Onderzoeksproject S0/00/071 (Onderzoeksactie S0)
Context and objectives
Soil Organic Carbon (SOC) stock have been the objects of a growing literature last decades due to their role in the conservation of soil quality and in the global C cycle. However, the study of their temporal evolution remains difficult due to a strong intra- and inter-field variability masking the signal with a lot of noise. As a result, the detection of SOC stock changes in soil monitoring or modelling studies require a high sampling that is rarely achieved without resorting to expensive and time consuming routine soil analysis methods. New analytical methods are needed that would allow a rapid sampling and instant determination of SOC contents.
Visible Near Infrared (VNIR) diffuse reflectance spectroscopy has been extensively explored in the laboratory as a rapid means to quantify various soil properties and OC in particular. For the time being, Portable field Spectroscopy (PS) and Imaging Spectroscopy (IS) have been less studied as SOC analytical tools.
The objective of this study is to compare the predictive ability of lab-, field- and airborne-based VNIR spectroscopy to determine SOC contents in cropland using Partial Least Square Regressions (PLSR) and try to map SOC with IS. The stability of the calibrations across time and space has also been addressed. To achieve these goals, several spectral datasets have been collected during three hyperspectral field campaigns financed by the BELSPO along the period 2003-2005.
Methodology
The flight campaign took place on June 20, 2005 using the AHS-160 hyperspectral sensor that acquired, under a clear sky, 7 South-North flight lines within the selected area. One hundred eighteen experimental plots of 7.5 by 7.5 m were cleared in 8 maize fields. The location of each plot was determined by DGPS. Soil samples were taken for analysis in the laboratory. The SOC content of each sample was determined with the so-called Walkley and Black method. The soil surface reflectance was measured with a contact probe in the laboratory, a hand held probe and the airborne sensor in the field.
Before the quantitative statistical analysis, a series of pre-treatments were applied to each spectrum to enhance the spectral response. Carbon contents were predicted from VNIR spectra through Partial Least Square Regressions (PLSR). The temporal and spatial robustness of the calibrations obtained are assessed by joining the dataset with previous samples collected during the CASI2003 flight campaign.
Results
Field spectroscopy (LS and PS) showed good performance when measuring SOC under specific surface conditions (low variability in soil moisture content, low roughness, absence of vegetation) and appropriate pre-treatments able to extract more information from noisier spectra (RMSE = 0.1-0.15 %C). The potential rapid in-situ sampling of this technique, requiring very little prior sample manipulation, outweighs the small loss of precision in comparison with traditional methods (Walkley-Black). The technique can thus be used for monitoring studies where their speed is a valuable advantage. Airborne techniques, such as IS, appear, for the time being, not able to predict SOC with an acceptable accuracy due to a low Signal to Noise Ratio. SOC Maps show a low variability and some differences between sampling points and predicted values by IS. Nevertheless, the range of SOC is consistent with the SOC content measured in the ground, suggesting that the fields can be well discriminated according to their spectral response. The greater potential lies in this technique and more effort have to be put in spectrum calibration. The use of different datasets from different study areas and field campaigns showed that calibrations are currently site-specific. The development of a regional calibration, valid for soils of the same physiographic region is thus one of the first future research priorities. Obviously, there is a need for the establishment of standard spectral measurement protocols in the field (surface conditions characteristics required, sampling needs in relation to spatial variability, robustness of the calibration across physiographic region) and soil spectral libraries covering broad zones.
