3.4. Class spectra generation for irrigated areas of India
The class spectra were generated using unsupervised ISOCLASS k-means classification (Tou and Gonzalez, 1975; Lieca, 2007) on the MODIS 500 m MFDC consisting of 952 bands for each of the seven zones (six zones in Fig. 4(a) and a single zone in Fig. 4(b)). Different classification methods (Lu and Weng, 2007) have their own merits. The ISOCLASS clustering has many advantages: (a) highly successful in finding spectral clusters in data, (b) interactive, (c) algorithm readily available and widely tested, (d) not biased to the top or the bottom of data, and (e) fastest-known method. The MFDCs of each of the segments were classified into 250 classes, and their time-series NDVI class spectra generated.