skyvista

3D gridded atmospheric data visualization in Python

skyvista provides scientifically accurate 3D visualization of gridded atmospheric science data (but likely applicable in many other disciplines). Skyvista's visualizations are primarily built on top of the pyvista library, and with the appropriate setup can be rendered directly in jupyter notebooks or IDEs, or written to disk. Pyvista is capable of creating interactive visualizations in pure HTML, making these visualizations conveniently portable. Skyvista also contains simplified functionality for creating animations of 3D data using pyvista, in addition to single visualizations.

Features

• Gridded data visualization: Create sets of isosurfaces, volumes, vectors, or planes (for things like land/ocean surfaces or cross-sections) from xarray datasets
• Multiple coordinate systems: Cartesian, geographic (lat/lon), spherical (radar range/azimuth/elevation), curvilinear, and unstructured grids with automatic detection
• Trajectory visualization: Visualize Lagrangian trajectory data, with options to show trajectories as continuous arrows or as particles at their instantaneous positions
• Animation support: Generate animations of time-evolving gridded atmospheric data
• Camera control: Advanced camera positioning and following for dynamic views
• Interactive HTML export: Export scenes to standalone interactive HTML files

Interactive example (download and open in browser)

Installation

First install skyvista, which will install pyvista as a dependency:

pip install skyvista

The Blender backend (physically based clouds, Scene.to_blender(), BlenderAnimation) needs the blender extra. It installs Blender itself as the bpy module, along with the Alembic and OpenVDB support and an ffmpeg for videos:

pip install "skyvista[blender]"

bpy has wheels only for Python 3.13, on Linux x86-64, Apple-silicon macOS and Windows. Elsewhere, set $SKYVISTA_BLENDER to a standalone Blender 5.1+ program. To run the example notebooks in examples/notebooks, install the notebooks extra instead, which adds JupyterLab and the model-output readers:

pip install "skyvista[notebooks]"

On a modern VTK (>= 9.3), rendering to disk on a headless machine works with no extra setup: skyvista auto-configures offscreen rendering (EGL on a GPU, OSMesa on CPU) when you import skyvista. To verify your environment, run the built-in doctor, which does a real offscreen render and prints an actionable fix for anything that's wrong:

python -m skyvista        # or:  import skyvista as sv; sv.doctor()

Live interactive plots inside Jupyter can still need a per-frontend nudge (and some setups need an OSMesa VTK build); see docs/pyvista_setup_tips.md for the details. You can also sanity-check PyVista directly with its sample bunny:

from pyvista import examples
dataset = examples.download_bunny()
dataset.plot(cpos='xy')

Quick Start

import skyvista as sv
import xarray as xr

storm_ds = xr.open_dataset("model_output.nc")

scene = sv.Scene()
scene.add_contour(storm_ds, "W", isosurfaces=[1, 3, 5, 10], cmap="Greens", opacity=0.8)
scene.add_contour(storm_ds, "RC", cmap="Blues", opacity=0.4)
scene.show()

For a full walkthrough of the API (convenience functions, factory functions, and VarSpec classes), see the API guide notebook.

For examples of working with different coordinate systems (geographic, radar/spherical, curvilinear), see the coordinate systems demo.

Data Format

Gridded data should be xarray Datasets with dimensions x, y, z, and optionally time for animations. Skyvista auto-detects the coordinate system from coordinate names or CF convention attributes. See the coordinate systems demo for details on supported formats.

License

MIT License

Citation

If you use skyvista in your research, please cite:

@software{skyvista,
  author = {Davis, Charles},
  title = {skyvista: 3D gridded atmospheric data visualization in Python},
  year = {2026},
  url = {https://github.com/cmdavis4/skyvista}
}