Read the sky
like a forecaster.
Every major weather model the moment each run is published, radar down to the raw Doppler scan, and storms tracked run by run, then checked against what actually fell. Self-hosted, built on public weather data. No ads, nothing to sell.
Radar down to the raw scan.
The national mosaic every two minutes, and single radars straight from the Level II feed: reflectivity and velocity, tilt by tilt, drawn on your phone.
Warnings, lightning and SPC outlooks sit on the same map, and when rain is close, a live card says when it starts at your place and how hard it will fall.
- MRMS · quality-controlled national radar, every 2 min
- Level II · velocity for spotting rotation, tilt by tilt
- Now · precipitation type and rate, and the next 3 hours
Every model. Every run. Side by side.
Each forecast hour is fetched the moment it's published, as any of 64 kinds of map: rain, snow, wind, severe weather and the upper air. Compare four models at once, step back through older runs, or subtract one run from the next to see exactly what changed.
ECMWF IFS, AIFS and EPS data © ECMWF, CC BY 4.0.
Explore Models
Any point, any hour, top to bottom.
Long-press anywhere on a map for a Skew-T from that model, run and hour: instant at your saved places, fetched on demand anywhere else. CAPE, the freezing level, precip type and a plain-English summary come with it.
It finds the storm, then watches it change.
Storms are picked out of every run days ahead. The desk follows each one run by run: how the range is narrowing, which ensemble scenarios are in play, and the one factor that decides between them, all in a briefing where every number is cited.
Deciding factor: where the low tracks past the 40/70 benchmark.Illustration
Explore the Storm DeskWhile it's happening, it goes live.
When a Storm Desk storm reaches one of your places, or severe storms or heavy rain come close, the app goes live: what's falling right now, how hard, the next three hours in 15-minute steps, and what's being reported around you. It switches itself on and off.
Then it checks its homework.
Every model is scored every day against real weather stations, and a blend weights them by how well they've done lately. After each storm, the forecasts are lined up against measured snow and rain: who called it, and how early.
Average miss in °F. ECMWF IFS data © ECMWF, CC BY 4.0.
Climb the atmosphere.
A meteorology course inside the app, taught on the weather where you are. Five levels, from "what does 40% chance mean?" to forecasting like a professional, with every lesson worked in the live app and a forecast contest alongside.
From raw model output to the map in your hand.
Reads each run's index and downloads only the GRIB records it needs, by byte range, the moment a forecast hour lands on the NOAA and ECMWF open-data servers. If a mirror throttles, it moves to the next.
Regrids every field with precomputed indices and stores it as a float16 grid, so any point forecast is a single array lookup. Soundings, CAPE, precipitation type and snow ratios are derived in plain numpy.
MRMS radar every two minutes, NEXRAD Level II decoded straight from the raw bytes, GOES GLM lightning, NWS warnings and SPC outlooks, each fetched once nationally and cut to the places in use.
An installable web app. The browser colors the model values itself with WebGL, so maps stay sharp at any zoom and the number under your cursor is the real one.
The forecast you'd build
if you could see everything.
WX is private and isn't open for sign-ups. This page shows what it does with nothing but public weather data.
Back to the top ↑