Mapping Mineral Prospectivity For Molybdenum Polymetallic Mineralization By Machine Learning Methods In Jining,Inner Mongolia,China

The Jining in Inner Mongolia,China,is one of the important Mo polymetallic mineral belts on the north margin of North China Craton.However,large amounts of Cenozoic basalts and sediments cover on the surface in the study area,which poses great challenges to mineral prospecting,especially in terms of few discovered mineral deposits and lack of ore-controling information under the covers.At first,a conceptual prediction model of Mo polymetallic mineralization was constructed based upon the knowledge of regional geology,structure and metallogeny.Furthermore,several stateof-the-art mathematical methods,such as machine learning and fractal models,were applied to extract multi-source ore-controling information and make mineral prospecting research.The main conclusions obtained in this study include:(1)The research of faults interpretationThe integration of potential field separation and boundary recognition methods were used to interpret the faults at depth and/or under the covers based upon gravity and aeromagnetic data with the scale of 1:200,000.The obtained results clearly identified the concealed faults at depth.Moreover,with the assistant of t statistic analysis,the best influence area of fault structures on the deposit production was quantitatively evaluated to be within the distance of 4 km.(2)The research of mapping intermediate and felsic intrusionsWith the stream sediment geochemical major and trace elements,gravity and aeromagnetic data,three methods,including principal component analysis(PCA),supervised support vector machine(SVM)and random forest(RF),were used to delineate intermediate and felsic intrusions,respectively.By comparison of the obtained results through t statistics and ROC curve,it was found that the inferred intrusional distribution from RF method has the best spatial correspondence with the exposed granitic intrusions and also can help locate the possible underlying granites;(3)The recoganition and extraction of multivariate geochemical anomaliesThe method of factor analysis was applied to integrate the multivariate geochemical anomalies based upon the mineralization indicator elements obtained from expert knowledge,which include W,As,Bi,Hg,Sb,Cu,Mo,Ag,Pb,Zn,and Au.The multifractal spectrum-area(S-A)method was further applied to separate the local anomalies from the background,thus allevating the screening of the basaltic cover and strengthing the weak anomalies in the covered areas;(4)The research of multi-source prospecting information mining and integrating based on machine learning methodsDue to few known deposits in the Jining area,the accuracies of the prediction obtained by machine learning methods are high,but the predicted results are meaningless in consinderation of economy and field exploration.In order to increase the trainging data to improve the prediction,this article explored to expand the number of negative samples(i.e.,non-deposit areas),and then balanced the dataset by Synthetic minority over-sampling technique(i.e.,incurease the number of deposit areas),which effectively improves the predicted accuracy and success rate at once.The semisupervised learning of positive unlabeled method(PU learning)was introduced to mineral prospecting as this method uses only positive and unlabeled data to implement tranining,avoiding the uncertainty caused by building negative samples.The predicted results obtained show better performance when comparied to the traditional supervised methods.By the comparison of the aboving 5 multi-source prospecting results,the predicted results from the over-sampling random forest method outperforms the other and produced six A-class,two B-class,and three C-class metallogenic prospective areas,which are helpful for guiding the future exploration of molybdenum polymetallic mineralization in the Jining area.(5)The integrated geophysical investigation of Quanzigou prospecting areaBy integrating geophysical data from gravity,magnetic,electrical,and seismic survey and geochemical data,main geological structures under the covers in Quanzigou area were identified,and the potential of mineral prospecting was further evaluated.The obtained post-stack migrated seismic profile clearly imaged the geometry of the concealed graben located at the north end of the Datong Basin,and the South-East dipping Kouquan normal fault controlling the depression of the graben and offseting Cenozoic sediments by several hectometers up to ?500 m.Joint modeling of seismic,gravity and magnetic data further provided evidence of the three buried granitic belts.Among the three concealed granits,one granitic belt in the middle of Quanzigou area was found to have both significant chargeability and Mo geochemical anomalies,indicating a high potential of hosting porphyry-type Mo ore deposits.As expected,Mo and W mineralized cores were discovered in this target area but limited to only one deep drilling hole.