[feldbruggem]

Hi,

I am currently interested in deformations regarding a slow moving landslide and I am processing different time series, however, during the APS processing I have noticed that when I use the automatically selected master image, the master image has a coherence of 1, but all slave images are low (0.2-0.4) but constant. When checking the data, I noticed that it had rained that day and when I manually change the primary image by a day or two, I get the reverse, with the master image having lower coherence but the secondary images having a high coherence.

To my understanding, low coherence in these plots means high noise, however, when the images show a constant pattern, the software approximates the APS correctly. Is it therefore better to have a high master coherence but constant low coherent slave images rather than having the master image’s coherence high and the slave images low? What exactly is the meaning of the master image being at the highest coherence when the slave images should result in a constant value?

I have attached two plot examples, highlighting the problem.

Thank you for your time and consideration,

Mauritz Feldbrugge

Attachments:

You must be logged in to view attached files.

[periz]

Hi,
here the concept is the following: when you take an image as a reference, if the image is noisy, the noise is propagated to all other images (being the reference image set to zero, since it’s the difference with itself). This is what you see in that plot.
However, Sarproz takes into account this phenomena and applies some corrections. So, changing the reference image at that point should not impact too much the final outcome (but you can check this yourself and share your feedback).
A cleaner job should be choosing an other Master image at the very beginning of the processing (because it could also impact on the coregistration). Notice that Sarproz does not choose the master image on the basis of weather conditions, but it gives you the chance to do so, if you observe such data.
The plots you reported (coherence as a function of time in the APS processing) are instead more useful to check the trend in time of the coherence. You could spot there possible accelerations or seasonal trends. You seem to have in fact a slight trend towards the end of your time series. If I were you I would proceed further in the analysis but keeping an eye on possible non-linearities.
To conclude, notice that the trend is similar in the 2 plots you posted. Just, in the first plot variations are more compressed while in the second one they look amplified. Ups and downs are corresponding in the 2 plots though.
best

[Author]

Posts