2.2. Tasselled cap transformation method For a sensor carrying four bands, Gram–Schmidt orthogonalization is used (Huang et al. 2002; Crist and Cicone 1984b; Kauth and Thomas 1976). For higher number of bands, after applying PCA, all three axes are rotated to separate different features from each other. It has been found that there is no standard mechanism in literature currently available for this post-PCA rotation, and thus rotation is subjected to interpretation (Lobser and Cohen 2007). Lobser and Cohen (2007) attempted to standardize derivation of transformation coefficients for future sensors. By following the standard developed by them, the below-mentioned approaches were followed:
(1) Firstly, dummy target was generated by applying Landsat TM TCTcoefficients on Landsat 8 data.
(2) Then, by using PR method, PCA transformed Landsat 8 data were rotated for the best alignment to the dummy target. (3) Finally, simple multiplication was done between the transformation coefficients developed through Procrustes and PCA coefficients to get the final TCT coefficients for Landsat 8.
Although these components are ultimately rotated, the 3D structure of the data space is retained through this rotation. Thus, the final transformation is highly dependent upon the chosen starting dummy target.