Support the Arctic Sea Ice Forum and Blog

Author Topic: Co-registering LandSat with bedrock depth  (Read 20432 times)

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Co-registering LandSat with bedrock depth
« on: June 26, 2014, 02:53:59 PM »
By popular demand ...  I'm posting now to the forum thanks to a memorable new password from Neven.

I took a look into what is involved in precise co-registration of Jakobshavn Isbrae imagery. Wipneus and Chris have gone round and round with this previously, relating the goofy coordinate system of Piomas to conventional Arctic projections.

For the calving front of Jakobshavn Isbrae, the scale is much smaller, just a few tens of km covers the calving front and descending ice. We'd like to have data co-registered to 15 meters or so when possible.

We have Landsat-8  and Modis (nadir), satellite radar (whiskbroom vs pushbroom), Operation IceBridge airplane radar transects (trailing antenna), projections such as mercator and stereographic, and coordinate manipulations done during image processing by satellite centers and later by journal authors, changes over time in archive coordinates, satellite drift, image resolution, and so forth.

All this data needs to be brought into a common coordinate system so it can be co-registered (stacked in pixel-perfect co-registered image layers) for purposes of arcGis-type analysis, for example overlaying the ice surface digital elevation map, bedrock DEM, surface velocity field, and a time series of images.

One problem I've encountered is the lack of fixed reference points on the ground. Historically a pattern of quarter circles would be emplaced on the ground along with a corner radar reflector. With 3-4 of these to anchor alignment, any imagery can be forced into pretty good registration.

While rock shows in most imagery, it can be obscured by clouds, snow-covered, have inland lakes frozen vs calving and melting, with shorelines alternately clear or obscured. The rest of the scene is worse with snow and ice blanketing features and the ice stream -- and indeed the whole Greenland ice sheet -- in motion.

In short, to tie everything onto a common lat, lon coordinate system, we are forced to root around in metadata for each data source. For example, the bedrock elevation transacts of Gogineni comes in x,y,z format, with the best DEM file for Jakobshavn an excel-busting 524,288 array. When sorted for depth in meters (z), the deepest parts look like this:

digital lat          digital lon      depth of bedrock below mean sea level
-48.49456221  69.20637076  -1512.200
-48.49656494  69.20637076  -1512.200
-48.46251853  69.20837349  -1508.124

I'll stop here. The idea is to get the precise lat,lon of major glacier overdeepenings and sills so they can be co-located precisely with lat,lon of the calving front and so brought into a predictive environment.

Wipneus

  • Citizen scientist
  • Young ice
  • Posts: 4009
    • View Profile
    • Arctische Pinguin
  • Liked: 644
  • Likes Given: 0
Re: Co-registering LandSat with bedrock depth
« Reply #1 on: June 26, 2014, 05:00:45 PM »
Hi A-Team

Welcome, glad to see you here.

I am looking forward to see your findings. My needs are simple: align the 15m Landsat images so the visible (or not) bedrock is stable and what is left is the movement of ice.
I use the  CORNER_UL_PROJECTION_X_PRODUCT and CORNER_UL_PROJECTION_Y_PRODUCT variables in the MTL.txt file, treating them as offsets into some grid.
That works excellent in images in the tropics that I follow, but seems to be only approximate in these high latitude images, a few pixels offset remain.

I should look at it again, could still be a bug on my side.
 

Espen

  • Young ice
  • Posts: 3251
    • View Profile
  • Liked: 177
  • Likes Given: 0
Re: Co-registering LandSat with bedrock depth
« Reply #2 on: June 26, 2014, 05:09:06 PM »
Welcome to the Forum A-Team, we need all capable persons and I really appreciate your approach to the stuff  ;) 
Have a ice day!

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #3 on: June 27, 2014, 02:07:25 AM »
Nice group of people here! Crowdsourcing knowledge really works ... I learn something from every page.

Today I took apart the Cresis ice-penetrating radar file for Jakobshavn Isbrae. It is in csv format (lon,lat,elevation), the first two in excessive decimal degrees and the last in meters. The DEM data is an array of size (1340,181) but linearized to 644,540 lines in the form (lon,lat,z). It comes sorted by decreasing latitude.

A whole lot of numbing numbers. I found that EarthExplorer will  display the coverage rectangle translucently over the high res Digital Globe image (google map default) -- in Search Criteria, just paste into the lat,lon popup from Add Coordinates.

-49.7104  69.1870 NW corner
-49.1171  69.0710 SE corner
-49.5943  69.1530 calving front (approx)

However I wanted to make a flythru showing just the bedrock channel with the ice gone. So to chop the file down, I deleted the two blocks of irrelevant latitudes from top and bottom of the file, then re-sorted on longitude and discarded top and bottom there. This got the file down to a more manageable 23034 rows or an 349 x 66 array.

A frame of the flythru then consists of a fixed longitude line (starting to the west of the calving front), with the z values associated with each latitude visualized (and smoothly interpolated) by a simple spreadsheet graphic that is then exported to a Gimp layer where 2x horizontal exaggeration (this fjord is really narrow), rock and stratified ice texture, etc are easily added.


Espen

  • Young ice
  • Posts: 3251
    • View Profile
  • Liked: 177
  • Likes Given: 0
Re: Co-registering LandSat with bedrock depth
« Reply #4 on: June 27, 2014, 06:24:55 AM »
Wow! Strange shape when considering having that huge grinding machine above?
Have a ice day!

LRC1962

  • Frazil ice
  • Posts: 431
    • View Profile
  • Liked: 5
  • Likes Given: 1
Re: Co-registering LandSat with bedrock depth
« Reply #5 on: June 27, 2014, 01:42:58 PM »
Good point espen. Living near what was left of ice age glacier works I should have seen it. Any water in motion tends to leave a flat bottom, so a square/rectangular shape would be better if that can be worked out. I live near the Niagara Escarpment and that is petty well vertical down to a flat plain. What is not can be attributed to receding lake shore line.
"All truth passes through three stages: First, it is ridiculed; Second,  it is violently opposed; and Third, it is accepted as self-evident."
       - Arthur Schopenhauer

sidd

  • Young ice
  • Posts: 4984
    • View Profile
  • Liked: 386
  • Likes Given: 0
Re: Co-registering LandSat with bedrock depth
« Reply #6 on: June 27, 2014, 07:22:55 PM »
a pointer to CRESIS data would be appreciated ..

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #7 on: June 27, 2014, 08:58:36 PM »
Sidd, that link to the x,y,z raw DEM data at Cresis is ftp://data.cresis.ku.edu/data/grids/old_format/ then 2008_Jakobshavn.zip then jakobshavn_bedmap_pts.txt after unzipping.

Wipneus and Espen, here's the variation in Landsat-8 orbital repeats -- the animation below shows all fifteen path 8 row 12's from 22 May 13 to 25 May 14, many cloudy.

As Nukefix noted, the satellite only repeats its orbit up to within some solid tube, so different scenes don't quite repeat perfectly in either ground footprint or geometry. There's definitely some deformation towards the sides, as well as a quantitatible shift of center (lat,lon) coordinate (assuming EarthExplorer implemented these faithfully).

Various online tools calculate distances between two (lat,lon)'s,  though if latitudes are the same, you could get by with  longitudes using a French meter (defined as one ten-millionth of the distance between the North Pole and equator).

Certain other (paths, rows) like the clear (83,233) 24 Jun 14 ascending repeat only once over the last year. And comparing (83,233)'s with (8,12)'s will probably have bigger issues, though life is better when calving front sits in the center of the image. (Almost all of these row,paths show it off in a corner.)

I just noticed that (8,11) and 8,12) not only tile -- that overlap sits over the calving front. So a stereo pair and more ... if only the weather would be clear. (Click second icon: 'Browse Overlay' to display footprints in EarthExplorer.)

It's worth noting that Gogineni (and others) lumped depth classes rather drastically in making their shaded reliefs. That plus jpging disconnect the color key from the image. The raw data is instructive -- sort on z depth for example.

