Confused about variance in readings

I have a reach system that I am testing (the original reach units not the RS) with confusing results.

I buried a ceramic tile with an X painted on it in my back yard and performed multiple readings with my base unit a few meters away and the rover placed precisely on the center of the X. The base is in averaging mode (set to 30 minutes for the most recent two test) and the rover is getting corrections via the base only (no NTRIP). The most recent tests were yesterday on v2.3.1 It was midday without a cloud in the sky. I left the system running long enough to get a fix and took a screen shot. I then power cycled both units without moving them at all, waited long enough to get a fix and took a screen shot. The values are grossly different. What am I doing wrong?

Here are the readings:

                    Rover					        Base

4/22 35.34010658 -82.54032916 514.859 35.33991112 -82.53966820 609.73
4/26 35.33997059 -82.53970600 599.683 35.33991015 -82.53968450 601.152
4/26 35.33997071 -82.53972864 605.995 35.33991291 -82.53969302 603.324

First output:

Second output:


This is 4 days earlier with the rover in the exact same position and the base within a meter of the same spot:

The base was set to 2 minute averaging when this was taken.

According to what Drone Deploy considers a pretty accurate map (RMSE 1.8ft) the location of the base unit should be: 35.33993, -82.5397

Each time you sample and and average, you will get a different base location. This is because the base is acting alone without any correction service when it averages its coordinate - just like a handheld GPS.

So, you need do ‘decide’ on a coordinate for your painted X on the ceramic plate. And manually enter that coordinate each time you place your base on the X. Then you will be able to get precise results every time with your rover.

By deciding on a coordinate for your painted X, I mean there is more than one way to do that, and this had been explained several times recently in the forum. Averaging is the least accurate, so maybe try PPP, or measure from a nearby known point, or get correction data from a third party, etc, etc.


I should have added that there is nothing wrong with averaging your base solution to get a coordinate. It really depends on what type of work you are going to do.

I see in the picture above that your base location may be in the middle of a field. Let us make the assumption that Reach will be used exclusively on a farm to:

  • locate fence lines
  • locate useable farm land
  • locate buildings
  • georeference aerial photos from a drone
  • steer a tractor
  • create a topographic map

All of these things can be done precisely by averaging the base coordinate, permanently recording that coordinate, and using it every time you perform RTK.

Now, let us talk about a situation that would create a discrepancy between measurements:
You buy a pair of Reach RS, and your neighbour also buys a pair of Reach RS. You generate a base coordinate by averaging over your painted X, and your neighbour generates his base coordinate by averaging from a pole on top of his grain silo. You each go out and locate all the things on your respective properties. You each locate the fence line that is shared. There is no doubt that each of you made precision measurements (~2cm). But if you compare each other’s coordinates for the fence line, then you will notice there is a discrepancy. It could be a few meters. This problem could have been avoided in two ways:

  • Your neighbour accepts your base coordinate for your painted X, places his base over your X and gets his new base location by placing the rover on the pole on top of his silo. Now both of your base locations are tied together.
  • You and your neighbour each go to the nearest point with published coordinates (a gov’t/ municipal survey marker) and place a base there and take your rover to your painted X and his pole. Now both of your base locations are tied together with the bonus that all your future coordinates will align with the gov’t/municipal maps.

I’m sure you also understand that this explanation could also be applied to your problem above where you averaged your own base coordinate on two different occasions.

In the end, I guess you have to ask yourself: Who am I going to be comparing my coordinates to?

  • Just myself?
  • My neighbour?
  • Local gov’t?
  • NASA?

So many of the published coordinates in the world are not even close in their accuracy compared to what we can accomplish with our Reach devices. So, also remember that if your coordinates aren’t lining up with other published coordinates or imagery in an online map - their coordinates might be less accurate then yours!.


I like bide’s explanation. Keep in mind ,also, that the World’s surface is not stable in GPS measurements. The Reference Stations are floating in random directions. Here in California US, the Reference Station GPS locations are used to monitor the plate tectonic drifts.

Suppose your neighbor uses a 1995 GPS Reference Station location on the east side of the San Andreas earthquake fault. Suppose you have a Reference GPS value on your side of the fault. An earth quake occurs. Now the Reference Station Coordinates have changed. As bide says, it is better to get a close reference.

As an experiment, I used Canada’s CSRS PPP service to plot a local reference station coordinates from 2000 to 2017 once a year(using the ref station OBS RINEX L1L2 files . The plot appeared to me to be quite close to the published tectonic drift for this area. A search of my forum posts should get you a plot.

As a side note, the USGS, Canadian PPP, and other precise final orbits,etc., are usually not final until a few weeks after survey date. For me as a novice, I look to Canadian PPP of local reference station
(mainly because their software determines appropriate files to input}. Then, you can use that location for the local ref station. However,I am still learning, so I’m always open to corrections…Hope I didn’t do anything to detract from bide’s excellent reply.

1 Like

Thanks so much for the replies!

My end goal is to increase absolute accuracy of maps generated from drone imagery. I initially started with putting the rover reach unit on a drone (pixhawk) with the reach acting as a second GPS. This seemed like an easy path since the drone would be working off of the RTK GPS data and thus would simplify geotagging the images. After a little testing it became apparent to me the reliability of this system was not such that I would feel comfortable trying to use it for a client out in the field. Now I am looking at simply using the reach system to generate GCPs which can then be added into the mix when generating the maps. After I test this I will look at the tuffwing method of achieving this goal which is to put the reach on the drone but not use it for nav but simply for getting precise GPS/picture times and rely on PPK. So far I have only taken time to stab at post processing 2x and did something wrong as after all was said and done my .pos file did not have any locations in it…only the header info. I will take time later to troubleshoot that later to see where I went awry. So much to do and so little time!

The first steps of this for me is to get an understanding of the accuracy/precision and to iron out a solid workflow to consistently produce reliable and repeatable results. From the responses above and several more hours of reading I understand now using the base in averaging mode by itself is not going to work.

Now I am receiving corrections at the base station via NTRIP using a VRS solution from the state CORS network. I have been collecting data on and off during the weekend (have not really looked at it yet) and hope to have some info to share early in the week :slight_smile:

Given this scenario I assume my best bet is still to use averaging on the base? These are the settings I am using now for reference:

For some reason my base would never get a fix yesterday evening so I did not get any data on 4/31 but here is a snapshot of the data for the past few days using VRS:

I can’t seem to give any help at this point. Unfortunately, I can’t seem to find any VRS stations in California that are not paid prescription. The Emlid guys may have better experience here.

I was able to run post processing successfully on a data set and am very happy with the results: