Study On The Metallogenic Prediction Models Based On Remote Sensing Geology And Geochemical Information Case Study Of Lalingzaohuo Region In Qinghai Province

Mineral resources assessment is a complex, high-dimensional and non-linearprocess to model and evaluate mineral resources.The main purpose of this processcan be divided into two steps. The first step is to identify and extract the mineralanomalies associated with mineralization from multiple sources of geological data.The data adopted here are usually geological, geophysical, geochemical, and remotesensing observation data. Based on these mineral anomalies, the second step is topredict the metallogenic information. In step two, researchers investigate the mineralanomalies in virtue of metallogenic theories and mathematical statistics methods.With the guidance of scientific prediction theory, they analyze the geologicalconditions of mineralization, summarize the metallogenic laws and establish thecomprehensive models of metallogenic prediction. These models could be used todelineate and evaluate the prospecting target area, and finally provide scientific prooffor regional prospecting and mineral resources development.Establishing mineral resources assessment model based on the anomalies derivedfrom a single source of data is inevitably limited and one-sided. In order to betterdistinguish ore and non-ore area, to infer probability of the existence of the concealedore deposits, and to guide the prospecting progress, the metallogenic information fromdifferent sources of data should be combined and analyzed comprehensively.Remote sensing geological information and geochemical mineralizationanomalies are two important components of comprehensive metallogenic information.With the improvement of the spatial and spectral resolution of remote sensing data,remote sensing technology shows more and more potential to be applied in the detailed prospecting process. Geochemical anomalies usually refer to the surface ornear surface supergene anomalies. The prospecting methods based on these anomaliesare effective in searching for concealed orebody and blind ore, because the media withdispersed metallogenic elements are diverse, and the element migration distance couldbe very long. Therefore, combining remote sensing with geochemical prospectingtechnology is a strategic and tactical method for searching for large and superlarge oredeposits.The growing popularity of digital geological information and computertechnology makes the wide spread of mathematical methods in metallogenicpredictions.Non-linear mineral predictions extract geophysical, geochemical andremote sensing information in virtue of the mathematical statistics methods. It focuseson the investigation of information extraction process, and helps the researchers trulyunderstand the law of singularity when the elements enrich or defect in themineralization process. This kind of prediction is advanced and accurate.When adopting the non-linear method to predict the mineral resources, theresearchers are aimed at finding ways to use geophysical, geochemical and RSinformation with high efficiency under complex geological conditions and predict themetallogenic prospect area automatically.In this study, the region of Lalingzaohuo was chosen as the study area. Afterfully researching the basic geological data and the metallogenic conditions in thestudy area, the methods extracted remote sensing geological information andgeochemical anomalies was analyzed. With the guidance of scientific predictiontheory and the comprehensive metallogenic information, different non-linear modelswere chosen to predict the mineral resources and divide the metallogenic prospectarea. Moreover, the results and reliability of different models was evaluated finally.In this paper, the following conclusions were drawn through analysis:1. Landsat ETM+images and the ZY1-02C images with a higher spatialresolution were combined together in this study. The middle level lineament out of ETM+images and secondary lineament out of ZY102C images were interpreted.Based on the existing interpretation information and analyzing length and frequencynumber of the lineament, the spatial distribution of lineament was studied throughconstructing density distribution graph. With the ZY102C data and geologic map,the lithology of geological body which related to the mineralization in the study areawas supplemented and modified. After fully analyzing the metallogenic conditionsand spectral characteristics of ore already known, the mineralization information wasderived from Aster images via ratio, PCA and SAM. The results were integrated thenand the most significant alteration information for prospecting was selected.2. Based on the geochemical survey for stream sediments of the study region at ascale1：50000, traditional iteration method and trend surface analysis method wererespectively used to extract the minimum of anomalies for various chemical elements.In addition, Youden index was also introduced into this process. Youden index was acommon parameter in medical science and is usually used to evaluate the authenticityof screen tests. It was redefined in this study to calculate the optimal minimum ofanomalies. The geochemical anomalies were delineated by extracting optimal minimaand the results were highly coincident with the actual mines. Then, the hierarchicalclustering method was adopted to delineate the anomalies for geochemical elements’combination. The result reveals the regularities in spatial distribution of thegeochemical anomaly in the study area.3. Based on the GIS technology, metallogenic information was transformed frommultiple sources into grids, and then the gridded information were overlaid andintegrated to build comprehensive metallogenic information. The weights of evidencemodel was chosen to extract contrast between the positive and negative weight. Thiscontrast was combined with distribution of the actual mining points and used tomeasure the correlation between metallogenic information and the output, and toselect the optimization among the metallogenic information.4. In this study, weights of evidence model, Logistic regression model and RBM model were introduced detailedly and respectively, and used them to predict mineralresources. In order to find out the optimal results for Logistic regression model andRBM model, the prediction results came from different model parameters duringmodeling were tested and analyzed. The results showed that iteration and trainingtimes have great influence on prediction results of Logistic model and RBM model,but both of them were not the more the better. In order to improve the performance ofRBM models, two metallogenic prediction indices ASC and ASE were defined basedon the trained RBM model, and evaluated by the ROC curve. It showed that it wasbetter to define the reconstruction error of RBM model by ASE.5. ROC-curve analysis is one of the commonly used methods in evaluating themodel’s classification effect in machine learning field. In this research, ROC-curvewas applied to evaluate the effect of the metallogenic prediction model. The basicprinciples of ROC-curve and the method of calculating the area below ROC-curve(AUC) were analyzed in details. AUC could be used as a comprehensive index toevaluate metallogenic prediction model and the range of value is between0.5and1.0.When the value was0.5, prediction result of the model equals to that of a randommodel, and when the value was1.0, the model gets the best performance. Theperformance of three models was compared by AUC values calculated from threedifferent metallogenic prediction models. All of the AUC values were over0.7, whichindicated that the prediction results for three models were much feasible. According tothe AUC value, Logistic model predicted best, followed by the weights of evidencemodel and RBM model.The distribution graph of metallogenic prospect area was described through theresult of these three models. It showed that the range of metallogenic prospect areawas similar to the spatial trend and actual metallogenic area. The result conformed tothe conditions of geological and metallogenic characteristics.