Combining visual interpretation and supervised classification technique with optical satellite data for classifying tropical forest cover

In this study, we applied a supervised classification technique for mapping the vegetation types in a tropical forest area. In applying the supervised classification technique, one of the problems is the selection of the training area. We used two methods of selection. In the first method the training area was selected directly from field observation. In the second method, we selected the training area from a paper printout of LANDSAT TM 542 using visual interpretation. The training areas from the two methods were then used for classifying the same image by the maximum likelihood classification technique. The maximum likelihood classification from the training area obtained by field observation gave a kappa accuracy of 90.7%. By comparison, the training area obtained by visual interpretation gave a kappa accuracy of 85.8%. The T test indicated that there was no significant difference in their kappa values, and through the overlay technique, the classification map obtained by both methods also showed similar consistency. Thus, the second method was considered to be an accurate and simple method applicable for tropical forests that are usually situated in remote areas with poor accessibility.