If anyone else was wondering about this, here are the answers to the following:
1.
Debian Manpage, and this
notebook for mintpy explains: "the average phase velocity using traditional stacking technique (Zebker et al., 1997) by dividing the unwrapped interferometric phase by their temporal baselines and averaging all interferograms." It is NOT the same as the linear velocity in m/yr from time‐series fitting.
2. demErr.h5 contains a 2D dataset /demErr (ℓ×w) of estimated DEM residual errors (meters) per pixel, computed via the baseline‐dependency method of Fattahi & Amelung (2013). These values represent the topographic error relative to the reference pixel (set to zero). The way it does this is to estimate each pixel’s DEM residual error by looking at the correlation between phase residuals and perpendicular baseline variations (
source)
3. It is simply just the raw time-series (timeseries.h5) after applying the DEM‐error correction. I guess there's no relevant documentation exactly, but if it follows the general
data structure, it can be inferred.
Sources definitely exist, but yeah, this should really be more clearly explained in the official documentation instead of being scattered around so much.