University of Exeter tests ML weather models in the IMPROVER blend
- Met Office
- Award under framework
- 16 October 2026 to 31 March 2028
The Met Office is paying the University of Exeter to find out whether more machine learning weather models would improve its IMPROVER forecast blend.
The question is whether the gain in forecast skill justifies the extra effort of running more models. The Met Office is most interested in wind, temperature and precipitation, and in lead times beyond about three to five days. Traditional forecast models start to lose skill there. Exeter proposed running up to five models, among them Microsoft Aurora and Google DeepMind GenCast, as 12-member ensembles over 10 years of past forecasts.
The university took the call-off after a mini competition on Lot 5 of the Met Office's MOAP SDF2 framework, in September 2026. The £297,014 contract runs to March 2028.
Contract ends March 2028
- What
- Technical research for standards and policy
- Buyer
- Met Office
- Supplier
- University of Exeter
- Value
- £297,014
- How it was bought
- Award under framework
- Dates and next
- Oct 2026 to Mar 2028