Accounting for spatial dependence in the processing of multi-temporal SAR images using factorial kriging
The interpretation or classification of multi-temporal satellite images that share large portions of redundant information can generally be improved by principal component analysis (PCA). A shortcoming of PCA is that the spatial structure of the images is ignored. Spatial variability often consists of several nested levels of variance and their blending could mask information that is dominant at a specific level or spatial scale. Factorial kriging (FK) is a geostatistical technique that allows the filtering of spatial components identified from nested variograms and is here used to extract scale-dependent information from satellite images prior to PCA. The benefit of this geostatistical pre-processing of multi-temporal images is investigated using a winter sequence of eight European Remote Sensing (ERS 1/2) Synthetic Aperture Radar (SAR) images. Each image was processed by FK to isolate the variation present at a 'regional' scale (between 289 and 700 m) prior to a PCA of the filtered images. Compared to an earlier study where PCA was performed on the original images, filtering enhanced the relation between the first three principal components and land characteristics associated with topography, soil drainage conditions and land use.