I can't decide between a fly-through or a mountain bike ride up the Jakobshavn Isbrae bedrock channel. I did manage to embed a moving yellow transect line over a Landsat scene so you can tell where the current bedrock DEM transect is sitting.

Cresis also provides nice grayscales of their actual radar returns. However I found it slow drilling given their many many  tracks and never did find the 2014 of Jakobshavn. Ditto Icebridge. (I'm animating 2008, there's been a big push to get these tight deep fast icestreams better.)




A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #8 on: June 28, 2014, 05:48:20 PM »
For everyone's convenience, I looked at 4"x4" thumbnails of all Landsat-8 coverage of the current calving region of Jakobshavn Isbrae for cloud coverage (rather lack thereof). Of the 70 available scenes, 38 appear useful. EarthExplorer accepts these scene identifiers under Additional Criteria searching.

There are 6-8 stereo pairs (same day, consecutive scenes, overlap) but no exact matches of dates between 2013 and 2014. However, the best matched date pairs can easily be found sorting on the Match column in the tab-delimited database below. Use Lsort and Csort to restore the current (very special!) order of rows.

Lsort   Landsat Scene Identifier   Match   #Day   Day   Mon   Year   Path   Row   clarity   Csort   Cloud   PaRow   Center Lat   Center Lon
3   LC80832332014175LGN00   2   175   24   06   14   83   233   good   1   33   83233   69.60667   -51.29052
6   LC80842322014166LGN00   2   166   15   06   14   84   232   cutoff   1   16   84232   68.27978   -51.15327
7   LC80080122014161LGN00   2   161   10   06   14   8   12   good SP   1   2   08012   68.27988   -49.46752
8   LC80080112014161LGN00   2   161   10   06   14   8   11   good SP   1   1   08011   69.6066   -47.7732
10   LC80100112014159LGN00   2   159   08   06   14   10   11   good SP   1   22   10011   69.6066   -50.86458
11   LC80090112014152LGN00   2   152   01   06   14   9   11   good SP   1   6   09011   69.60662   -49.31691
12   LC80842322014150LGN00   2   150   30   05   14   84   232   cutoff   1   4   84232   68.27948   -51.15202
17   LC80080122014129LGN00   2   129   09   05   14   8   12   good SP   1   12   08012   68.27992   -49.60601
18   LC80080112014129LGN00   2   129   09   05   14   8   11   good SP   1   4   08011   69.60666   -47.9235
19   LC80100112014127LGN00   2   127   07   05   14   10   11   good   1   28   10011   69.60623   -51.0104
21   LC80100112014111LGN00   2   111   21   04   14   10   11   good   1   10   10011   69.60624   -51.0072
22   LC80090112014104LGN00   2   104   14   04   14   9   11   good   1   10   09011   69.60643   -49.45147
23   LC80080122014097LGN00   2   097   07   04   14   8   12   good   1   17   08012   68.27955   -49.59887
24   LC80080112014097LGN00   2   097   07   04   14   8   11   good   1   22   08011   69.60652   -47.91558
27   LC80080122014081LGN00   2   081   22   03   14   8   12   good   1   4   08012   68.27951   -49.59312
28   LC80080112014081LGN00   2   081   22   03   14   8   11   good   1   4   08011   69.6065   -47.9094
34   LC80090112014056LGN01   2   056   25   02   14   9   11   good   1   6   09011   69.60626   -49.43474
37   LC80090112014040LGN00   2   040   09   02   14   9   11   good   1   12   09011   69.60632   -49.42945
38   LC80080122013302LGN00   0   302   29   10   13   8   12   goodSP   1   9   08012   68.27983   -49.51267
39   LC80080112013302LGN00   0   302   29   10   13   8   11   goodSP   1   11   08011   69.60661   -47.82377
41   LC80090112013293LGN00   0   293   20   10   13   9   11   good   1   5   09011   69.60646   -49.38038
42   LC80080122013286LGN00   0   286   13   10   13   8   12   so-soSP   1   12   08012   68.27961   -49.52841
43   LC80080112013286LGN00   0   286   13   10   13   8   11   so-soSP   1   32   08011   69.60642   -47.83909
44   LC80100112013284LGN00   0   284   11   10   13   10   11   so-so   1   23   10011   69.60661   -50.93055
46   LC80080122013270LGN00   0   270   27   09   13   8   12   goodSP   1   61   08012   68.27948   -49.5069
47   LC80080112013270LGN00   0   270   27   09   13   8   11   goodSP   1   39   08011   69.6064   -47.81722
49   LC80090112013261LGN00   0   261   18   09   13   9   11   so-so   1   9   09011   69.60647   -49.36928
54   LC80080122013238LGN00   0   238   26   08   13   8   12   so-soSP   1   29   08012   68.2798   -49.52364
55   LC80080112013238LGN00   0   238   26   08   13   8   11   so-soSP   1   22   08011   69.60668   -47.83321
58   LC80100112013172LGN00   1   172   21   06   13   10   11   good   1   30   10011   69.60634   -50.87524
59   LC80100112013156LGN00   1   156   05   06   13   10   11   so-so   1   40   10011   69.60661   -50.91349
60   LC80822332013149LGN00   1   149   29   05   13   82   233   good   1   7   82233   69.60659   -49.75361
62   LC80080122013142LGN01   1   142   22   05   13   8   12   goodSP   1   9   08012   68.27962   -49.52037
63   LC80080112013142LGN01   1   142   22   05   13   8   11   goodSP   1   13   08011   69.60664   -47.82795
64   LC80100112013140LGN01   1   140   20   05   13   10   11   good   1   45   10011   69.60623   -50.91778
65   LC80100112013124LGN01   1   124   04   05   13   10   11   so-so   1   25   10011   69.60624   -50.87119
69   LC80100112013108LGN01   1   108   18   04   13   10   11   good   1   27   10011   69.60654   -50.88116
70   LC80090112013101LGN01   1   101   11   04   13   9   11   good   1   12   09011   69.60663   -49.54324
1   LC80080122014177LGN00   0   177   26   06   14   8   12   useless   2   67   08012   68.27959   -49.44788
2   LC80080112014177LGN00   0   177   26   06   14   8   11   useless   2   27   08011   69.60633   -47.75372
4   LC80100112014175LGN00   0   175   24   06   14   10   11   useless   2   72   10011   69.60629   -50.84766
5   LC80090112014168LGN00   0   168   17   06   14   9   11   useless   2   34   09011   69.60667   -49.31732
9   LC80832332014159LGN00   0   159   08   06   14   83   233   useless   2   28   83233   69.60634   -51.30726
13   LC80080122014145LGN00   0   145   25   05   14   8   12   useless   2   30   08012   68.27973   -49.45283
14   LC80080112014145LGN00   0   145   25   05   14   8   11   useless   2   27   08011   69.60661   -47.75806
15   LC80100112014143LGN00   0   143   23   05   14   10   11   useless   2   66   10011   69.60656   -50.8458
16   LC80090112014136LGN00   0   136   16   05   14   9   11   useless   2   60   09011   69.60651   -49.48273
20   LC80080122014113LGN00   0   113   23   04   14   8   12   useless   2   51   08012   68.27967   -49.60016
25   LC80100112014095LGN00   0   095   05   04   14   10   11   useless   2   34   10011   69.6063   -51.00663
26   LC80090112014088LGN00   0   088   29   03   14   9   11   useless   2   39   09011   69.60654   -49.44342
29   LC80100112014079LGN00   0   079   20   03   14   10   11   useless   2   52   10011   69.60654   -51.00212
30   LC80090112014072LGN00   0   072   13   03   14   9   11   useless   2   6   09011   69.60658   -49.45018
31   LC80080122014065LGN00   0   065   06   03   14   8   12   useless   2   37   08012   68.27984   -49.57592
32   LC80080112014065LGN00   0   065   06   03   14   8   11   useless   2   11   08011   69.60628   -47.89226
33   LC80100112014063LGN00   0   063   04   03   14   10   11   useless   2   30   10011   69.60624   -50.97462
35   LC80080122014049LGN00   0   049   18   02   14   8   12   useless   2   15   08012   68.27988   -49.57734
36   LC80100112014047LGN00   0   047   16   02   14   10   11   useless   2   11   10011   69.60666   -50.9785
40   LC80100112013300LGN00   0   300   27   10   13   10   11   useless   2   13   10011   69.60653   -50.91439
45   LC80090112013277LGN00   0   277   04   10   13   9   11   useless   2   74   09011   69.6065   -49.37297
48   LC80100112013268LGN00   0   268   25   09   13   10   11   useless   2   27   10011   69.60643   -50.90873
50   LC80080122013254LGN00   0   254   11   09   13   8   12   useless   2   53   08012   68.27957   -49.52488
51   LC80080112013254LGN00   0   254   11   09   13   8   11   useless   2   54   08011   69.60667   -47.83454
52   LC80100112013252LGN00   0   252   09   09   13   10   11   useless   2   82   10011   69.60663   -50.92559
53   LC80090112013245LGN00   0   245   02   09   13   9   11   useless   2   79   09011   69.60641   -49.37792
56   LC80100112013236LGN00   0   236   24   08   13   10   11   useless   2   63   10011   69.60625   -50.92145
57   LC80090112013229LGN00   0   229   17   08   13   9   11   useless   2   76   09011   69.60639   -49.36156
61   LC80842322013147LGN00   0   147   27   05   13   84   232   useless   2   17   84232   68.27988   -51.15388
66   LC80090112013117LGN01   0   117   27   04   13   9   11   useless   2   29   09011   69.60627   -49.3174
67   LC80080122013110LGN01   0   110   20   04   13   8   12   useless   2   26   08012   68.27985   -49.47715
68   LC80080112013110LGN01   0   110   20   04   13   8   11   useless   2   7   08011   69.60659   -47.78475

