The use of hyperspectral data to estimate soil properties

Recently there has been a significant increase in the amount of satellite and airborne remotely sensed data (e.g., multi- and hyperspectral) available for use in engineering. The use of these types of data for the assessment of soil properties requires the identification and quantitative estimate of key parameters from soil spectral measurement. This study focuses on soil water content and soil composition. Spectral response of soils with different water content was studied. Artificial neural networks were developed to predict water content of soils from their spectral response. The predicted water contents are in good agreement with the actual water contents. The study could serve as an initial step for developing in-situ water content determination techniques. Spectral un-mixing was also discussed in this study. The N-FINDER algorithm was adapted to find the end-members in a hyperspectral data set. The linear mixture model and the Orthogonal Subspace Projection (OSP) algorithm were used to calculate the fractions of end-members. It is shown that the linear model and OSP can give a fairly good estimate of the end-member abundances. A hyperspectral imaging software tool was used to analyze an AVIRIS data set following a proposed processing routine. The abundance maps of end-members were generated. It is expected that the spectrally determined soil information can be used along with other soil engineering correlations to help engineers characterize a site.