Every season. One map.

A year-round US weather platform, built in public.

Most weather sites pick a lane — radar apps for storms, model viewers for forecast nerds, seasonal pages for foliage or ski reports. seasonmap's premise is that the atmosphere doesn't work that way: the product should rotate with the seasons, the way weather actually does. Severe tools when spring convection fires, hurricane tracking through the warm season, foliage projections in fall, snowfall and ice products all winter — all riding the same always-on core of model maps, radar, and verification.

What's on the map

Forecast models — GFS, ECMWF, ICON, GEM, HRRR, NBM, the AI models (AI-GFS, AIFS), and the GEFS ensemble with mean and spread, rendered as smooth GPU-animated fields or WPC-style filled contour bands.
Single-site NEXRAD radar — Level II reflectivity, velocity, and correlation coefficient with split and quad panels and scan-by-scan scrubbing.
Click-anywhere soundings — a full skew-T/log-P with hodograph, CAPE/CIN shading, the dendritic growth zone, and hover readouts at any pressure level, for multiple models.
Global drivers — the MJO phase wheel, a Rossby wave-train Hovmöller, and an East Asian mountain-torque index computed in-house: the week-2/3 context behind the map.
Season products — severe parameters (STP, SRH, shear), fire weather, snowfall/ice/p-type suites, ski-resort conditions, allergy & pollen, fall foliage projections, river gauges, drought.

Where the data comes from

Source	What we take from it
NOAA Open Data (NODD)	GFS, HRRR, NBM, AI-GFS, GEFS, RTMA — raw GRIB2, byte-range subset and decoded by our own pipeline
ECMWF Open Data	IFS and AIFS forecasts under ECMWF's open-data license
DWD Open Data	ICON global — native icosahedral GRIB2, regridded by our own pipeline. Datenbasis: Deutscher Wetterdienst
ECCC MSC Datamart	GEM (GDPS) forecasts. Data source: Environment and Climate Change Canada
Met Office (AWS open data)	UKMET — Global Deterministic 10km NetCDF, regridded by our own pipeline. Contains Met Office data © Crown copyright, used under the Open Government Licence
NWS / SPC / NHC	Alerts, convective outlooks, tropical products (live feeds)
NEXRAD Level II	Single-site radar volumes
Bureau of Meteorology (AU)	The Wheeler–Hendon RMM index behind the MJO wheel

seasonmap reads agency data directly wherever a public feed exists. The exception: UKMET, which has no free feed and arrives via the Open-Meteo mirror, as do several derived daily layers.

Consensus — our in-house forecast, scored in public

seasonmap publishes its own forecast, Consensus: a lead-weighted blend of every model we carry, built each cycle from the exact grids this site serves. It competes on the public board against its own members — same RTMA reference, same hours, same lead buckets. If it loses, you'll see it lose. Forecast products usually ask for trust; this one publishes its verification scores, including the losses.

The deeper tier of the program — CAESAR, in-house AI-model rollouts (Pangu-Weather) with adaptive recalibration and a perturbed-member ensemble — ran through development and verification here and is paused pending dedicated compute. Its scored history is retained; it returns to the board when it can run without competing with the streaming pipeline for the same cores.

Learn the tools

The field guide covers the mechanics behind the map in short, practical explainers: reading a skew-T sounding, how verification scoring works, what each forecast model actually is, and the MJO.

Who builds this

seasonmap is an independent project, built lean on open agency data and free-tier infrastructure, improving continuously. Bugs and ideas: use the in-app feedback button — it goes straight into the triage queue.

From the same builder: The Space Weather Observatory — live aurora nowcasts, 3-night forecasts, and measured reconstructions of historic geomagnetic storms (Gannon 2024, Carrington 1859), fused from DMSP/SSUSI, POES, and AMPERE data with the same measured-first, limitations-stated approach as the verification board here.

Open the map →