1

Topic: General Info

If, in a process, i did all correctly but the final results are not realistic, which could be the cause?

I used two aster image for pre and post event but in one of this there were lots of clouds and the overlapping area was little between the two images. I used an Aster GDEM.

The irrealistic results could be due to noise and the A. GDEM?

Dan

2

Re: General Info

Hi Dan,

Your question is too vague, you should check the result of each sub-process and track the pb. You should also better define and quantify what you call "not realistic" and better describe your observations so that we can help you. You can check that the GCP optimization gives reasonable results before going further in the processing. If the GCP optimization fails because of clouds or poor distribution, select new GCP, or don't use any. Then you have to learn how to filter outliers from the correlation results.

Sebastien

3

Re: General Info

If I had two images and the overlapp area between them is little and the centre of earthquake is in this overla area i have to selecet tie points beetween them in the area where there are the bigger displacements. In this conditions is not real do an analysis? I have to study an eartquake but i have got only these two image with a little area of overlap and the in all overlap area there are big displacements.

Thank you

dan

4

Re: General Info

Hi Dan,

Please try to quantify your description.

How much offset on the ground are you expecting, how wide is the overlap between the 2 images? What is the EQ magnitude? How is the fault oriented? EW, NS, or diagonally? If the fault is EW oriented and the overlap is at least 15-20km and the expected fault offset is more than 3m for the horizontal component, correlation should work well. What is the level of noise in your correlation? What is the convergence of your GCP between the 2 images? How did you ortho-rectify the first image? Did you use GCP? Did you optimize them? what was the convergence and how many did you use? etc...

Sebastien

5

Re: General Info

The fault is oriented W/E. The overlap area of the two images is around the 35% of an image. The convergence of the GCP between the two aster image orthorectified is obteined after 5 loop but is around 3 meters for the standard deviations.

The noise is great especially if i use SRTM DEM. If i use ASTER GDEM the noise is lower but is big.

6

Re: General Info

I will expect a max displacements of six meters. The problem is that the only zone where i can choos gcp is the zone of the centre of the eartquake. I would know if the choose of tie points in the zone of bigger displacements is a big trouble for the analysis or if it can be done!

7

Re: General Info

Hi Dan, I'm not sure I understand your sentence. As a general rule, you shouldn't use tie points in the fault rupture zone. You can try processing the images with, and without any GCP, and compare the results.

Sebastien

8

Re: General Info

I'm sorry for my bad english. Now i will try to explain better my trouble. I know that GCP should not be chosen near the fault and near the area where i will expect biggest displacements but i have to study an earthquake, and about this earthquake i have got only two aster image. The pre-event image has got lots of clouds while the post event image is ok. The overlap area of these image is about 30% of an image. And in the centre of the overlap area there is the earthquake epicentre. Moreover the morphology of the area forces me to choose tie points near the epicentre.

9

Re: General Info

The epicenter area may not be the zone of largest displacement. Try to use GCP. If it doesn't work, don't use any GCP, it may work with ASTER.

Sebastien

10

Re: General Info

Ok, tomorrow i will try this. For try with any GCP i have to not select the GCP file when COSI-Corr sais to me to upload this?

11

Re: General Info

Sorry but... If i don't chose any GCP How can i orthorectified the post event image? This is the problem!

12

Re: General Info

Read the documentation, it's optional.
Sebastien

13

Re: General Info

Yes, i read it but I'm perplexed beacause i don't understand how can the program estimate the transformation coefficients. Tomorrow i will try this.
Many thanks

Dan

14

Re: General Info

From the ancillary data only.

15

Re: General Info

Ok, many thanks.
Tomorrow i try with and without GCP and then i will say the differences.
Thousend thanks

Dan

16

Re: General Info

I did the correlation making what you said but results are not valid. I obeteined the vector (in vector field) all oriented in the same direction but i think that is impossible because there is the fault that goes frome west to est. Also the displacements are big: bigger than the real displacements. I think that the overlap area is too small and i think this is the real trouble of this analysis.

dan

17

Re: General Info

Looking at the vector field is not the way to look at the correlation, you should try to understand the result of each subprocess. You have to check visually that the two ortho-images are well aligned. If this is not the case, it's useless to go any further in the processing.
Then you look at the correlation looking separately on each band, and you try to estimate the noise level of each band. You'll need to threshold the values with low SNR and large unphysical displacements. You may see the ASTER attitude undulations.

When you said that the results are not valid, you should try to quantify them. Talk about the range of the correlation values, their spread, distribution function, etc...

If you didn't use any GCP, you may have a general offset between the 2 images, corrupting the measurements. You may need to be removing some first order polynomial in the correlation.

Sebastien

18

Re: General Info

I understand what you said but it's to difficoult to me. I'm only a student and i use cosi-corr from little time. After many analysis i did well the analysis of the example of the two spot image. Now i'm trying with two aster image. The two orthorectified image seem to me aligned. I considered also the SNR after the analysis and so i have a general idea of what is appening.

I understand also that if i don't use GCP it could be an offset. I also chase to select GCP not near the fault.
I'm sorry, i'm putting the maximum engagement.

19

Re: General Info

Nice information i got from here. Thanks for the best explanations.