done. The two approaches complement each other and can be used jointly to provide a more robust algorithm for characterization of hyperspectral scenes with clouds and active fires.