|M.Sc Student||Agmon Ori|
|Subject||Change Detection Using Unmixing in Transition Zones between|
Mediterranean and Semi-Arid Areas
|Department||Department of Civil and Environmental Engineering||Supervisor||Professor Maxim Shoshany|
|Full Thesis text|
Change detection (CD) is the process of identifying differences in a surface's conditions by observing it at different times. The basic premise when using remote-sensing data for CD is that changes in land cover must result in changes in the surface's reflectance.
CD plays a major role in the identification of variations in land use and land cover that are the consequence of economic, ecological, and social processes. Better understanding of past processes may improve planning strategies for land use. A variety of CD techniques have been developed by researchers. Mediterranean and climatic gradient regions complicate the CD process due to their spatial and temporal heterogeneity. The aim of this study was to develop a method for detecting change in Mediterranean and climatic gradient regions. For this purpose, Landsat TM imagery of central Israel, for the years 1984 and 2003, were used and the images were separated into (a) regions with relatively large edges that represent boundaries between different land covers and infrastructures or (b) regions that are relatively homogeneous. CD in homogeneous regions was performed using a linear mixture model (LMM). The most dominant end-member represented the pixel land cover and its change, if occurred. In regions with relatively large edges, edge detection was performed using a procedure developed specifically for the study area whereby change is expressed by the appearance or disappearance of edges.
Research results show that this process of separation improves CD. The LMM approach is highly efficient for CD in homogeneous regions of the study area (error=0.04). Comparison of these results with results obtained using an image differencing method showed fewer "false alarms" for the LMM method, in addition to the elimination of the need to set threshold values. CD in edge regions was not as good as for homogeneous regions (error 0.29) but was still better than the results of LMM in edge regions (error=0.34).
These results indicate that using spatial and textural information to separate different parts of the image and applying a suitable CD method for each part can improve CD quality in regions of high spatial and temporal variability. Further research is needed to improve the separation of images into regions with relatively large edges and regions that are relatively homogeneous by either improving the proposed method or developing a better one.