[bridgetwang]

Dear Prof. Daniele,

I have some questions when using APS for my dataset. I got 83 images from Jan 2017 to Dec 2018 covering a center city.

As the city has many buildings, and the images are dense in 2 years, I assume the deformation will be non-linear, and for buildings there will be thermal deformation. So I use 3 different models in APS trying to see which model gives the best result: 1. linear model 2.nonlinear model 3.a linear model with thermal expansion.
It seems that the result from linear model with thermal expansion gives the best coherence after APS removing. I’ve upload the JPG of my results.

My first question is for my case, should I use the result of the linear model with thermal expansion as my final APS result? I remember you have mentioned before that the model used in APS should better be a simple linear model. And you also mentioned that the coherence will always be high when Inverse Parameter is used , but this may not be a truly good result. So I’m not sure with my results and how to make the decision in my case.

My second question is about the temperature data import. When I do APS estimation, I actually didn’t download temperature data and I just click the “thermal expansion”. Am I doing right or should I import temperature data via “Load Temp” first, then use the “thermal expansion”?

Also, what the difference between using “Weather Module” and “Load Temp”? Since the weather data can’t be download without a Java version, is it the same if I download temperature data and load in via “Load Temp”?

Attachments:

You must be logged in to view attached files.

[bridgetwang]

the result of linear model is added here

Attachments:

You must be logged in to view attached files.

[bridgetwang]

finally the linear model with thermal expansion is added.

By the way, the temporal coherence with index “_1” is the coherence before removing APS.

Looking forward to your reply. Thanks very much.

Best regards

Attachments:

You must be logged in to view attached files.

[periz]

Hi,
1. in the APS module the aim is estimating the APS, not the model parameters.
2. the APS is estimated from phase residuals. phase residuals = original phase – model. if the model is wrong, phase residuals will be wrong and the APS will be wrong.
If you have non linear signals, given the 2 above statements, you should either estimate as well as possible the non linear model OR you should use only linear targets to estimate the APS.
If you do not have precise temperature data, then you can make the estimation in 2 steps and e.g. discard points affected by thermal expansion. Note that this is actually done also when using the coherence as weight: if you use a linear model, non linear signals will lead to lower coherence, but using higher thresholds you will discard such points. Doing it in 2 steps should help better connecting the graph. This can also be applied with the graph analysis and refinement.
Alternatively, get the best temperature data you can find. The weather module tries to collect data from online repositories. If it fails, you can import your own data, just put one temperature value in degrees for each image in a text file (in temporal order) and load it.
Do not look only at the coherence to judge whether the analysis is correct or not. Plot also the APS data and check the coherence trend in time. If there are temporal patterns, it will be an indication of some residual quantity, not well estimated by your model.
best

[bridgetwang]

Thank you so much for the rapid reply!

As for “estimate as well as possible the non linear model”, I wonder what’s the difference between using a non-linear model built by myself(e.g. include deformation induced by temperature or water level) and using the “smart” non-linear model?

Additionally, I know that I can use higher threshold for point selection in APS module, and then complete the APS processing. But I don’t quite understand how to implement the 2 steps estimation in APS module. Could you please explain further on how to implement step1 and step2 in the software?

Best regards

[periz]

the smart function is a non parametric model, basically a time filter that smooths residuals. thermal expansion is a function of temperature, which has a smooth component in time (like a sinusoid with 1 year period), but also higher frequencies fluctuations. the smart model would compensate only for the smooth component, while actual temperatures would be more accurate. the smart model works well in case of accelerations.
to do the aps processing in 2 steps, you could e.g. save the aps partial results and then load them with load mask, make a subselection in scatter plots, save it and use it in the aps processing as a second step.
as anti says, this can also be done with the graph analysis and refinement (if your subselection is just about selecting points with high coherence). what you could do more in the 2 steps approach could be e.g. selecting points at low elevation (points on the ground) which, ideally, should not be affected by thermal expansion…
best

[antig]

Additionaly, I suggest to use Graph Analysis and Refinement module. It is between APS and SPP button. There is possible to remove low coherence connections and improve APS graph.

[Author]

Posts