Discrimination of Senescent Vegetation Using Thermal Emissivity Contrast

A remote sensing method utilizing multiband thermal ble to determine in detail these characteristics directly, they are in turn estimated from land surface classifications infrared (8–12 lm) imagery that discriminates between obtained from a combination of ground observations and senescent vegetation and bare soil is described. This disremote sensing data. crimination is achieved by computing thermal band emisIt is difficult, however, to create and maintain accurate sivities from a temperature-emissivity separation algoland surface classification data sets, particularly over agrithm, and then classifying surface features based on ricultural regions. While it may take weeks to develop and spectral emissivity contrast. In a study of rangelands and verify land use maps based on both ground observations winter wheat fields in central Oklahoma, the contrast, or and remote sensing data, the actual land use patterns and range, of these spectral emissivities is diagnostic of the crop conditions will potentially change, perhaps several presence or absence of surface vegetative cover. A large times, over a period of a few days. For example, a wheat range of emissivities, approximately greater than 0.03, is crop could mature, be harvested, and the field plowed all indicative of bare soil, while a low range, less than 0.02, after the time of classification. is indicative of vegetative cover. When knowledge of the One way to improve surface classification is to reemissivity range is combined with a vegetation index, such motely collect diagnostic bands from the visible and nearas NDVI, the surface may be classified by a ternary system: infrared (VIS/NIR) bands simultaneously with brightness bare soil, green vegetation, and senescent vegetation. Distemperature data from thermal infrared bands. This will crimination between bare soil and soil covered with senessoon be feasible with a new sensor—ASTER (Advanced cent vegetation using emissivity contrast should be feasible Spaceborne Thermal Emission and Reflection Radiometer; in other settings. The benefit of this technique is that heat see http://asterweb.jpl.nasa.gov)—recently launched on flux predictions can be based on a more accurate surface the EOS-Terra satellite (Yamaguchi et al., 1998). ASTER representation than otherwise provided by visible and simultaneously collects visible, near-infrared, and thermalnear-infrared land classification schemes. Published by infrared data at spectral and spatial resolutions preElsevier Science Inc. viously unavailable. In our preparation for ASTER, we have been investigating the performance of heat flux models by combining INTRODUCTION remotely sensed Thermal Infrared Multispectral Scanner The accurate parameterization of land surface characteris(TIMS; Palluconi and Meeks, 1985) data from aircraft and tics is critical to reliably estimate surface heat fluxes from LANDSAT TM with ground observations. We have noticed remote sensing platforms. In typical flux modeling that the land use images we develop using conventional schemes, characteristics such as vegetation canopy height, VIS/NIR classification schemes fail to distinguish between density, and albedo are related to surface roughness, fracsenescent, or harvested vegetation, and bare plowed soil. tional cover, and intercepted radiation. Since it is not feasiThis inability has been noted elsewhere. Daughtry et al. (1995), for example, report that plant litter in general cannot be discriminated from bare soil based on any one * USDA/ARS Hydrology Laboratory particular visible or near-infrared reflectance. Under some Address correspondence to A. N. French, Hydrology Lab, Building circumstances litter can be discriminated by using three 007, BARC-WEST, USDA/ARS, Beltsville, MD 20705, USA. E-mail: 50-nm bands over 2.0 to 2.2 microns (Nagler et al., 2000), [email protected] Received 15 November 1999; revised 10 March 2000. but the technique is somewhat sensitive to moisture con-