1

Topic: Landsat imagery

Hi guys,

I am a master student working on my dissertation with dynamics of glaciers. I am looking  at velocity vectors from crevasses on glaciers on the Antarctic Peninsula, and for this I am using Landsat TM / ETM + Imagery.
I would like to know if the cosi-corr has input for the my Landsat imagery, because I only find input for other sensors.

If I can’t ortho-rectified my Landsat imagery so can’t I use this in cosi-corr?

Thank you Aline

2

Re: Landsat imagery

Dear Aline,

Landsat images should already be georeferenced when you get them, hence ancillary information is lost and Lansat imagery cannot be ortho-rectified in COSI-Corr. Your only solution is therefore to directly correlate the Landsat images in the georeferenced geometry, but the displacement field may be corrupted by topography errors. However, topography errors may not account for much bias if glaciers flow at a high rate.

If you have pb directly correlating the georeferenced Landsat imagery, use the ENVI tool: Basic Tools -> Stack Layers and resampling using bi-cubic to make sure all images belong to the same footprint.


Sebastien

3

Re: Landsat imagery

Hello

I am trying to use COSI-Corr on some Landsat 7 imagery over Greenland from 2009.  I have a number of images from throughout the year and I was hoping to extract some velocity profiles from some outlet glaciers.  Average ice flow during the summer months varies from 100-400m on a number of different outlets.  I have carried out correlations using Cosi-corr using Frequential and statistical methods and with a range of window sizes.  The problem I have is trying to mask out the black lines in the images from the scan-line correction failure.

I have tried interpolating between the lines before correlation and then later masking out the affected areas and I have tried correlations with the blacklines left in and later masking these out, however, I'm unsure about the results as profiles running up-glacier still have show evidence of large fluctuations, mostly likely as a result of the initial black lines running through the images.

Is there any way I can run the correlation whilst masking out areas affected? or can anyone suggest an alternative method?

Thanks in advance

cheers

Andrew

4

Re: Landsat imagery

Hi Andrew,

COSI-Corr doesn't offer the possibility to mask the Lansdsat SLC errors. These errors can potentially be minimized if you interpolate missing information before correlation, but even that will not be great as interpolation doesn't increase frequency content. I'm afraid that obtaining reliable measurements where scan-line error occurred is unlikely to be successful with COSI-Corr.

Let us know if you mitigate the pb,

Cheers,
Sebastien

5

Re: Landsat imagery

Hello
Previously it was mentioned in a post that Landsat imagery gives good results for correlation results if the surface movement rate is high. I'm analyzing Landsat-5 TM imagery over Gangotri Glacier whose mean slope is 0.097. Is this slope steep enough to support high rates of surface displacements ? Can you please quantify as to how high the rate should be ?

Also, I have calculated the displacements for 2 images for the years sept. 2009 and sept. 2010 using 4 pixels as step size and 32-by-32 as correlation window size. I have also applied masking so that pixels with SNR values less than 0.9 can be neglected.
Then I applied NLM filtering process on the displacement fields (N/S and E/W respectively) and chose the standard deviation to be 1/10th the image pixel size i.e., 3. But regardless of filtering the results did not change drastically. Can you please tell me if my process chain is correct ?

Thanks in advance
Prateek

6

Re: Landsat imagery

The process looks correct. I cannot give you an estimate of the velocity over Gangotri Glacier, you will have to look at the literature on that subject. If you have too much noise, you can also try 64x64 pixel windows for the correlation.

NLM filtering will reduce a bit the noise, but will not transform poo measurements into good ones. If filtering power is no enough, increase the H parameter.

Sebastien

7

Re: Landsat imagery

Hello

Thanks for your help smile but I'm sorry I did not ask about the surface velocity estimate of the glacier. Can you please tell in general how fast should be the surface deformation rate of any event so that it can be correctly estimated using Landsat imagery ?

Thanks again
Prateek

8 (edited by prateek20111989 2013-01-31 09:16:14)

Re: Landsat imagery

Hello Sebastien

If the slope of topography for a glacier (along the main trunk) changes from 20 degrees at the accumulation region to around 5 degrees in ablation region over a length of 30.2 Km. Will we be able to observe topography artifacts in the displacement map generated by correlating 2 Landsat-5 TM images ?

