Chen, X, Warner, TA, Campagna, DJ (2007). "Integrating visible, near-infrared and short wave infrared hyperspectral and multispectral thermal imagery for geologic mapping: simulated data". INTERNATIONAL JOURNAL OF REMOTE SENSING, 28(11), 2415-2430.
Hyperspectral and thermal infrared (TIR) multispectral remote sensing have great potential for surface geological mapping. This paper investigates the potential impact of combining these data on the comparative accuracy of different classification methods. A series of simulated datasets based on the characteristics of Airborne Visible/ InfraRed Imaging Spectrometer (AVIRIS) and MODIS/ ASTER Airborne Simulator (MASTER) sensors was created from surface reflectance and emissivity data derived from library spectra of 16 common minerals and rocks occurring in Cuprite, Nevada. System noise, illumination effects, the presence of vegetation, and spectral mixing were added to create the simulated data. Five commonly used classification algorithms, minimum distance, maximum likelihood classification, binary encoding, spectral angle mapper (SAM) and spectral feature fitting (SFF), were applied to all datasets. All the classification methods, excluding binary encoding, achieved nominal to significant improvement in overall accuracy when applied to the combined datasets in comparison to using only the AVIRIS dataset. Furthermore, certain classification methods of the combined datasets show a marked increase in individual rock or mineral class accuracies. Limestone, silicified and muscovite, for instance, show an improvement of almost 30% or greater in either producer's or user's accuracy using the combined datasets with SAM. SFF provides a great improvement in accuracy for limestone, quartz and muscovite. In terms of overall comparative accuracy for the individual and the combined datasets, maximum likelihood classification shows the best performance. For the simulated AVIRIS data, SFF was generally superior to SAM, although the accuracy of SAM applied to the combined datasets was slightly better than that of SFF. SAM applied to the combined datasets increases classification accuracy for some minerals and rocks which do not exhibit distinct absorption feature in the TIR region, while for SFF, only the accuracy of minerals and rocks with characteristic absorption features in the TIR region is improved.