Question about processing sampling planes

[passboy666]

Dear all,

I am currently working with AMR-Wind and using the plane sampling feature as described in the documentation. I have successfully defined a sampling plane in my input file (e.g., at hub height), and the simulation runs without any issues.

However, I am unsure how to extract the velocity field data from this sampled plane. My objective is to perform POD (Proper Orthogonal Decomposition) on a series of velocity snapshots from this plane.

I would greatly appreciate your guidance on the following:
I would greatly appreciate your guidance on the following:

Where exactly are the sampled plane velocity data stored after the simulation?

In what format is the data saved (e.g., .csv, .nc, .plt, etc.)?

Is it possible to export the sampled data to CSV format directly or through a post-processing script?

Is there a recommended way to extract and organize these plane snapshots over time for POD analysis (e.g., in MATLAB or Python)?

Will enabling plane sampling significantly affect the simulation speed or memory usage?

Any suggestions, examples, or scripts would be extremely helpful. Thank you very much for your time and support!

Best regards

[marchdf]

Hi, thanks for reaching out. I think we answer most of those questions in our documentation. I refer you to the following links:

• https://exawind.github.io/amr-wind/user/post_processing_examples.html
• https://exawind.github.io/amr-wind/user/inputs_Sampling.html
• https://github.com/Exawind/amr-wind/tree/main/tools

I hope this helps.

Will enabling plane sampling significantly affect the simulation speed or memory usage?

No, particularly if you use the native sampling.

[hgopalan]

Format depends on whether you specify native or netCDF. There is a Python script to load the files if you save them in amrex particles format. For netCDF, you can use netCDF4 or xarray.

You need to use glob or other directory readers in Python to sample all the planes in a directory and then reshape them into a 2-D array using numpy and then perform POD or other surrogate model studies.

[lawrenceccheung]

Hi @passboy666,

We have also looked at doing some (Spectral) POD analysis from AMR-Wind sampling planes and our code to do that is available online. In case it helps, you can look at the AMR-Wind frontend postprocessing engine, and our method for SPOD analysis of rotor plane sampling data https://github.com/Exawind/amr-wind-frontend/blob/main/postproengine/doc/spod.md. We're still working on the full documentation for this, but there's a pre-print of the analysis that @gyalla did here: https://wes.copernicus.org/preprints/wes-2025-14/.

Lawrence

[passboy666]

Hi all,

Thanks so much for providing your suggestions earlier — I’ve tried them out on my case setup.

I'm currently using PlaneSampler with native format in AMR-Wind, and my amrwind.inp sampling section looks like this:

incflo.post_processing = sampling
sampling.type = Sampling
sampling.fields = velocity
sampling.output_frequency = 100
sampling.labels = yz1D yz2D yz3D yz4D yz5D yz6D yz7D yz8D yz9D

sampling.yz1D.type = PlaneSampler
sampling.yz1D.origin = 808.3 -200.0 0.0
sampling.yz1D.axis1 = 0.0 400.0 0.0
sampling.yz1D.axis2 = 0.0 0.0 400.0
sampling.yz1D.num_points = 200 200
sampling.yz1D.normal = 1.0 0.0 0.0

... (similarly for yz2D to yz9D)
On the postprocessing side, I’ve written a Python script using amrex_particle.AmrexParticleFile (inspired by the provided tools) to classify the planes by their x-coordinate and write CSVs accordingly. The coordinates and num_points match exactly — each plane should have 200 × 200 = 40,000 points.

However, only the yz1D, yz8D, and yz9D planes have any data, and even those don’t contain the expected 40,000 points — typically far fewer. All other planes return empty after duplicate removal, or are entirely missing.

Has anyone encountered this issue before? I’ve checked the following:

The origin and axes are correctly defined for each plane.

The simulation domain should cover all the planes.

The output directory does contain the binary samplingNNNNN files for the correct time steps.

Any ideas on what might be going wrong? Could it be AMR level coverage, domain truncation, or something else?

Thanks again for your help!

Best,

[passboy666]

podinp.txt
inp and py.
native to csv.txt
there no any warnings and every sampling plane created.

[passboy666]

the sampling csv is too large inthat I can‘t put in here
csv_output (3).zip
and I put part of it

[marchdf]

Can you post one of the raw sampling plane output? It's per time step and 40k points isn't much.

[passboy666]

Thank you very much for your kind suggestion!

I’d like to ask a quick follow-up: for a 300m × 300m plane, how many sampling points would you recommend as appropriate?

Also, just to confirm — are you suggesting that I should start by sampling a single plane first, rather than setting up sampling on all 9 planes, in order to better identify the issue?

Thanks again for your help!

[marchdf]

how many sampling points would you recommend as appropriate?

No idea. Totally depends on your use case. I would assume there are POD guidelines out there.

are you suggesting that I should start by sampling a single plane first, rather than setting up sampling on all 9 planes, in order to better identify the issue?

Yes, start small, make sure you get what you expect after a single time step, build from there.

[justhawk98]

Unrelated to all the comments before this but I use paraview to view the sampling planes. Though I remember some newer versions had trouble reading the sampling files. Check out this old discussion: #1305 . I use paraview 5.10.1. Also, once you have the sampling file open in paraview, you can click the save data button and output it as a csv file. That's mostly how I've worked with them, save to csv then use a different language (julia) to process the data. Though I imagine python would also be well suited for this.

[passboy666]

Yes, I understand that it's possible to directly export a snapshot of the flow field on a single plane at a specific time using ParaView. However, in my case, I need to export data from multiple planes — as you can see, there are 9 planes in total.

The simulation runs for 300 seconds, with sampling every 2 seconds. So if I were to repeat the export process manually for each plane and each time step, it would be very inefficient and impractical.

[marchdf]

You can script paraview so @justhawk98's workflow would totally work for you. If you don't know how to script paraview, it's a useful skill and a good opportunity to do so. Lots of tutorials online.

[justhawk98]

Yeah I use scripting. Paraview has a button somewhere to start recording an equivalent python script for your analysis. So if you start recording, do your analysis on one file, then stop recording you'll get a python script of the commands you just executed. Then you can take that and modify it to be in a loop. Here's one I made that loops through sampling files and converts them to csvs.

make_csvs.py.txt

Then when you want to run it the command looks like:
/path/to/paraview/bin/pvpython make_csvs.py