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #9 on: June 30, 2014, 09:23:33 PM »
Next we would like to automate the download, cropping and alignment of the entire series. To do that, fix one date (say 24 June 14). Then in a spreadsheet, subtract from its lat, lon center from those of the other cloud-free dates.

This gives the translation vectors necessary to move the centers so the photos would align on their overlap, which by construction includes the calving front area. Really we just want to use this vector to get crop coordinates for the 34 images and not move large files into a stack until file size is seriously reduced.

Normally the crop box rectangle is specified by (w,h) pixel coordinates in the fixed image. So it boils down to converting decimal lat,lon coordinates provided by EarthExplorer into pixel coordinates.

For 30 m Landsat-8 images of Jakobshavn Isbrae, the dimensions are always the same, 8591 x 8641. That puts the central pixel at (4296,4321) which for the 24 June 14 center (lat,lon) is (69.60667,-51.29052). Entering the upper left and lower right pixel coordinates of a chosen crop view into a spreadsheet, convert those into lat, lon.

Now add the translation vectors to get lat,lon crop corners for the 34 images and convert them back into pixel coordinates. This gives the specifications necessary to crop them without opening the files. (As Wipneus notes elsewhere, these are sometimes called the offsets.

Open them all in ImageJ rather than Gimp and apply the two contrast options because this will process the whole stack to a uniform standard making full use of 12 bit resolution. Save and open in Gimp for further processing -- 8-bits per channel at this point will not lose information.

In Gimp, view the stack in a fast animation to see if the images are properly aligned. If not, the move tool can nudge them into register with the fixed date.

In choosing the crop box, it is a very good idea to include fiducial points. For Jakobshavn Isbrae calving front, a lot of the scene is snow-covered moving ice sheet and ice stream. So no reference points there. However, I found two well-separated fixed points in exposed rocks that are locatable in radar as well as visual imagery.

Their lat,lon can be determined with great precision because EarthExplorer zooms down to Digital Globe imagery of much higher resolution than LandSat-8 or TerraSar. The first picture below shows the two features as they appear in a variety of imagery along with assorted scale conversion factors.

Recall Cresis DEM files come in lon,lat,z format. To view depth slices along a latitudinal gradient, these have to be located on imagery via pixel coordinates. Or vice versa, we would like to look at bedrock depth around the calving front, sills and troughs and need to find our location within the DEM. In the path up the main icestream channel used by Joaghin, those coordinates are provided over the TerraSar peak retreat image of 20 Sep 13 in the upper right corner of the second image.

The DEM data only allows vertical or horizontal depth profiles and sometimes these cut obliquely across the icestream, distorting the profile. Thus the image also shows the angles needed to rotate the latitudinal depth profile to be orthogonal to flowlines of the calving front, the two troughs and the first still (the second is well served by a vertical slice of the DEM).

Here picture two vertical profiles flanking the oblique cut (ie for which the cut is the diagonal). To interpolate the DEM onto the diagonal cut, simply weight the z values  of the two flanking vertical profiles point by point in accordance to distance.








jimbenison

  • New ice
  • Posts: 13
    • View Profile
  • Liked: 0
  • Likes Given: 0
Re: Co-registering LandSat with bedrock depth
« Reply #10 on: July 02, 2014, 03:35:26 AM »
A-Team, I like where you are going here. I have also contemplated doing something like this.

I'm thinking we should convert the dataset into geojson or import it into a PostGIS database. Openlayers could be used for the front end. Then we could use a web framework like Django or Symfony2 to create a REST server for getting the data to the front end. That way it could be fully interactive.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #11 on: July 02, 2014, 07:13:33 PM »
Jim, that would be great contribution if you could figure out how to dish out something more interactive in terms of raster GIS layers... this glacier is its own news site the way imagery and flight line keep rolling in.

http://boundlessgeo.com/solutions/solutions-software/postgis/
http://openlayers.org/
https://en.wikipedia.org/wiki/Django_%28web_framework%29
https://en.wikipedia.org/wiki/Representational_state_transfer

A couple more posts and I'll have this particular example built out starting with adam and eve. To de-mystify the process.

We've got quite a few people posting raster base layers (photo time series animations) but only a few (Wipneus) doing actual GIS analysis on the stacks (ice thickness DEM).

Photoshop/Gimp/ImageJ/ImageMagick are actually GIS software in disguise, glitzy visual front ends for an underlying spreadsheet (one pixel per cell, one color per sheet) so you could ditch them and do it all in Excel or by command-line on coupled arrays. Including extruding a grayscale as DEM heights in perspective view.

Dedicated GIS software today has unbelievable capabilities. That's great in terms of producing eye-popping illustrations but problematic when limitations in the raw experimental data are forgotten. That's what I'm checking here: the channel is 5 km wide but how many radar bedrock soundings do we have over that 5 km?

I looking at the beyond-fantastic http://earth.nullschool.net/ to see if it could be tweaked to show glacier flow lines instead of weather over wind.

Actually David Podrasky already did a nice time series of speed, slope, and solar aspect for the Jakobshavn icestream. http://glaciers.gi.alaska.edu/pubs/theses And Cindy Starr over at NASA Visualizations did a low resolution version of nullschool gremlins for both all-Greenland and just Jakobshaven flows. http://svs.gsfc.nasa.gov/cgi-bin/details.cgi?aid=3374

This amounts to first computing the gradient (downhill, direction of steepest descent) of a high resolution digital elevation map and then drawing the flowlines.

Contours and flowlines then provide locally orthogonal coordinates for the surface (or for that matter, bedrock under the ice) respectful of the gravity that ultimately drives ice sheet flow and meltwater drainage.

Crevasse fields being the tangents to contours, we would like to isolate them directly from the imagery (via wavelet or fourier decomposition) because that's the test of whether computed flow lines are really consistent with observation.

We need to produce catchy, intuitive illustrations to communicate global warming to the world outside this blog -- and somehow retain scientific accuracy.