Thanks

prateek

9

Re: Landsat imagery

It's hard to tell because topo effects depend on the variation of elevation, not really the change in slope along the glacier. For sure, Landsat imagery is not the right imagery to look at elevation changes as most images are acquired in near-nadir conditions. If you have significant topo changes, you will most likely be able to differentiate between a horizontal motion, or the effect of the elevation change. To do so, you usually need a combination of several high angle imagery to resolve all 3 component of the displacement field.

Sebastien

10 (edited by Jiangpeng Cui 2013-04-25 03:52:02)

Re: Landsat imagery

Hello,
I correlated Landsat 7 images, Sep. 2002 and Sep. 2003. They result looks good. However, the velocity profile along the central line fluctuated frequently with a velocity of 4 m/a(average velocity 46 m/a, maximum 60 m/a).
What the reason of fluctuation? Is the fluctuation true?
The  correlation error was 3.6 m/a( velocity in the non-glacier area)
PunBB bbcode test

11

Re: Landsat imagery

This type of fluctuation is consistent with the noise level on Landsat imagery.

Sebastien

12

Re: Landsat imagery

Thanks a lot.

13

Re: Landsat imagery

Hi all,

I'm struggling with correlating Landsat 7 imagery on the Getz Ice Shelf in West Antarctica.  I imagine it ought to work fairly well - as ice shelves are more or less flat, orthorectification is not a huge issue.  In addition, the velocities I'm looking at are fairly high - anywhere between 200 and 1200 meters/year along the shelf - so the signal to noise ratio should be decent.  On top of that, I've run correlations on these images using IMCORR, and the vectors look great, so I know the features are trackable.

So far, though, the correlations I've been getting from Cosi-Corr end up looking like nonsense.  Either I get an image that is completely noisy, or I get a blocky image that shows large regions of the same velocity, which does not match the physical characteristics of the region at all.  I can't figure out what I'm doing wrong.  I take two Landsat 7 images or subsets of Landsat 7 images (between 1999 and 2003, so they don't have missing stripes), enter them into the Cosi-Corr correlation window, and choose the frequency correlator.  I've tried both the pan-chromatic band 8 (in hopes of correlating a higher-resolution image) and the NIR band 4.  In the frequency correlator options, I've tried starting with a window size of 512, 256, and 128, and going down to a window size of 8.  I've tried 2 and 3 robust iterations, and I disable the gridded output.  Some of the correlations I've tried kind of vaguely show the general shape of the dominating features of the original image, but the whole thing is still so noisy that it's unusable.  Maybe these images just don't work?

If you have any ideas that could help me make this work, I'd greatly appreciate it!

-Karen-

14

Re: Landsat imagery

Minimum usable window size is 32x32, maybe best to stay with 64x64 at the minimum.

15

Re: Landsat imagery

The time span may be a little large. One or two year apart between the two images will produce good correlation with low noise.

16

Re: Landsat imagery

Thanks for the responses - I've now tried a whole lot more combinations of window sizes, with the minimum no less than 64.  Still no luck - the output image looks bizarre and blocky, with many blocks of values that don't make any sense for the image.  The images are actually only 3 years apart, with very distinct features that I've correlated with no problem in other programs.  Any other ideas?  Are there some images that just don't work for one reason or another?  Or do Cosi-Corr outputs from Landsat just tend to be blocky and not very reliable because they can't be orthorectified within Cosi-Corr?  I've seen Cosi-Corr outputs from WorldView imagery, and since that's such a high resolution, maybe I'm just expecting too much from a Landsat correlation.

Thanks for the help!

17

Re: Landsat imagery

Send me your correlation results,

Sebastien

18

Re: Landsat imagery

Hello
Is it possible to correlate 2 Landsat images belonging to different path and row ? I'm trying this but I'm unable to proceed further. I thought of reprojecting the images into a common geometry but even that did not work neither did georeferencing. Any help will be great. Do we need to reregister them ?
Thank you
Prateek

19

Re: Landsat imagery

Yes it's possible, why the question? Landsat images are already orthorectified.