World Congress of Soil Science Logo 18th World Congress of Soil Science
July 9-15, 2006 - Philadelphia, Pennsylvania, USA
International Union of Soil Sciences

Monday, 10 July 2006 - Friday, 14 July 2006
131-13

This presentation is part of 131: 1.5A Diffuse Reflectance Spectroscopy, Soil Sensing, Remote Sensing and Image Analysis - Poster

Determining the composition of mineral-organic mixes using UV-VIS-NIR diffuse reflectance spectroscopy.

Raphael A. Viscarra Rossel1, Alex McBratney2, and Rob McGlynn1. (1) Faculty of Agriculture, Food & Natural Resources, Australian Centre for Precision Agriculture, McMillan Building A05, The University of Sydney, Sydney, Australia, (2) The University of Sydney, Faculty of Agriculture, Food & Natural Resorces, JRA McMillan Building A05, Sydney, NSW, 2006, Australia

In this paper we address the need for a simple, quantitative, non-destructive and inexpensive methodology to characterise the mineral composition of soil using diffuse reflectance spectroscopy (DRS). Although there are studies that qualitatively characterise soil minerals using DRS, few studies quantify their composition in soil. Hence, our aims were: (i) to model mixtures of common soil minerals and a humic-fulvic acid mix together with various other end-member minerals as a function of their UV-Vis-NIR DRS and (ii) to use these models to predict the mineral-organic composition of independent mineral-organic mixes. Our eventual goal is to use these models to predict the mineral composition of soil and to incorporate them in a spectral soil inference system (SPEC-SINFERS) that may be used in the field, in real-time and eventually on-the-go. Our experiments used a three-factor simplex lattice design with three levels, corresponding to kaolinite (K), illite (I) and smectite (S). To this simplex we then added two levels of goethite (G) and two levels of a 50/50 mix of humic and fulvic acids (H-F). Finally we also added 3 levels of quartz (Q) to the mixes, by mass. We modelled the data using the partial least squares regression 1 algorithm (PLSR1) and made predictions using bootstrap aggregation-PLSR (or bagging-PLSR). We successfully modelled the mineral composition of our calibration data and accurately predicted the amount of K, I and S in the (independent) validation mixes (RMSEs of 3.7, 3.3 and 3.5 %, respectively). Predictions of goethite and the H-F mix (used to represent soil organic matter) were biased. Predictions of Q were invalid as quartz does not have spectral response in the UV-Vis-NIR. In future work we plan to improve our spectral libraries and our models.

Back to 1.5A Diffuse Reflectance Spectroscopy, Soil Sensing, Remote Sensing and Image Analysis - Poster
Back to WCSS

Back to The 18th World Congress of Soil Science (July 9-15, 2006)