Looking at Podrasky's overlays, I would say logarithmic color (for speed) is hard for the average citizen to grok and circularly permuted color (for slope angle) is not immediately intuitive either. To get the photo-realistic effect, the DEM is shown as shaded relief. The final images have flattened the layers but fortunately they can be resurrected by decomposition in hue, saturation, grayscale (RGB --> HSV color space).

I did that for surface glacier speed over the Jakobshavn gorge, contouring by grayscale select and coloring with a recent Wipneus palette that distinguishes 'adjacent' colors rather than more 'logical' spectral stepping. This came out quite well if you believe the ice stream slows from wall friction along its edges.









Espen

  • Young ice
  • Posts: 3251
    • View Profile
  • Liked: 177
  • Likes Given: 0
Re: Co-registering LandSat with bedrock depth
« Reply #12 on: July 02, 2014, 10:21:00 PM »
A-Team pretty interesting and fascinating stuff you are digging into. Keep up the good work ;)
Have a ice day!

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #13 on: July 03, 2014, 06:23:41 PM »
Thanks! Now if only I could wrap up all the loose ends above.

I did a couple more versions of surface ice speed contours in the Jakobshavn Isbrae drainage. It's probably ok to muddle the distinction between speed and velocity because the direction of motion is clear (downhill).

The first image is regional. It shows that motion is dramatically faster -- but not uniformly -- in the main icestream channel. The ice is moving fastest along the central flowline and the speed falls off towards the walls. This region markedly accelerates as it nears the calving front.

The north branch is also moving but not as rapidly. It drains a much more limited iceshed. Note too a 'middle' branch that we see on 30 m imagery as an icefall. It has a larger drainage than the north branch and may speed up its discharge when the calving front retreats out of its way.

The second image shows more detail near the calving front and big bend. I've faded this in to help co-locate the underlying icestream. From this, you could figure out what ice will arrive at the calving front a year from now (remembering that this particular DEM dates from 2008).

That is, the velocity vectors have to be parallel along any straightaway or the ice would become impossibly jumbled. Integrating instantaneous velocity as a function of position yields the time to arrive at the calving front.

Ice arriving a year from now does not correspond to a channel transect at all but rather is bent like the contours. It is rather reminiscent of Hawaiian lava tube flows to the ocean.

These contour maps are easily made by overlaying a grayscale gradient, picking a gray, extending it over the whole image, and replacing it with some palette off the internet. The ideal palette (haven't located!) makes sensible use of color, both in distinguishable but related incremental local steps and in an overall fade in saturation towards pastel.




A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #14 on: July 03, 2014, 06:28:34 PM »
For some reason, the animation above is not working today -- preview here does not include images. Have to click on it, works fine. Here is the final frame:

Espen

  • Young ice
  • Posts: 3251
    • View Profile
  • Liked: 177
  • Likes Given: 0
Re: Co-registering LandSat with bedrock depth
« Reply #15 on: July 03, 2014, 06:39:06 PM »
Very impressive A-Team!!!!!!!!!!!! ;)

To get those "over"sized animation please include:

Please click on the image for animation!
Have a ice day!

SteveMDFP

  • Nilas ice
  • Posts: 1463
    • View Profile
  • Liked: 196
  • Likes Given: 19
Re: Co-registering LandSat with bedrock depth
« Reply #16 on: July 03, 2014, 11:27:17 PM »
Very impressive A-Team!!!!!!!!!!!! ;)
 

Seconded.  Among the best of our citizen-scientist work presented here.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #17 on: July 06, 2014, 05:09:45 PM »
Here is another contour map, this time bedrock elevations rather than icestream speed. As you can see, the 'Bristol' map had really crummy resolution and could not produce a good outcome. I then tried Bamber 2013 but they only made public a badly colored 100x100 pixel postage stamp.

Today I wrote JPL to see if they could release their newest tomographic DEM. That is described in an abstract from last July: http://trs-new.jpl.nasa.gov/dspace/bitstream/2014/44242/1/13-0268_A1b.pdf

It sounds like very satisfactory resolution despite some radar return voids. These arise from surface and shallow sub-surface pockets saturated with water. Highly reflective, they greatly reduce energy reaching and reflected back from the bedrock surface.

If the resolution is 100x80 km at 50,50,10 meters resolution, that amounts to 2000 x 2000 data points. Since the z value ranges from 500 to -1500 m, it has 200 possible values at 10 m resolution. So the 255 possible values of 8 bit grayscale seems adequate. So I just asked for that as a tif or png image.

The main channel is ~ 5 km wide north to south so there would be 100 data points in a transect, with adjacent transects separated by 50m. That would be sufficient to make a nice colored contour map of any under-ice bumps, holes, ridges, sills and troughs.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #18 on: July 07, 2014, 05:29:26 PM »
I've been wondering now about the total kinetic energy of this icestream, the ice sheet that feeds it, and indeed that of all of Greenland's ice. And how to display the energy density over a map of the calving front.

Surely this conversion of gravitational potential energy to motion will all end up somewhere as heat. Will this heat provide a runaway feedback loop (melting or making temperate ice at the interface with bedrock, further accelerating motion) or 'just' go into warming Baffin Bay as it calves?

If the icestream has speeded up 40% in the last few years, then its kinetic energy has doubled as it goes as the square of velocity, 0.5m(1.4v)*(1.4v) = 2(0.5mv*v).

The glacier is moving rather slowly but there is a lot of it. Suppose a cubic meter of ice at the surface is moving at 10 km per year. That is a mass of 916.7 kg moving at 0.00032 m/s which is 94 µjoules.

This cubic meter of surface ice is sitting on top of a column of ice moving with it. On average, the depth of the ice stream is 1167 m below sea level and 402 m above for a total of 1569 meters (calculated from fig.3 of Joaghin 2014 by counting pixels, see figure). So this column has 147 mjoules of energy. Suppose the calving front is 5 km wide. The kinetic energy carried by this 1x5000 m vertical slice is then 735 joules. Enlarging this to include ice up to a kilometer upstream gives 7.3 x 10^5.

It takes 21,080 joules to bring a kilogram of ice at -10 C to 0 C and a further 334,000 joules to melt it to 0 C water, so 3.5 10^5 joules.

Thus the kinetic energy of this giant block of ice converted to heat is barely enough to melt two kg (liter) of cold ice. Even if some gross mistake was made in the above calculation, the outcome will still be trivial. The explanation: the ice is moving very slowly so when squared, an exceedingly small number overwhelms the enormous mass term.

The good news is the velocity map above which is logarithmic in its scale, can double as the energy density map (for what little that is worth). That is, a given color corresponds to a fixed velocity and so to a fixed velocity squared which is consistently proportional to pixel kinetic energy (approximating thickness as constant). So all that is needed is a second very different scale.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #19 on: July 08, 2014, 04:56:32 PM »
Here is my initial foray into displaying the Jakobshavn Isbrae bedrock as rendered 3D. I am using PovRay 3.7, a free ray tracing program that is capable (if the operator knows what they're doing, not applicable here) of very exquisite imagery. The most common scientific applications of PovRay is to crystallographic structures of proteins (see PDB http://www.rcsb.org/pdb/home/home.do) or to NASA-type visualizations.

It has a conventional interface but as a practical matter is driven by easily edited scripts. However it takes quite a bit of experimentation to get the parameters to do what you want.

Below, I used a grayscale DEM and eventually got an instructive display though it is upside-down, ie looking upwards at a mold of topography. This wasn't what I intended but it does bring out the troughs and sills that lie just upstream of the calving front. The scale <26,7,26> means 7x vertical exaggeration.

