VIII. CONCLUSION |
Spectral linear mixture analysis provides an efficient mechanism for the classification of superspectral RS data (e.g. from MODIS). It aims to identify a set of reference signatures (endmembers) that can be used to model the reflectance spectrum at each pixel of the original image. These endmembers are extracted from the images using techniques such as PPI or scatter plots. Thus, the modelling is carried out as a linear combination of a finite number of ground components and their reflectance spectra. An abundance map helps in estimating the proportion of each endmember in a pixel. The performance of linear unmixing technique for identification of endmembers, interclass variability and presence of adjacency effect using 250-500 m MODIS data was evaluated. MODIS LC outputs are comparable to LISS-III classification results, evident from validation, as most endmembers were correctly classified while mixed pixels were within 10-15% of the values obtained in LISS-III classification. The MODIS data tested here have an advantage over alternative high or medium resolution datasets, given their coverage, high temporal resolution, cost-free availability, and utility for LC mapping as shown in this study. The challenges associated with this data type are
(i) georegistration of the pixels when no or limited identifiable GCPs exist,
(ii) misclassification of pixels due to LC mixtures, and
(iii) mapping of LC in heavily fragmented areas with highly contrasting nature.
One way of overcoming problems (ii) and (iii) is to use MODIS data with high spatial resolution images.
Even, in case MODIS ceases to function, the technique would still be applicable for datasets with low spatial resolution.
We are grateful to the Ministry of Science and Technology, Government of India for the financial assistance.
