Samuel Edem Kodzo Tetteh, MSc
S. E. K. Tetteh1, S. Brueckner1, J. Ayer1, J. Harris1, R. Sherlock1, L. Bonhomme2
1Harquail School of Earth Sciences, Laurentian University, Sudbury, Ontario, Canada,
2International Explorers & Prospectors Inc., Timmins, Ontario, Canada
The rate of mineral deposit discoveries has declined globally in recent years as most shallow targets have already been discovered. This holds for the Kamiskotia area, a volcanogenic massive sulphide (VMS) district in the Abitibi greenstone belt. Past-producing VMS deposits in this area are hosted by the Neo-Archean Blake River assemblage and share several similarities (e.g., comparable bimodal host lithologies, alteration signatures, deposit morphology, stratigraphic position, and ore assemblages) to suggest further potential. However, no significant deposit has been uncovered since the initial discoveries of the past-producing mines in the 1920s and traditional prospecting is inhibited by relatively low outcrop exposures. This study uses mineral prospectivity mapping (MPM) as a relatively new approach to identify potential exploration targets in the study area. To achieve this aim, random forest (RF) was used as a modelling technique to integrate predictor maps from lithologic, structural, geophysical, and geochemical data prepared in both continuous and binary surface map formats. Forty-four base metal occurrences and 44 non-VMS locations were used to train the RF model, whereas 22 showings were used as test data. Additionally, RF feature ranking was used to constrain the most important predictors of mineralization.
The probability map from continuous predictor maps (continuous MPM) and binary predictor maps (binary MPM) showed high overall classification accuracies (i.e., > 85 %), success rates of classification and prediction, and area under the curve (AUC) on efficiency curves. The success rates and AUCs obtained were higher for the binary MPM than the continuous MPM, suggesting that binary predictor maps outperform continuous maps. The binary MPM was, therefore, selected as the best performer. Ten areas with probabilities greater than 90 % were highlighted as the most prospective areas, out of which six were interpreted as new potential targets away from past-producing mines that may be prime for follow-up. RF ranks predictor maps from subvolcanic-synvolcanic intrusions and faults, mafic and felsic volcanic lithologies, Bouguer gravity and its derivatives, high-Zr rhyolites, evolved mafic rocks, Cu, Zn, and chloritization indices as the most important parameters to consider for follow-up studies in these areas. The results underscore the usefulness of RF MPM in integrating multiple geoscience datasets to map VMS prospectivity and exhibit the potential for new discoveries in the Kamiskotia area.