Multivariate statistical comparison of Northern Apennines Paleozoic sequences: a case study for the formations of Monti Romani (Southern Tuscany-Northern Latium,Italy)

The Paleozoic quartzites and graphite-rich metapelites cropping out in the area of Monti Romani, in Central Italy, show very common mineralogical and petrographical features and are lacking of fossils. For this reason, their stratigraphic attribution in the general picture of the Tuscan Paleozoic is still debated.In this work 23 rock samples have been analyzed for major and trace (Li, Ni, Cu, Zn, Rb, Sr, Y, Zr, Nb, Ag, Ba, La, Ce and Pb) elements, in order to give a contribution to their stratigraphic interpretation. Using linear discriminant analysis (LDA) on major elements, the paleozoic formations of Monti Romani have been compared with 4 “reference groups”, representing the formations of: (1) Cambro-Ordovician; (2) Middle Ordovician; (3) Late Ordovician-Silurian; (4) Permo-Carboniferous age, cropping out in Tuscany (in total 136 samples).The application of LDA to real geological data is generally limited by (a) the deviation of the data from the normal (Gaussian) distribution model, which is an important basic assumption of LDA, and (b) the closure of compositional data (e.g., major elements sum to 100), which can induce false correlations between elements. With regards to both problems, the log-ratio transformation of the data suggested by Aitchison has resulted in a satisfactory solution for the application of LDA.The results of LDA show that most Paleozoic formations of Monti Romani closely resemble the Tuscan Permo-Carboniferous formations: the samples are in fact attributed to this group with probability ranging from 0.632 to 0.995. Only the data of the “Valle Tegolaie Carbonatic-Phyllitic Complex” give contradictory indications; this unit is in fact believed to be Devonian in age, on the basis of field evidence, and should be eventually compared with the other Silurian-Devonian formations of the Northern Apennines.Correlations between elements have been studied with the Spearman rank correlation coefficient ϱ, which is much less influenced by the non-normal distribution of the data than the Pearson coefficientr. The analysis of the Spearman's matrix allowed some inferences about mineralogy. The high correlation of Fe_tot and Na with K, Al and LOI should account for a phengitic-paragonitic composition of sericite. The same mineral strongly controls the distribution of most trace elements, namely Li, Ni, Rb, Sr, Y, Nb, Ag, Ba, La and Ce (coefficients between 0.600 and 0.877). In contrast, low coefficients for Cu, Pb and, to minor extent Zn, suggest that other minerals such as graphite or sulphides are probably important for the distribution of these elements. In the same way, Zr does not show any significant correlation, as expected, being concentrated mainly in the granoblastic layers (detrital zircon).