vertical aggregation

[otzi5300]

Added vertical mean, max, min

user can now do:

# In [1]:
fpath = Path("C:/Users/lajo/offlineSpace/tmp/dfs_testfiles/HD3D.dfsu")
dfs = mikeio.open(fpath)
ds = dfs.read(items=3)#, time=range(0,1000))

temp_vmean = ds[0].mean(axis="z")
temp_vmax = ds[0].max(axis="z")
temp_vmin = ds[0].min(axis="z")

# In [2]:
temp_vmean.plot()

# In [3]:
temp_vmax.plot()

# In [4]:
temp_vmin.plot()

# In [4]:
# Subset top 2 layers, take max
ds.sel(layers=[-1,-2]).max(axis="z")[0].plot()

# In [4]:
# Take vertical mean then max to get time, depth average
ds.mean(axis="z").mean(axis="time")[0].plot()

test

Two test files added:

On simple with 3 sigma and 2 z. No sigma stretching
On more complex with 3 sigma and 2 z. On second timesteps, surface level is increase and sigma streched. Bottom layer in element 2 is nan

Computation speed

the loop: for col_idx in range(n_cols) for sigma-z models is slow.

On a laptop, a mesh with:
number of nodes: 131263
number of elements: 192568
number of timesteps: 9504 records (396 days with hourly data)

takes around 1 minutes to do vertical mean on all-layers for one item

min and max takes only a few seconds

[Copilot]

Pull request overview

This PR adds vertical (layer-collapsing) aggregation support to the mikeio DataArray/Dataset APIs by introducing axis="z" for mean, min, and max on layered FM 3D geometries, returning a 2D result. It also extends FM 3D geometry reduction to explicitly support reducing over "z", and adds regression tests covering sigma-z behavior, dynamic zn, NaN/inactive layers, and layer subsetting.

Changes:

• Add axis="z" handling for DataArray.mean/min/max and Dataset.mean/min/max (collapsing layered 3D to 2D; mean is thickness-weighted).
• Extend GeometryFM3D.reduce() to support reducing over "z" and combinations with "element".
• Add tests for dynamic sigma stretching, NaN/inactive layers, and layer selection/aggregation.

Reviewed changes

Copilot reviewed 5 out of 8 changed files in this pull request and generated 10 comments.

Show a summary per file

File	Description
tests/test_dataset.py	Adds dataset-level vertical aggregation tests (min/max/mean over "z"), including NaN/inactive layer coverage.
tests/test_dataarray.py	Adds dataarray-level tests for thickness-weighted vertical mean with dynamic zn, NaN masking, and layer selection behavior.
src/mikeio/spatial/_FM_geometry_layered.py	Adds GeometryFM3D.reduce() support for reducing over "z" (to 2D) and "z"+"element" (to 0D).
src/mikeio/dataset/_dataset.py	Adds axis="z" dispatch for dataset min/max/mean to call the corresponding dataarray vertical operations.
src/mikeio/dataset/_dataarray.py	Implements vertical mean/min/max for layered FM geometries via _mean_z() and _extremum_z().

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[ecomodeller]

@miab-dhi please take a look at this.

[miab-dhi]

Hi Henrik, I saw this and reviewed the PR. It seems like it was already completed, tests passed. Did you want me to test on some unseen data or something?

[ecomodeller]

