Monday, November 5, 2007
19-9
Characterization of Soil Shrink-Swell Potential using the Texas VNIR Diffuse Reflectance Spectroscopy Library.
Katrina Hutchison1, Cristine Morgan2, and Charles Hallmark1. (1) Texas A&M University, MS 2474 TAMU, College Station, TX 77843-2474, (2) Department of Soil and Crop Sciences, Texas A&M University, College Station, TX 77843-2474
Shrinking and swelling soils cause extensive infrastructure and economic damage worldwide. Shrink-swell soils are of great concern in Texas for two reasons, 1) Texas has the most acreage of shrink swell soils in the United States, and 2) yearly evapotranspiration rates exceeds those of precipitation creating optimal conditions for soil wetting and drying cycles. This study was conducted to determine if visible near infrared diffuse reflectance spectroscopy (VNIR-DRS) can be used to predict the coefficient of linear extensibility (COLE) of soils. If successful VNIR-DRS would provide a means to rapidly quantify a soil’s shrink-swell potential real-time. Using soils that have been previously analyzed and archived in the Texas A&M Soil Characterization Laboratory, our objectives were to: 1) Predict the coefficient of linear extractability (COLE) using the VNIR spectra, 2) Predict COLE using a measurements of total clay, cation exchange capacity (CEC) and pH, 3) Compare the two models. A total of 2342 soil samples were scanned to create the Texas spectral library. Of these samples, 900 had COLE measurements. Seventy percent of the COLE samples were randomly selected to build a calibration model using partial least squares regression. The remaining thirty percent were used to validate the calibration model. The coefficient of determination (r2), root mean square deviation (RMSD), and ratio of prediction to standard deviation (RPD) were calculated to assess the prediction models. The COLE prediction using spectroscopy had an r2, RMSD, and RPD of 0.63, 0.028, and 1.6, respectively. Using stepwise regression and backward elimination, we determined that CEC alone was the best predictor of COLE with r2, RMSD, and RPD of 0.79, 0.021, and 2.2, respectively. According to the RPD, using spectroscopy to predict COLE has some predictive value, while CEC is more effective and stable. However spectroscopy data collection is more rapid and has fixed costs.