Sorry, I was thinking you want to generate DEMs because I was only looking at the topic’s title.
So for DInSAR the perpendicular baselines are good, but probably the time between the image pairs is too long for the extraction of reliable displacement.
Masking low coherence areas before unwrapping sounds a good idea to me. You can follow this guideline: Filter out low coherence pixels before phase unwrapping
But as your study area is in the tropics, there might still be large atmospheric contributions. Did you consider using persistent scatterer interferometry using StaMPS or PyRate?