Rock Protolith Prediction App

This app tries to predict a rocks protolith from major element geochemistry, using an algorithm trained on over half a million labeled global geochemical data. When you input your major element data, the model will try to predict if your sample is either igneous or sedimentary.

The rock will then be classified into a lithology based on the protolith prediction, using either the TAS igneous classification (Middlemost, 1994) or the SandClass sedimentary classification after Herron (1988)

How to use

You can either enter a single sample via an online form or upload a csv file containing multiple samples. Data should be in weight% oxides and must contain the below elements, even as 0% if they are missing.

Example minimum data required in input csv, including sample identifier in column 1 and header row. You must include either one of FeO or Fe₂O₃, or both. All Fe will be converted to FeO

SampleID	SiO₂	TiO₂	Al₂O₃	FeO	Fe₂O₃	MgO	CaO	Na₂O	K₂O	P₂O₅
873827	73.43	0.41	14.48	0.55	0.00	0.44	1.72	3.24	5.58	0.12

Single sample prediction

Batch prediction

Caveats

This predictor is a trained balanced random forest model based on major element geochemistry. Because of the chemical similarity between some rock types (such as felsic igneous rocks and arkosic sediments) there will always be the potential for misclassifications. The model performs better on some compositions than others. The classified lithology is also based on major element geochemistry and not mineralogy and is therefore subject to all the limitations associated with chemical classification of rocks, particularly sediments.

To asses the quality of the model please visit the GitHub repo and view the model_assessment notebook.

Contact, references and source code

Contact me @RADutchie on twitter or GitHub for comments or issues, and check out my web site GeoDataAnalytics.net

https://github.com/RADutchie/Rock_protolith_predictor for model and source code

This predictor is a reformulation of the original work published by Hasterok et al 2019. Chemical identification of metamorphic protoliths using machine learning methods. Computers and Geosciences. 132, 56-68