Hi Henrik, I saw this and reviewed the PR. It seems like it was already completed, tests passed. Did you want me to test on some unseen data or something? Best regards, Michael Allen Bell Marine Data Scientist Marine & Coastal - US [DHI logo] The expert in WATER ENVIRONMENTS DHI Water and Environment Inc. 16-2095 King Kamehameha BLVD Pāhoa, HI 96778 USA Tel: +1 303 937 4488 Direct: +1 720 439 4395 @.@.> www.dhigroup.com<http://www.dhigroup.com/> Privacy policyhttp://www.dhigroup.com/privacy This message contains information which may be confidential and privileged. Unless you are the intended recipient (or authorised to receive this message for the intended recipient), you may not use, copy, disseminate or disclose any information contained in it thereof. If you have received this communication in error, please advise the sender appropriately by reply e-mail, and then delete it. Thank you very much. From: Henrik Andersson @.> Sent: Wednesday, June 3, 2026 10:29 PM To: DHI/mikeio @.> Cc: Michael Allen Bell @.>; Mention @.> Subject: Re: [DHI/mikeio] vertical aggregation (PR #984) [Image removed by sender.]ecomodeller left a comment (DHI/mikeio#984)<#984 (comment)> @miab-dhihttps://github.com/miab-dhi please take a look at this. — Reply to this email directly, view it on GitHub<#984?email_source=notifications&email_token=ANAQ3QSZ2I267QY6F6CZBD346EXMLA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINRSGAZTQMZTGA3KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLDGN5XXIZLSL5RWY2LDNM#issuecomment-4620383306>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ANAQ3QR3HYIINQI43TABASD46EXMLAVCNFSM6AAAAACZYZOK3KVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DMMRQGM4DGMZQGY. Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOShttps://github.com/notifications/mobile/ios/ANAQ3QRFAK752CAWVBBOHML46EXMLA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINRSGAZTQMZTGA3KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJKTGN5XXIZLSL5UW64Y and Androidhttps://github.com/notifications/mobile/android/ANAQ3QSBZLO7AZUE2W7I5D346EXMLA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINRSGAZTQMZTGA3KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLTGN5XXIZLSL5QW4ZDSN5UWI. Download it today! You are receiving this because you were mentioned.Message ID: @.***>

Just wanted to get your opinion if this is useful for you, and make sure that the syntax is optimal, before releasing this into the wild😉.

[otzi5300]

@tile-timley any comments/suggestions?

[oceandigitalis]

I think this is a very useful contribution and the syntax is clear. Looking forward to start getting rid of my own very time and memory limited solutions.

Does it work to work when sigma-layers crossing border into z-layers (if sigma depth isn't deep enough for ones purposes)?.

Great stuff!

[otzi5300]

I think this is a very useful contribution and the syntax is clear. Looking forward to start getting rid of my own very time and memory limited solutions.

Does it work to work when sigma-layers crossing border into z-layers (if sigma depth isn't deep enough for ones purposes)?.

Great stuff!

Yes (and no).

Yes: It works with any combination of sigma and z-layers. It performs a vertical aggregation of the layers in the Dataset/DataArray regardless of whether they are sigma, z, or a mix of both.

No: Currently it only supports aggregation across full layers — you cannot slice at a specific depth (e.g., "give me everything above -5m"). If you need depth-based extraction, you'd first need to select layers manually before applying this operation.

[miab-dhi]

@otzi5300 I tested this out on a different dataset, 3D sigmaZ dfsu with Phytoplankton results. It worked nicely.

I think this a useful addition, is a huge time saver from the traditional ways of achieving this.

[tile-timley]

Terrific and compared to doing calculation with PFS I think it is fast ("mean" operation included).

We often calculate eg. "vertical mean, 0-8 meter depth". If there was a functionality that allowed to "ds.sel(z-range=(z_min, z_max))" in addition to the existing "ds.sel(layers=(layer_min, layer_max))", that would be amazing.

This is related to the response from @otzi5300 :
"> No: Currently it only supports aggregation across full layers — you cannot slice at a specific depth (e.g., "give me everything above -5m"). If you need depth-based extraction, you'd first need to select layers manually before applying this operation."

[miab-dhi]

We often calculate eg. "vertical mean, 0-8 meter depth". If there was a functionality that allowed to "ds.sel(z-range=(z_min, z_max))" in addition to the existing "ds.sel(layers=(layer_min, layer_max))", that would be amazing.

I created a method to extract data from z-val, which extracts a slice at a depth of water either looking from the bottom or down from the surface. We can discuss it more, maybe find a way to implement it in.