GLIMS "Round-Robin" Comparative Image Analysis Experiment



Summary

In order to quantify the uncertainties in GLIMS results due to analyses taking place at different institutions and by different people, we propose an experiment whereby each Regional Center (RC) will analyze the same subset of an ASTER image, using whatever tools they wish, to produce glacier outlines and various scalar attribute data. These results will be sent to the GLIMS team at the U.S. National Snow and Ice Data Center (NSIDC) for compilation and analysis. The results will be published, with input and co-authorship from the participating RCs.

Description

In order to quantify the uncertainties in GLIMS results due to analyses taking place at different institutions and by different people, we propose that each Regional Center (RC) analyze the lower-right corner portion of the image shown in Illustration 1. This portion of the image contains a variety of glaciers, including clean and debris-covered glaciers, glaciers terminating on land and in water, and an ice field. Several boundary types (e.g. water/ice, veg/ice, rock/ice) are also therefore represented. We ask that analysis be carried out using whatever tools and algorithms you think appropriate. However, orthorectification should not be done, since the differences between results from orthorectified images versus images not orthorectified are well known, and, if present in this experiment, would cloud the results. This experiment is focused on quantifying the uncertainties stemming from use of different algorithms, as well as variations in human interpretation.

The glacier indicated by a green dot (at approximately line 2275, pixel 3900) should be digitized by hand. This will help gauge variability due to differences in glaciological interpretation.

What should be delivered

  1. Glacier outlines, in latitude/longitude/WGS-84 coordinates. Please use the geolocation data in the HDF metadata to determine the mapping between line/sample coordinates and lon/lat, even if you know it's incorrect. This is so that lon/lat coordinates from different groups will be comparable. (We could also work exclusively in pixel coordinates, but GLIMSView is not designed to output pixel coordinates, and some may want to use GLIMSView.)

  2. Glacier outlines, in pixel (line/sample) coordinates (optional).

  3. As much GLIMS metadata as you can include. The primary goal of this test is to quantify variability in the glacier outlines, but variability in WGMS classifications are also of interest.

  4. Data should be in the GLIMS Data Transfer Format if possible. If this is not possible or too inconvenient, then we can accept ESRI shapefiles of the outlines and attributes, text files for the metadata, etc. We could also accept ESRI .E00 files, or GMT multi-segment files.

  5. A completed checklist documenting the analysis steps (see Resources below).

Time Line

We would appreciate prompt attention to this experiment. I (Bruce) intend to present whatever results are available at the IGS Symposium on Arctic Glaciers in Geilo, Norway, in August 2004. We realize that everyone is busy, and possibly doing field work, but the results of this experiment are of paramount importance to the quality and usability of the GLIMS glacier database.

Therefore, we require your results by July 9, 2004 in order to be included in the presentation at the IGS Symposium.



Resources



Frame2