Journal of Economic Geology (Sep 2023)
GIS-Based Identification of Promising Porphyry Copper Mineralization Areas in Shahre Babak (Kerman Province, Iran) using Machine Learning Method
Abstract
Producing mineral potential model using GIS software has been increased over the past years. In this study, predictive map consisted of argillic alteration, philic alteration, iron oxide alteration, reduction to pole of aeromagnetic data, lineaments, cu geochemistry anomaly, and principal component analysis (component 3) were prepared from Shahre Babak area. For training model, 37 mineralized points were used. Point pattern analysis was used as well for making non-deposit points and for training model, percepteron artificial neural network with two layers was applied. The training model was used to prepare the final mineral potential model. Based on the mentioned model, the main promising areas were identified to be in the northwest and eastern part of the studied area. Moreover, two areas in the northern and southwestern parts of this area were identified for additional studies. For evaluating the model, ROC curve was used. ROC curve shows high precision of the produced model. For more evaluating, sensitivity, specificity, positive predict value, negative predict value, accuracy, and kappa were computed. The coefficients confirm the high accuracy of the mineral potential model. Introduction Mineral prospectivity mapping (MPM) is a multicriteria decision-making task that aims to outline and prioritize prospective areas for exploring undiscovered mineral deposits of the type sought (Carranza and Laborte, 2015; Yousefi and Carranza, 2015; Sun et al., 2019). In the early stages of exploration, if there are enough known indices in an area, data-driven modeling is proper for mineral potential prospectivity. In this method, at first, all the characteristics of the known indices, of the type of mineralization sought, are collected and the relationship of these characteristics with evidence and spatial patterns is quantified. Then, points with similar characteristics are searched in those areas. Shahre Babak as the studied area is a part of Urumieh-Dokhtar zone. Urumieh-Dokhtar zone is proper for porphyry copper deposits. In this study, at the first stage, conceptual model was defined for porphyry copper modelling. Then, based on the model, some predictive layers were made ready and the data were imported to the trained model of artificial neural network in MATLAB 2021. At the next stage, final model was presented. Material and methods For constructing mineral potential model, a conceptual model was defined. Based on this model, some predictive layers consisted of argillic alteration, iron oxide alteration, phillic alteration, reduction to pole of airborne magnetic map, cu geochemistry anomaly, principal component geochemistry anomaly, intrusive units, lineaments structures, and digital elevation models were made in ARCGIS in raster formats. The pixel size of the raster files is 100m*100m. After fuzzification of raster files, these features were extracted to ASCII formats. Geology data (Intrusive body, faults, and dykes) Shahre Babak geology map in 1:250000 scale was used for extracting geological information. The intrusive bodies, faults, and dykes were extracted from Shahre Babak geology map. After extracting geological information, based on the Euclidean distance, the distance maps were made in ARCGIS. Then these maps became fuzzy. Airborne magnetic data The airborne magnetic data were surveyed by Atomic Energy Organization of Iran (AEOI) during 1977 and 1978. The flight lines distance and the sensor altitude were about 500 and 120 m, respectively. The reduction to pole filter was applied on total magnetic intensity map. Geochemistry data Geochemistry data in 1:250000 scale was used for geochemical interpretations. The cu geochemistry anomaly was drawn from the data. Principal component analysis method was applied on geochemical data. Component 3 was extracted from the data. Aster data Band ratio method was used for extracting the alterations. Iron oxide alteration, philic alteration, and argillic alteration were drawn in ENVI software in raster format. The iron oxide, argillic, and philic alteration files were imported to ARCGIS software and transformed to shapefile format. The distance maps were drawn based on the Euclidean distance. Then these maps became fuzzy. Digital Elevation Model Digital Elevation Model (DEM) was extracted from Aster data. The data became fuzzy. Training dataset For training model, 37 deposit points were selected. Point pattern analysis was used for non-deposit points. Based on this method, 37 non-deposit points were extracted of the Shahre Babak (the studied area). Each of the labels was located in a unique pixel. The features of these points were extracted from the predictive maps. Then these points were imported to artificial neural network (perceptron neural network with two layers). 70% of data were used for training model and 30% were used for testing model. Then the trained model was applied on the ASCII format. The resulting model was drawn using ARCGIS. Artificial neural network ANN is a modelling approach that simulates human brain system inspired by biological neural networks (Celik and Basarir, 2017). ANN can be effectively applied for pattern recognition in a wide variety of geoscience investigations. In this network, the neurons of different layers are interconnected to exchange information in a unidirectional way starting from the input layer through hidden layers to the output layer (Rodriguez-Galiano et al., 2015; Celik and Basarir, 2017). The flow of information is performed by assigning weights to the connections of different neurons (Rodriguez-Galiano et al., 2015). The back-propagation algorithm is employed to ensure the learning capability of ANN. This algorithm computes the error between the outputted value and real target value, then feeds back it to ANN in order to adjust the weights and biases (Celik and Basarir, 2017). Results Mineral potential map of studied area was produced by artificial neural network. Based on resulting model, the first-class promising areas were detected in north western and eastern parts of the studied area. Moreover, two areas in north and south western parts of studied area were identified. For evaluating the model, ROC curve was used. This curve shows model accuracy with high precision. For further evaluation, sensitivity, specificity, positive predictive value, negative predictive value, accuracy, and kappa were calculated with 94.7%, 91.8%, 92.3%, 94.4%, 93.3%, and 89%, respectively. These coefficients also confirm the high accuracy of the mineral potential model.
Keywords
