About a week ago, I posted about an OPUS-RS solution that I had to temporarily settle for on a control point of a project I'm working on. The OPUS-RS solution was necessitated by a battery that only allowed a 1:15 session.
Being unfamiliar with OPUS-RS results, I was trying to estimate the uncertainty of the solution and thought that it was remarkable that the OPUS-RS solution quoted standard errors for the various components (X,Y,Z or N,E,U) of the solution that differed by more than an order of magnitude.
Anyway, later in the project I logged more data on the same control point. It was on a couple more days, nearly 5 hrs. on one and more than 9 on another. Those sessions provided excellent OPUS solutions to compare to the OPUS-RS solutions.
Out of curiosity, I cut the 9 hr RINEX file into twelve 20-minute non-overlapping pieces and submitted them all as separate files to OPUS-RS.
Of the twelve, only five generated nominally acceptable solutions. Those five did, when weighted using their variance-covariance matrices supplied with the OPUS-RS solution, give a weighted mean that differed by 2mm or less from the weighted mean of the OPUS solutions from the 5hr and 9hr sessions, but that agreement was better than the scatter of the individual OPUS-RS solutions would predict.
The ellipsoid height components of the OPUS-RS solutions were not nearly as good, of course, but none were in error by more than 7cm.
What struck me about the whole business is the apparent need to get at least 4 hrs. of data in order to end up with five, good 20-minute sessions.
What am I missing about OPUS-RS? At the face of things, it seems like a rather small enhancement over regular OPUS for land surveying purposes.