Supervised classifications of Landsat TM band ratio images and Landsat TM band ratio image with radar for geological interpretations of central Madagascar Jennifer Inzana a , Tim Kusky b, * , Gary Higgs c , Robert Tucker d a Center for Remote Sensing, Boston University, Boston, MA 02215, USA b Department of Earth and Atmospheric Sciences, St. Louis University, 3507 Laclede Ave., St. Louis, MO 63103, USA c College of Public Service, St. Louis University, St. Louis, MO 63103, USA d Department of Earth and Planetary Science, Washington University, St. Louis, MO 63130, USA Received 27 February 2003; accepted 8 July 2003 Abstract Landsat TM and radar JERS-1 SAR (L-Band) imagery of the Itremo area, central Madagascar, were processed to emphasize structural geology features including folded quartzite ridges and plutons. TM band ratios 5/7, 5/1, 5/4*3/4 were assigned to RGB. Band 5/7 highlights pelitic schist, band 5/1 emphasizes mafic igneous rocks, and 5/4*3/4 distinguishes mafic from non-mafic rocks. In a second technique, band 5/7 was replaced with registered L-band radar imagery because radar is useful for differentiating between granite, granodiorite, diorite and serpentinite. The last technique evaluated in this study used the spectral information from the radar image as well as the 5/7, 5/1, 5/4*3/4 band ratio bands. Supervised classification training sites were selected using nine classes (clouds, quartzite, schist, gneiss, gabbro and basalt, granite, vegetation, water, and cloud shadows). The band ratio classification results are fairly accurate (a confusion matrix shows an accuracy of 89.346) and correspond well with geologic maps of the area showing complexly refolded nappes of quartzite, carbonate, schist, gneiss and gabbro, intruded by late granites. The radar, 5/1, 5/4*3/4 classification (accuracy of 89.04) shows significant differences from the band ratio classification, with fewer schist pixels displayed in the radar, 5/1, 5/4*3/4 classification, but with greater resolution of structural features including faults, fold nappes, and foliations. More pixels are displayed as mafic gneiss, and fewer quartzites appear in the radar classification. Some areas classified as quartzite in the first classification (and on the geologic maps) were classified as clouds in the radar/band ratio classification. This indicates that the 5/7 band contains significant spectral information that the radar band does not contain, which aided in mapping quartzite. This comparison illustrates that combined use of TM band ratioing merged with radar imagery can emphasize both spectral and textural features that aid geologic mapping using supervised classifications. A third technique was examined where a supervised classification was performed on an image containing the 5/7, 5/1, 5/4*3/4, and radar bands. The confusion matrix for this classification produced an accuracy of 91.23 which was better than either the 5/7, 5/1, 5/4*3/4 or the radar, 5/1, 5/4*3/4. It is preferable to keep all band ratio bands and the radar band to produce the most complete supervised classification image for geological feature discrimination. Ó 2003 Elsevier Ltd. All rights reserved. 1. Introduction Madagascar is the world’s fourth largest island, consisting of 627,000 km 2 of area, little of which is un- derstood in detail (Ashwal and Tucker, 1999; Windley et al., 1994; de Wit, 2003). Unlocking the secrets to the geology of this region could reveal information about the formation, break-up, and dispersal of several su- percontinents including Rodinia, Gondwana and Pan- gea. French colonial geologists pioneered by Moine (1968) mapped and described the principal geologic elements of Madagascar. Numerous investigations have been undertaken since then, including several by the authors (Tucker et al., 2001, in press). This paper aims to enhance and differentiate geologic units and structures and advance the understanding and definition of specific remote sensing data combinations useful for structure discrimination in Madagascar and * Corresponding author. Fax: +1-314-977-3350. E-mail address: kusky@eas.slu.edu (T. Kusky). 0899-5362/$ - see front matter Ó 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0899-5362(03)00071-X Journal of African Earth Sciences 37 (2003) 59–72 www.elsevier.com/locate/jafrearsci