The present study compares the effectiveness of two common preclassification change detection (CD) methods that use two-dimensional data space of spectral-textural (S-T) change information. The methods are principal component analysis (PCA) and change vector analysis (CVA) in the Gorgan Township area, Golestn Province, Iran. A series of texture-based information was calculated mainly to separate those land use/land cover (LULC) conversions that are spectrally indistinguishable and also to provide a basis for automatic classification of S-T data space. Both methods were evaluated in terms of accuracy and the required time and expertise. Having the two-dimensional S-T data space generated, support vector machine (SVM) classifier was implemented to automatically extract changed pixels and the receiving operator characteristic (ROC) was employed to assess the accuracy of the output. According to the results, the study area has witnessed substantial mutual transformations between various LULCs among agricultural lands were the most dynamic category in the region. The PCA method applied to the S-T information achieved a ROC of 0.90-indicating an acceptable performance-while the S-T CVA method achieved a lower value of 0.75. The S-T PCA method was considerably less time-consuming and less expertise demanding as well as more accurate in our study area.
Keywords: Change detection; Change vector analysis; Iran; Principal component analysis; Texture information.