fdiblen Scientific visualization
Scientific visualization turns meshes, fields and volumes into understanding - and, done as engineering rather than clicking, into REPRODUCIBLE artifacts: a figure or animation anyone can regenerate from data plus a script. The reference stack is VTK (the rendering and data-processing library) and ParaView (the application built on it, scriptable in Python); the practices below outlast any one tool.
Reproducibility first
An interactively composed visualization is a dead end the moment it is needed again. Make visualizations regenerable:
- Script the pipeline: ParaView's Python tracing records interactive work as a pvpython script - clean it, parameterize the input path, and commit it next to the analysis code (rseng-version-control-review).
- State files capture a full session as a restorable artifact; scripts beat state files for review and parameterization, state files beat nothing.
- Treat visualization scripts as code: inputs and camera/colormap parameters in configuration, outputs written to a results directory, runnable headless in the pipeline (rseng-workflows) so figures regenerate when data changes.
- Record the tool version with the output - renderers evolve, and a pinned environment (rseng-reproducible-environments) keeps animations regenerable years later.
Honest representation
- Colormaps: perceptually uniform by default (viridis-class); rainbow/jet-class maps create false boundaries and mislead - flag them on sight. Diverging maps only for data with a meaningful center; always show the colorbar with units (rseng-scientific-file-formats' unit discipline pays off here).
- Respect the data's structure: do not interpolate across discontinuities, do not volume-render categorical data, state isovalue choices - an isosurface at an arbitrary threshold is an editorial decision and should be a labeled parameter.
- Accessibility applies: colorblind-safe maps, readable annotation sizes in videos and figures (rseng-ux-accessibility).
Scale: large data and in-situ
- Larger-than-memory results: use parallel/distributed rendering (pvserver) or level-of-detail decimation for interaction, full resolution for final renders; chunk-friendly file layouts (rseng-scientific-file-formats) decide how painful this is.
- In-situ visualization (ParaView Catalyst-style) renders DURING the simulation instead of writing everything to disk - the escape hatch when output volume makes post-hoc analysis impossible; it changes I/O planning (rseng-hpc-computing).
- Batch renders of animations belong on the cluster as jobs, not on laptops overnight (rseng-hpc-computing).
Sharing and interaction
- Web delivery lets collaborators explore 3D results without installing anything: trame-style Python apps expose a VTK/ParaView pipeline in the browser; a hosted viewer is a research service with operational needs when it outlives a demo.
- For talks and papers, render key frames as static figures with the same scripted pipeline - one source of truth for interactive and print outputs (rseng-science-communication for the framing).
Working with this skill
This skill is source-independent: its authority is the VTK and ParaView documentation linked below.
Learn more (verified):
- https://docs.paraview.org/en/latest/ - ParaView documentation
- https://vtk.org - the Visualization Toolkit
- https://kitware.github.io/trame/ - trame web framework for VTK/ParaView apps
Related skills
Check whether any of these applies before moving on:
- rseng-hpc-computing - parallel rendering and in-situ output
- rseng-reproducibility - regenerable figures from scripts
- rseng-science-communication - framing figures for talks and papers
- rseng-scientific-file-formats - mesh and volume data layouts
- rseng-ux-accessibility - colorblind-safe maps and annotations
- rseng-workflows - figures regenerate inside pipelines