My issue with bad initializations isn't 'fear', it's an experience related knowledge that my practice has to eliminate them from my data. The method used depends on the circumstances at the point being observed. In general we use repeat observations. This is primarily due to the other advantages of redundancy.
Our first visit to monuments with RTK is a '30 by 30'. This eliminates the error of the bubble being out of adjustment and most of the operator bias. If the relative positions of monuments prove to match recent work we generally call it done. If not we revisit. When you factor in improved results, this method has proven time efficient as opposed to forcing a new init on every point.
When I compare two sets of '30 by 30' observations I get a very real check on the true error to be expected in local, real world conditions. Actual results tend to run at about a 30% reduction in residuals when checked by independent methods (i.e. total station ties). I will take that over testing that does not mimic the way data is gathered in the real world.
I have watched the developments in our local network and various publications. I integrate tests into the work flow hoping to streamline processes and improve results. For all of the hoopla about how much things have gotten better, the results don't bear much of it out. I still follow folks in RTN and VRS based subdivisions and shake my head. It is not possible to set on a corner, sight another and turn to a third hoping to check well. A good bit of boundary law no longer applies as the errors are random or follow a roller-coaster pattern that cannot be retraced or recreated. While I acknowledge we need to adapt, much of what I see is grotesque slop.
At the end of the day we have to use procedures that give results we can stamp and the owners can live with. There will be a lot of variation in what that means. From what I see locally we are not a point where repeat observations can be completely eliminated.
My .02, Tom
> Actually, the constellation repeats itself every 24 sidereal hours (23h56m solar time). But, as you say, 12 hours is important because the same constellation is on the other side of the earth, but we have spun around to there, so we are seeing the same sats twice a day. Good point.
John,
The 24 hour constellation repeat is relative to any given point on the globe. So at any given time the constellation repeats itself every 23h56m. Here is a plot of both GNSS constellations for a 24 hour period over Oregon.
When working critical control I always shoot to begin a second observation 4 sidereal hours (accounting for the 4 minute shift).
I read the example somewhat differently. Note that the difference is from either point to the average of the coordinates of the two points. To me, that says that if you want 0.03, then the two occupations can be 0.06 apart. In the example, the two occupations are 0.033 apart, but the horizontal passes. I haven't looked at what the statistical basis for such a criterion might be, but there's not much that can reliably be inferred from a sample of 2.
You would think that as more satellites (especially with different technological approaches on board different constellations) are added in the near future one would be able to accomplish this goal at the data collector with ONE observation.
What I am saying is, you would think with enough given satellites and a long enough RTK epoch (say 10 minutes) the DC could watch if the RMS rises a certain amount over a given time period. If it gets too high it could (probably) mean that one ore more SVs are subject to multipath, made evident as they travel through their orbit.
Then, it could prescribe how much longer you would need to occupy to reach the desired level of precision.
You would think there are enough SVs (at most times) available to accomplish this now...
Currently most DC RTK collection programs simply say that the RTK solution is getting better and better the longer you occupy because your N size is going up, and we all know that it is pretty much hogwash.