1. ASTER VNIR‐SWIR Based Lithological Mapping of Granitoids in the Western Junggar Orogen (NW Xinjiang): Improved Inputs to Random Forest Method.
- Author
-
Zhou, Yarong, Zheng, Shuo, An, Yanfei, and Lai, Chunkit
- Subjects
RANDOM forest algorithms ,NONFERROUS metals ,PRINCIPAL components analysis ,SUPPORT vector machines ,COHEN'S kappa coefficient (Statistics) ,ORE deposits - Abstract
Although advanced spaceborne thermal emission and reflection radiometer multispectral analysis for lithological mapping has been widely applied, traditional methods such as band ratios (BR) and principal component analysis (PCA) are still hampered by cumbersome data processing and poor classification performance. In this study, we utilize improved data inputs for random forest (RF) to extract lithological information of granitoids, which are the predominant rock type for intrusion‐related polymetallic ore deposits in the western Junggar Orogen (NW Xinjiang). Based on spectral absorption features of minerals (e.g., orthoclase, K‐feldspar, hornblende, biotite, plagioclase, and oligoclase), image statistical information and textural features, we tested different combinations of bands, BR, PCA, and texture using RF method, and found that the combination of B13678 + T1 (Mean texture) achieved the highest weighted‐F1 score for granitoids, with an accuracy of 87.32%. Compared to the support vector machine, RF effectively distinguishes lithological differences between different types of granitoid and wallrocks, especially the granite, granodiorite, and alkali granite in the Akebasito intrusion, as well as the alkali granite, plagiogranite and biotite granite in the Karamay intrusions. Moreover, the large number of rare metal deposits (including Cu, Au, Mo, etc.) distributed near the granitoid intrusions in the western Junggar, our result facilitates the analysis of regional tectonic evolution and mineralization controlling. Plain Language Summary: Random forest (RF) is a machine learning method widely used in various fields due to its high classification accuracy and ability to handle high‐dimensional data without overfitting. Advanced spaceborne thermal emission and reflection radiometer data provide important information of related minerals for lithological identification because of its characteristic values for different bands and textures. In this study, ArcGIS was used to extract information from different bands and textures as input data for RF model. Then, the RF model was trained using Python, and the lithological mapping were generated through test data set, and the weighted‐F1 score, and kappa coefficient were compared. The combination of B13678 + T1 yielded the optimal result. Key Points: Based on the spectral data, image statistics and texture, select the most suitable bands combination for lithological mappingLithological identification of granitoids subclasses is very important for the exploration of metal deposits in the western Junggar Orogen [ABSTRACT FROM AUTHOR]
- Published
- 2023
- Full Text
- View/download PDF