The Application of Factorial Kriging and Fourier Analysis for Remotely Sensed Data Simplification and Feature Accentuation

High spatial resolution satellite imagery provides information on landsurface properties and features over a wide range of spatial scales. Depending upon the application, and the spatial resolution of the sensor with respect to the size of landsurface features, we may wish to discard or isolate certain components of spatial variation, corresponding to specific spatial frequency ranges. This is an important step in image simplification, data reduction and feature accentuation. From a theoretical point of view factorial kriging (FK) and Fourier analysis are identical image filtering techniques, however they differ in the method of application. We compared the results provided by applying both to multi-temporal synthetic aperture radar imagery (ERSSAR) and a single channel SPOT-XS image. Choosing which spatial frequencies to suppress is the most important part of filtering. When using the fast Fourier transform (FFT) it was not evident which frequencies to suppress. Factorial kriging uses the variogram model to both identify and separate short-range and long-range components of spatial variation. Variogram analysis and FFT can be combined to produce similar results to FK alone. However, the quality of the filtered image relies heavily on the quality of the chosen nested variogram model. For SAR imagery, the long-range spatial component images represented an important simplification of the originally very complex images. Applied to SPOT-XS images we were able to separate short-range spatial structures related to fields, from the smoothly varying properties of the landsurface, thought to be a function of soil and geomorphic variation. For large images a significant data reduction was achieved either simultaneously by kriging to a larger pixel size using FK, or by degrading the FFT filtered image.