Characteristics of Mesozoic magmatism in the North China Craton based on machine learning of petrological geochemical data: Implications for destruction processes of the craton

The North China Craton, one of the world's oldest cratons, has experienced significant lithospheric thinning and destruction since the Mesozoic. However, due to regional variations in magmatic rocks and limited rock geochemical data, there remain disagreements regarding the timing, mechanisms, and processes of destruction basing on studies of Mesozoic magmatic rocks. Compared to traditional rock geochemical research methods, machine learning offers the capability to handle large, complex, multidimensional datasets, yielding multidimensional indicators to resolving geological issues. This study utilizes geochemical data from Mesozoic mantle-derived magmatic rocks in the North China Craton, as well as global island arc and oceanic island magmatic rocks, to establish discriminant models for source provenance and source characteristics. By applying the Latent Dirichlet allocation algorithm to extract features from Mesozoic mantle-derived magmatic rocks in the North China Craton, two endmembers were identified in the rock geochemical data. Subsequently, the logistic regression algorithm was employed to discriminate between these two endmembers, revealing that they indicate mantle-derived magmatic rocks originating from the lithospheric and asthenospheric mantle, respectively. Machine learning analysis of the geochemical data from Mesozoic mantle-derived magmatic rocks in the North China Craton shows that the destruction of the craton persisted from the Triassic to the Cretaceous, involving two destruction mechanisms: delamination and thermo-chemical erosion. The spatiotemporal variations of these two mechanisms were influenced by the closure of the Paleo-Asian Ocean, subduction of the Yangtze Plate, and the alternating effects of subduction and retreat of the Paleo-Pacific Plate, leading to spatiotemporal heterogeneity in the destruction of the cratonic lithosphere. The findings are in substantial agreement with existing understandings of magmatic rock formation in various geological settings. Nevertheless, they offer a superior level of detail and precision. Machine learning applied to geochemical data of Mesozoic mantle-derived magmatic rocks is of great significance for deepening the understanding of the destruction process of the North China Craton.