Figure 2.

(a) X-ray transmission map of a thin mudrock slice. The scale bar represents 1 mm. (b) L-curve resulting from principal component analysis. It displays the proportion of variance explained as a function of principal component PC_j, for , where m is the number of considered principal components. (c) Evaluation of the optimal number of clusters. From the silhouette criterion, the data set is best classified into four clusters. (d) Classification of WAXS signals into four clusters. The clusters are not isolated. Thus, transition regions prone to misclassification are observed. (e) Segmentation of scanning WAXS data according to clustering results. (f) Representative signals extracted as the nearest point to the cluster’s centroid. For readability, the signals are shifted along the y axis. (g) Representative signals extracted as an average of the furthest points from the centroids in each cluster. These regions are represented by the dashed circles in (d). For readability, the signals are shifted along the y axis.