Storm as a 4D Object — Greenfield EF4

Where the severe-weather suite shows the whole continent for 72 hours, this demo goes deep on one storm: the supercell that produced the Greenfield, Iowa EF4 of 21 May 2024. Instead of flattening the radar to a 2D mosaic, every NEXRAD Level II gate from the Des Moines radar (KDMX, 78 km away) is kept as a 3D point — placed at the altitude the 4/3-earth beam model says the beam sampled — so the storm becomes a time-animated volumetric cloud you can orbit and dive through. A render-mode toggle switches the volume between NWS reflectivity bands and dealiased radial velocity, where the mesocyclone reads as bright inbound green pixels beside bright outbound red — the couplet.

The timeline is the story: the SPC particularly-dangerous-situation tornado watch at 18:10 UTC, touchdown near Villisca at 19:57, and the crossing of Greenfield around 20:26–20:32. Rated EF4 at 185 mph from the damage survey, the tornado was also sampled by a Doppler-on-Wheels mobile radar, which measured 263–271 mph winds at 44 m above ground — analyzed to roughly 309–318 mph instantaneous near the surface, making it only the third tornado ever radar-measured above 300 mph. Five people died. NOAA’s experimental Warn-on-Forecast system (WoFS) had highlighted the Greenfield area with roughly 75 minutes of lead time; the NWS Des Moines warnings rise here as translucent wireframe prisms the moment they were issued, shrinking with each SVS update and vanishing on expiry.

Around the volume, the context arrives in painter order: county power outages grow as dark-red columns behind the storm, the GOES-16 C13 cloud-top "anvil canopy" floats at its brightness-temperature height, multi-level HRRR winds thread the scene at four pressure levels, one-minute ASOS stations gust beneath it, local storm reports strike the ground trailing the radar, GLM lightning flickers additively, and the 18Z Omaha radiosonde climbs through the whole scene as a tiny bright trail. All ten archives ride one playhead behind the playback governor, and every altitude-bearing layer shares a single 4× vertical exaggeration so the scene never lies about relative heights.

A wider cut of the same idea streams from the fullscreen viewer: the continental MRMS volume trades this one radar’s gate detail for every storm in CONUS at once.

python nexrad_volume.py --start 2024-05-21T17:30Z --end 2024-05-22T03:00Z --out-dir examples/showcase/public/data

Ten archives on one clock: nexrad_volume.py builds the storm4d-volume gate cloud (plus the storm4d-couplet shear markers) with Py-ART dealiasing; the companion storm4d generators in stt:scripts/data-generation build storm4d-warnings, storm4d-reports, storm4d-stations, storm4d-outages, storm4d-cloudtop, storm4d-wind3d (multi-level hrrr_advect extension) and storm4d-sounding; goes-glm-lightning is the existing continental archive, subset by the demo time range at render — no rebuild. Exact schemas + knobs: docs/roadmap/storm-4d-greenfield-2026-07.md §9.

Full recipes for every dataset live in the data-generation guide.