![[Visit Client Website]](/img/gsa/banner.jpg)
Paper No. 8-23
Presentation Time: 9:00 AM-5:30 PM
A COMPARATIVE HYPERSPECTRAL STUDY OF HYDROCARBON SEEPAGES NEAR UVALDE, TEXAS: AN ANALYSIS OF CLASSIFICATION METHOD ACCURACY
The purpose of this study is to use high resolution, ground-based hyperspectral imagery to map and study the distribution and alteration patterns of migrated hydrocarbons in the Anacacho Limestone. Both Visible Near Infrared (VNIR) and Shortwave Infrared (SWIR) scans were obtained from the inactive Dabney asphaltic limestone pit mine near Uvalde, Texas. This study seeks to use comparative methods of hyperspectral pixel composition classification in order to maximize the advantages of hyperspectral data acquisition while minimizing the limitations. Techniques compared in this study include the Mixture-Tuned Matched Filtering (MTMF), Spectral Angle Mapper (SAM), and Neural Network (NN) methods of classification. Data were processed with band subsetting, spectral smoothing, and Minimum Noise Fraction (MNF) transformation, with endmembers derived from laboratory-obtained spectroradiometer data and training pixels identified through MNF analysis. Training regions were produced from ground truth data, and confusion matrices against the ground truth regions were generated to assess the accuracy of each method. Neural Network was shown to have the highest accuracy and best resilience against error due to shadows. Classification maps show moderately cohesive high-asphalt seepages in discontinuous layers throughout the limestone matrix. The outcrop is punctuated by thin, continuous bentonite ash layers reflecting volcanically active periods of deposition. Utilizing ground-based hyperspectral scans allows for fast and continuous mapping of the distribution of target deposits at high resolution, although somewhat lower precision than discrete physical sampling.