If you work with Sentinel-1 data, short temporal baselines should be preferred, especially in areas with vegetation cover.
As your area of interest is a dam site, probably surrounded by vegetation, traditional DInSAR approaches are likely to fail, because unwrapping of the phase needs large areas with high coherence and clear patterns of fringes. This is not the case for small areas or objects, and you end up with rather randomly unwrapped phase which has no significant information at a specific area (= the ramp you currently observe).
Please keep in mind that DInSAR is used to get deformations of large areas and unwrapping only works if these are visible in the interferogram.
A nice example is shown by ASF:
Source: How to Phase Unwrap an Interferogram – ASF
If you have small targets and high proportions of phase decorrelation, approaches based on persistent scatterers might be more promising. StaMPS - Detailed instructions - #2 by ABraun