Mantle sources of Cenozoic volcanic activities around the South China Sea revealed by geochemical and isotopic data using the principal component analysis (PCA)


 Principal component analysis (PCA) was conducted to analyze geochemical and isotopic data and interpret the characteristics and types of mantle sources of Cenozoic volcanic activities around the South China Sea (SCS). Fifteen trace element indicators and five isotopic indicators were surveyed from 623 volcanic rock samples obtained from the SCS, Hainan Island, Fujian–Zhejiang, Taiwan, Vietnam, and Thailand to characterize the geochemical properties of the volcanic rocks, determine the types of mantle sources, and assess the influence degree of each mantle source. Two principal components (PCs) were extracted by PCA based on trace elements and Sr–Nd–Pb isotopic ratios, which are an enriched oceanic island basalt-type mantle plume and a depleted mid-ocean ridge basalt-type spreading ridge. In the Southeast Asian region, the influence of Hainan mantle plume on younger volcanic activities (< 13 Ma) was greater than that on older ones (> 13 Ma) in the same location. PCA was used to verify the mantle plume–ridge interaction model of volcanic activities beneath the expansion center of the SCS and refute the hypothesis that the tension in the SCS is triggered by the Hainan plume. The results of this study demonstrate the efficiency and applicability of PCA to the discussion of mantle sources of volcanic activities and provide a new method with which to analyze geochemical data.


Introduction
Extensive and voluminous Cenozoic basalts are widely distributed within the Southeast Asian region ( Fig. 1; Sun et al., 2009;Ho et al., 2000;Fedorov and Koloskov, 2005;Zou and Fan, 2010;Wang et al., 2012;Zhang et al., 2018), including the South China Sea (SCS) Basin (Yan et al., 2008;Zhang et al., 2017Zhang et al., , 2018, Hainan Island, Leizhou Peninsula (Wang et al., 2012;Li et al., 2013;Liu et al., 2015;Zou and Fan, 2010), Fujian-Zhejiang (Huang et al., 2017;Ho et al., 2003), Taiwan (Tian et al., 2019), Vietnam Hoang et al., 2018), and Thailand . The detailed geological setting of Southeast Asia is present in Supplementary I. The SCS shows evidence of a diverse array of spatially and temporally complex tectonic processes, including continental rifting, sea oor spreading, subduction, and terrane collision, all of which make up what is called a complete Wilson cycle (Li et al., 2014a, b;Zhou et al., 2009). The SCS is a highly complex research area worthy of in-depth study. Previous scientists proposed a variety of tectonic dynamic hypotheses to explain the Cenozoic volcanic activities in the SCS and its surroundings on the basis of multiple geochemical and geophysical study methods; these hypotheses include the upwelling of the mantle plume Zhou et al., 2009), the retreat and withdrawal of the subducted Paleo-Paci c Plate (Shi and Li, 2012), tectonic extrusion related to the India-Eurasia collision (Briais et al.,1993), sea oor tension as a result of the subduction of the proto-SCS (Hall, 2002), and the mantle plume-spreading ridge interaction model Zhang et al., 2018). However, the genesis of volcanic activities in the SCS and its surrounding areas remain controversial and ambiguous. For example, the dynamic mechanism triggering the expansion of the SCS is yet unknown, and the connection between the Hainan mantle plume and the expansion of the SCS is incompletely understood. Moreover, the in uence of the Hainan mantle plume on volcanic activities in Southeast Asia has not been studied in detail. Given these knowledge gaps, analysis of the geochemical data of Cenozoic volcanic rocks covering a wide range of areas surrounding the SCS is necessary. Geochemical data contain multiple correlated indicators, such as trace elements and isotopes. Hence, extracting information from these data, such as the independent factors contributing to the rock contents, requires an advanced method. The present study employs a novel research method, i.e., principal component analysis (PCA), to explore the characteristics and properties of the mantle sources of volcanic activities in the SCS and its surrounding regions.
PCA, an important method of multivariate statistical analysis, is an effective dimension-reduction technique and a comprehensive evaluation strategy (Abdi and Williams, 2010). This method can easily identify the most "main" elements and structures of a dataset, remove noise and redundancy, reduce the number of dimensions of the original data, and reveal the simple structures hidden behind complex data (Abdi and Williams, 2010). Compared with other techniques, PCA presents the advantages of simplicity, no parameter limitation, and easy application to various analyses. Therefore, it is widely used and regarded as one of the most valuable applications of linear algebra (Zhao, 2016). Improvements in theory and the advancement of computer technologies have enabled the use of PCA to solve geoscience problems. As a powerful analytical tool, PCA may be applied to various branches of earth science, especially in the elds of meteorology and remote sensing (e.g., Abdi and Williams, 2010). For example, the PCA method could be applied to visualize multiband remote sensing data (Shimizu et al., 2010), solve the hydrochemical characteristics of groundwater systems (Peng et al., 2015), analyze atmospheric aerosols in Mexico City (Miranda et al., 2000), and describe the recurrent snowmelt pattern of multiyear remotely sensed snow cover (Woodruff and Qualls, 2019). Indeed, the application of PCA to geoscience represents a novel research approach characterized by diversi ed development and integration with new methods.
This study applies the PCA method to the analyses of geochemical data (e.g., trace elements and isotopes) and discover the characteristics of the mantle sources of the SCS and its surrounding areas.
Determination of independent potential sources is often challenging because one chemical or isotopic indicator may originate from different sources, and one source can bring multiple trace elements and isotopes (Chen et al., 2020). Thus, we adopted PCA to replace the original variables (i.e., the contents of trace elements or isotopes) with a smaller number of derived variables that are easier to explain. These derived variables, called principal components (PCs), are linear combinations of the original variables.
Each PC usually represents one independent geochemical or isotopic source for an indicator (Chen et al., 2020). The success of our analyses proves that the PCA method is effective, feasible, and applicable for dissecting geochemical data and understanding the nature of mantle sources of volcanic activities. More importantly, we introduced a new study method for the analyses of geochemical data.
- (1) The rst group includes 11 trace elements (Rb, Nb, Hf, Th, U, La, Ce, Nd, Sm, Eu, and Tb) with correlation coe cients exceeding 0.6; in fact, most of the correlation coe cients obtained reached 0.9. These ndings indicate that the 11 elements are strongly positively correlated and generally enriched or depleted simultaneously. In general, an oceanic island basalt (OIB)-type magmatic source likely causes the enrichment of these 11 trace elements (Sun and McDonough, 1989). Thus, we can preliminarily speculate that this group represents an OIB-type mantle source.
(2) The second group includes three trace elements, including Yb, Lu, and Y, the correlation coe cients of which exceed 0.8. This group may represent enriched mid-ocean ridge basalt (E-MORB)type mantle sources because E-MORB can generally cause the enrichment of Yb, Lu, and Y contents (Sun and McDonough, 1989). (3) The third group contains only Ba element, which is not signi cantly correlated with any of the 14 other elements (correlation coe cients < 0.21). This result reveals that Ba element is a good independent indicator. Thus, Ba element may re ect a subduction-related mantle source (e.g., subduction-related uids or sediments) (Leeman et al., 1994;Brian et al., 2009;Hanyu et al., 2012).

