Improving Mineral Prospectivity Maps Applying Support Vector Machine on restored data: The Kupferschiefer case

Pablo Mejía-Herrera and Jean-Jacques Royer and Jürgen Hartsch. ( 2013 )
in: IAMG 2013 Annual Conference

Abstract

This work explain the procedure to obtain better assess potential maps of high copper concentration values from geo-datasets in restored 3D models. The potential copper zones within the Kupferschiefer (Zechstein base) are predicted for the North-Sudetic Trough in Germany, with geological and geochemical datasets from unfolded/unfaulted 3D models using SKUA® Kine3D®. The blue-prints in the training dataset were selected on known high copper values located in the Fore-Sudetic region in Poland, while the barren dataset were selected randomly in other parts of Germany and Poland. Then, the Support Vector Machine (SVM), with the R package e1071, is applied for classifying mineralized and barren locations. We compared the results using geo-datasets from restored and non-restored models in the training protocol, finding the first one more accurate to predict mineralized occurrences. We used the geo-datasets from the restored models of the Fore-Sudetic region for the final assessment in the North Sudetic Trough and compared the results using the Weight of Evidence (WofE) procedure, obtaining improved predictions than WofE. The final SVM model and training geo-datasets can even predict other high copper concentrations at the base of Zechstein in Germany. The above research received funding from the European Union's Seventh Framework Program under grant agreement 228559 (ProMine project).

Download / Links

BibTeX Reference

@inproceedings{mejiaherrera:hal-04062380,
 abstract = {This work explain the procedure to obtain better assess potential maps of high copper concentration values from geo-datasets in restored 3D models. The potential copper zones within the Kupferschiefer (Zechstein base) are predicted for the North-Sudetic Trough in Germany, with geological and geochemical datasets from unfolded/unfaulted 3D models using SKUA® Kine3D®. The blue-prints in the training dataset were selected on known high copper values located in the Fore-Sudetic region in Poland, while the barren dataset were selected randomly in other parts of Germany and Poland. Then, the Support Vector Machine (SVM), with the R package e1071, is applied for classifying mineralized and barren locations. We compared the results using geo-datasets from restored and non-restored models in the training protocol, finding the first one more accurate to predict mineralized occurrences. We used the geo-datasets from the restored models of the Fore-Sudetic region for the final assessment in the North Sudetic Trough and compared the results using the Weight of Evidence (WofE) procedure, obtaining improved predictions than WofE. The final SVM model and training geo-datasets can even predict other high copper concentrations at the base of Zechstein in Germany. The above research received funding from the European Union's Seventh Framework Program under grant agreement 228559 (ProMine project).},
 address = {Madrid, Spain},
 author = {Mej{\'i}a-Herrera, Pablo and Royer, Jean-Jacques and Hartsch, J{\"u}rgen},
 booktitle = {{IAMG 2013 Annual Conference}},
 hal_id = {hal-04062380},
 hal_version = {v1},
 title = {{Improving Mineral Prospectivity Maps Applying Support Vector Machine on restored data: The Kupferschiefer case}},
 url = {https://hal.univ-lorraine.fr/hal-04062380},
 year = {2013}
}