Supplementary files

1.Tables:

Table-S1_Parental traning dataset: We compiled modern arc and non-arc basalts from the GEOROC and PetDB databases.

Table-S2_Training dataset for XGBoost modeling: The major and trace elements were extracted from the "Parental training dataset", which can be applied to train the XGBoost machine learning model.

Table-S3_Validation-Northeast China, Fangcheng, Vanuatu, EPR: Validation dataset for the XGBoost machine learning model.

Table-S4_Alteration test-Hainan Island: Alteration test for the XGBoost machine learning model.

Table-S5_Parental application dataset: The application dataset comprises global basalts (~0.85 Ga to the present).

Table-S6_Application dataset for prediction: We extracted major and trace elements from the "Parental application dataset", which can be predicted by the trained XGBoost model.

Table-S7_Calculated melting PT of primitive arc basalts: We calculated the meting pressure and temperature conditions of global primitive arc basalts.

Table-S8_Petrological modeling for arc mantle melts: We applied the partial melting model to eatimate the sub-arc mantle fO2 (see Methods in the main text).

Table-S9_Calculated mean V-Sc and PT of primitive arc basalts: The estimated mean V/Sc ratios and melting pressure and temperature conditions of primtive arc basalts since ~0.8 Ga.

Reference key from EarthChem dataset: References information for EarthChem dataset.

Reference key from KS2012 dataset: References information for KS2012 dataset (Keller and Schoene, 2012).

2.CodePlease put the Excel file and the python code in the same path. All the python codes were implemented on the Jupyter Notebook platform.

If you want to run the "Python_arc_classification ML model_ctliu2024", you can use the Excel Table-S2 and Table-S6. See details in the python file.