Comparison Between Supervised and Unsupervised Classifications for Mapping Land Use/Cover in Ajloun Area

This research was undertaken in Ajloun area, located north west of Jordan, to assess the accuracy of Land Use/Cover maps derived from supervised and unsupervised classification techniques of Landsat TM imagery. Digital image processing techniques included geometric correction, resampling, registration, subset of study area, application of unsupervised classification with ISODATA algorithm and identification and merging of the output classified image into four major classes. Output from the previous stage was used to identify the number of training areas and to collect ground data to implement supervised classification for the same image. Accuracy assessment was made for both classifications with confusion matrices based on ground visits. Results of digital classification showed that unsupervised classification was more accurate than supervised classification with high classification accuracy for forested area, with variations in accuracy among classes and between the two classifications. These variations were mainly attributed to the complex and fragmented patterns of cultivation and the scattered pattern of urban area. Therefore, the study recommends the use of higher spatial and spectral resolution data to increase the obtained level of accuracy