Stefanov, W.L., P.R. Christensen, M.S. Ramsey, Geology Department, Arizona State University, Box 871404, Tempe, AZ 85287

Mineralogical analysis of soils is important for understanding soil development, paleoclimate, and soil mechanics. Energy emitted from mineral mixtures in the mid-infrared (7-25 micrometers) combines linearly as a function of areal percentage of each mineral present. Using a linear, least squares fitting technique and spectral endmember library allows retrieval of the mineral (endmember) spectra and the percent contribution to the mixed spectrum. The use of this technique for mineral identification and quantification has been established for bulk rock samples and mineral separates. The present work investigates the applicability of the technique to soils.

Mid-infrared soil reflectance spectra (7-14 and 7-15 micrometers) were obtained from the USGS online soil spectral database and converted to emission using Kirchoff's Law. A total of 36 soils (representing seven soil orders) with detailed mineralogical data were selected to test the deconvolution technique. An endmember library with 198 spectra was created using the following grain size ranges (in micrometers): 1000-710, 500-125, 125-45, and <45. The endmember spectra were obtained from direct measurement at Arizona State University and from the Jet Propulsion Laboratory online spectral library.

The average retrieval accuracy of all reported minerals is 48%, major minerals (>5% reported abundance) is 71%, and clay minerals is 68%. Percent error between reported major mineral abundance and calculated spectral abundance ranges from 1-98%. Higher deconvolution errors are correlated with decreasing dominant grain size, presence of organic material, and coatings (clay, organic, and oxide) on grains. Because organic materials and rhyolitic to andesitic composition glasses are not represented in the endmember library, it is likely that this is an additional source of deconvolution error. Despite the high variability of quantitative error, identification of major minerals and clays present in the soil samples is encouraging. Work is ongoing to refine the technique and address the sources of error. This involves acquisition of emission spectra from mineralogically characterized unwashed and washed grain size separates of soils collected from the McDowell Mountains, Scottsdale, AZ.

Key words: soil, emission, deconvolution, mineral, spectra

------------------
Presented at: Geological Society of America Annual Meeting
Date: 1998