•  
  •  
 

Turkish Journal of Earth Sciences

DOI

10.3906/yer-1401-9

Abstract

This study applied coded spectral ratio imaging and a self-organizing map (SOM) neural network model for subpixel classification of Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data to map alteration minerals of the Masahim volcano, SE Iran. Spectra of samples were measured using a FieldSpec spectroradiometer, and were interpreted for finding key alteration minerals. Key minerals including sericite and kaolinite were selected. Thirty-six nonreciprocal spectral ratios of 9 ASTER reflective bands were created. Each spectral parameter (of 9 spectral bands and 36 spectral ratios) for the field and library mineral spectra was divided into deciles. Each decile of spectral bands or ratios was labeled from 9 for the highest down to 0 for the lowest decile. Spectral ratio code signature patterns of field samples were then compared to the US Geological Survey (USGS) spectral library to find 3 spectral ratio images for each key mineral. Based on these 3 ratio images, color ratio images were created and training areas were selected by a geographic information systems (GIS) overlay approach and were used to train the SOM model. This model was applied to the 5, 15, and 36 ratio image data sets as chemical composition images, and the results were compared using confusion matrices and coefficient of determination (R^2). The results of SOM applied to 15 and 36 ratio image data sets showed good identification of key minerals. We achieved overall accuracies of 87% for SOM applied to the 36 ratio images, 86% for 15, and 85% for 5 ratio image data sets. The R2 produced from 15 ratio image data sets for sericite and kaolinite were 0.35 and 0.52, whereas they were 0.64 and 0.68, for 36 ratio image data sets. It was concluded that using SOM applied to ratio images could, therefore, be useful in alteration mapping of a volcano.

First Page

627

Last Page

644

Share

COinS