Combined visible-near infrared and multi-band thermal infrared data for discrimination of vegetated and bare surfaces

Land classification data sets that are required for energy flux modeling of landscape scale remotely sensed data are highly reliant upon vegetation indices for estimating fractional vegetation cover. The visible-near infrared band based index, NDVI, was found inadequate in a previous study of rangelands at El Reno, Oklahoma (SGP97). Bare soil and senescent vegetation, surfaces with quite different energy transport properties, could sometimes not be distinguished. A practical, remote sensor based remedy to this problem was developed from multi-band thermal infrared data, which was used to create an image of spectral emissivity contrast. Vegetative cover, whether green or brown, showed low emissivity contrast, while bare soil surfaces showed high emissivity contrast. When emissivity contrast data were combined with NDVI data, harvested winter wheat could be discriminated from plowed fields, which improved two-source model estimates of surface heat flux. This approach will have wider application when multi-spectral thermal infrared data becomes available from the ASTER instrument.