Search

Abstract The subaerial exposure of the modern continental crust through time remains intensely debated, with estimates of the first exposure ranging from the late Archean to the Neoproterozoic. To constrain when and how much of the continental crust was exposed subaerially during Earth's history, we trained a supervised machine learning model on the compositions of modern subaerial and submerged basalts. Then, we applied this well‐trained model to a refined worldwide data set of basaltic compositions and calculated the mean proportions of basalts erupted subaerially since 3.8 billion years ago (Ga). Our results suggest that ∼20% of the basalts were exposed subaerially in the early Archean, which may have driven the synthesis of biopolymers crucial to the origins and evolution of life. The proportion of subaerial basalts increased markedly during two stages between the late Archean and the Paleoproterozoic before reaching the present‐day level no earlier than ∼1.8 Ga.

Abstract Subduction processes play a pivotal role in facilitating material exchange between the crust and mantle, contributing to the growth of continents. However, the onset and evolution of subduction remain hotly debated. Here, we developed a high‐dimensional machine learning (ML) model to use multiple compositional data (e.g., Nb/La, Nb, Ti, Nb/U, Pb/Nd, and Nb/Th) to distinguish arc‐type from non‐arc basalts worldwide, then applied this well‐trained ML model to identify and delineate the secular occurrence of Archean arc‐type basalts. Our findings indicate that basaltic arc magmatism can be traced back to only a few early Archean cratons. A noticeable increase in the prevalence of arc‐type basalts during the Mesoarchean and Neoarchean suggests the synchronous formation of Archean subduction across different cratons. This observation hints at transitioning from the predominantly stagnant geodynamic regime to the early, more mobile tectonic regime.

Abstract Comprehending the temperature distribution within Earth's lithospheric mantle is of paramount importance for understanding the dynamics of Earth's interior. Traditional mineral‐based thermobarometers effectively constrain temperature and pressure for particular compositions, but their application is limited at the global scale. Here, we trained machine‐learning (ML) algorithms on 985 published high‐temperature and high‐pressure experiments for use as thermometers and barometers to overcome the limitations of classic methods. We compared our ML models to classic thermobarometers to assess the accuracies of predicted pressures and temperatures. The comparison shows that the ML models outperform classic methods and better fit various mineral pairs. Global application of the ML models unveils mantle conditions beneath cratons. Furthermore, depths to the lithosphere‐asthenosphere boundary (LAB) calibrated based on the ML thermobarometry results are generally deeper by ∼40 km than those derived geophysically, implying the existence of melt‐bearing or hydrated mineral zones at the LAB.