Suitable GSD/accuracy for watershed analysis with M300 RTK_P1

Good day,

I have a client who intends on doing a watershed analysis of app. 800 ha. Usually clients say they want the surface as accurate as possible but don’t really understand how GSD and it’s associated relative accuracy translates to the general accuracy of a site and how quickly the amount of data increases. i.e. if I say we fly at an elevation giving a GSD of 2 cm giving an relative /.absolute accuracy of 2-6cm in Z at each pixel they say its not accurate enough because they think the entire survey is off with that amount (It could be, if I don’t understand correctly either). Which when you compare to a GNSS survey where say every 20-30 metres a spot shot is taken with a precision of 2cm at best the interpolation across the surface is surely more prone too inaccuracies across the site when averaged. We will be using a fair amount of GCPs for the transformation to our local projection and checking as well as PPK and RTK to have redundant control.

So I’m not sure the accuracy actually required for a watershed analysis ( First time for this specific request, and the area only has a gentle slope to one side) where they require 25cm interval contours for planning a solar farm. I usually try to get my GSD between 1-1.2cm for any reasonable size site but in this case it amounts to crazy amounts of images. We use a M300 RTK and the P1 so the gigapixels for this project will be absurd. Also, regarding the overlap is it safe to assume we can use 80 even 75 or 70 front and 60 side lap due to the quality of the camera sensor being full frame and size of the image footprint because we can fly higher as opposed to a phantom 4 we have used in the past where you usually require 80 and at least 65-70 to get good matches when trying to get a GSD close to 1cm flying extremely low. I’m trying to find a balance between GSD and overlap, preferably reducing overlap as much as possible and pushing the GSD as high as possible without risk of creating a model which cannot even be best fit to the local control after processing.

Any advice will be appreciated as I can try process 15k + images ( we have a decent workstation) and survey the site in 2-3 days (We have a lot of batteries) but is it necessary as the data handling and presentation to the client will be so much more challenging especially on their side.

A suitable setup for parameters I think should be sufficient is a GSD of 1.5cm and Front lap of 75, side lap of 60. This equates to app. 9000 images.

Thanks in advance.

From my extensive testing using a WingtraOne with Sony RX1RII 42mp camera, lower altitude doesn’t necessarily result in higher absolute vertical accuracy. I found no noticeable difference in vertical accuracy between 300 and 400 feet agl. And 200 feet agl was actually slightly less accurate than 300-400 feet. Higher altitudes result in larger areas covered per frame. This produces more keypoints per image and a more accurate model. I also tested various image overlaps 85/85, 85/70, 70/85, and 70/70. No discernible differences in accuracy between any of those. So I only fly at 400 ft agl for topographic missions with 70/70 overlap.

With a lower quality camera I would expect different results. But the DJI P1 has similar specs to the Sony RX1RII. So it should perform similarly.

Regarding UAS for topographic mapping, you should be able to achieve absolute vertical accuracy of 0.08 ft. That is very comparable to GNSS RTK surveying. But even if you concede slightly less accuracy on a point by point basis for UAS data, the increased point density more than makes up for that.

You will want to use Pix4DMatic to process this amount of data. I tried to process 22,000 images. I had to break it into two projects. Also couldn’t create a single dsm or ortho. Had to merge the tiles in ArcGIS Pro. Matic will handle it with no problem. But you will need a powerful workstation with a lot of memory.

Awesome stuff.

Thanks for the quick response. I’m glad my understanding leans towards your answer.

I will have to look into Pix4Dmatic then. Although I think I might have used the trial when it came out to compare to Mapper which we have a perpetual license for but I’ll see. The project with new parameters comes out to approximately 9500 images… much better than 15k.

The merging of tiles we will do in GlobalMapper then.

Regarding our system it might fall short, might not. We have:

i9 7900x, clocked to 4GHz
Quadro P4000
128Gb Quad channel RAM (3200Mhz)
2 TB M.2 SSD

Hi terrass358,
You might want to visit our site on the hardware requirements for PIX4Dmatic.