Need GCP advice from the community

We are attempting to deliver our first survey grade map and have hit wall…Our best RMSE is 3.9ft.
Flight Mission Details:
75 front
80 side
drone avgd about 11mph over the 15 acres

GCP placement:
5 pins read by licensed crew using Trimbul TCS2 and R10 base station
CSV file from them was NAD83 VI/EPSG:2230

The easy answers may be fly lower, increase side/front lap…that we know, but we need to be efficient too. Is there any suggestions from someone who has successfully completed a map with 1ft accuracy or better?


I do not use GCP and still get RMSE’s of 5ft.
Flight Mission Details:
90 front
84 side
9 mph over 20 acres of which 17 acres is the area of interest. I manually set upper limit to 9 mph.
Flight area is 1200’ x 726’ for map area of 1070’ x 700’
Overcast day
Virtually no wind
Long E-W legs of 1200’, short N-S 70’ legs
60% trees 30% more open area with buildings
Time: 19 minutes, 70% battery used, 30% remaining after landing. To maximize battery efficiency, I used an even number of long legs (10) to get the same short distance from takeoff point to start and from end point back to takeoff point.

I suggest you try a mission with parameters closer to these and see what you get.


Hi @C3D,

It’s Andrea here. I have been in conversations with Jeffrey in regards the accuracy of this map and I suggested him to re-upload the information and tag it with GCPs again. Once this processing is complete, we can review closely the accuracy results.

I look forward to seeing your map complete soon.


Thanks @Andrea, not trying to step on your toes, Jeff asked me to get some peer input so I decided to solicit some examples from here. Trying to deduct anything else we could do different so that we can deliver to our client in a timely manner. I believe my coworker is on his way home now to re-upload.

@SolarBarn–thanks for that input, definitely provides a different insight on what we might fly next time.

Hi @C3D,

Totally, we would love to hear from other Forum users their suggestions to improve the accuracy of the GCP maps. In the meantime, Jeffrey and I are troubleshooting with this specific GCP map. We hope to see the results soon.

Have a wonderful day!

Are you sure your not at 3 inches. Make sure you are reading the report right. 36 inches?

Flight plan elevation: 250’
Side Lap: 65%
Front Lap 70%
Speed 15-20mph
RMSE. 0.7 - 1.0 inches (although I have had less)

Make sure:

  • Good lighting
  • Set your flight plan with and against the wind. Never fly a crosswind.
  • Make sure the coordinates in your GCP’s are better than the error you expect to accomplish both from a survey quality perspective and a rounding factor. If using state plane you need thousandths and lat/lon should reach at least 1 hundredth of a second. As a whole when the GCP’s are localized the RMSE will be no better than your worst GCP. I am sure your points are better than inches though…
  • Find a good photo editor. Something that can batch edit images to ensure an even exposure and contrast across the set.

Follow these and you should be sub 2-inches. If not I would really suspect your control network and the GPS data you are collecting.

I would be very careful when stating that accuracy without GCP’s. How do you know you are getting RMSE of less than 5’? Are you surveying on the ground? Didn’t you mention that you are using artificial GCP’s n Rhino? How does that RMSE compare to the algorithms that DD is using for there accuracy reports?


The RMSE is what DroneDeploy reports. We need the DroneDeploy guys to explain the details of this accuracy. The DroneDeploy RMSE I quoted is for a mission without my ACT corrections (which will only get the elevation perfect at one spot on the map). I am impressed that the DroneDeploy RMSE report showed such a good number for my last mission. But I am getting better at flying missions and the DroneDeploy RMSE’s reflect this.

Isn’t your RMSE the DroneDeploy reported RMSE just like mine?


The RMSE reports that I get are slightly different because I use ground control points. The standard accuracy reports only have one set of numbers. When using ground control points the XYZ rmse is reported on each one. That being said, I am not really understanding what the standard report really means. Your report of 5ft is definitely the best I have seen with a non gcp map. I think the best that I have seen in our Maps is around 10ft. I’ll have to go through my maps to see.


All my missions fly the long 1200’ leg aligned with a latitude line(y) so only the longitude (x) is changing. There are no pictures taken on the short 70’ leg that runs along a longitude line (x). Perhaps this helps and is reflected in the RMSE report:


Just a thought.


