Assessment of two reflectance techniques for the quantification of field soil organic carbon

2003

Raphael A. VISCARRA ROSSEL^1,2, Christian WALTER^2,3 and Youssef Fouad³

¹ Australian Centre for Precision Agriculture, University of Sydney, NWS 2006, Australia
² INRA Laboratoire de Science du Sol, 65 rue de St Brieuc, 35042 Rennes, France
³ Ecole Nationale Supérieure Agronomique, Rennes, France

In J.V. Stafford and A. Werner (Eds), Precision Agriculture, Fourth European Conference on Precision Agriculture, Berlin.

ABSTRACT

The primary aim of this work is to compare predictions of soil organic carbon using two different reflectance techniques; specifically, soil colour measurements derived from digital images and spectrometric measurements made using a visible-range spectrometer. Digital images of forty-three different soils, collected from various locations across Brittany, France, were acquired in the laboratory under ‘ideal’ lighting conditions. Soil colour was represented using RGB and CIE tristimulus values. The RGB pixel data was extracted from each digital image and median R, G, B pixel values calculated. These data were also transformed to CIEL*a*b* and CIEL*u*v colour coordinates. Relationships between R, G, and B image-intensities, CIEL*a*b* and CIEL*u*v values and SOC were derived for predictions of SOC content. The visible and near (near) infrared spectra (400 nm – 1100 nm) of these soils were also measured using a spectrometer and partial least-squares regression (PLSR) implemented on the spectra for predictions of SOC. Both colour and PLSR calibration models were used for site-specific determinations of SOC in two agricultural fields in Brittany. Predictions of field SOC using simple soil colour models were good (root mean squared-error (RMSE) values of 0.36 and 0.34 dag/kg for each field respectively) and comparable or better than spectrometric PLSR predictions (RMSE 0.36 and 0.54 dag/kg for each field respectively).