An image-sharpening method to recover stream temperatures from ASTER images

Linear unmixing of spectra from daytime Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) images can be used to improve the spatial resolution of temperatures calculated for streams that are not fully resolved in the 90-m thermal infrared (TIR) data. We first examine ASTER 15-m Visible-Near Infrared (VNIR) data to select three endmembers using a simple automated technique. These endmembers correspond to vegetation, shade/water, and other scene components (e.g. urban/soil/non-photosynthetic vegetation). Then the 15-m VNIR data are unmixed into the three corresponding fraction images. Threshold and adjacency tests are used to separate the shade and water fractions creating a total of four fraction images that together are used to specify the amount of the scene components in each 90-m TIR pixel. The emitted thermal radiance (ETR) from each of the scene components can be estimated if we assume that it is the same as for a "pure" area nearby. In practice, we estimate the ETR for each component from an image kernel of surrounding TIR pixels using a simple linear regression of the ETR against the endmember fraction, since truly pure pixels rarely exist. The ETR in each 90-m pixel is partitioned into the 36 corresponding 15-m pixels. Temperatures and emissivities are derived from the sharpened ETR images using the ASTER Temperature-Emissivity Separation algorithm. Application of this algorithm to ASTER scenes in the Pacific Northwest (USA) shows results that appear good visually and correspond quantitatively to field measurements of radiant temperatures. Our image-sharpening method addresses the need for spaceborne thermal images at higher spatial resolutions than measured today.