Estimates of spectrometer bandpass, sampling interval, and signal-to-noise ratio required for identification of pure minerals and plants were derived using reflectance spectra convolved to AVIRIS, HYDICE, MIVIS, VIMS, and other imaging spectrometers. For each spectral simulation, various levels of random noise were added to the reflectance spectra after convolution and then each was analyzed with the Tetracorder spectral identification algorithm (Clark et al., 2003). The outcome of each identification attempt was tabulated to provide an estimate of the signal-to-noise ratio at which a given percentage of the noisy spectra were identified correctly. Results show that spectral identification is most sensitive to the signal-to-noise ratio at narrow sampling interval values but is more sensitive to the sampling interval itself at broad sampling interval values because of spectral aliasing a condition when absorption features of different materials can resemble one another. The bandpass is less critical to spectral identification than the sampling interval or signal-to-noise ratio because broadening the bandpass does not induce spectral aliasing. These conclusions are empirically corroborated by analysis of mineral maps of AVIRIS data collected at Cuprite, Nevada between 1990 and 1995, a period during which the sensor signal-to-noise ratio increased up to sixfold. There are values of spectrometer sampling and bandpass beyond which spectral identification of materials will require an abrupt increase in sensor signal-to-noise ratio due to the effects of spectral aliasing. Factors that control this threshold are the uniqueness of a material's diagnostic absorptions in terms of shape and wavelength isolation, and the spectral diversity of the materials found in nature and in the spectral library used for comparison. Array spectrometers provide the best data for identification when they critically sample spectra. The sampling interval should not be broadened to increase the signal-to-noise ratio in a photon-noise-limited system when high levels of accuracy are desired. It is possible, using this simulation method, to select optimum combinations of bandpass, sampling interval, and signal-to-noise ratio values for a particular application that maximize identification accuracy and minimize the volume of imaging data.

HTML and pdf versions of the entire JGR publication are being served from the JGR pub server. Click here to go to the JGR pub server. Use the back button or a link to come back to speclab. (There are no links back to speclab from the JGR pubs server.)

Pre-print PDF Version:

Text and figures (6 Mbytes).

---

U.S. Geological Survey, a bureau of the U.S. Department of the Interior
This page URL= http://speclab.cr.usgs.gov/PAPER/snpaper
This page is maintained by: Dr. Gregg A. Swayze gswayze@speclab.cr.usgs.gov
Last modified Jan. 18, 2008.