Add xugrid extension for TriMesh plots from unstructured grid data

[rsignell]

Holoviz folks, as you can tell, I made this with Claude Code, and I've been using it for the past few months for many applications. But I have no idea how well it's coded.

@Huite did give it a quick look and didn't raise any red flags.

There is discussion on the xugrid repo here: Deltares/xugrid#204 (comment)

Please let me know if there is anything else I can do -- I'd love to see trimesh supported by hvplot!

Summary

• Adds hvplot/xugrid.py — registers .hvplot on xugrid.UgridDataArray and xugrid.UgridDataset, enabling uda.hvplot.trimesh() for unstructured triangular mesh plots
• Extends HoloViewsConverter (converter.py) with a trimesh() method that supports rasterize=True, geo=True, tiles=, cmap=, etc.
• Extends hvPlot in plotting/core.py with the hvPlotXugrid subclass; _get_converter() extracts mesh topology (node coords, face-node connectivity) and stores it in kwds
• Defers face-to-node interpolation until after groupby slicing, so extra dimensions (time, sigma layers) become interactive widgets rather than requiring explicit selection
• Adds hvplot/tests/testxugrid.py with 243 lines of tests covering node-data, face-data, time/depth groupby, geo projection, and rasterization
• Includes two example notebooks (FVCOM_xugrid.ipynb, STOFS_xugrid.ipynb) demonstrating real-world FVCOM and STOFS coastal ocean model output

Usage

import hvplot.xugrid

# UgridDataArray with node data
uda.hvplot.trimesh(clim=(0, 2), cmap='viridis', rasterize=True, tiles='OSM', geo=True)

# Face data with time/depth sliders (auto groupby)
uda_3d.hvplot.trimesh(rasterize=True)  # sliders for time + siglay

Test plan

• pytest hvplot/tests/testxugrid.py — all tests pass
• Manual verification with FVCOM/STOFS example notebooks
• Confirm import hvplot.xugrid is a no-op (clean ImportError) when xugrid is not installed

🤖 Generated with Claude Code

[ahuang11]

Looks awesome! Took a quick glance at the code and maybe https://holoviz-dev.github.io/holoviz-skills/contributing-to-holoviz/cleanup/ could help clean it up (imports)

I think pixi dev may need updating to run these tests in CI.

This does increase maintenance burden for a fairly niche library, but I think it's worth having, if this is the one package for reading and parsing unstructured grid? However, I vaguely recall xarray natively supporting unstructured grids too with its new refactor of indexing?

Alternatively, could maybe use https://github.com/panel-extensions/copier-template-panel-extension to create an hvplot-extension as well and put this there under hvplot-xugrid.

[rsignell]

Thanks @ahuang11 for taking a look!

Since xugrid follows the same integration pattern as hvplot's other data backends (xarray, geopandas, etc.) and this is the canonical Python package for unstructured mesh data, I'd prefer to keep it as a PR to hvplot if you're open to it.

On xarray advanced indexing — it'll help xugrid work more efficiently, but it won't replace the need for a package that understands how to subset unstructured meshes, interpolate face data to nodes, and so on -- xugrid is that package!

I have updated the PR after telling Claude to run the "cleanup". Hope this looks better!

[ahuang11]

It sounds fine to me, but I'll defer to @maximlt for his suggestions!

[rsignell]

@maximlt , just a note here that I'm still very interested in this -- we would love to share this capability with users in the CoastPredict project that does a lot of work with the SHYFEM triangular mesh coastal ocean model.

[maximlt]

Noted @rsignell, I'll review this week and get a release soon.