Technical Presentation
Title: Regional Prospectivity Analysis Applying Fuzzy Logic and Machine Learning - Extending Into 3D
Date: 8 April 2025
Time: 15:00 - 15:30
Abstract
SRK Consulting (SRK) in partnership with DeepIQ and BHP have undertaken a recent regional scale (~25,000km2) prospectivity study integrating both data driven (machine learning) and knowledge driven (fuzzy logic) workflows targeting large porphyry copper deposits in northern Chile. The project combined regional geological and structural interpretations, development of the porphyry mineral systems targeting framework and implementation of advanced data processing and machine learning to develop targets for follow-up field investigations. This study expanded the typical 2D prospectivity analysis by applying the analysis in both 2D and 3D. Following this initial pilot study, the application of this integrated data driven and geological knowledge approach to prospectivity analysis was subsequently applied for a greenfield porphyry copper deposit study at a similar scale and an intrusion-related nickel deposits at a much larger scale. This abstract focuses on the 3D aspects of the northern Chile study with readers directed to Woodfull et al., (2023) for a more comprehensive overview of the study.
The initial study stages included data cleaning, integration and interpretations of a wide range of datasets, with reprocessing and development of data enhancements to feed into the 2D and 3D targeting stages. A mineral systems approach was used to identify key feature elements specific to porphyry deposits for ingestion into the prospectivity workflows. Sparse/non-uniform data (e.g., geochemistry, drilling) were ultimately not used as inputs to eliminate potential spatial bias, instead using this data as confirmatory layers during target review and validation. Over fifty 3D elements were created for the 3D prospectivity analysis, replicating input features created for the 2D workflow. 3D modelling integrated several datasets including solid geology, cross sections, regional drill hole data and 3D geophysical inversion datasets (electromagnetics, magnetics and gravity). Features created included structural architecture (e.g., faults, crustal lineaments), lithology and intrusions (sub-split by age and intrusive type), alteration proxies, depth of cover, paleodepth and training sites (known porphyry deposits). A phase of 3D gravity and magnetic forward modelling was conducted to highlight areas of geophysical misfit with geometric updates conducted to resolve misfits. All raw data, 3D elements and mineral system proxies were attributed into regularized (500m) block model and then ingested into DeepIQ’s cloud-based Data Studio platform where machine learning and fuzzy logic analysis was conducted. Structured learning using known deposits as the training datasets was conducted and machine learning completed using random forest and deep learning algorithms to make predictions across the model space.
Outcomes from both the fuzzy logic and machine learning analysis in both 2D and 3D showed good correlation with known mineral deposits providing good confidence in results. The machine learning method enabled rapid interrogation of large amounts and provided more objective outcomes, minimising human bias of more traditional knowledge driven techniques. By integrating the parallel knowledge driven approach allowed mineral system elements to be tested and provided an additional layer of validation to the machine learning results. Interestingly the machine learning process was ultimately found to only rely on only a few key exploration layers driving prospectivity results. Both the 2D and 3D outcomes were also found to show good correlation, however, clear variances were apparent reflecting geometric and spatial differences when integrating the depth dimension. The use of 3D modelling enabled better understanding of the region from a geometric perspective particularly geometries of intrusions as well as paleodepth for preservation of the porphyry system. The process highlighted the importance of several key mineral system elements and 3D interactions, enabling the interplay of targeting elements to be evaluated in real space and highlighting favorable target areas. The use of the machine learning and fuzzy logic workflow in 3D ultimately proved to be a successful method, complementing the 2D workflow in identifying target areas for follow-up field investigations with several highly favorable target areas identified.
