A comparison of the classification of wetland characteristics by linear spectral mixture modelling and traditional hard classifiers on multispectral remotely sensed imagery in southern India

Abstract

An attempt has been made to compare the recently evolved soft classification method based on linear spectral mixture modelling (LSMM) with the traditional hard classification methods based on iterative self-organizing data analysis (ISODATA) and maximum likelihood classification (MLC) algorithms, in order to map and monitor the coastal wetland ecosystems of southern India, using Indian Remote Sensing Satellite (IRS) 1C/1D LISS-III and Landsat-5 Thematic Mapper (TM) image data. ISODATA and MLC methods were attempted to produce maps of 5, 10, 15 and 20 wetland classes for each of Pitchavaram, Vedaranniyam and Rameswaram sites. The accuracy of the derived classes was assessed with the simplest descriptive statistic technique called overall accuracy and the discrete multivariate technique called KAPPA accuracy. We found that MLC classification method produced maps with higher accuracy than ISODATA classification method. However, there was a systematic decrease in overall accuracy and KAPPA accuracy, when more number of classes were aimed to be derived from IRS-1C/1D and Landsat-5 TM imagery by ISODATA and MLC, caused by the limited spectral separability and instantaneous field of view (IFOV) of the sensor. The later one inevitably caused occurrence of number of mixture pixels (mixels) in the image and its effect on the classification process was a major problem to derive accurate wetland cover types, inspite of the increasing spatial resolution of new generation Earth Observation Sensors (EOS). To achieve higher accuracy of deriving wetland cover information from these data, a soft classification method based on linear spectral mixture modelling was presented. This method considered number of reflectance end-members that form the scene spectra and determined their nature and finally decomposed the spectra into their end-members. Because of the limited number of spectral bands, we collected only three spectral end-members (vegetation, soil and moisture) ideal to the accurate estimation of their sub-pixel fractions from the image data. The resulted fractions from LSMM were compared with normalised difference vegetation index (NDVI), ground truth data, as well as those estimates derived from the traditional hard classifier. NDVI values exhibited a positive correlation with vegetation fractions and negative correlation with soil fraction. Comparison with field data demonstrated higher reliability of the LSMM than the traditional approach of using predefined classification schemes with discrete numbers of cover types. The LSMM would seem to be well suited to locate the small wetland habitats that occurred as sub-pixel inclusions, and to represent continuous gradations between different habitat types in the study sites.