A comparison of change detection techniques for monitoring land-cover change in an arid urban environment.

Abdullah Almutairi, Department of Geology and Geography, West Virginia University, aalmutairi@geo.wvu.edu

This study compares five change detection techniques to study land cover change associated with the growth of Las Vegas, Nevada, one of the fastest growing areas in the U.S. The data used were two MSS images covering a twenty year period (1972 and 1992), and one TM image from 1993. The change detection methods used were red band image differencing, principal component analysis, post-classification comparison, change vector analysis, and multi-temporal compositing and classification. Each change detection method was applied to a pair of Landsat MultiSpectral Scanner (MSS) images and a pair of MSS and Thematic Mapper (TM) images. Image normalization for the change vector analysis and image differencing was carried out using a pseudoinvariant feature approach. It was assumed that detecting changes using data from a single sensor would be preferable to an analysis based on two or more different sensors. However, this study showed that change detection using a combination of TM and MSS produced results comparable to those using only MSS data. It was also found that each of the five methods had their own strengths and weaknesses. Red band image differencing successfully differentiated the two major types of changes. Disturbed areas were associated with increased red radiance, and areas of new or denser green vegetation were associated with decreased red radiance. Principal component analysis gave a much more complex perspective of land use changes, though it requires considerable knowledge of spectral cover types in order to interpret the eigenvector loadings. Post-classification comparisons were found to be intuitive to understand, but complex in the sense that the number of potential land use transitions equals the square of the number of land use classes in the original classifications. Results obtained from change vector analysis showed that the large magnitude spectral changes using the MSS-TM image pairs were comparable to those using only MSS images. For the multi-temporal composite and classification approach, the best results were obtained using the combined TM and MSS data.