An interactive Skew-T log-P diagram for real radiosonde soundings and forecast profiles.
Time × height section of a model run — wind speed field with barbs, isotherm overlays, and the model-CAPE / 2 m T/Td strip chart:
- Observed — University of Wyoming upper-air archive (
weather.uwyo.edu/wsgi). Twice-daily balloon launches from ~550 stations worldwide. BUFR-sourced stations provide 1-second native resolution (4000–7000 levels per flight, including per-second balloon GPS positions, frost point, and RH over ice). FM35/TEMP stations provide classic significant-level data. The bundled Node proxy adds the CORS headers the archive lacks and caches immutable past soundings. - Forecast — Open-Meteo pressure-level API (19 levels, surface data, and model CAPE/CIN for cross-checking) with a 7-day hourly scrubber across GFS, HRRR, ECMWF, ICON, GEM, ARPEGE, and UKMO.
All from scratch in src/met/ (validated against Wyoming's own indices):
- Bolton (1980) moist thermodynamics: vapor pressure, mixing ratio, θ, θe, LCL temperature/pressure; RK4 pseudoadiabatic ascent/descent
- Parcel analysis for surface-based, 100-hPa mixed-layer, and most-unstable parcels: CAPE/CIN with virtual-temperature correction, LFC/EL, lifted index, cap strength
- DCAPE, precipitable water, freezing level, wet-bulb zero, lapse rates, K index, Total Totals, SWEAT, Showalter, θe minimum
- Kinematics: bulk shear (0–1/3/6/8 km), Bunkers storm motion, SRH (right and left movers), critical angle
- Effective inflow layer (Thompson et al. 2007) with ESRH, effective bulk wind difference, and effective SCP/STP alongside the fixed-layer variants
- CCL and convective (trigger) temperature; soaring thermal tops
- Fire weather: Haines index (elevation variant), mixing height, transport wind
- Winter: precipitation type via the Bourgouin (2000) energy-area method (SN / RA / FZRA / IP and mixes, with warm-nose and melt/refreeze energies), Kuchera snow-liquid ratio, dendritic growth zone depth and mean RH
- Cloud and airframe-icing layer detection (drawn as bands on the diagram)
- SHIP composite
- SARS analogues — the current sounding is matched against the SPC
proximity-sounding databases (1,148 hail events, 1,875 supercells) using
the two-tier SPC algorithm, giving sig-hail / tornado probabilities and a
list of historical quality analogues. Databases by Ryan Jewell and Rich
Thompson (NOAA SPC), distributed with SHARPpy
(BSD 3-Clause — see THIRD_PARTY_LICENSES.md);
the matching logic is a TypeScript port of SHARPpy's
sars.py. The ~160 KB of database text is code-split into a lazy-loaded immutable chunk, so it never weighs on initial page load. - PW climatology — precipitable water shown against the station's monthly mean in ±σ (SPC climatology via SHARPpy; US and some Canadian stations)
Hand-rolled canvas rendering (no chart library): skewed isotherms, log-p isobars, dry/moist adiabats, mixing-ratio lines, CAPE/CIN area shading, parcel path, wet-bulb and virtual-temperature overlays, standard wind barbs, km height ticks, LCL/LFC/EL markers. Wheel-zoom in ln-p space, hover readout with the full per-level state (θ, θe, Tw, frost point, RH over ice, seconds since launch). Companion panels: auto-fitted hodograph colored by height band with storm-motion markers, θe and storm-relative-wind profiles, and a balloon-drift map drawn from the per-second GPS track.
Time × height view (forecast mode) — a BUFKIT-style section of the whole model run: selectable heatmap field (RH / θe / T / wind speed) on a time × log-p grid, isotherm overlays with a highlighted freezing line, wind barbs, day gridlines, and a surface strip chart of model CAPE bars with 2 m T/Td traces. Hover for a full readout at any hour/level; click a column to jump the forecast hour (the Skew-T view and analysis panel follow).
Multi-model comparison (forecast mode) — overlay any set of models on the primary profile at the same valid hour as thin color-coded curves (solid T, dashed Td) with an in-diagram legend; scrubbing the hour keeps them in sync, so model spread is visible at a glance.
Also: sounding modification (pencil-toggle, then drag the T/Td curves —
Gaussian nudge, full re-analysis on release), pin-as-reference comparison
(overlay any sounding on any other: obs vs. model, run vs. run), PNG export,
URL permalinks (#m=obs&st=71722&c=2026-08-09T00), forecast hour
animation, click-anywhere forecast points, favorite/recent stations, and an
archive inventory calendar — a month grid showing exactly which launches
exist for the station (via the archive's INVENTORY pages, back decades);
click any cycle chip to load it.
npm install
node server/index.mjs # proxy + static server on :8642
npm run dev # Vite dev server on :5799 (proxies /api → :8642)Production: npm run build && npm start — serves the built SPA and the proxy
from the single Node process on :8642 (set PORT to override).
The live instance runs at https://aerologic.fly.dev. The repo ships the
whole deployment: a multi-stage Dockerfile (Vite build →
slim Node runtime with server/ + dist/) and fly.toml
(port 8642, HTTPS forced, health check on /api/health, machines auto-stop
when idle so an unused instance costs ~nothing).
To deploy your own copy:
fly auth loginfly apps create <your-app-name>Then point fly.toml at your app — edit the first line (app = "<your-app-name>")
and pick your primary_region — and ship it:
fly deploy --remote-only--remote-only builds the image on Fly's builders, so you don't need Docker
locally. Subsequent releases are just fly deploy again. No secrets or env
vars are required — both upstream data sources (University of Wyoming,
Open-Meteo) are public APIs, and the app is fully stateless (the proxy cache
is in-memory, so it simply warms back up after a machine restart).
server/index.mjs dependency-free proxy + static server
src/met/ thermodynamics, parcel physics, indices, kinematics
src/data/ Wyoming CSV/JSON parsers, Open-Meteo client
src/skewt/ coordinate transform + canvas renderer + interaction
src/timeheight/ BUFKIT-style time × log-p section of a forecast run
src/ui/ panels: stations, indices, hodograph, drift, chrome
scripts/validate.ts physics check against Wyoming's published indices
© 2026 Andre Paquette · MIT · third-party data and code: THIRD_PARTY_LICENSES.md