WeatherMesh-5c
Traditionally, global weather models are produced on fixed cycles four times per day, commonly 00Z, 06Z, 12Z, and 18Z (UTC). For much of the day, that means the most recently available guidance can already be hours old. While some regional systems (e.g., hourly rapid‑refresh models) update more frequently, they typically have limited domains and shorter forecast horizons.
With WeatherMesh‑5c, WindBorne introduces Continuous Forecasting, the first continuously updated global weather model. WM‑5c runs every 20 minutes, assimilates the latest observations, including a real‑time stream from WindBorne's Global Sounding Balloons (GSBs) and delivers forecasts that are both fresher and more accurate. Instead of waiting for the next fixed cycle, WM‑5c narrows the gap between observed conditions and the published forecast to roughly 20 minutes.
1. Model Metadata
| Name | WeatherMesh-5 continuous (WM-5c) |
| Status | Operational |
| Model Code | wm-5c |
2. Forecasting Regime
| Initialization Frequency | Every 20 minutes |
| Forecast Horizon | 15 days |
| Ensemble | 24 members with mean and spread |
| Data Latency | ~20 minutes from observation to forecast |
| Update Mode | Continuous (not tied to 0z/6z/12z/18z cycles) |
Traditional global models run 4× daily at fixed UTC times. WM-5c forecasts are available up to 6.5 hours before ECMWF and incorporate up to 8 hours more recent observations.
3. Data Assimilation & Observations
WM-5c uses WindBorne's proprietary AI-based data assimilation (AI DA) system. Unlike other operational AI models that depend on ECMWF or NOAA analyses published every 6 hours, WM-5c's independent assimilation enables continuous updates and remains operational during external delays or outages.
4. Model Resolution
| Spatial Resolution | 0.25° × 0.25° global grid |
| Vertical Resolution | Surface + 25 pressure levels |
| Variables | See Variables API or Gridded Forecast |
5. Outputs & Products
Gridded Forecasts
| Forecast Time Steps | 3-hourly |
| Atmospheric Coverage | Surface + 25 pressure levels |
| Delivery Format | netCDF via cURL, PyPi package, or CLI |
| Upper-level Variables | Mean + members 0–5 only, 6-hourly, forecast hours 0–195. Depending on the initialization_time, this may mean upper-level data is available at forecast hours 0, 6, 12, etc, or it may mean upper-level data is available at 3, 9, 15, etc. See the run_information endpoint for details. |
See Gridded Forecast API for usage.
Point Forecasts
| Forecast Time Steps | 1-hourly |
| Spatial Resolution | Any latitude/longitude pair |
| Atmospheric Coverage | Surface only |
| Delivery Format | JSON via cURL, PyPi package, or CLI |
See Point Forecast API for usage.
6. Benchmarks
WM-5c demonstrates approximately equal skill to WM-4 and WM-4.5 while running continuously.
Live benchmarks comparing WeatherMesh to ECMWF are available at benchmarks.windbornesystems.com.
7. Historical Data
Historical data is not available via the API and is provided upon request. Please contact us for access.
| Property | Value |
|---|---|
| Available Period | October 1, 2024 – September 30, 2025 |
| Validation Hold-out | 2020 data excluded from training (standard practice) |
| Available Time Blocks | |
| Oct 1, 2024 – Jun 13, 2025 | One run every 6 hours (00, 06, 12, 18 UTC). Simulated real-time input; within 1% of operational skill. |
| Jun 14, 2025 – Sep 30, 2025 | One run every hour. Archived real-time input; allows tracking forecast evolution within 6-hour cycles. |
| Gridded Forecasts | 0.25° × 0.25° global, netCDF delivery |
| Point Forecasts | 364 pre-determined locations (primarily CONUS), JSON delivery |