Wetland change detection using Landsat-5 Thematic Mapper data in Jackson Hole, Wyoming

This study examined the digital change detection method using Landsat-5 TM imagery between 15 August 1985 and 23 August 1988. Emphasis is on the detection of wetland-cover type changes in the floodplain of the Jackson Hole, Wyoming study area. A new change detection algorithm, Euclidean Distance Analysis, was developed in an attempt to utilize the multiple-band information in a selected band-combination, as an alternative to the conventional single-band analysis methods. The other seven change detection techniques investigated include post-classification comparison, image differencing, image ratioing, image regression, principal components analysis, tasseled cap transformation, and vegetation index analysis. The results of the eight methods were compared to each other and to ground-referenced information.; The nature of the change detection problem in general is first addressed in order to demonstrate the complexity of the digital change detection task. Change detection algorithms used with satellite imagery were then reviewed and compared. A threshold technique was tested at various levels of 0.1 intervals in order to discriminate the change and no-change pixels in the transformed images of the different image analysis techniques. The use of different accuracy indices was also examined in determining the optimal threshold level for each change image. As the standard measure for classification accuracy, the Kappa coefficient of agreement was used for evaluation.; The highest classification accuracy in terms of change and no-change was provided by both the image ratioing and the Euclidean distance analysis. The new technique of Euclidean distance analysis holds more promise for change detection than the other enhancement algorithms evaluated in this study, since it summarizes the multiple-band information on the cover-type changes and reduces the data dimensionality. The monitoring of wetland condition by the tasseled cap transformation was confirmed to be one of the most effective methods and it presented slightly better change detection results than the principal components analysis. Two different types of vegetation indices (VI and NDVI) were ineffective in this change detection analysis. Also, image differencing and image regression methods did not provide good results in change/no-change classification accuracy. Finally, the multicomponent approach, incorporating both enhancement and classification, appeared to offer the most promising change detection technique since it provides more reliable information on the location and characteristics of the change while minimizing the classification errors. In conclusion, the multicomponent approach for change detection, which would combine the strengths from both procedures of enhancement and classification is recommended.