PCA calculations and dimensionality reduction
After data normalization and spatial projection calculations, we obtain 15 PCs with linear expressions for the trace element contents. The proportions of data information explained by each PC are respectively 65.9%, 19.3%, 6.5%, 3.2%, and 1.8%, etc. PCs accounting for over 5% of the data information are selected for analysis; thus, only the rst three PCs, which could explain 91.7% of the data information, are considered in this work. Table 1 shows the trace element coe cients of these three PCs, which are designated PC1, PC2, and PC3. The Rb, Nb, Hf, Th, U, La, Ce, Nd, Sm, Eu, and Tb coe cients corresponding to PC1 have larger positive values compared with those corresponding to PC2 and PC3. The Yb, Lu, and Y coe cients corresponding to PC2 have larger positive values compared with those corresponding to PC1 and PC3. The Ba coe cient corresponding to PC3 has a large positive value (Table   1). Thus, the trace elements affecting PC1 are Rb, Nb, Hf, Th, U, La, Ce, Nd, Sm, Eu, and Tb, those mainly affecting PC2 are Yb, Lu, and Y, and that affecting PC3 is Ba. This analysis is consistent with the ndings described in Section 2.1.1. Thus, PC1, PC2, and PC3 may represent an enriched OIB-type mantle source, a depleted MORB-type mantle source, and the involvement of subduction-related uid/sediment, respectively.  - We use cluster analysis to understand the similarity between the combined and standard samples. In the cluster maps obtained (Fig. 2), samples with similar geochemical features are clustered on the same branch. Most volcanic rocks from the SCS and its surrounding regions are clustered on the same branch as the OIB standard sample (Fig. 2), which indicates that deep enriched OIB-type magmatic sources widely affect the volcanic activities of the SCS and its surrounding regions. 1, 19-23, and 25 combined samples are clustered on the same branch as the N-MORB and E-MORB standard samples (Fig. 2), thus revealing that these volcanic samples are mainly affected by MORB-type spreading ridge mantle sources.
Scatter diagrams of the 110 combined and ve standard samples (Fig. 3) are constructed on the basis of the PC values (Appendix 3) and clustering analysis results (Fig. 2). These diagrams intuitively show that most volcanic rocks in the SCS and its surrounding regions (black crosses in Fig. 3) are clustered together, which is basically consistent with the PC values of OIB. This nding con rms that volcanic activities around the SCS are closely related to an OIB-type mantle source.

