Suivi du carbone organique des sols agricoles par télédétection hyperspectrale

Executive summary Conventional sampling technique are often too expensive and time consuming to meet the amount of quantitative data required in soil monitoring or modelling studies. The emergence of portable and flexible VNIR sensors could provide the large amount of spatial data needed. In particular, the ability of imaging spectroscopy to cover large surfaces in a single campaign and study the spatial distribution of soil properties with a high spatial resolution represents an opportunity for improving the monitoring of soils. However, some challenges still remain to be solved concerning disturbing factors and the accuracy of the SOC analysis. Disturbing factors, especially soil roughness and moisture content, must be taken into account to produce good calibration models. These factors induce a spectral variability not related to the studied property (here, SOC) and degrade the accuracy of the image-based predictions. The use of hyperspectral remote sensing as a fast analysis of SOC stocks could lead to a loss of precision, which should be evaluated because it may be incompatible with the accuracy needed by end-users in the evaluation of the impact of agricultural practice on SOC stocks. Until now, imaging spectroscopy has been generally applied over small areas or homogeneous soil types and surface conditions. During the MOCA project: · Five hyperspectral images acquired with the AHS-160 sensor were analysed to predict Soil Organic Carbon (SOC) in an area in Luxembourg characterized by different soil types and a large variation in SOC contents. · The effect of soil Relative Shadow (RS, the percentage of shadowed soil of the surface studied) on SOC prediction from spectral data under field conditions was quantified. First, the impact of RS on reflectance and SOCp is briefly described. Then, a methodology to measure RS and correct its impact on field reflectance measured with an ASD FieldSpec Pro spectrometer and the AHS-160 sensor is proposed. Finally, SOC content is predicted with uncorrected and corrected reflectance values to evaluate the enhancement in SOC prediction accuracy. · The results of the investigations both in the laboratory (wet chemical SOC analysis (CONVIS), LECO CN analyzer (calibration and validation dataset) and with remote sensing via airplane were compared Reflectance data were related to surface SOC contents of bare croplands by means of 3 different multivariate calibration techniques: Partial Least Square Regression (PLSR), Penalizedspline Signal Regression (PSR) and Least Square Support Vector Machine (LS-SVM). The performance of the methods was tested under different combinations of calibration/validation sets (global and local calibrations stratified according to agro-geological zones, soil types and image number). The results demonstrated that PSR and LS-SVM performed better than PLSR using global calibrations. The Root Mean Square Error in the Predictions reached 5.6-6.2 g C kg- 1. Under local calibrations, this error was reduced by a factor 1.3 to 1.9, depending on the stratification scheme adopted. Pixels of two agricultural fields were extracted from the data cube and predicted for SOC with the best models. Intra- and inter-field variability of SOC contents were observed related to topography and land management. In the future, the mapping of SOC over the entire study area will constitute a database used as input in digital soil mapping and SOC monitoring. Tests under laboratory conditions showed that the prediction of SOC decreases when the relative shadow increases. A methodology for correcting the effect of relative shadow on reflectance spectra measured with ASD or AHS during field campaign was elaborated and tested. Results show that the methodology enables to significantly enhance SOC prediction in all cases studied. Correction always improves the prediction of SOC (and increase of 25 % in RMSEP for raw reflectance) when using non pre-processed reflectance. The best prediction of SOC is always achieved with corrected pre-processed reflectance. From the point of view of an agricultural extension organization in the field of fertilization planning as well as of maintaining and improving soil fertility such as the CONVIS s.c., the results presented above have to be considered positive and the investigations of the MOCA project a successful experiment. The results and the related calibration models appear to be able to deliver in most cases values of SOC which are precise enough to be used in agricultural extension. In order to optimize the calibration models of the remote sensing investigation, traditional chemical analysis of other fields of the investigated air corridor should be made and the results compared with the SOC values derived from imaging spectroscopy value. This could deliver more information about the strong points and the limitations of the applied method.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.