Speciation of Energy Critical Elements in Marine Ferromanganese Crusts and Nodules by Principal Component Analysis and Least-squares fits to XAFS Spectra

Marine ferromanganese crusts and nodules ("Fe-Mn crusts") are considered a potential mineral resource due to their accumulation of several economically-important elements at concentrations above mean crustal abundances. They are typically composed of intergrown Fe oxyhydroxide and Mn oxide; thicker (older) crusts can also contain carbonate fluorapatite. We used X-ray absorption fine-structure (XAFS) spectroscopy, a molecular-scale structure probe, to determine the speciation of several elements (Te, Bi, Mo, Zr, Pt) in Fe-Mn crusts. As a first step in analysis of this dataset, we have conducted principal component analysis (PCA) of Te K-edge and Mo K-edge, k³-weighted XAFS spectra. The sample set consisted of 12 homogenized, ground Fe-Mn crust samples from 8 locations in the global ocean. One sample was subjected to a chemical leach to selectively remove Mn oxides and the elements associated with it. The samples in the study set contain 50-205 mg/kg Te (average = 88) and 97-802 mg/kg Mo (average = 567). PCAs of background-subtracted, normalized Te K-edge and Mo K-edge XAFS spectra were performed on a data matrix of 12 rows x 122 columns (rows = samples; columns = Te or Mo fluorescence value at each energy step) and results were visualized without rotation. The number of significant components was assessed by the Malinowski indicator function and ability of the components to reconstruct the features (minus noise) of all sample spectra. Two components were significant by these criteria for both Te and Mo PCAs and described a total of 74 and 75% of the total variance, respectively. Reconstruction of potential model compounds by the principal components derived from PCAs on the sample set ("target transformation") provides a means of ranking models in terms of their utility for subsequent linear-combination, least-squares (LCLS) fits (the next step of data analysis). Synthetic end-member models of Te⁴⁺, Te⁶⁺, and Mo adsorbed to Fe(III) oxyhydroxide and Mn oxide were tested. Te⁶⁺ sorbed to Fe oxyhydroxide and Mo sorbed to Fe oxyhydroxide were identified as the best models for Te and Mo PCAs, respectively. However, in the case of Mo, least-squares fits contradicted these results, indicating that about 80% of Mo in crust samples was associated with Mn oxides. Ultimately it was discovered that the sample from which Mn oxide had been leached was skewing the results in the Mo PCA but not in the Te PCA. When the leached sample was removed and the Mo PCA repeated (n = 11), target transformation indicated that Mo sorbed to Mn oxide was indeed the best model for the set. Our results indicate that Te and Mo are strongly partitioned into different phases in these Fe-Mn crusts, and emphasize the importance of evaluating outliers and their effects on PCA.