Here are some tutorials on drawing these heightmaps:
http://www.povray.org/documentation/view/3.7.0/279/
http://homepages.xnet.co.nz/~calculus/HFTUT/HF101SizeMatters.html


camera{location <-7, 12, -7> look_at <0,0,2>  translate <-6,-1,0>}
light_source {<70, 80, -40> rgb 1}
light_source {<150, 50, 40> rgb 0.7 shadowless}

#declare selectSize = 1;
#switch(selectSize)
  #case(1)
    #declare SourceImage = "jakobshavn_bedmap-HSV.png";
  #break
  #case(2)
    #declare SourceImage = "jakobshavn_bedmap-HSVinvert.png";
  #break
#end 

height_field {
  png SourceImage
  smooth
  translate -0.5
  scale <26,7,26> 
  pigment{color rgb <1,1,1> }
}

Espen

  • Young ice
  • Posts: 3251
    • View Profile
  • Liked: 177
  • Likes Given: 0
Re: Co-registering LandSat with bedrock depth
« Reply #20 on: July 08, 2014, 05:26:31 PM »
Brilliant work A-Team, now you just have to turn it, right! ;)
Have a ice day!

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #21 on: July 08, 2014, 07:44:32 PM »
Ok, that can be done by inverting the grayscale in Gimp so lows <-->highs. But, monkey at the keyboard again, here is a view looking west up past the calving front, 4x vertical exaggeration.

TerryM

  • First-year ice
  • Posts: 5434
    • View Profile
  • Liked: 531
  • Likes Given: 5
Re: Co-registering LandSat with bedrock depth
« Reply #22 on: July 08, 2014, 08:48:48 PM »
A-Team that's A-Mazing


Is it possible to extend these techniques to other glacial valleys?


Terry

Espen

  • Young ice
  • Posts: 3251
    • View Profile
  • Liked: 177
  • Likes Given: 0
Re: Co-registering LandSat with bedrock depth
« Reply #23 on: July 08, 2014, 08:55:30 PM »
A-Team that's A-Mazing


Is it possible to extend these techniques to other glacial valleys?


Terry

By the way A-Team and Terry, I do have a list? ;)
Have a ice day!

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #24 on: July 09, 2014, 01:16:30 AM »
Here someone would need to chase down and post a DEM for each glacial valley of interest before I could process them. A DEM consists of a grayscale images in which the tone of gray indicates the z height above (or below) sea level for a region than includes the valley.

These are often found inside journal articles or their supplements. Often you would need to use something like http://www.extractpdf.com/ to get the best resolution the authors had. I think the minimum size should be something like w,h = 400x300 pixels as postage stamp size just will not render details well

It is not good practice just to take a screenshot of a pdf as this will likely be crudely upsampled in hardware. In some cases, I can decompose a colored DEM in HSV space to find the underlying elevation grayscale.

I was also looking for a good map of Arctic Ocean ice thickness (wipneus makes these, but where). If the map is not 'too busy', it may be possible to display that thickness directly and do away with the color scale.

Here is one that contours glacier velocity. Promising, not there yet. So far I can tint uniformly by any RGB but haven't come across a scheme yet that really lays on a height color scheme. (Note to self: maybe actually read height map tutorials.)

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #25 on: July 09, 2014, 05:59:28 PM »
Ok ... more or less have elevation color scale figured out ... a little garish but can fix shortly. Povray scene descriptor looks like this:

include "colors.inc"
  camera{
    location <-25,18,-2>
    look_at 0
    angle 35
  }
  light_source{ <1000,1000,-1000> White }
  height_field {
    png "jakobshavn_bedmap-HSVinvert.png"
    smooth
    pigment { White }
    translate <-.5, -.5, -.5>
    scale <17, 2, 17>
      texture{pigment{
                     gradient y
                     color_map {
   [ 0.00000 rgb < 1.00000, 0.01111, 0.05829> ]
   [ 0.10000 rgb < 0.99457, 0.43569, 0.12356> ]
   [ 0.20000 rgb < 0.99457, 0.65239, 0.12521> ]
   [ 0.30000 rgb < 0.99457, 0.87889, 0.06625> ]
   [ 0.40000 rgb < 0.83409, 0.99457, 0.02736> ]
   [ 0.50000 rgb < 0.38010, 0.99457, 0.07204> ]
   [ 0.60000 rgb < 0.05287, 0.99457, 0.64586> ]
   [ 0.70000 rgb < 0.04840, 0.95938, 0.99457> ]
   [ 0.80000 rgb < 0.09890, 0.49815, 0.99457> ]
   [ 0.90000 rgb < 0.16623, 0.08811, 0.99457> ]
   [ 1.00000 rgb < 0.47403, 0.02386, 0.99457> ]
} }}  }

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #26 on: July 13, 2014, 04:11:05 AM »
If the kinetic energy of the glacier seem low, as estimated above, how much 'should' there have been? Here mgh is the gravitational potential energy that would have been acquired by the icestream during its 'fall' from interior to coast.

Here m, the mass of the glacier is a very large number (enough islandwide to displace the viscoelastic mantle), g the gravitational constant at the surface of the earth is not small at 9.8 m/sec^2, and the height h is substantial for a 3.2% grade over 100 km (3200 meters).

For comparison the Grand Canyon, a serious whitewater river, has an average gradient of 7.0% over 386 km http://www.durangobill.com/GCriverProfile.html This implies Jakobshavn Isbrae would be quite a fast-flowing river were it liquid water.

Recalling the definition of joule, under mgh a cubic meter (917 kg) of ice falling a km amounts to 917 x 9.8 x 1000 = 9x10^6 joules. Thus a meter-wide slice of ice 1400 m thick and 5000 m across making its way downstream to the calving front acquires (9x10^6)(7 x 10^6) = 6.3 x 10^13 joules.

However the speed of glacier is so slow in meters per second that very little of this arrives in the form of kinetic energy. The slice has lost almost all this energy in friction with the bedrock and walls (and internally, as deformation).

The lost energy could go into heating ice or cold rock at their interface (recalling the reverse slope on much of the channel) with the long time frame allowing (per heat equation) for ample dissipation and conductive loss.

There is very likely some liquid water at the bottom despite a large latent heat of transition. However this may originate from surface meltwater. Pressures are enormous so the phase diagram needs to be considered, along with freeze-up and the existence of an ice-water state distinct from either ice or water seen in drill cores.

In summary, potential energy considerations alone will not go very far towards understanding the state of the ice and how that might account for glacier motion (and especially recent speedup). For that we are largely dependent on the interpretation of ice-penetrating radar profiles and modeling of surface observables.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #27 on: July 13, 2014, 03:57:47 PM »
Continuing, if that slice of ice took 100 years to move 100 km, that represents 100*365*24*60*60 = 3.2 x 10^9 seconds, the divisor necessary to convert the 6.3 x 10^13 joules above to watts (energy to power, or rate of energy delivered), here 20 kw. Not a lot of light bulbs shining on a slice of ice thick enough to fill the Grand Canyon.

Had all the potential energy been converted to kinetic, then mgh =  0.5 mv^2 implies a glacier velocity today of v = sq rt (2gh)= 1.4*9.8*3200, a very respectable 0.177 km/sec (396 miles/hr) compared to the maximal velocity observed at the calving front of 17 km/year (0.0005 m/s), a ratio of 354:1.

The earth's geothermal gradient (from radioactive decay) provides another very substantial source of energy, especially given the steepness of this gradient in view of overdeepening to a depth of 1500 m below sea level by a hundred thousand years of previous glacier grinding of bedrock. (The actual history of the Jakobshavn channel has not been reconstructed outside o Holocene moraine dating).

The time-invariant geothermal gradient and fixed boundary condition provided by the ice above set the stage for a steady-state solution of the heat equation, further constrained by the observed temperature profile in drill holes (none of these reached bedrock in the channel itself).

The drainage of this glacier acts a solar collector. Sunlight warming the ice surface has little direct influence on deep rheology; it is drainage of melt lakes to subglacial channels that puts this energy where it has, arguably, potential to create more 'temperate' ice lubricating the bed and so speed up glacier deformation and discharge.

Two AGU posters take a more serious look at heating issues. Dating from 2010-11, these have not yet emerged (for unknown reasons) as peer-reviewed journal papers.

The latter poster asks if softening ice at the channel side walls influences glacier speed, concluding that ice shear could be enough to raise to raise temperatures by 9º C. The stress field is deduced from surface velocity transects and assumption of a weak bed unable to absorb basal shear.

