1

Topic: Interpreting correlation results

I got the E/W, N/S, and SNR images from correlating ASTER 2001 and 2007 images for a sand dune area. Could you, please, provide some clue to interpret these images? In addition, is it possible to geo-reference the vector field? So, I can conduct some overlay analysis with other datasets.

2

Re: Interpreting correlation results

Hello,

The EW and NS bands correspond to the projection of the 2D displacement field measured, along the EW and NS directions. Values are given in meters, positive in the North and East directions. If at a point you have NS = -10 and EW = 0, it means the ground at this location moved 10m toward the South between the 2 acquisition dates (considering no other bias exists, such as topo bias, undulation from satellite roll/pitch, outliers from decorrelation, etc..), and neglecting the noise...
If you measure EW = 10 and NS = 10, the ground moved by 10*sqrt(2) = 14.14 meters, in the E-NE direction.

When you use the vector field tool, it is advised to first "clean" the correlation data. You can discard most of the outliers by thresholding  the NS and EW bands to remove nonphysically large values, or values with low SNR (in practice, discard values with SNR < 0.9).
You may also need to destripe the correlation data, if undulation from satellite wobbling are present. When you display the vector field, adjust the parameters to obtain a good visualization (scale, smoothing, sampling). You can save the vector field as an ENVI layer (File -> Save as -> ENVI layer). You will be asked for projection parameters. You should input the projection parameters of your correlation file by clicking on "Change Proj...". You will have to select the UTM projection with WGS-84 datum, units in meters, and input the appropriate UTM zone (same as your ortho-images). Then you will be able to display the vector field on top of your ortho-images. When you open your ortho-image in a display, go to Overlay -> Vectors, and select the vector file you just created. Adjust color, etc.. to your liking for better visualization.

You should be able to find similar information in the COSI-Corr manual. Pls refer to ENVI user's guide for non-COSI-Corr functions.

Hope this help,
Best,
Sebastien

3

Re: Interpreting correlation results

Dear Sebastien,
I am having some difficulties in masking SNR band. I need to create a mask identifying all the displacements whose ROI is less than 0.95, and excluded these pixels from the analysis (of N/S and E/W bands). Can you please tell the steps? The masking tool gives binary output.How do I use it to mask the same areas in N/S and E/W bands?

Also, for example, If we measure EW = 5 and NS = 10, what is the displacement?

Regards,
Lydia

4

Re: Interpreting correlation results

Lydia,

You can mask poor correlation values by using the tool "discard/replace values", these values will be replaced by NaN. If you using the ENVI masking function, you have to apply the mask after it's computed. I'll refer you to the ENVI documentation if you're using ENVI functions.

COSI-Corr returns displacement values in meters if your images are georeferenced with UTM projection. EW= 5 means 5m to the East, NS = 10 means 10m to the North.

Sebastien

5 (edited by lydia 2012-05-01 20:56:23)

Re: Interpreting correlation results

Hello all,

When creating North-South Displacement (range: -44 to +43), East-West Displacement (range: -42 to 43) there are linear patterns/lines in the displacement maps. Why is it occurring? I used another image processing software for ortho-rectification and co-registration as I did not have ancillary data to do it in COSI-Corr, but did correlation in COSI-Corr. In addition, the subsequent results for the study, i.e. sand flux is coming very high/incorrect is it because of correlation noise in slow moving dunes? Please advise.

Thanks,
Lydia

6

Re: Interpreting correlation results

Hi Lydia,

Are you still using ASTER images? If so, are you noticing patterns such as in fig 1 of this paper:
http://www.tectonics.caltech.edu/slip_h … SS2007.pdf

If so, you're seeing jitter artifacts from spacecraft unrecorded attitudes. You can try to remove this using the destriping tool in COSI-Corr. More details in this paper too:
http://www.tectonics.caltech.edu/slip_h … SE2008.pdf

If the artifacts are different, please send you data.

Sebastien

7

Re: Interpreting correlation results

Hi Sebastien,

No I am not using ASTER image, but Cartosat-1 image, and I did ortho-rectification in another image processing software since I did not have ancillary data needed for processing in COSI-Corr, but did correlation in COSI-Corr.

The East-West displacement map (range: -42 to 43)is as follows:
http://i48.tinypic.com/osqxaf.jpg

And the North-South Displacement map (range: -44 to +43) as follows:
http://i45.tinypic.com/21o1tu8.jpg

The ortho images used to calculate correlation have no such linear patterns. Pls advise.
Regards.
Lydia

8

Re: Interpreting correlation results

Lydia,

All the errors you observe are due to improper ortho-rectification. The large discontinuity is due to improper tiling of images, and residuals correlated with the topography can be topography residuals due to poor DEM, poor ortho-image/DEM registration, or too large incidence angles of the images.

Sebastien

9

Re: Interpreting correlation results

hello
i am nwe cosi corr and envi user, i am studing aeolian features in mars planet, i have 2 orthorectified hirise and i have realized the displacement map with cosi corr, now i wuold to export this map into a gis project, may you explain in some steps how to arrive to this point?
thank
Marco

10

Re: Interpreting correlation results

Export to geotiff format?

Sebastien

11

Re: Interpreting correlation results

hi Sebastien, as i wrote to you before, i have two hortorectified images, and with cosicorr i have realized the displacement map . now i would like to export this map into a arcgis project, may you help me in this???thanks
Marco

12

Re: Interpreting correlation results

As I said before, export in geotiff.

Sebastien

13

Re: Interpreting correlation results

Dear Sebastien,
I am comparing two ASTER data 1 month apart for glacier velocity mapping. I have got a good orthorectification. The expected velocity is 40-60 m/y.
Problems I am getting are:
1.    Glacier is Oriented South-North. General flow should be in north direction and the at some tributary glacier it should be flow from East and west toward North (giving flow in North east and north west direction). But after velocity field creation I observed that many arrows are pointing in south directions. Can I use a directional filter of bias here to remove such points ? Or can I use only NS velocity field.
2.    I have seen the papers you have mentioned in earlier posts but I am not quite getting the  “Removal of residual attitude effects” please explain the process.
3.    Can we filter the values with low SNR and then interpolate the data to fill the gaps in velocity raster.
4.    What should be ideal correlation window size in my case I have used 16*16 with the step of 2 to compare the ASTER data which is one month apart? And please suggest the window size if I am comparing ASTER data which is one year apart. Velocity of glacier is around 40-60 m (field observation). Size of glacier 9 km width 0.7-2 km, main glacier orientation S-N. Tributary glaciers in east and west. 
Thanks
Reet

14

Re: Interpreting correlation results

Your windows are too small, 16x16 usually don't achieve much. If the glacier moves at 60m/year and the images separated by 1 month, offset will be sub-pixel only, use 32x32 windows. You can also try 32-16, but with the level of noise of ASTER images I'm not optimistic. Attitude effects are wave artifacts from the sensor, please refer to the documentation to see how destripping works.

You should of course remove points with low SNR and erroneous large values. Use the discard value tool.

Yes, it's advised to use a directional filter, not just one component. If one component is wrong, there's no reason the other one should be trusted.

Sebastien