Predicting soil engineering parameters from spectral data using PLSR

Expansive soils give rise to major problems that need due attention in geo-technical investigations. Their characteristic to undergo large volume changes upon wetting and drying, has serious implications on the planning, design, construction, maintenance, and overall performance especially of lightweight civil engineering infrastructures. For this study, 80 soil samples were collected from the eastern part of Addis Ababa city. Specific expansive soil engineering parameters namely; Atterberg limits (liquid limits, plastic limits and plasticity indices), free swell and cation exchange capacity were measured in a soil mechanics laboratory. Reflectance spectra of each soil sample were acquired using the ASD fieldspec spectrometer. A multivariate calibration method, partial least squares regression (PLSR) analysis, through simple wavelength approach, was used to construct empirical prediction models that enable the estimation of engineering parameters of expansive soils from their reflectance spectra. Correlation coefficients obtained showed that a large portion of the variation in the engineering parameters (using full spectral band sets R=0.90, 0.87, 0.71, 0.81 and 0.71 for CEC, LL, PL, PI and FS respectively) could be accounted for by the spectral parameters. The results of the prediction models are high, indicating the potential of spectroscopy in deriving engineering parameters of expansive soils from their respective reflectance spectra, and hence its potential applicability in the geotechnical investigations of such soils.