While ice rheology is indeed significantly affected by temperature, the putative 9º C change needs evaluation in a full-fledged model to see if it is enough to account (or even over-account) for observed icestream velocity changes. The speedup is sudden and recent suggesting a threshold being crossed, yet marginal wall warming effects are gradual and continuous.

Rather than a threshold, a run-away positive feedback could provide an explanation. The first poster considers bed friction leading to more temperate ice (at the melting point) on the bed, leading to more deformation, more speedup and yet more temperate ice. The poster concludes that while this is definitely going on, it can only account for 1% of observed speedup.

http://students.washington.edu/kpoinar/poinarCV_2014.pdf

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #28 on: July 15, 2014, 06:48:58 PM »
Here is the 15 m panchromatic Landsat image with an accurate grid overlay on 11 Jul 14 (LC80812332014192LGN00_B8, available at EarthExplorer). The Jakobshavn forum explains how fixed reference rocks and very high resolution imagery are used to anchor and scale the grid.

For digital latitude, longitude over a small area it is more convenient to use milli-degrees (mD). In those terms, kittycorner latitude changes are 10 mD while those of longitude are 20 mD. By counting over from the labelled reference cell, the coordinates of other points are easily determined without obscuring the image with numberings.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #29 on: July 17, 2014, 03:46:24 PM »
The EarthExplorer interface to Landsat-8 images presents its preview images in polar stereographic coordinates north of 63º latitude (ie for Jakobshavn Isbrae). However the main GeoTiff download package has converted the images into Universal Transverse Mercator projection using zone 22 W and WGS 84 for earth shape and cubic convolution.

In the meta data file, the four corners of the image are provided in both systems for example:

CORNER_UL_LAT_PRODUCT......................70.81070......(latitude of upper left corner)
CORNER_LR_PROJECTION_X_PRODUCT......739800.000...(easting of lower right corner)

Note the resolution of the images in Band 8 panchromatic is 15 m yet the northing and easting UTM data is provided to the nearest millimeter. This is puzzling -- a factor of 15000 too much precision -- when the dimensions of each pixel represents 15 x 15 m on the ground.

Fixed points on the rocks to the immediate SW of the calving front are important in co-registering data that often arrives without a scale. For this, a grid of latitude and longitude points can be very precisely placed on the high resolution Digiglobe image used at Google Earth via the coordinate converting utilitiy http://www.latlong.net/lat-long-utm.html.

Note incremental bumps in lat,lon do not correspond to integral pixel value differences whereas this holds for UTM coordinates. It is more convenient to resample the 15 m imagery to 10 m as then 100 pixels amounts to 1 km (rather than 66.6667. Landsat-8 images can be usefully upsampled to 7.5 m in the Jakobshavn calving region.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #30 on: July 22, 2014, 08:12:43 PM »
To follow up on ice thickness and bedrock depth from Jakobshavn Isbrae east to the top of its drainage basin, I've posted below the maps in Bamber et al 2013.

This is a much-cited but scientifically unacceptable article that reflects very poorly on The Cryosphere -- a huge amount of important work accompanied by utterly incompetent graphics (Greenland upside down, lines and text obscuring crucial data, preposterously coarse resolutions, browser-breaker supplemental pdf) and above all no link to the data justifying figures and conclusions. For that you have to beg an author ("the complete set of grids, metadata and documentation are available in netcdf and geotiff format from the lead author (JLB)").
 
It's useful however for seeing how sparse Ice-penetrating radar tracks are in upper Greenland east of Jakobshavn relative to measurement error in depth to bedrock. Even if the two best tracks could be located within the voluminous Cresis data set, no interpolative technique (eg Morlighem 2014) could make a dent in the spacing. 

The volume of ice in the basin amounts to potential contribution to sea level rise (overlooking continuing snowfall etc). But what exactly is meant by the JI drainage basin and how is it determined?

If defined as surface flow that winds up exiting the fjord, those flow line boundaries are difficult to determine from remote sensing in the upper basin because the ice is featureless and moving  slowly. Indeed, there's a huge hole in the SAR map of Joughin 2010; perhaps it's been updated but note velocities are less than a meter per year. Surface drainage may or may not be a satisfactory proxy for unobservable englacial flow.

I took a look at computing the gradient from very precise Spot clinometry but the topography is just too subtle. The icestream cannot be followed up to the summit ridge on Landsat either, even using extreme azimuths and sun elevations for the best shadowing. It all reminds me of the time I came within hours of getting creamed by a huge avalanche low on Mt Rainier that had crossed into another glacier's drainage.

Those surface ice basin boundaries wouldn't necessarily be the same in any case as a moulin-defined (bedrock) drainage basin. That's somewhat hypothetical for now because melt lake drainage to bedrock is a mid-to lower elevation phenomenon. Higher up, it's yet to be explained why water wouldn't just puddle up in the very extensive bedrock region below sea level -- or even drain up the big canyon out Petermann in the far NW.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #31 on: July 22, 2014, 08:19:39 PM »
Grrr, blog software took it upon itself not to display the first graphic at the submitted dimensions 700 x 1370. Some secret rule about too tall, I guess. Below I broke it into two pieces.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #32 on: July 23, 2014, 04:43:11 AM »
It is very common in fluvial geomorphology for one stream to capture a watershed from another  over time and in doing so enlarge its catchment. Sometimes called stream piracy, a drainage can be diverted from one basin to another as the weaker stream is intercepted by headward erosion of the more active one.

Will the glacial counterpart of this occur in west Greenland? The flow lines initially being parallel, as the much more rapid motion of Jakobshavn icestream engages the ice sheet on its sides, lowering by thinning gradually rotates the latter's gradient so that more ice sheet flow enters the Jakobshavn catchment.

If the 2014 Nasa data visualization below is accurate, this is already underway. Estimates of the total ice delivered to the ocean are then too low because the ultimate catchment of Jakobshavn is much larger than usually depicted which will deplete the modeled contribution of slower moving icestreams, for example those in the Eqip area, replacing them with greater sea level rise earlier.

The interior 'reverse topography' of Greenland surely has a role to play. The improved bedrock map of Morlighem 2014 -- though its resolution (already pixellated as published but enlarged still further below) still needs better details -- shows Jakobshavn overall as an open basin without the mountainous atoll barrier of other parts of Greenland.

Indeed if the ice were gone, the sea would flow into the interior basin which is several hundred meters below sea level. After a certain amount of glacial thinning and melt lakes moving inland, it becomes hard to see why moulins would drain to the sea, even as basin buoyancy becomes an issue.
« Last Edit: January 16, 2016, 09:32:03 AM by A-Team »

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #33 on: July 23, 2014, 04:52:32 PM »
Here is a more recent determination of Greenland ice velocites (Rignot 2012 http://onlinelibrary.wiley.com/doi/10.1029/2012GL051634/pdf), that integrates data from  Envisat Advanced Synthetic-Aperture Radar (ASAR), the Advanced Land Observation System (ALOS)‚  Phase-Array L-band SAR (PALSAR) and RADARSAT-1 SAR. That map still has many glitches along the ridge.

After some healing of holes in Fig.2 with nearby pixels and cubic upsampling to slightly smooth, velocities upslope of Jakobshavn can be contoured. The animation below shows these marching up to the divide. Note isovelocity 'lines' can be patchy where little change occurs. That is due in equal parts to imperfections in velocity determination, to the color scale not defaulting uniformly to grayscale, and choice of color picking radius in Gimp.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #34 on: July 23, 2014, 05:02:04 PM »
Nope, not animating unless you click on image. And then it takes a cycle or two before it gets up to 100 ms per frame design speed, depending on your internet connection and caching. I must say the secret programming rules of this blogging software are rather opaque. This file is only 585 pixels wide, nowhere near their secret boundary of 700. It's difficult to figure out the rules when the rules do not permit previewing (unlike the ASIF which does allow previewing of attachments).

