A comparative study of independent component analysis with principal component analysis in geological objects identification. Part II: A case study of Pinghe District, Fujian, China

• ICA and PCA are applied to a geochemical data set for geological interpretation and geo-objects identification.
• Independence components (ICs) and principal components (PCs) are analyzed and compared.
• ISODATA clustering is applied in ICs and PCs and the results of classification show associations with geo-objects.
• The 3D scatterplots are used to validate the theory in Paper I.
• The influences of input data's log-normal-like distribution of are studied and an embryo solution is proposed.

In a sister paper (Paper I), emphasis was on method comparison between Principal Component Analysis (PCA) and independent component analysis (ICA), suggesting that the main components generated by PCA usually represent dominant populations of samples such as large geological bodies with extensive coverage, whereas the main components of ICA may represent the directions along which there is more divergence of populations of samples. This Paper II contains further demonstration of the differences and similarities between ICA and PCA in applications to a regional stream sediment geochemical data set for geological object identification in the Pinghe district, Fujian Province, China. Considering the input elements have log-normal-like distributions that may significantly influence the results of ICA, the raw data were log-transformed and standardized prior to ICA and PCA analysis. The results show that the second and fourth principal components (PC) obtained by PCA may indicate five geo-objects. Similarly, the second, fifth, sixth, eighth and tenth independent components (IC) obtained by ICA may indicate eleven geo-objects. The first several PCs are most likely to represent geo-objects, and the ICs with low order in kurtosis rank are more likely to reflect some geo-objects. An unsupervised classification method (ISODATA clustering algorithm) applied to scores of the PCs and ICs shows that classification of geo-objects based on the results obtained by ICs is relatively more accurate than that obtained by PCs. The 3D scatterplots for three components show that the samples belonging to the same rock unit have more clear structure in the space of ICs than in the space of PCs. Moreover, IC results show geochemical element patterns depicting different associations between hydrothermal system of Zhongteng plutonic complex and hydrothermal mineralization in the northwestern and southeastern parts of the study area. The loadings of PC indicate hydrothermal systems at different temperatures occurred in the Nanyuan Group and the intrusions. In general, ICA is more effective than PCA for characterizing geo-objects in the study area.