Deciphering the “rock record” of the Hadean crust: Insights into the early Earth evolution by using the machine learning

The early history of the Earth, especially the Hadean Eon (> 4.0 Ga), marks a crucial stage of the formation and evolution of the Earth. Yet, due to the almost complete loss of rocks, which were formed in the Hadean Eon, by the post-Hadean geological processes, the compositions and formation mechanisms of the Hadean continental crust are poorly understood. How to restore geochemical characteristics of the early crustal rocks of the Earth has long been a thorny problem faced by researchers on the precambrian geology. Recent advances in the wide application of big data and machine learning have provided new tools for exploring the intrinsic connection between the composition of magma zircons and their parental magma. The advantage of the machine learning lies in its capability to effectively handle multi-dimensional geochemical features and capture the hidden nonlinear relationships in those data. Using the most comprehensive zircon–whole-rock paired database collected and organized to date by us, in this study, the machine learning was applied to restore the compositions of various major and trace elements in the Hadean continental crust of the Earth. The results indicate that the Hadean continental crust of the Earth was mainly composed of felsic rocks, including the tonalite–trondhjemite–granodiorite (TTG) suite and potassic granite, rather than andesitic rocks. Characteristics of major and trace elements of “rocks” suggest that the significant compositional diversity had already emerged in the continental crust in the early days of the Earth, and its formation mechanism was more likely related to the crustal thickening in convergent settings rather than the typical deep subduction environment. This study demonstrates that the machine learning has great potential in the field of Earth science and provides new evidences for our understanding of the evolution of the early crust of the Earth.