Petrogenesis of Basalts: Examination of REE Data Using Pattern Recognition Approach Through Principal Component Analysis
Abstract
A pattern recognition (PR) approach based on the principal component analysis (PCA) is adopted for the petrogenesis of basalts using the relative behaviour among the REE, on the data-sets of tholeiites (MOR, ocean island and continental) and alkalic basalts (melilites, nephelinites and alkali olivine basalts). It is observed that the overall variability in the REE data requires two or three principal components; with the first component accounting for 80% of the variability. The coefficient patterns of these components reveal distinct features. The dominant (first) component of tholeiites, exhibits a poor covariance of La and Ce with MREE. Coefficient patterns of the dominant component of tholeiites show breaks at Sm and at Dy/Tb similar to those observed in high pressure partition coefficients for cpx. The coefficient patterns in this major component appear to describe the REE behaviour during partial melting and high pressure fractionation of opx or cpx and eclogite.In contrast with the tholeiites, the dominant component in alkalic basalts, shows excellent covariance among the LREE and MREE. Among the alkalic basalts, the alkali olivine basalts, show a different HREE behaviour. Comparison of the coefficient patterns of the dominant component of tholeiites and alkalic basalts suggests distinct differences in their source/mechanism of formation.
Theoretical computations involving partial melting and fractional crystallisation of only the major mineral (ol, opx, cpx, garnet/spinel) prove to be inadequate to simulate the observed coefficient patterns. The inclusion of 87Sr/86Sr isotopic ratios along with REE in the PCA, generates 'features' suggestive of additional sources (interstitial liquid, accessory minerals in the mantle and/or granitic crustal contamination) as contributing factors.
The PR approach described in this work allows one to extract more information from the REE data-set than what is possible by simple chondritic normalisation.