ParaView

ParaView is a scientific visualization application built on VTK, aimed at large three-dimensional and volumetric datasets. It handles isosurfaces, volume rendering, streamlines, slicing through volumes, and animation over time, and it scales to data too large for a workstation through parallel rendering.

When it is the right tool

Most atmospheric analysis is two-dimensional — maps, cross-sections, time series — and matplotlib or a GIS handles that better and faster. ParaView earns its complexity when the structure being examined is genuinely three-dimensional.

Convective cell structure is the clearest case. A cumulonimbus in a convection-permitting run has an internal organisation — updraft core, entrainment at the flanks, detrainment into the anvil — that a horizontal slice at one level cannot show. Any single slice misses the thing. Volume rendering and isosurfaces show the structure whole.

The same applies to urban flow. Building wakes and channelling in a PALM-4U domain are three-dimensional by construction, and understanding why a street is poorly ventilated means seeing the flow around the buildings rather than a plane through them.

The costs

The learning curve is steep. The pipeline model — sources, filters, representations — is powerful and unfamiliar, and the interface exposes a great deal at once. It is a real investment rather than an afternoon.

It is a GUI, with the same reproducibility problem: a visualization built by clicking cannot be re-run or reviewed. ParaView’s Python scripting interface addresses this properly, and anything going into a paper should be scripted rather than clicked. The GUI is for exploring; the script is for producing.

It is heavy. Startup, memory use, and responsiveness on large data all reflect what it is built for, and using it for something matplotlib could do is a poor trade.

Practical notes

It reads netCDF and can be pointed at model output directly, though the readers are sometimes particular about how dimensions and coordinates are declared — CF-conforming files load far more reliably, which is one more argument for the conventions.

For a static publication figure, programmatic plotting usually wins on control and reproducibility. ParaView is for understanding the data and for animation, where nothing else does the job as well.

See also: QGIS, netCDF, PALM-4U, and large-eddy simulation.