Stratification is a procedure for subdividing the heterogeneous population into subpopulations which are internally homogeneous. Spectral stratification, i.e. stratification based on spectral characteristics can be used for classification of multispectral remotely sensed data. In this paper an approach is suggested for estimating the areas under different cover types by taking samples from spectral strata.

The basis of spectral stratification i s the multidimensional frequency information from the data. The size of a sampling unit is decided by the coefficient of variation within the data and the size of the samples can be obtained either from the allowable error or the level of significance. The samples are allocated on the basis of the proportional geographical area in each stratum. The samples from each stratum are classified using an unsupervised method namely, the Iterative Self Organizing Clustering system and the area under each class is estimated.

The procedure is tested with the Indian Space Research Organization-MSS data for two test sites each covering approximately 20 Sq. kms. One test site pertains to a planned agricultural research farm and the other belongs to a normal agricultural area. Areas under different cover types are obtained for both the test sites. These results are compared with the results obtained by a supervised approach on pixel by pixel basis using maximum likelihood quadratic discriminant function. In the majority of the classes the results are in agreement where as there are little discrepancies in few cases.

Date of this Version