Precision description for remote sensing glacier velocity data

A large amount of velocity data is now becoming available through portals, pipelines and repositories. Typically the error characterisation for these individual velocity fields or mosaics is done through sampling statistics, resulting in a proxy of precision for the whole dataset. However even within a scene pair, the appearance can change considerably, or be stable at nearby locations. For example, think of regions close to the transient snowline, or an elongated moraine band, a crevasse train after a bump or a shear zone. Here the precision of localising an exact image match is clearly anisotropic. If such anisotropic precision estimates are taken into account, it is possible to provide a more correct error-propagation. The merit of velocity data can be found in the help for inversion for thickness estimates (as it is related to the fourth power), or shear and strain rates. Here we introduce a simple and fast methodology to generate an individual dispersion estimate, based upon the similarity surface of an image match. A linear least squares adjustment of the neighbouring similarity scores is sufficient to fit an oriented gaussian peak. This setup makes the computation fast and is easy to implement into already available processing pipelines. We demonstrate its effectiveness on two glaciers, Sermeq Kujalleq, a large outlet glacier of the Greenland icesheet, with strong shear margins and Malaspina Glacier a piedmont glacier with looped moraines. We find directionality within an image subset to be the dominant factor influencing the correlation dispersion. This stems from crevasses and moraine bands within the imagery, while a relation to differential flow, such as shear, is less pronounced. It is our hope, this methodology will narrow the integration gap between models and measurements.