Can hyperspectral techniques improve estimates of carbon stocks in agricultural soils?

Soil organic carbon (SOC) represents one of the major pools in the global carbon cycle. However, fluxes of CO2 from soils into the atmosphere by respiration or inversely sequestration of CO2 through photosynthesis and subsequent immobilisation in the form of humus are difficult to quantify. In principle changes in SOC stock over time reflect CO2 fluxes. The detection of these stock changes, however, require intensive sampling mainly due to the large spatial variability of SOC both within individual fields and larger units with similar soils and land use. The aim of this paper is to evaluate the potential of airborne-hyperspectral techniques using a CASI sensor and hand held Near Infrared Spectroscopy (NIRS) with an ASD spectrometer to conduct SOC inventories of individual parcels. During a field campaign in the Belgian Ardennes during Octobre 2003, more than 120 sites on a regular grid within 13 freshly ploughed fields were selected. At these sites, field spectra of the bare soil have been measured and samples from the topsoil were taken. SOC content (Walkley and Black), soil moisture and bulk density of these samples have been determined. As a first step, the soil reflectance has been transformed (log (1/R), Savitsky-Golay smoothing and derivative, gap derivative, moving average) in order to filter the spectral responses and to eliminate noise. Then, we used both stepwise and partial least square (PLS) regression analysis to relate these spectra to measured SOC contents. Regression models performed much better when the data were divided in two sub-groups representing different moisture conditions of the soil surface. These statistical model calibrations were validated on an independent data set. Standard Error of Prediction (SEP) ranged from 0.19 to 0.24 % carbon for the field spectra determined using the ASD depending on soil moisture of the surface layer. This is a little bit more than the reproducibility error inherent to the Walkley and Black analysis. Airborne CASI techniques performed less well mainly due to the narrow spectral range. Tests on airborne CASI+SASI hyperspectral data from a previous field campaign [1] showed better results. Overall, low bias allowed the use of spectral techniques to estimate population means with a high confidence level. The spectral techniques have a strong potential in determining changes in carbon tock change studies. The large within field variability of SOC content precludes the assessment, using conventional soil sampling, of SOC changes as a result of management (1 t C ha yr) over a reasonable time period (5 years). Depending on the variance of the SOC content measured in the field (σ = 11-166 t C ha), we need 16-210 samples to detect a change. Since this number of samples is rarely available for individual fields, conventional sampling methods can only be used for larger spatial units containing many fields. In contrast, the airborne-hyperspectral technique and portable NIRS are able to supply these large amounts of data, and can thus improve the accuracy of SOC stock assessments of individual fields. This in turn will result in a smaller detection limit of SOC stock changes.