Description of proposed changes

1. There is a bug in the calculation of design matrix for ifgram inversion using minimum norm phase velocity. Currently the matrix only works for a network with sequential interferograms where the secondary images are always after reference date. So basically it does not work for some networks, for example a single reference network with the reference image in the middle. Incorrect timeseries are resulted where they are zero for dates before reference. Also weighting does not work in that case.

Proposed changes:

• Correct design matrix calculation
• Correct reference date based on interferograms
• Remove the condition where rules out weighted inversion for non redundant network

2. The second issue is regarding the memory and failure of Dask during inversion. The default of ifgram inversion is to not calculate/write covariance matrix of the timeseries which is a huge 4D matrix. But the problem is that it allocates memory for covariance matrix during inversion even when you don't want to calculate it

proposed changes:

• Release memory occupied by covariance matrix if it is not opted to calculate for.