Evaluating the VNIR-SWIR datasets of WorldView-3 for lithological mapping of a metamorphic-igneous terrain using support vector machine algorithm; a case study of Central Iran

Abstract

• Evaluating the WV-3 data in lithological mapping of an area through confusion matrix. • Successfully discrimination of lithological units and boundaries with high accuracy. • Verifying the produced map by field observations, petrography, and spectroscopy. • Defining effect of each band of WV-3 in classification accuracy of individual class. • Finding SWIR-7 as the most significant band in raising the classification accuracy. WorldView-3 (WV-3) is the first high spatial resolution and super-spectral commercial satellite which has the capability in improvement of geological mapping. The VNIR-SWIR data of this satellite are evaluated in this study for mapping the lithological units of a metamorphic-igneous terrain in Chadormalu area, Central Iran, by the use of support vector machine (SVM) classification method. Applying principal component analysis (PCA) as an image transformation technique on the WV-3 data, and interpretation of its results, supported by field observations, petrography, and spectroscopy, lead to identifying training areas which are input to the SVM algorithm. This method not only produces a detailed lithological map of all exposed rock units, but also well discriminates diorite-gabbro diorite from granite, and gneissic granite from green schist and garnet mica schist, which are not revealed in 1:100,000 geological map of the area, published by Geological Survey of Iran (GSI), with overall accuracy of 88.36% and the Kappa coefficient of 0.86. Furthermore, the efficacy of each applied band of WV-3 is assessed in promoting classification accuracy using SVM method. Results show that band 7 of SWIR region has increased the overall accuracy 4.81% relative to SWIR dataset, and improves the classification accuracy of green schist and diorite. Moreover, bands 5, 6, and 8 of SWIR dataset are more efficient in improvement of the overall accuracies than other bands. This study concludes that WV-3 data provides a good facility to generate large scale geological maps owing to its high spatial and radiometric resolution and appropriate SWIR bands.