1

Topic: Issues with GCPS Optimization

Hi guys, I am very new to Cosi-Corr software so please bear this in mind. I am a university student currently working on my dissertation. I am looking at velocity changes that may or may not have taken place over the Susitna glacier since the 2002 Denali earthquake. I am using Landsat 7 ETM+ imagery and an ASTER DEM.

So far I have created 20 tie points between my first image and the DEM. I have then converted this to GCPS, however when it comes to the GCPS optimization it asks for a "Ancillary file". I don't have one of these, how do I go about creating one??

Can anyone please help me!

Thanks

Duncan

2

Re: Issues with GCPS Optimization

Dear Duncan,

You re not explaining what you are trying to do. Are you trying to ortho-rectify Landsat? COSI-Corr doesn't deal with Landsat ortho-rectification. Please give us more information.

Sebastien

3

Re: Issues with GCPS Optimization

I am using Landsat imagery and an ASTER DEM together. I am on part 6 of the Cosi-Corr user guide "Tie Points and GCPS". I have used a Landsat 7 ETM+ image, along with ASTER DEM data to link the two with the tie points. This was done successfully, so I moved onto 6.2 "tie points to GCPS". I created the GCPS file fine. However this is where I have run into a stubbling block, as I can't optimize the GCPS (part 7) as I don't have an Ancillary file.

Judging from what you're saying above Sebastien this isn't possible either as Cosi-Corr can not optimize these points?? I know that the Landsat iamgery I downloaded has had some form of orthorectification done on it, however my dissertation supervisor said to use Cosi-Corr because I need to makew sure the images fit perfectly over each other with exact precision.

1. Can I even use Landsat imagery to look at displacement measurements with Cosi-Corr?

2. If yes, what do I have to do in order to get the desired displacement maps?

I hope you all can understand what I am writing here.

Duncan

4

Re: Issues with GCPS Optimization

Hi Duncan,

Landsat imagery is already ortho-rectified when provided, so there's no way you can re-orthorectify a badly orthorectified image. Landsat images cannot be ortho-rectified in COSI-Corr. You can try to optimize the orthorectification of a raw ASTER image with respect to an existing Landsat ortho, but not the opposite.

What is your application, what are you trying to measure? Landsat imagery can be used for measuring ground displacement but uncertainties are usually large. Hence it only makes sense to use Landsat for ground displacement measurement if the displacement to estimate is also large.


Sebastien

5

Re: Issues with GCPS Optimization

I am looking at the change in ice velocity through a change in surface albedo (due to the rockfalls that resulted from the Denali fault earthquake in 2002) of the Susitna glacier in Alaska. I am looking at 2 images before the Denali 2002 earthquake, creating a displacement map, and then the same again with two images after the earthquake. I want to then see how much (if any) change in velocity have taken place.

So what do I need to do to get the information I require? Do I simply input the images for correlation and displacement measurements and skip the other section on tie points, GCPS, orthorectification etc etc?

Duncan

6

Re: Issues with GCPS Optimization

Hi Duncan,

If you want to look at surface displacement from a change in surface albedo, what you're doing is probably photoclinometry. For this, you'll only access the glacier slope variation in the axis aligned with the Sun incidence, and assuming the surface didn't change between acquisitions. I'm skeptical this method can work in your case.

The ideal situation to measure glacier displacement is to have two-pairs of stereo imagery acquired before the Denali EQ, and 2 pairs of stereo acquired after the Denali EQ. If you're looking at velocity variations, you need to create a time series before 2002 and after 2002. Unfortunately, the glaciers in Alaska have sustained a large amount of thinning over the last few decades, and DEM errors will bias you measurements if 3D topography changes are not taken into account. You would then ideally need at least 8 images. You could reduce this number if you can get images with very small incidence angles (<1 degree), so that you're not sensitive to ice thickness change. Unfortunately, no such imagery exists (to my knowledge) over the Denali area.

You can look at this paper, where we incidentally observed glacier velocity:
http://www.tectonics.caltech.edu/slip_h … r_2008.pdf

I've continued the study later with five SPOT images acquired before and after the Denali EQ. I spent a large amount of time on this study and at some point I thought I was able to find a difference in ice velocity due to the EQ. Unfortunately, I realized that my observations were only due to large ice thinning and that horizontal and vertical changes couldn't be disambiguated. I wrote a paper that I never sent for review because I realized that the whole study was flawed. If you're interested, I could send you the paper I wrote on the subject and that I never sent for review because it's all wrong. At least you won't redo the same mistakes. We could discuss it offline.


Sebastien

7

Re: Issues with GCPS Optimization

Thank you Sebastien so much for your reply. The paper you have posted I have actually read and have used in my literature review. I will go over it again though. In terms of your other paper, I would very happy for you to send me that paper that you haven't sent for review as it would give a great understanding into my paper.

I have around 12 sets of Landsat images. 6 from before the earthquake, and 6 from after the earthquake occurred. I wanted to work out the velocity from before the EQ using two images, and then the same with two images afterwards. Is there any possible way of doing this? Even if it is flawed I will still need to come up with some sort of results that I can rip apart in my discussion and conclusion sections.

Where would you suggest I go from here then Sebastien?

