Comparing the utility of image algebra operations for characterizing landscape changes: The case of the Mediterranean coast

The aim of this study is to compare various image algebra procedures for their efficiency in locating and identifying different types of landscape changes on the margin of a Mediterranean coastal plain, Cukurova, Turkey. Image differencing and ratioing were applied to the reflective bands of Landsat TM datasets acquired in 1984 and 2006. Normalized Difference Vegetation index (NDVI) and Principal Component Analysis (PCA) differencing were also applied. The resulting images were tested for their capacity to detect nine change phenomena, which were a priori defined in a three-level classification scheme. These change phenomena included agricultural encroachment, sand dune afforestation, coastline changes and removal/expansion of reed beds. The percentage overall accuracies of different algebra products for each phenomenon were calculated and compared. The results showed that some of the changes such as sand dune afforestation and reed bed expansion were detected with accuracies varying between 85 and 97% by the majority of the algebra operations, while some other changes such as logging could only be detected by mid-infrared (MIR) ratioing. For optimizing change detection in similar coastal landscapes, underlying causes of these changes were discussed and the guidelines for selecting band and algebra operations were provided.

► Landsat TM Band 3 algebra efficiently detects changes on the Mediterranean coast. ► Algebra of Landsat TM Bands 3,4,5,7 detects agricultural encroachment on sand dunes. ► Landsat TM Bands 1 and 2 fail to detect coastal retreat and advance. ► Infrared bands fail to detect change from bare soil to plastic greenhouses.