A Test of the Hypothesis That Syn‐Collisional Felsic Magmatism Contributes to Continental Crustal Growth Via Deep Learning Modeling and Principal Component Analysis of Big Geochemical Datasets

The origin, way of growth, and compositional transition (from basaltic to andesitic) of continental crust remain enigmatic. To better understand the evolution of the Earth's crust, geoscientists have hypothesized two competing models, one is the widely accepted island-arc model, the other is the newly proposed collision-zone model that continental collision produces and preserves syn-collisional Mantle-derived Bulk-continental-crust-like Granitoids (MBGs), and hence maintains net continental crust growth. Here, we tested the collision-zone model by investigating the existence, temporal-spatial distribution, geochemical signatures, and possible sources of the syn-collisional MBGs. We applied deep learning (DL) algorithm and principal component analysis (PCA) to the database GEOROC and Tibetan Magmatism Database. DL successfully built a regression model of whole-rock element compositional data and mean zircon epsilon(Hf)(t) data of igneous rocks. This can not only assign values to the missing Hf data, but statistically unveil the potential relations between the compositions (both isotopic and geochemical) and the possible sources of the igneous rocks. The DL and PCA enabled to recognize the MBGs and define their geochemical and isotopic fingerprints differing noticeably from arc magmas (e.g., Kohistan arc type and Tibetan adakite-like type). Besides, our observations suggest that MBGs are common in collisional settings as a response to known collision events. Moreover, the MBGs' distinct geochemical and isotopic signatures indicate that they are likely sourced from subducted ocean crust. Our results therefore generally support evident contribution of syn-collisional felsic magmatism to net continental crust growth. However, further refinement of the petrogenesis and estimation of the (relative) volume are critically needed.