4. Discussion |
Hard classification technique performs well when the actual ground cover is heterogeneous with high spatial resolution data (i.e. LISS-3 MSS data). However, classification accuracies depend on adequate number of training sites, their distribution, and overlap of season (if not day) of training data collection and remote sensing data acquisition. MODIS classified image with coarse spatial resolution had many pixels that were misclassified as is clear from the Accuracy Assessment (table 2, 3 and 4).
Classification accuracy at different spatial scales (administrative boundary level and pixel level) revealed that GMLC on MODIS bands 1 to 7 is superior to other techniques for mapping agriculture as well as built up followed by SAM on MNF for mapping agriculture. Also, a certain algorithm may be good for mapping a “particular class” but at the same time may not be equally good for mapping “all other classes” – GMLC on MODIS band 1 to 7 could map agriculture, built up, forest, plantation and waste land well. However, water bodies could be well mapped with GMLC on PC’s compared to other techniques.
Pre-processing techniques such as Principal Component Analysis (PCA) and Minimum Noise Fraction (MNF) on MODIS data had different effects on the accuracy of the hard classification algorithms. Spectral Angle Mapper and Maximum Likelihood classification performed badly on the PC’s and the MNF components when assessing accuracy with ground truth. Though, the class separability was very good, yet it was difficult to classify each pixel based on signature, since the image was slightly pixelated. However, at the pixel level, the two techniques performed moderately better on PC’s but relatively poor on MNF components maintaining the same level of accuracy as obtained by accuracy assessment using ground truth in the range of 42% to 49%. The pixels in the MNF components were not very distinct and were clustered into sub groups comprising of two or three pixels, leading to inaccurate classification results. The interesting point here is that both the techniques performed well at various spatial scales on MODIS band 1 to 7 products. This reveals that highly preprocessed MOD 09 data (level 3) takes care of all the atmospheric disturbances whereas the 36 band data, MOD 02 at level 1B requires further preprocessing to actually represent a good estimate of the surface spectral reflectance as it would have been measured at ground level without atmospheric scattering or absorption.
The accuracy assessment showed that GMLC on MODIS bands 1 to 7 has highest overall accuracy followed by SAM on the same 7 bands (table 4). However, some errors may have occurred since the signal of the pixel is ambiguous, perhaps as a result of spectral unmixing, or when the signal is produced by a cover type that has not been accounted for in the training process. Also, with coarser resolution, due to mixed pixels, chances of high accuracies in hard classifications decline.
