Horizon AI Global

Weather drives risk and cost for utilities, energy and commodity traders, insurers, airlines, logistics operators and government agencies, and gaps in observations, especially over the oceans, limit the accuracy of standard global forecasts. Horizon AI Global is Climavision's global numerical weather prediction (NWP) model. It combines targeted AI techniques with NWP: AI is used for data assimilation and quality control of high-resolution observations, which come from government and exclusive datasets including satellites, ground stations and weather balloons. Climavision states the model is fuelled by 1.5 billion observations captured daily, and the company has access to proprietary GPS radio-occultation (GPS-RO) satellite data that fills gaps over the ocean. The model runs on in-house and cloud-based supercomputing.

Forecasts extend to 360 hours (15 days), hourly to 120 hours and then every 3 hours, at 8.5 km horizontal resolution at the equator (7.5 km at mid-latitudes) with 127 vertical levels and more than 100 forecast parameters, including high-altitude winds for aviation. Output is delivered through an interactive map (DASH) and an API in GRIB2 format. Climavision published comparisons for Hurricanes Helene and Milton (2024) in which Horizon AI Global's tracks were closer to the observed paths than GFS and ECMWF, and forecast 14-15 inches of rain for Asheville, NC.