Optimization of Cross Correlation Algorithm for Annual Mapping of Alpine Glacier Flow Velocities; Application to Sentinel-2

Nowadays, satellite observations cover most of the Earth’s surface in a repetitive manner. This information is crucial for documenting variability and environmental changes such as glacier surface velocity. With this in mind, digital image processing has been developed and improved over the past decades. The processing challenges are now related to optimizing parameters that account for the high variability of natural processes, as well as filtering and aggregating the results to provide useful products to end-users. Based on the normalized cross correlation (NCC) method applied to Sentinel-2 optical satellite observations up to 400 days apart, we present a series of tests to derive optimal parameter values for the quantification of alpine glacier ice velocity that we have applied to the Mont-Blanc massif where in situ measurements are available. We found that a search distance adapted to the temporal baseline, a $16\times 16$ pixel window size, and a $5\times 5$ pixels sampling provide an appropriate combination of parameters to process Sentinel-2 with the NCC method when applied to small alpine glaciers. Combining several spatial and temporal filters applied to a large set of more than 18000 displacement maps obtained between 2015 and 2021, then aggregating these filtered maps using statistical or linear regressions into annual maps, yields near-complete maps of the test region with a root mean square error (RMSE) reduced to about 10 m.yr $^{-1}$ compared to in situ measurements.