Combining wavelets with statistical inference to map the mineralogical composition of pedological features from synchrotron X-ray diffraction data

Abstract

Clay translocation is among the most widespread processes in soils. It is generally identified by the presence of clay coatings at the macroscopic or microscopic scales. Nevertheless, several authors demonstrated that clay coatings have different origins, which renders the attribution of individual coatings to a particular process difficult. Therefore, their characterization at the microscopic scale is necessary. Modern synchrotron techniques allow mapping of the mineral composition of soil thin sections by X-ray diffraction with a lateral resolution of a few micrometers that is compatible with the size of the clay coatings. However, the use of this technique raises a certain number of technical difficulties when clay minerals and small pixel size are considered. Therefore, a preliminary feasibility analysis was performed on soil coating obtained experimentally in a soil column experiment. We developed a mathematical/statistical method to automatically identify and map the minerals present in synchrotron X-ray diffraction maps. This method combines a subtraction of the background signal, an identification of the diffraction peaks and the attribution of the obtained peaks to the different minerals phases. The robustness of the method was tested for the two first steps that were critical. We then showed that most of the minerals present in the considered samples could be identified, including clay minerals and Fe oxides; however, their relative proportions were difficult to estimate within the experimental conditions. As a conclusion, we proposed here a new mathematical method for the data treatment of synchrotron X-ray diffraction. Our study shows that this method could be applied on natural heterogeneous samples with data of poor quality to perform robust qualitative analysis. Moreover, with the recently available fast data acquisition schemes, large amounts of datasets (diffractograms) can be rapidly acquired and are expected to benefit the automatic data treatment approach proposed here.