Integration of Landsat TM, Gamma-Ray, Magnetic, and Field Data to Discriminate Lithological Units in Vegetated

Image classification of geological units in vegetated granite-gneiss terrain from multispectral data is hampered by vegetation cover and limited spectral contrast of its lithological variations. In this paper an alternative methodology is presented, employing airborne gamma-ray spectrometry and magnetic data. The methodology includes the selection of combinations of data channels and their transformations to enhance the discriminative power of the lithology classification. An analysis of the training set compiled from 2,795 field stations showed that the potassium, thorium, and uranium gamma-ray spectrometry and total and residual magnetic field channels provided an overall classification success rate of 65% for a total of 10 lithological units. Parametric classifications based on this training set yielded 67% coincidence with the geological map, whereas a neural network classifier provided only 23% coincidence. Exploiting the spatial autocorrelation of the geophysical signatures by adding averaged filtered channels slightly improved the coincidence percentage to 70% and enhanced the continuity of linear enclaves within larger lithological units. The contribution of magnetic data to the classification depends on the extent to which the anomaly spectrum and its processed derivatives reflect surface geology. We found that both the short and long wavelengths in the spectrum contributed to the classification performance. This is explained by the geological structure of the area, where both broad and narrow units extend downward subvertically. A comparison of our results to regional geological map patterns identified targets for map refinement and exploration.