Tips for flowers and grassland, only orthomosaic needed

Hi team,
I need to generate high-res (as possible from my data) orthomosaics of mostly grassland and wildflower growth. Growth up to around a metre.

I am hoping you can give some tips on how to tweak the settings to get the best ortho possible or which parameters are the best to tweak to try to improve the map?

Some detail for my use case:
I used a Sentra 6x multispectral sensor with 80% overlap at 30 m. GSD ~1.3 cm but morphometric variation changes this as the FieldAgent app flies at 30 m from the Drone home point. That being said, most of the area is relatively flat.

I have found weird results relying on the WebODM radiometric calibration (ortho looks like a rainbow) but I can use the Sentera calibration prior to using WebODM so no problem there.

I have:
64GB RAM and Intel i7-7700K
Sentra 6x has R,G,B,NIR,RedEdge (and a RGB that is useless and discarded).
Link to a screenshot/TIF of a previous ortho and a photo of one of the fields from ground level.

The other issue I have is that some of the projects are BIG. Minimum about 55GB (5000 photos), most around 100GB. Radiometric calibration seems to vastly increase the file size for some reason. My issue is running out of virtual disk space, is there an easy way to increase this? Using WSL2. I have plenty of HD space.

Best wishes,

1 Like

When were those processed? What WebODM/ODM version?

Piero did a ton of work to tighten up the alignment between different sensors/lenses flown at the same time that should reduce banding/rainbowing greatly.

You may need to expand your VHDX that WSL2 is using. Microsoft has docs here:

Hi Saijin,
The ortho in google drive was made 4/12/21 using the most recent version of Docker (v20.10.10) and WebODM (1.9.11) on a fresh install of both as I had issues with getting docker/WebODM to run once the disk space was full.

The rainbow effect occurred on the dataset that used the WebODM radiometric calibration. The effect does not occur if I add the data to WebODM after I have converted it myself using the Sentera executable and turn off this setting in WebODM. The downside of this is the significantly larger file sizes, and some issues with analysis as some values from the ortho need tidying up.

So tweaks I use are:
dem-resolution: 2
feature-quality: ultra
ignore-gsd: true
min-num features: 12000
orthophoto-resolution: 2
pc-quality: ultra,
radiometric-calibration: (varies as I pick the best output)
texturing-skip-global-seam-levelling:true (I think I need this and local to be true for appropriate quantitative analysis but I am currently testing to see how results differ)

Example for small size project below
Average GSD: 1.65 cm
Area: 52,614.03 m^2
Reconstructed Points: 120,478,185

Are there other parameters I could alter that may help?
Are any of these settings pointless for an ortho?

Thanks for the link. :+1:

1 Like

I would absolutely drop --ignore-gsd as it is likely just massively bloating your intermediate products and processing requirements for no real gain.

Similarly, you could use --pc-sample 0.01 if you wanted to thin your pointcloud to 1cm resolution to help speed things up and keep intermediate product size more reasonable.

It is odd that it is still rainbowing, though. Can you provide sample data?

1 Like

Does dropping --ignore-gsd: true prevent --theme-resolution and --dem-resolution from working? Does thinning --pc-sample 0.01 have no impact on orthophoto?

I would provide some data, but they are big projects and I am using the PC for multiple jobs and that upload to google drive will take hours. I will provide some sample data when possible at a later date but I did notice another user having the same issue with his multispectral dataset in a recent forum post.


Dropping it prevents you from forcing the GSD to go lower than the calculated GSD. Basically, if your GSD is 1.65cm and you set it to 1.0cm without --ignore-gsd it will go as low as possible (1.65cm).

Similarly, using --pc-sample set to something close to your actual calculated GSD ensures that you’re not reconstructing tons of points that are being interpolated between the actual finest spacing possible at your GSD. By setting it to 0.01 in your case, it will still be finer than your sampled GSD of 0.0165, but will not be reconstructing interpolated points down to the milimeter/nanometer (or beyond!) spacing level that can happen without --pc-sample and with --ignore-gsd on.

So, will it affect the orthophoto? Possibly, yes. But it should not be noticeable and it should be closer to “real”.

Any time you can upload a sample is great. We need the sample data to test and work with multispectral/multi-camera systems to try and get things working smoothly.


Not sure how well this will turn out on the forum but below are screenshots of some orthomosaics to show the differences of radiometric calibration using the additional settings mentioned above.

First, ortho with radiometric calibration using Sentera executable without data from light sensor. Sentera staff have mentioned this is often preferred when weather is constant for the whole flight.
no light sensor reflectance

Second, radiometric calibration with light sensor using Sentera executable.
sun and light reflectance

Using WebODM radiometric calibration shows some odd (reduced) rainbow effect again. Not sure what is going on here.
webODM spec reflectance sun and light

Any tips on how to get the apple trees to come out a little better? Below is a screenshot zoomed in.
orchard tree polygons

Images are screenshots from QGIS to keep sizes down.
Any other tips for improving the ortho? I will be moving on to wildflower fields soon (once I have been given admin rights to increase wsl2 virtual disk space). Uploading the sample data is not realistic at the moment (over 24 hrs to upload the data to google drive).

1 Like

Thanks for the extra examples/info.

I wonder if these changes here would stand to improve our behavior:

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.