GLIMS Working group on algorithms

[Niels Reeh]

Dear Andi

Response to your mail of 2. November.

>> (2) It will be the RC's decision whether to use the delivered orbit
>> information or own precise control points for geo-registration. In the
>> first case, immobile points on the image are needed for multitemporal
>> comparison (for details cf. GLIMS-homepage!)

AS A START, WE WILL USE THE DELIVERED ORBIT INFORMATION FOR CO-REGISTRATION

>> (4) a standard multispectral classification for clean ice. This algorithm
>> should be a more or less simple and uniform one for entire GLIMS,
>> easily applicable for every RC. Within the Swiss Glacier Inventory we
>> (mainly the work of Frank Paul) have good experience with ratio-images
and
>> subsequent thresholding.

DUE TO THE DELAY WITH THE RELEASE OF ASTER SCENES (I ASSUME THERE WILL BE NO
GREENLAND ASTER SCENES THIS YEAR) WEE HAVE STARTED TO ORDER AND WORK ON
LANDSAT 7 SCENES OF SELECTED GREENLAND ICE-MARGIN AREAS. WE  HAVE NO
EXPERIENCE WITH MULTISPECTAL CLASSIFICATION OF GLACIER COVERED AREAS. CAN
YOU GIVE US A REFERENCE TO THE WORK OF FRANK PAUL AND/OR OTHER RELEVANT
REFERENCES TO RATIO-IMAGE CLASSIFICATION?


>> (5) the case is more difficult for debris covered ice. The debris-covered
>> ice is certainly important for a inventory. Monitoring glacier
>> fluctuations from debris-covered ice, however, is a difficult thing
>> anyway. So, perhaps, we need less accuracy for these cases. As you
>> know, Bruce Raup did some work on detecting the debris-coverd ice
boundary
>> by geomorphometric DEM-analysis.

DEBRIS COVERED ICE IS NOT A BIG PROBLEM IN GREENLAND. HOWEVER, MAPPING
TRIM-LINES COULD BE USEFULL, SINCE IT WOLD GIVE GIVA AN IDEA OF THE ICE
MARGIN RETREAT SINCE THE END OF THE 19TH CENTURY/ BEGINNING OF THE 20TH
CENTURY. DOES ANYBODY HAVE SUGGESTIONS FOR MULTISPECTRAL OR OTHER
CLASSIFICATION OF TRIM LINES?

>> (3) besides the above "GLIMS-standard" algorithm, we should allow every
>> group to improve such results by enhanced classification, manual or
>> semi-automatic digitising, or field work, or... depending on the
>> available experience and equipment.
>> >>> Shall we introduce two groups of boundaries in the GLIMS data base, a
>> standard one (11) and a individually improved one (10), or simply take
the
>> best (RC-decision) ? <<<

TO MY OPPINION, STANDARDIZATION SHOULD HAVE A HIGH PRIORITY IN GLIMS. I
THEREFORE FAVOUR TWO GROUPS OF BOUNDARIES (STANDARD AND INDIVIDUAL).

>> (8) cf. GLIMS-homepage. (There is presently a discussion on the
>> subdivision of contiguous ice masses and related ID-assignment. Related
>> to that is a discussion whether to introduce different types of
>> boundaries: distinct (e.g.ice/rock), indistinct (e.g.firn-divide) ,
>> arbitrary).

FOR THE GREENLAND ICE SHEET, WE ARE CERTAINLY FACING THE PROBLEM WITH
INDISTINCT BOUNDARIES. WE WILL GENERATE THE INDISTINCT BOUNDARIES
AUTOMATICALLY AS TRAJECTORIES TO ELEVATION CONTOURS. AS THE INDISTINCT
BOUNDARIES ARE LIKELY TO BE  MUCH MORE UNCERTAIN THAN THE DISTINCT ONES, AND
PROBABLY SUBJECT TO MAJOR CHANGES AS THE KNOWLEDGE ON SURFACE TOPOGRAPHY
AND/OR FLOW PATTERN IS IMPROVED, THERE ARE GOOD REASONS TO INTRODUCE THESE
MARGINS AS A SPECIAL TYPE. ALSO, IT IS NOT FEASIBLE TO ASSIGN A GLACIER-ID
TO EVERY INDIVIDUAL FEATURE AT THE ICE-MARGIN THAT COULD BE CONSIDERED A
SPECIFIC GLACIER. AS A START,  ONLY MAJOR GLACIER OUTLETS SHOULD BE
CLASSIFIED AS INDIVIDUAL GLACIERS, WHEREAS THE QUIET ICE MARGIN SECTORS
SEPARATING THE OUTLET GLACIERS SHOULD BE TREATED AS A UNIT WITH  ONE
GLACIER-ID. THESE SECTORS CAN LATER BE SUBDIVIDED INTO INDIVIDUAL GLACIERS
IF A DEMAND FOR MORE DETAIL ARISES.

>> (9) One thing which has to be deduced manually (or semi-automatically) is
>> the (or several) central flow lines. These are of some importance for
>> deriving glaciological parameters (14).

WE INTEND TO USE A SEMIAUTOMATICAL PROCEDURE TO GENERATE CENTRAL FLOWLINES.

>> (14) There are some (simple) algorithms needed to derive glaciological
>> parameters from the above results (glacier area, altitudinal range,
>> length, slope, direction etc.). I suggest, it should be a task of the
>> parameter group to define these parameters and the way to compute it in a
>> GLIMS standard way. Many of these parameters can be both calculated from
>> the entire glacier and a flow line. For most parameters a intersection of
>> the planimetric boundaries and flow lines with a DEM is necessary.

MANY OF THE GLACIOLOGICAL PARAMETERS (E.G. ALTITUDINAL RANGE, LENGTH,
SLOPE ) THAT MAY BE USEFUL FOR CHARACTERIZING MOUNTAIN GLACIERS, DON'T HAVE
MUCH MEANING FOR THE GREENLAND OR ANTARCTIC ICE SHEETS. THEORETICALLY, MOST
OF THE GREENLAND ICE MARGIN SECTORS HAVE ACCUMULATION BASINS THAT EXTEND
RIGHT TO THE CENTRAL OR SOUTHERN  DOMES OF THE ICE SHEET. SO ALTITUDE RANGE
AND SLOPE ARE NOT USEFUL PARAMETERS. A SUGGESTION IS TO DEFINE AN UPPER
ALTITUDE LIMIT AS THE LIMIT BELOW WHICH E.G. 90% OR 99% OF THE DRAINAGE
BASIN AREA IS LOCATED. THIS LIMIT WILL DEPEND ON THE ESTIMATED BASIN
BOUNDARIES, AND THUS BE SUBJECT TO CHANGE AS OUR KNOWLEDGE OF THE POSITION
OF THE BOUNDARIES IMPROVES.

>> (16) The data base things are not our task. Presently definition of
>> maximum data base fields and the minimum required for data base entry is
>> underway.


THE POSSIBILITY OF LATER SUBDIVISION OF ICE MARGIN SECTORS ORIGINALLY
CLASSIFIED AS A SINGLE  UNIT SHOULD BE CONSIDERED IN THE DESIGN OF THE
DATABASE.


Best regards

Niels Reeh