Insights on Source Lithology and Pressure‐Temperature Conditions of Basalt Generation Using Machine Learning

• Published in: Earth and space science (Hoboken, N.J.) Vol. 11; no. 7
• Main Authors: Cheng, Lilu, Yang, Zongfeng, Costa, Fidel
• Format: Journal Article
• Language: English
• Published: Hoboken John Wiley & Sons, Inc 01.07.2024; American Geophysical Union (AGU)
• Subjects: Basalt; Investigations; Lithology; Machine learning; Temperature
• ISSN: 2333-5084; 2333-5084
• DOI: 10.1029/2024EA003732

Identifying the origin and conditions of basalt generation is a crucial yet formidable task. To tackle this challenge, we introduce an innovative approach leveraging machine learning. Our methodology relies on a comprehensive database of approximately one thousand major element concentrations derived from glass samples generated through experiments encompassing a wide range of source lithologies, pressure (from 0.28 to 20 GPa) and temperature (850–2100°C). We first applied the XGBoost classification models to assess the compositional characteristics of melts from three principal mantle source categories: peridotitic, transitional, and mafic sources. We obtained an accuracy of approximately 96% on the test data set. Furthermore, we also employ an XGBoost regression model to predict the pressure and temperature conditions of generation of basalts from diverse lithologic sources. Our predictions of temperature and pressure exhibit remarkable precisions, of about 49°C and 0.37 GPa, respectively. To enhance accessibility of our model, we have implemented a user‐friendly web browser application, available at (https://huggingface.co/spaces/lilucheng/sourcedetection). The web application allows users to swiftly recover the source lithology as well as pressure and temperature conditions governing basalt generation for a broad array of samples within a matter of seconds. Plain Language Summary The depth and temperature of basalt generation are critical for understanding of the thermal structure of the earth and how plate tectonics works, topics that are still intensively investigated and not well agreed upon. Here we use a database of experiments of basaltic phase equilibria with Machine Learning models to infer the source and pressure and temperature of formation of basalts. We found that the results of our models are mainly in agreement with but more precise than previous works. We have built a user‐friendly web‐based application that allows users to quickly calculate the source, pressure and temperature of formation of basalts. Key Points We use machine learning to infer the source, pressure and temperature conditions for the generation of basaltic melts We obtained accuracies of about 96% for our predictions, which are within 50°C and 0.4 GPa of experimental values We produced a user‐friendly app to easily model the conditions of basalt generation