Robust glacier displacements using knowledge-based image matching

Matching of repeat optical satellite images is a powerful tool for the analysis of glacier movement. Due to the rapid growth of earth observation data archives there is an increasing potential for regional glacier flow studies on a decadal scale. However, exploitation of these image collections is hindered by the currently manual pre- and post-processing steps required. Automating these steps, and using content from all available imagery will help tap the wealth of information present within the data. In this study we explore this possibility with the aim to extract velocity fields that are both accurate and reliable. We propose a bottom-up approach in which simple image matching estimates from three images are pierced together, to give rise to the far more complex time-dependent glacier flow. Taking advantage of our knowledge of glacier behavior, rather then crude statistical inferences. We first loosen the threshold of the matching, estimating dispersions for every candidate match. Then by implementing the procedure for multiple time periods we constrain the solution making it possible to apply probabilistic testing. Our approach is alternative to common top-down implementations and instead derives from first principles. In summary, we introduce the concept of precision and reliability of image matching which enhances the analysis of velocity fields with hypothesis model testing to investigate its driving forces.