Data fusion of gamma and x‐ray spectra via Common Dimensions for soil erosion characterization

Abstract This study explores the synergistic application of energy dispersive x‐ray fluorescence (EDXRF) and gamma‐ray spectrometry (GRS) raw spectra to characterize soil redistribution. The evaluation employs Common Dimensions (ComDim), a multiblock non‐supervised machine‐learning technique, to identify shared latent patterns across nuclear and atomic data blocks from 31 soil samples (0‐ to 3‐cm depth) collected at an agricultural mega‐parcel. ComDim analysis revealed three common dimensions (CDs) explaining 68.34% of the cumulative variance. CD1 was primarily driven by the EDXRF block (salience ), effectively distinguishing samples by sampling location, while CD3 was dominated by the GRS block (salience ) and discriminated samples by collection depth. Quantitative validation using Spearman's rank correlation () demonstrated significant links between CD scores and analytical variables, particularly between CD3 and beryllium‐7 activity, providing a statistical basis for soil redistribution assessment. These findings suggest that the ComDim multiblock framework offers enhanced insights into soil dynamics compared to individual block analysis or conventional data concatenation, minimizing preprocessing assumptions while preserving original spectral information.