Impact of Surface Anisotropy on Classification Accuracy of Selected Vegetation Classes: An Evaluation Using Multidate Multiangular MISR Data Over Parts of Madhya Pradesh, India

Abstract

An empirical analysis of the effect of surface angular anisotropy on classification accuracy is presented in this paper using seven-date multiangle imaging spectroradiometer (MISR) data acquired during the period of October 2005-March 2006. The effect of surface anisotropy on classification accuracy was assessed for those classes that show spectral mixing at nadir for three dates, viz., December 18, January 3, and January 19 at red and near-infrared (NIR) wavelengths, using three different methods: (1) using two off-nadir sensor angles (70.5deg and 60deg); (2) using the second component of principal component analysis (PC2); and (3) using the second component of the Rahman-Pinty-Verstraete model (k); the latter two methods represent the directional components of MISR-observed reflectance. The parallelepiped classifier with a three-sigma threshold was used for all classifications, and the classification accuracy was assessed using overall accuracy and the kappa coefficient. In the red band, it was observed that classification using off-nadir sensor angles improves the classification accuracy by about 10%-50%, depending on the vegetation stage, with respect to nadir. A consistent significant increase in classification accuracy for the three dates (December 18, January 3, and January 19) was found using the directional component (PC2) compared with the spectral component (PCI) in the red band, whereas for the NIR band, the classification accuracies were consistently lower compared with that of PCI. Classification using the three different anisotropy measures previously defined as well as nadir, in combination with multispectral (green, red, and NIR) information, resulted in high classification accuracies for all the three dates (December 18, January 3, and January 19). This implies that the multispectral component of reflectance is by far the most important determining factor influencing the classification accuracy.