Correlation analyses of the original indicators (isotopic ratios)
Analysis of the correlation matrix of ve isotopic ratios of 623 volcanic samples (Appendix 4) reveals that these ratios could be divided into two groups. (1)

PCA calculations and dimensionality reduction
After data normalization and spatial projection calculation, we obtain ve PCs with are linear expressions for the isotopic ratios. The proportions of data information explained by each PC are respectively 79.9%, 13.6%, 3.7%, and 2.4%, etc. PCs accounting for over 5% of the data information are selected for further analysis; thus, only the rst two PCs, which could explain 93.5% of the data information, are considered.  (Sun and McDonough, 1989;Zindler and Hart, 1986). PC1 may re ect a slightly enriched OIB-type mantle plume in uenced by an enriched mantle 1 (EM1)-type mantle source (Sun and McDonough, 1989;Zindler and Hart, 1986).    -4, 6-7, 9-12, 16-27, 32-43, 45-46, and 49-61 combined samples have relatively signi cantly higher PC1 values relative to PC2 values, which indicates that these samples are affected by the OIB-and EM1-type mantle source (Appendix 5; Fig. 4). Scatter diagrams and cluster analysis directly illustrate that these samples are clustered together, consistent with the PC values of OIB and EM1 (black cross in the Fig. 5). These results indicate the samples are related to OIB-and EM1-type mantle sources.

Discussion
Trace elements were analyzed using the PCA method, and three PCs that could explain 65.9%, 19.3%, and 6.5% of the variance found were extracted, which are an enriched OIB-type mantle plume source with fairly high trace element contents (PC1), a depleted MORB-type mantle source featuring spreading ridges with enriched Yb, Y, and Lu contents (PC2), and subduction-related uids/sediments with large variations in Ba contents (PC3; Table 1). PCA was used to analyze the Sr, Nd, and Pb isotopic ratios, and two PCs were respectively found to explain 79.9% and 13.6% of the variance observed, which are a relatively enriched OIB-type mantle plume source containing large amounts of EM1-type components with high typical depleted MORB-type mantle source with relatively low Pb and Sr isotopic ratios and a high Nd isotopic ratio (PC2; Table 2).
The OIB-type mantle plume revealed by the PC1 values determined from PCA calculations of trace elements and isotopic ratios is the most important magma source dominating volcanic activities in the SCS and its surrounding areas. OIB-type volcanic activities around the SCS are most likely due to the Hainan mantle plume (Hoang and Flower, 1998;Zhou et al., 2009;Zou and Fan, 2010;Wang et al., 2012Wang et al., , 2013Huang et al., 2013;Liu et al., 2017;Yan et al., 2006Yan et al., , 2018 Wang et al., 2012;Liu et al., 2015). This result indicates than an enriched OIBtype mantle plume plays a more signi cant role in Hainan volcanic activities during the period 0-12.6 Ma than in those during the period 4.6-33 Ma (Appendix 6). The above observations reveal that younger combined samples collected in the same area have higher PC1 values than older samples. Similar phenomena have been observed in the SCS and its surrounding areas, including the seamounts of the SCS, the expansion center of the SCS, Zhejiang-Fujian, Thailand, and Vietnam (Appendixes 6, 7; Supplementary I), thus indicating that the phenomenon is not accidental but, instead, quite common in our study area. We speculate that the in uence of the Hainan mantle plume on the Cenozoic volcanic activity in this area gradually strengthened. Speci cally, the in uence of the Hainan mantle plume on nearby young volcanic activity during the period of <13 Ma is much stronger than that during the period of >13 Ma (Appendixes 6, 7; Supplementary I).
PCA is a powerful tool that could reveal the potential mantle sources of Southeast Asian volcanic activities on the basis of geochemical or isotopic indicators. PCA of the trace elements of volcanic samples from Hainan Island, seamounts in the SCS, and expansion center of the SCS yielded two PCs that explained over 85.2% of the variance observed; these PCs are represented by the red (PC1; enriched OIB-type mantle plume) and blue (PC2; depleted MORB-type spreading ridge) lines in Fig. 6, respectively. The ordinate of Fig. 6 re ects the relative degree of in uence of the two PCs on the trace element compositions of the volcanic samples. The gure clearly shows that <13 Ma volcanic samples from Hainan Island (0-12.6 Ma), seamounts of the SCS (3-8 Ma), and the expansion center of the SCS (7.4-12.8 Ma) have similar and high PC1 values (0.38-4.62, 1.38-7.63, 5.14), which indicates that these sample are similarly affected by the OIB-type mantle plume to the same degree (red lines in the Fig. 6A). The <13 Ma samples from the expansion center of the SCS (7.4-12.8 Ma) have high PC1 (5.14) and PC2 values (2.25), thus indicating that these samples are simultaneously in uenced by the OIB-type mantle plume (red lines in the Fig. 6A) and the MORB-type spreading ridge (blue line in the Fig. 6A). Therefore, the role played by OIB-type mantle plumes in the formation of <13 Ma volcanic activities in the expansion center of the SCS should not be ignored or underestimated (Fig. 6A). Volcanic activities in the expansion center of the SCS are affected by MORB-type spreading ridges and the OIB-type mantle plumes, which con rms the validity of the mantle plume-ridge interaction model in the expansion center of the SCS. In addition, volcanic activities in the expansion center of the SCS during the periods of 7.  Fig. 7, respectively. The ordinate of Fig