Salient gemAI is a proprietary vision transformer model that generates natively calibrated, probabilistic daily ensemble forecasts (200 members) up to 126 days ahead. It maintains spatial and temporal covariance across variables enabling better uncertainty quantification of multivariate risk and tail events. It is an independent model, initialized with with atmospheric and ocean conditions from ECMWF HRES . It is natively trained on CRPS (continuous ranked probability score) and ECMWF ERA5 historical data.
API model: gem
The Salient gemAI forecast has a global extent and provides forecasts for latitude-longitude grid points at 0.25 x 0.25 degrees (25 km) resolution. Spatial debiasing is available to downscale 0.25deg ERA5-style data to a specific location within a grid cell at a granularity of 100m.
| Version | Timescale (UI) | API timescale |
Horizon | Update Frequency | Initialization Time | Availability post-initialization |
|---|---|---|---|---|---|---|
| v2 | Daily | daily | Days 1-126 | daily | 00z | 0600z* |
*Historically, the forecast has been available by 0600z, but is not guaranteed. Availability is dependent on early delivery of ECMWF HRES.
<aside> 📌
Note:
GEM2 is a left-labeled daily aggregation for all variables.