Calculation of lithosphere composition based on machine learning of multiple physical parameters

The lithosphere holds critical clues to Earth’s formation and long-term geodynamic evolution. Constraining its composition through rock physical properties not only deepens our understanding of tectonic processes but also enhances our ability to interpret geophysical observations and evaluate resource potential. By training machine learning models using a compiled dataset of rock physical properties, including P-wave velocity (Vp), S-wave velocity (Vs), and density (ρ) obtained from experimental measurements and calculations at 600 MPa and 25 °C, we predict the chemical composition of the lithosphere. The machine learning model for predicting SiO2 content achieved a coefficient of determination (R2) value of 0.90 on the test dataset, indicating good predictive performance. Following temperature and pressure corrections of the geophysical data, this model was then applied to the South China Block lithosphere, revealing depth-dependent spatial variations across four tectonic units. The Mg# prediction model, which achieved a test R2 of 0.86, was also applied to the South China Block, revealing spatial variations in the Mg# of the mantle lithosphere. Our findings challenge the traditional view that cratonic lithosphere consists of high Mg# (∼92) refractory peridotite, showing instead that the mantle lithosphere beneath the Sichuan Basin and surrounding areas has lower Mg# (91−87), consistent with peridotite xenolith data. Likewise, the domination of fertile peridotites in the mantle lithosphere of the Cathaysia Block reflect rejuvenation, as suggested by geochemical studies. Unlike prior geochemical studies limited to coastal regions, our model reveals a fertile mantle lithosphere across the Jiangnan Orogen and Cathaysia Block, suggesting widespread lithosphere thinning and chemical erosion. Extending predictions to 10−250 km depths, this study demonstrates the model’s applicability for inferring 3-D lithosphere compositions globally, offering new insights into mantle lithosphere properties.