Many thanks

Duncan

8

Re: Issues with GCPS Optimization

Hi Ducan,

I'll send you the paper. The most important for your study is to understand the geometry of the images acquisition. Unfortunately, it's not well known for Landsat images...

Sebastien

9

Re: Issues with GCPS Optimization

What do you mean by the geometry exactly Sebastien?

10

Re: Issues with GCPS Optimization

I'm talking about the acquisition geometry, look at Fig.2 of this paper:
http://www.tectonics.caltech.edu/slip_h … SE2008.pdf

Sebastien

11

Re: Issues with GCPS Optimization

Hi Sebastien, I am trying both the stacking method you told me about via email, as well as the COSI-Corr method of adding in the pre-Eq image and post-EQ image and seeing what the results will show. I will then need to do some levels of filtering on each output image, what filtering methods do you recommend I do to give me the best results?

Duncan

12

Re: Issues with GCPS Optimization

Hi Duncan,

I suggested putting al images in the same reference grid, and the stacking tool is a "quick and dirty" way to do it. Once the images are stacked, then correlate the bands appropriately.
You'll then have to use the "discard/replace value tool", and also the destripe tool in manual mode.

Sebastien

13

Re: Issues with GCPS Optimization

Will I need to do some sort of filtering though? I mean after the correlation section? What will the discard and value tool do exactly?

Duncan

14

Re: Issues with GCPS Optimization

Please look at the user's guide to answer this question, it'll help you getting rid of bad correlation points. Regarding filtering, it depends on your results. You'll have to destripe your results as well because Landsat will produce a striping effect. Further filtering will only be relevant if after doing all this your measurement are not biased by the topography, you'll decide when you get there.

Sebastien

15

Re: Issues with GCPS Optimization

Slightly off topic Sebastien, but do you know when roughly the DEM used to orthorectify Landsat 7 ETM+ was obtained? It would be good to know for when it comes to writing my dissertation discussion.

Duncan

16

Re: Issues with GCPS Optimization

Hi Duncan,

No, I have no clue, that's why it'll be difficult to use Landsat in your case. For my study in Alaska that I didn't publish, I used the USGS DEM, that I later realized was off by up to 80m on the glacier elevations. That introduced huge errors, hence erroneous conclusions.

Sebastien

17 (edited by Duncan 2010-04-11 03:42:33)

Re: Issues with GCPS Optimization

Hi Sebastien

I have completed the stacking, now my supervisor is saying I need to use Band Math in ENVI to work out the necessary incidence angle and length to give me the overall velocity of the glacier. Is this necessary or not? If it is, what equation do I need to use in order to work this out correctly in ENVI?

Also, do you know of any other papers that have looked at possible velocity changes across the Alaskan mountain range as a result of the Denali EQ of 2002?

Thanks

18

Re: Issues with GCPS Optimization

Hi Duncan,

After the stacking, you have to so the correlation.

Sebastien

19

Re: Issues with GCPS Optimization

Was that an incomplete post Sebastien? I didn't really understand. I have to do the Band Math? If so what equation do i need to do to find the incidence angle and length?

Duncan

20

Re: Issues with GCPS Optimization

I meant to say that you have to do the correlation between the different images you have stacked to deduce glacier motion. I have no clue what you're advisor is talking about.
Regarding the Landsat incidence angle, it may be available in some image metadata, I don't know, it's you to find out.

Sebastien

21

Re: Issues with GCPS Optimization

Hi Sebastien, do you know know what would consistute a low SNR? I've destriped the imagery and now I need to use the discard and replace value tool. I'm looking over the glacier of interest and the values I'm getting are between -1 and -38 give or take. I'm assuming this is poor SNR?

Many thanks

Duncan

22

Re: Issues with GCPS Optimization

Hi Duncan,

I would expect you to use the replace value tool BEFORE and AFTER destriping. Which correlation parameters did you use? If you use the  frequency correlator, discard all values with SNR < 0.9. Poor SNR occur when the images don't look alike.

Sebastien

23

Re: Issues with GCPS Optimization

Its a great informative paper you shared with us. Interesting to read the paper and it is well explained too.

Thanks for nice sharing.

24

Re: Issues with GCPS Optimization

Hi
This discussion was very useful for me.
I'm using some Landsat images to obtain the surface displacement field of a glacier. I obtained some results that seems good at the first glance but i doubt them!! firstley i registered the slave image relative to master image using Image-to-Image registration tool in Envi and after registration i went to correlation step. the displacement image had a pattern like 3 vertical parallel strips. so i tried to measure the correlation between two landsat images with no registration and the result have no pattern or strips.
i would like to know if i can trust the secound results? or manual image to image coregistration will improve the accuracy of displacement measurement?

Thanks in advance

25

Re: Issues with GCPS Optimization

Hello,

If you're using Landsat imagery, I assume you're not using COSI-Corr for the registration, but only the embedded ENVI functionality. Note that the image-to-image tie point selection is not part of COSI-Corr but simply a link to the same ENVI function. Unfortunately, we often noticed incorrect implementation of ENVI functions, and what you observe might be one example.

The COSI-Corr function is the GCP optimization tool, which will only work using raw imagery and supported sensors.

Cheers,
Sebastien