Ian Szumila, Dustin Trail, Miki Nakajima, Veanessya Vazquez-Lopez, Rasmus Haugaard, Taus Jørgensen, Michael R. Ackerson, Steven B. Shirey, Nicolas B. Litza, Scott Hull, James Darling, Shweta Narkar
Abstract excerpt: "The source lithology of many Archean and Hadean zircons is unclear either because grains are separated from their source rock (i.e. detrital) or hosted in a pervasively metamorphosed conglomerate. Therefore, we developed a machine learning model using a support vector machine trained on the U/Yb, Th/U, Yb, and Gd/Yb contents of zircons from the Sudbury impact melt sheet and more typical igneous processes, so that it could distinguish between them. Here, we show that zircons from three >3.2Ga localities did not crystallize from a shock-induced impact melt sheet by comparing with zircons from the melt sheet layers of the 1.85 Ga Sudbury impact structure9, which contains Earth’s largest, best preserved melt sheet10. " SudburyZirconAnalysisNotebook_.ipynb will conduct the machine learning analysis, using a support vector classifier, and predict whether a suite of early earth zircons are from impact melt sheets or endogenous melting. It will also produce the geochemistry plots in the main text. Supp_OceanicContinentalZirconAnalysis_JupyterNotebookCode.ipynb will produce the analysis and figures related to the oceanic and continental zircons discussed in the supp. Reproducible Run will run all associated code via the bash script. Going to an single jupyter notebook and using 'Restart Kernel and Run All Cells' will allow to run an individual jupyter notebook. The code will produce the analysis and plots (before stitching into composite figures) for this manuscript.