Comparison of feature-tracking methods for glacier-velocity evaluation in different regions of the Earth.

Several feature tracking methods for glacier velocimetry calculations has been proposed in the recent years for the increasing interest in these features, being the most significant signals of Climate Change. In the present work, we compare different modules of feature tracking, i.e., IMCORR, ImGRAFT and GIV. To these existing modules, we add for testing and comparing a new Machine Learning-based method, which couples rigid image registration and correlation methods in order to disentangle the real glaciers movement from the artifacts caused by the image acquisition and provide an accurate estimation of the glacier velocity and direction. The evaluation of this last model and the cross-comparison with the previous methodologies is possible by validation with field available data, in four regions of the Earth: Antarctica (David Glacier), Italy (Miage G.), Chile (Exploradores G.) and Pakistan (Baltoro G.). This field comparison allows to evaluate existing feature tracking modules in different regions of our planet, periods and satellite (Landsat-family, Sentinel-2 and ASTER imagery) sources and to provide and validate a new method from Machine Learning techniques.