Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
4458863 | Remote Sensing of Environment | 2015 | 19 Pages |
•We compare algorithms to create glacier area, elevation change and velocity products.•Round robin experiments were used for each product to select the best one.•The accuracy of glacier area depends on the correct interpretation of debris cover.•For elevation change accurate co-registration of the two DEMs is most important.•Velocity is best obtained by feature/offset tracking for optical/microwave data.
Glaciers and their changes through time are increasingly obtained from a wide range of satellite sensors. Due to the often remote location of glaciers in inaccessible and high-mountain terrain, satellite observations frequently provide the only available measurements. Furthermore, satellite data provide observations of glacier characteristics that are difficult to monitor using ground-based measurements, thus complementing the latter. In the Glaciers_cci project of the European Space Agency (ESA), three of these characteristics are investigated in detail: glacier area, elevation change and surface velocity. We use (a) data from optical sensors to derive glacier outlines, (b) digital elevation models from at least two points in time, (c) repeat altimetry for determining elevation changes, and (d) data from repeat optical and microwave sensors for calculating surface velocity. For the latter, the two sensor types provide complementary information in terms of spatio-temporal coverage. While (c) and (d) can be generated mostly automatically, (a) and (b) require the intervention of an analyst. Largely based on the results of various round robin experiments (multi-analyst benchmark studies) for each of the products, we suggest and describe the most suitable algorithms for product creation and provide recommendations concerning their practical implementation and the required post-processing. For some of the products (area, velocity) post-processing can influence product quality more than the main-processing algorithm.