Channel selection for soil spectrum reconstruction in 8-13 μm region

On the basis of the method developed by Price [1990] for selecting a limited number of channels to best represent the high-resolution spectra of materials within a spectral region, this paper presents a simplified procedure to reconstruct such spectra directly from their spectral measurements in the selected channels. Applying this procedure, it is shown that spectral reflectivities of more than 50 soil and vegetation samples measured in the laboratory at Johns Hopkins University (JHU) can be reconstructed using six selected channels in the 8-13 mu m spectral region with an uncertainty of 0.005. It is also shown that the process of spectral channel selection proposed in this paper minimizes the propagation of measurement error to the whole reconstructed spectrum. Thus, if the reconstruction of spectrum is nearly insensitive to a small change in the center wavelengths and widths of the selected channels, the resulting errors on this reconstructed spectrum due to the measurement errors are increased by such a change. In order to validate this approach, the channels selected using the JHU data set are used to reconstruct the spectral data measured at the University of California at Santa Barbara for 43 types of soils. The results showed that the soil reflectance spectrum could be reconstructed by the channel reflectance measured in these six channels with their basis functions to within 0.005 almost over the full spectral range except for wavelengths around 8.6 mu m and 9.5 mu m for which the reconstruction is within 0.009. It should be kept in mind that these results refer to laboratory spectral data but not to remote sensing data where additional uncertainties will come from radiometric noise, errors associated with radiometric calibration, atmospheric corrections, and temperature/emissivity separation. Appreciable future work therefore has to be done with remote sensing data.