Information extraction from spaceborne SAR data of up to 64-image depth image stacks.

Abstract

The spatial statistics of synthetic aperture radar (SAR) images of the Earth’s surface have been extensively studied in the past. Recent commercial projects from several companies will provide a lot of high quality and importantly high quantity of SAR satellite data, which enables the analysis of the SAR images in time (temporal) instead. This thesis has investigated two different image stacks, containing 28 and 64 images respectively, of SAR intensity data. The investigation has focused on how the statistical distributions in time can be characterized. The temporal SAR data have

been tested against eight different target distribution models with the Anderson- Darling goodness-of-fit test. The use of the Anderson-Darling test together with

looking for spatial patterns have allowed for comparison of how well the target distribution models fit, even with a limited number of images available. Both statistical and visual interpretations of the results have been made through tables and false-colored images. The results have shown that the distribution of temporal SAR intensity data follows a Gamma distribution irrespective of surface area type. Furthermore, none of the target distributions performed specifically well on areas containing buildings and structures. There was a difference how well the Gamma distribution model performed between the smaller and larger image stack, where the Gamma distribution model performed better for the larger stack, which could point to even better fit to the distribution model if an even larger image stack was used.