Climate Sight Introduces ForecastWire

Introducing ForecastWire

Forecasts are now live on Climate Sight.

ForecastWire is our new forecasting service, a purpose-built pipeline that continuously ingests the most major weather forecast models available, at the highest resolution possible, for all temperature markets. Every model run, streamed in as fast as it publishes.

This has been a significant engineering effort. Getting forecast data right isn’t just about fetching files. Each model runs on a different grid projection, uses different coordinate systems, and publishes at different spatial resolutions. ForecastWire handles all of it: proper grid-to-station projection math, correct spatial interpolation for each model’s native grid format, and the accounting adjustments needed to map a grid cell to the exact point where a market actually settles.

All while providing processing times measured in milliseconds oweing to its Rust based processing pipeline and incorporating low level libraries direct from NOAA and EMCWF. This means forecast readings are available in seconds from when they are published.

What You Can See Now

The initial launch focuses on what matters most for temperature markets: current forecast high and low temperatures across all supported markets.

As each model run completes, the latest forecast high and low gets surfaced immediately. More detail on how to read and interpret each model’s output including update frequency, forecast horizon, bias characteristics, etc. is coming alongside a more complete forecast view. For now, the data is live and available on every market page for Early Supporters.

The Model Lineup

ForecastWire ingests most of the major operational forecast models. Here’s a rundown:

US / NOAA

• HRRR — High-Resolution Rapid Refresh. 3km CONUS grid, 24 runs per day (every hour), full hourly forecast coverage out to 18-48 hours. One of the highest-resolution operational models available.
• GFS — Global Forecast System. 0.25° global coverage, 4x/day, 120-hour range.
• GFS MOS - Global Forecast System Model Output Statistics, 4x/day, provides bias adjusted values for each station where available with some of the overall lowest Mean Absolute Error currently available today.
• NBM — National Blend of Models. 2.5km CONUS statistical blend, 24x/day with ensemble spread. A new effort by NOAA to combine many different models to adjust for individual model biases and account for seasonal variations and anamolies, one of the lowest MAE currently available.
• GLMP — Gridded LAMP (Local Aviation MOS Program). 2.5km CONUS statistical guidance, 24x/day.
• GEFS — Global Ensemble Forecast System mean + spread. 0.25° global, 240-hour range.
• RRFS — Rapid Refresh Forecast System. The next-generation 3km North America model, the eventual successor to HRRR.
• AI-GFS — NOAA’s AI/GraphCast-based global model. 6-hourly output, 384-hour range.

ECMWF

• IFS — ECMWF’s Integrated Forecasting System at full native resolution (~9km). Hourly output out to 144 hours. This is the real native IFS — not the regridded version.
• AIFS — ECMWF’s AI Forecasting System. ML-based deterministic global forecast, operational since early 2025.

International

• HRDPS — Environment and Climate Change Canada (ECCC). 2.5km pan-Canadian domain, 4x/day, 48-hour range.
• ICON-EU — DWD (German Weather Service). 7km Europe-wide coverage, 4x/day, 120-hour range.
• ICON-D2 — DWD. 2.2km Germany and Central Europe, 8x/day, 48-hour range.
• ICON-D2-RUC — DWD. 2.2km Germany rapid-update cycle on the native icosahedral grid. 24 runs/day, 12-hour range.
• AROME — Météo-France. ~2.5km France and surrounding area, 8x/day, 51-hour range.
• AROME-HD — Météo-France. 1km France, 8x/day, 51-hour range.
• AROME-PI — Météo-France nowcast. 15-minute output, 6-hour horizon, updated every hour.
• UKV — UK Met Office. 2km UK coverage, 8x/day, 54-hour range.

The complete model reference, with details on how to interpret each one, is being finalized and will ship as its own set of guides.

Resolution Matters

The 0.25° global grid that most people think of when they picture a forecast model covers an area roughly the size of New York City’s five boroughs in a single cell. Settling a market on a single station inside that box means the grid-cell value and the actual reading can diverge significantly.

At 3km (HRRR), or 9km (native IFS), the picture is substantially sharper. At 1km (AROME-HD), sharper still. ForecastWire ingests these at their native resolutions and applies the correct projection math for each grid. Lambert Conformal for HRRR, the octahedral Gaussian grid for IFS, the icosahedral grid for ICON-D2-RUC, before mapping to each station. Ensuring all forecast data is provided accurately.

What’s Next

More is coming. Detailed per-model documentation, a fuller forecast view, and continued additions to the model lineup. The infrastructure is in place — everything else builds on top of it.

Stay tuned.