I’ll make a correction before proceeding. You appear to be getting a different report than what I get with non-gcp maps. I thought my error was higher because it reports inches. I am getting approximately the same results though.
Now for gcp vs non-gcp reports, the figures are reporting global vs. local accuracies respectively. Basically non-gcp maps rely solely on the GPS of the drone and the deviation over the entire site. GCP accuracy is the combination of the drone, survey grade GPS and a localization of the GCP’s. The GCP’s determine the correction of the camera location and those are the RMSE values for that type of report.

…and the RMSE doesn’t have anything to do with the change in coordinates along the flight path unless I misunderstood your meaning.


The RMSE could be smaller if the drone generates better GPS coordinates for an axis that is not changing (which it could judge by looking at its IMU data). If the drone was really smart and compared the compass direction (E-W) with what the IMU is saying for the perpendicular direction (N-S), then it could average the GPS y-reading as long as the IMU keeps saying there is no N-S change. Then after a turn (two 90-degrees), it starts over and finds once again its going E-W, IMU says there is no N-S change so it averages the GPS y-reading again. But maybe the drone is not this smart. I would have to fly a mission with 1200’ N-S legs to confirm this. Certainly the RMSE report seems to indicate this is what’s happening. All my missions have shown over 2 times X error compared to Y error. If this is actually what is happening, then all missions flown parallel to longitude or latitude lines for the long legs will have lower RMSE. And missions flown at 45 degrees relative to longitude or latitude will have the worst RMSE. Is this what your missions show? Of course one needs to be careful to sort out other factors that can have a large influence (wind, lighting conditions, altitude, flight speed) before drawing conclusions. I try to carefully control these variables for my missions plus use better and better flight plans and have been rewarded with smaller and smaller RMSE. But currently I cannot prove that flying long E-W legs reduces the RMSE. But it not, then something else strange is going on.

This is what I was getting at.


Unless you are running RTK the accuracy of the GPS in the drone is not good enough to track a repeatable error in coordinate change. You can see that by your RMSE values when not using GCP’s. Looking through various maps of all different orientations my numbers are consistently below 3 inches regardless. Wind is the only factor that I see affecting the accuracies and I can note that from my logbook. I can also see that in flight by the drone’s ability to stay on track. I am able to plot this point at approximately 18 mph before I start seeing any degradation.

Here are the RMSE reports from yesterday with SE-NW, E-W and N-S flight directions. Wind was approximately 12mph with gusts to 20. The first report is the SE-NW flight.


Its great that your GCP missions provide such good results.

But I was looking for non-GCP mission results. The errors are so tiny with GCP that finding a way to improve them is beyond my current horizon of interest. It is for those missions without GCP, with much larger RMSE, that I can imagine improvements are possible. This could benefit me since most of my missions are over trees with no possibility for GCP in the areas of interest unless trees are removed. Getting the best non-GCP RMSE would help my efforts to track tree growth over time. I look forward to the availability of affordable RTK-equipped drones that log the cm-level accurate GPS coordinates to the EXIF data of the photos. The DJI M210-RTK drone flies cm-level accurate but tags the photos with ordinary GPS data, at least for now. I have heard that they have a plan to fix this. Then we could have a discussion about whether it is affordable (>10K currently).



Can you think of a reason why the X-error is always more than 2X the Y-error on my missions? This is what I am trying to understand. At some point the rain should let up so I can fly back-to-back missions with 700’ long legs (most that fits in site) with N-S, E-W and maybe NW-SE and NE-SW long legs. Perhaps these will provide some insight.


My first guess would be something dimensional on the drone. I wonder if it would be the same with a different one? Or aspect ratio? I could try shooting 16:9. The algorithm? I’ll have to look back at some P3P reports.


Your comments got me to think of another factor, overlaps. I am now flying with 90% front overlaps and 84% side (more side requires an undesired 2 battery mission). But this interacts with the aspect ratio you mentioned. The aspect ratio gives more Y (N-S) information on my E-W long legs while the greater front overlap provides more X and Y information. So overall there seems to be a bias in favor of Y. Hard for me to sort this without knowing more details about the photogrammetry. But there is definitely more angular Y information than X for the photogrammetry to chew on. How this reflects back into the X and Y components of the RMSE I am not sure.