[anarizka]

dear prof.Daniele

Could you give me suggestion and opinion from my PS results?
1. Why residual height results still rather high? In APS parameter I put -50 x 50

2. Why thermal expansion is not linear with building height ?
I have assumption, I dont know it’s true or not.
-Right side area is more density than left area, so in right area thermal higher than left area.
-Right area sparse of vegetation than left area

Best Regards,

Ana

Attachments:

You must be logged in to view attached files.

[periz]

I am afraid this is an overfitting case.
I suppose you do not have many images, so, in such case I do not suggest you to estimate the thermal expansion.
Also, keep height and velocity ranges small in the APS estimation.
Better you remove high points when estimating the APS.
The APS estimation process should be kept as simple as possible. Later in the sparse points analysis you can estimate more things if the system is stable. But not in the APS.
If your estimated results (vel dem and so on) look correlated, there is surely something wrong (in the APS inversion you are basically integrating errors…)

[antig]

What is the “normal” APS values for trend and H if I dont want to overfit?

[periz]

overfit is caused by trying to fit too many parameters in few data. the range of v and h does not cause overfit. range of v and h should be chosen depending on the analyzed area

[ic350450]

Hi, I am having a difficult time understanding what to look for when evaluating the APS results and notice that in one of the comments above, correlation was one way of determining if the data looks incorrect. What exactly qualifies as correlation between DEM, vel/displ, integrated height etc.? After processing the data in the APS module, I typically look for peaks at zero for each parameter in the data (similar to what was discussed in the tutorials) However, I often get irregularities in height which I feel is alright. Thank you.

- Isaias

[periz]

hi,
to become able to recognize correlated results or to spot possible problematic signals, you need training and experience. difficult to help you much with theoretical statements. we should look at your results together and then we could give you some suggestions.
however, here some general hints:
+ it is correct to check that the peaks of the estimated parameters are centered at zero, in this way you make sure your reference point is adequate
+ check the APS final test (the coherence in space and time after integration of the estimated parameters and removal). if the coherence is high and homogeneous in space and time, this is the best sign that everything went well and that you can trust the estimated APS
+ if the test did not go well, you can play a bit with the non-linear weights hoping to stabilize in this way the spatial integration. however, this is a delicate operation and you need some experience to modulate it correctly
best

[Author]

Posts