TerryM

  • First-year ice
  • Posts: 5434
    • View Profile
  • Liked: 531
  • Likes Given: 5
Re: Co-registering LandSat with bedrock depth
« Reply #35 on: July 23, 2014, 08:25:39 PM »
A-Team
Wonderful work as usual!


In Northern Quebec I was fascinated by large, rocky formations left after the ice sheets had melted away. These drumlin shaped forms in granite were in some cases very large and I think they were carved during periods of rapid melt far beneath the ice surface.
Is it possible that Greenlands topography is undergoing large changes even as we are mapping it. Could studies of regions formerly covered by large ice sheets help in determining how Greenland will be affected by the breakup of the GIS.


Terry

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #36 on: July 24, 2014, 04:51:00 PM »
Excellent suggestion, Terry. The image below shows grooves on an ice-penetrating radar reconstruction of surface bedrock just south of Jakobshavn Isbrae the authors quite plausibly compare to an existing exposed surface feature in northern Canada.

It is not so clear though whether these Greenland grooves are old, still in the process of being made, or a mix of both. The history of the channel overdeepenings is equally opaque (to me). Quite a bit of work has been done on dating Jakobshavn fjord moraines but this is an ice sheet with rather a long history of advances and retreats.

http://onlinelibrary.wiley.com/doi/10.1029/2010GL045519/pdf

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #37 on: July 24, 2014, 05:42:02 PM »
Just to wrap up the Landsat shadow story, some of the scenes are shot at very low sun angles and so cast good shadows (the poor man's lidar) on a landscape with very subtle topographic relief.

There is a choice too of azimuths (--> shadow directions). That is, Landsat-8 is by and large in sun-synchronous orbit with most scenes having the sun 'due south'. Our site has none at 180º and averages 173.7º azimuth and 35.4º sun elevation.

However 5 cloud-free scenes in 2014 were quite different (descending orbit?), having very low sun angles and azimuths of about 45º NW which is -45º ccw of north in Landsat convention, meaning shadows are cast to the SE.

I pulled these exceptional scenes out below, calculating in the last columns how high an object would have to be to cast a 1 pixel long shadow in 15 m and 30 m Landsat resolution. The highlighted example shows the 19 July sun would cast a 7.7 m shadow from a 1 m stick.

However at 15 m ground resolution, it takes a 1.9 m stick to cast a 1 pixel shadow which would be tough to pick out visually, though a short ridge, hump or depression of that relative height wouldn't escape the attention of image processing software.

The ice is not pristine so semi-exposed rocks, debris carried by the ice stream, melt features, ablated ice and so forth give darker pixels that are not shadows. Those might be sorted out using a second scene with higher sun elevation because these will still be dark unlike true shadows.

Except it might have snowed or ablated between scenes --  18 cloud-free scenes as of day 206 amounts to an 11 day average gap.

Jim Hunt

  • Young ice
  • Posts: 4289
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 274
  • Likes Given: 27
Re: Co-registering LandSat with bedrock depth
« Reply #38 on: August 06, 2014, 03:36:09 PM »
My apologies for raining on your parade A-Team, but nobody else has stepped up to the plate as yet. Would you mind giving me a quick GIMP tutorial over in the Developers Corner?

Reverse Engineering "Goddard's" Real (un)Scientific Method
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #39 on: August 06, 2014, 11:37:56 PM »
Jim, sure. I am of the opinion scientific graphics are a very potent tool for communicating climate change. Better than straight verbiage. The number of people posting effective graphics on the forum might be 10-12; it would be great if we could bring that up to a couple dozen.

We could profitably offer a lot more exposition on how to use Gimp and ImageJ over at that forum. Wipneus is providing some excellent coverage of the third free option, which is command line, ImageMagick. PovRay is also very powerful but primarily for advanced lighting effects.

Just thinking out loud, what is going to be useful? Already when a gimp command is googled, the official manual and 4-5 youtube tutorials pop up. So it might be better instead to start with a few specific product types people want for the forums and work backwards to the steps needed to get there.

I'll have to poke around the forums to see what those are, plus see what people request. All the tools are frustrating at first, no getting around that.

Gimp especially needs an enforced six month moratorium on new advanced features until they can correct the hundreds of interface violations and long-standing bugs. ImageJ is clean and minimalist, just don't expect undo.

One common forum use is clearly sat photo acquisition, contrast improvement, rescaling to the same size and projections, rescaling to fit in the blog window, and a time series animation that keeps file size within reasonable bounds.

Another is bagging and repackaging graphics from journals. These are often in front of their paywalls and reproducible by us under Fair Use especially when value is added but have various issues such as postage stamp size (resp. wall hanging), legends that don't track with pictures, tons of wasted white space, lack of subject integration across journals, gratuitous overlaid text, lossy compression and so on.

I made quite an interesting check list this morning, unfortunately while taking a walk, of a dozen things *not* to do in map design -- things I recall covered in high school geography but that I see every day in peer-reviewed
Arctic science papers.

While a handful of researchers are producing spectacular, sophisticated graphics, the rest appear to have bought expensive software on a grant and never found time for its learning curve. The focus is entirely on getting something past peer review; graphics, no matter how bad, are not show-stoppers for that.

Short term, I will try to add a small methods paragraph at the end of each post if any noteworthy processing steps were involved. As easy byproduct, I might take notes and intermediate shots as I go and save that separately as instructional tutorial for the Dev Corner forum.



A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #40 on: August 13, 2014, 11:34:08 PM »
Ever wonder what was going on down below on that long wide dark patch east of the rock zone north and south of Jakobshavn that is visible in every Landsat image?

It's an isochron of local Holocene dust derived from local tundra, not Icelandic volcanoes or Asian deserts. It has been the subject of 4 incredibly thorough recent papers by Wientjes et al (3 free online).

The dust has low albedo, it warms, melts the ice below, attracts cyanobacteria and becomes known as cryoconite. The dust has emerged as ice moved east from the summit area and entered the ablation region where it was left, consolidated from many layers of melting or wind-removed ice.

An explanation for the dark region in the western melt zone of the Greenland ice sheet
www.the-cryosphere.net/4/261/2010

Dust from the dark region in the western ablation zone of the Greenland ice sheet
www.the-cryosphere.net/5/589/2011/

A study of the dark region in the western ablation zone of the Greenland ice sheet (2011)Open access
http://dspace.library.uu.nl/handle/1874/210558

Carbonaceous particles reveal that Late Holocene dust causes the dark region in the western ablation zone of the Greenland ice sheet among others
http://dx.doi.org/10.3189/2012JoG11J165 August 1, 2012

The image below shows the wondrous effect of a very special filter they used to decorrelates RGB layers. That is, in imagery like Landsat bands over an ice sheet, if you had the R layer, the G and B don't come as too much of a surprise. The filter, developed primarily for enhancing petroglyphs, seems available only as an ImageJ plugin (for which you have to write Jon the coder). On Landsat, with bands 1-6, an alternative is RGB from the principal component analysis which also takes out correlated pixels.

http://www.dstretch.com/

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #41 on: August 15, 2014, 01:26:27 AM »
Landsat server is back -- with two clouded over shots from 13 Aug. LC80832332014223LGN00, the low angle one I wanted, was evidently too low angle to bother with. Next up: 20 Aug.

Last year, there were 13 useful Landsat scenes between today's date and end of season 29 Oct but 8 of those were subsequent to peak calving front retreat. Let's hope that Sentinel kicks in with regular repeat visits so we don't miss that.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #42 on: August 19, 2014, 05:03:37 PM »
After a long snooze (21 years), Greenland researchers have embarked on analyzing all the ice-penetrating radar data beyond just ice thickness bedrock topography and ice thickness. There is a tremendous amount of information about post-Eemian rates of snow accumulation and past, present and future ice sheet dynamics in the isochron radar horizons if only those could be digitized and interpreted.

Easier said than done give 394,000,000 km of tracks from 1993-2012. However the Cresis group has been able to  determine all the regions containing at least 25 not too badly deformed isochronous reflectors: over half the flight path data points had them at 1000-3000 m depth plus more in cold older ice. 

