On the Potential of Fully-Automated Glacier Mapping for Area Change Assessment
Paul, Frank; Bolch, Tobias; Moelg, Nico
University of Zurich, SWITZERLAND

Automated glacier mapping from optical satellite data with a simple band ratio is straightforward and provides very accurate results in case of clean ice. But so far time-consuming editing work is required to correct the debris-covered glacier parts, improve the mapping in regions with clouds, shadow and seasonal snow, and also the selection of the optimal threshold for the band ratio takes time and requires manual intervention. While the interaction by an analyst guarantees creation of a high quality product, it hinders fully automated and thus quick or operational processing. In times of rapid glacier changes, change assessment should be performed much more frequent than re-quired to update a glacier inventory. In this study we have investigated the possibilities of automated data processing for rapid change assessment and its impact on the quality of the results.

The underlying assumptions for these investigations are that most of the area changes took place for clean ice regions (i.e. the debris-covered area remains rather unchanged for a couple of years) and that the mapped glacier area is rather insensitive to the thres-hold value used to transform the ratio image in a binary glacier map. Moreover, changes for debris-covered glaciers are more difficult to determine as the uncertainty in the re-peat manual delineation can be in the same order as the observed change (a few per-cent). To provide quantitative numbers on the impacts of fully-automated glacier map-ping on the quality of the final product, the following questions were investigated:

- What is the impact of a fixed vs a manually selected threshhold on glacier area and derived changes in area?
- Can debris-cover mapping be neglected for change assessment?
- What is the smallest time period between two scenes that provide usefull results?
- How can commission (e.g. lakes, seasonal snow) and omission (e.g. debris, ice in shadow) errors be reduced in automated processing chains?