Increasing value of displacement with increase height

How to mask out pixels is described here:

Of course you need the long temporal baseline in order to observe changes. But the coherence won’t be high over the full image.
Regarding the atmospheric content: Maybe you have the chance to investigate another image pair within a similar period. If it contains the same ramping as your example above, it is probably caused by the topographic phase removal (did you use the 1Arc Sec SRTM instead of the 3Sec?). If it is different atmospheric effects dominate in the distortion of your displacement values and you could have more success with another image pair.