Isochronous information in a Greenland ice sheet radio echo sounding data set
Louise C. Sime et al Geophysical Research Letters DOI: 10.1002/2013GL057928 (paywall)
http://tinyurl.com/o9dlw25 (free supplemental)

In the slow slide show below, the first one summarizes the data set. Subsequent image pairs show the outcome of automatic image analysis from 2010-12 fights. Here green indicates no suitable reflectors, blue so-so regions of too few horizons, red regions of too much slope variability, and black the bedrock profile.

I sure hope they retained something better than these clumsy dot texture overlays -- masking should have been done in gimp with a translucency layer because the next step will be automatic tracking of individual horizons on what's left, probably digitizing them with some level of human supplementation in fiducial transects.

In Antarctica, they have a long-established numbering system for radar horizons. I have not seen this for Greenland but now that calendar dates are established and correlated across the five main cores, a sensible way to number radar horizons would be by isochron date b2k because that allows new horizons detected at future higher resolution to be intercalated without upsetting numbering.

In the case the reflector layer arose from a global deposition event of volcanic ash and sulfate on the ice sheet surface, its date will transfer to Antarctica as well.  Although many radar striations don't correspond to anything in the ice core annual layer read-outs, they still allow transfer of those dates to remote regions of the Greenland ice sheet that will never have drilled cores, hence the need for a detailed post-Eemian timeline.

The cores were drilled along the plateau ridge in the first place because the layers are thick (rarely melt) and flat (deformation stress ~ downhill slope ~ 0). Thus a composite flight track connecting up the drill sites is the top priority in establishing an island-wide horizon numbering system.

Lots of these have been flown over the years; no one has compared them for consistency and optimal resolution (though technology improved over the years). It may be certain radar configurations have better resolution say at Holocene depth and others were more tuned to reaching bedrock or determining basal melt status.

Once the primary summit line profile has been established, secondary tracks crossing it can extend horizon dates to the ice sheet margins in favorable cases. It's very clear to me that simple pattern matching can extend many horizon sets past basal topography bumps and bottom freezeup deformations -- the horizons do not really need to be continuous if there is sufficient quality flanking disruptions.Tertiary tracks -- those intersecting secondaries -- then extend the dates to tracks parallel to the primaries and to dense track sets along the coasts.

This still does not result in volumetric filling since the radar sectioning is only on the order of 100 m wide. However it does allow for isochron hypersurface construction by interpolation of the surface/bedrock DEMs. ImageJ is all set up to do this -- horizons correspond to electrophoresis gel lanes of molecular biology -- as well as display them interactively.

These hypersurfaces will have exceedingly efficient analytic descriptions as second order polynomials in two variables which is convenient for re-gridding. Because the Greenland surface DEM is well approximated by a cylinder with north-south axis, we are going to see a lot of downstream cylindrical harmonics (Bessel functions) in the analysis of ice sheet dynamic flow. Separation of variables very much levels the playing field in modeling as main-frame numerical methods won't be needed -- anyone can play.

Give it 10 seconds, then click if it doesn't animate:

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #43 on: August 21, 2014, 06:36:34 AM »
The August 20th Landsat LC80090112014232LGN00 is a total bust so nothing to do but wait another week for the August 27 scene, LC80100112014239LGN00, and hope that is cloud free.

Sentinel radar can see through the clouds but has not stopped by since mid-July, though September may provide 11-day repeat passes in time to catch the time of maximal Jakobshavn Isbrae retreat and give us some velocities (via movement of identifiable features).

Modis at worldview is as always a couple bricks short of a wall in terms of adequate resolution. However if you squint at the 19 Aug 14 scene long enough, it appears that a calving front lobe has appeared in the south branch, at a very similar location to last year's maximum.

Decent bit of interface: a tinyUrl: http://1.usa.gov/1s1MPuC

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #44 on: September 04, 2014, 06:02:14 PM »
Here are various Bouguer gravity maps of Greenland. Although not at the greatest resolution, it is quite fascinating to see what jumps out: Jakobshavn and NEGIS but not so much Petermann but rather a system to its southwest. The top of the map shows bedrock elevation and stream drainages modeled from that. Note the strong (but not perfect) correlation with both gravity lows and bottom freezeups (black markups, top)) reported by Bell 2014.

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #45 on: September 14, 2014, 05:08:03 PM »
Although dramatic interior deformations, below sea level isostatic compression, and past mile-deep overdeepenings get most of the attention, most of interior Greenland is layer cake -- regular radar reflection isochron horizons that can extend unbroken for hundreds of km along flight paths.

Because these carry dates with them from the main summit ice cores where annual layers can be counted to places that will never have cores, it is important to sharpen layer cake regions to the maximum extent that the raw data allow. This is important for tracking horizons across disruptions as well as finding more subtle layers and measuring thinning between layers as horizons approach the coast (as this has the history of Greenland icesheet movement).

Cresis radar imagery can be sharpened by various digital image processing techniques. C Panton has previously looked at gaussian blur noise removing using local isochron slope; blog posts here have looked at wavelet noise removal and conventional high and low band pass Fourier filtration. Emboss and bump mapping enhance radar images perceptually but have not been examined for improvement in the statistical sense.

Radon transform takes another approach to enhancement of layer cake, one with a commonsensical basis: in dimension 2, radon transforms take the ice cross-section into a hyperspace of all possible lines and their associated line integrals of grayscale density. Since the ice horizons in layer cake are very nearly straight lines, they are exceedingly favored when that direction comes to be considered. (There is also a generalization to parabolic lines.)

Consequently noise filters applied to the initial Radon transform should be very effective returning the sinogram with the inverse Radon back transform to a cleaned profile . A number of helpful web pages show how this works with examples but in practical terms, it takes but a moment to install the Radon transform into the ImageJ plugin jar folder and experiment what various filter and setting choices accomplish on layer cake (eg, the vast region east and north of Jakobshavn). So far I've not seen any jaw-dropping improvements but a lot of setting combinations remain unexplored.

http://rsb.info.nih.gov/ij/plugins/radon-transform.html
https://en.wikipedia.org/wiki/Radon_transform

Inferring ice-flow directions from single ice-sheet surface images using the Radon transform
JL Roberts
http://conference.antarctica.gov.au/program/posters-day-2/posters/inferring-ice-flow-directions-from-single-ice-sheet-surface-images-using-the-radon-transform

A Semiautomated Multilayer Picking Algorithm for Ice-Sheet Radar Echograms
VdP Onana
http://gsfcir.gsfc.nasa.gov/authors/publication/228116/

A-Team

  • Nilas ice
  • Posts: 2448
    • View Profile
  • Liked: 301
  • Likes Given: 21
Re: Co-registering LandSat with bedrock depth
« Reply #46 on: September 14, 2014, 05:50:30 PM »
I've seen quite a bit of whining in the Greenland scientific literature about how hard it is to follow radar isochron horizons over long distances of layer cake. Actually it is easy:

Pattern matching can bridge isochrons across gaps, indistinct regions, radar glitches, and deformations. Here ImageJ provides vertical profile graphing averaged over a rectangle: the spacing and intensities of peaks and troughs provide a very strong signature that allow objective continuation from a dated profile. If optimal matching requires a uniform stretch of one cross section relative to reference, that probably represents thinning away from the summit ridge.

Peak and trough sliders have been used for many decades in dendrochronology to transfer dates from dated cores to new timber.

Even dramatic freezeups can still have enough well-defined striated regions to be dated despite steep slopes, multiple recumbent folding, and complete disconnection with other striated horizons in the image. However, provided the patch within the freezeup is represented elsewhere in the orderly striated part of the radar image (or something tiled to it), the patch can still be reliably dated.

The example below gestures at the procedure in the case of a bedrock induced deformation. ImageJ is asked to provide two profiles and the numeric grayscale values associated with them. These are then aligned (not shown) and the best match located, allowing bridging of the discontinuity at the deformation.