Analysis of Spatial Patterns in Remotely Sensed Data Using Multivariate Spatial Correlation

Abstract

Multivariate Spatial Correlation (MSC) (Wartenberg, 1985) is a method for quantifying spatial autocorrelation in multiband data. Analysis of simulated data shows that the MSC matrix is more robust in the presence of noise within groups of correlated bands than conventional covariance or correlation matrices. The first three MSC components of 10 band, simulated data explained 99% of the multivariate spatial correlation, whereas for the principal component analysis (PCA) the first three components explained only 75% of the total variance. HYDICE imagery of Cuprite, Nevada is used to illustrate the value of the method with real data. The MSC matrix shows much less correlation between bands, suggesting that although the HYDICE data is highly correlated, it is not necessarily redundant information. MSC component images tend to be less sensitive to noise, unlike the PCA components.