Most weather services build on a single global model. We teach machine learning to blend the world’s best ones, correct them with local data — and keep correcting, every ten minutes.
ML picks the best of 40+ models for each situation and place.
Regional models with detailed geography and 3D orography.
Nowcasting corrections every 10 minutes, minute-by-minute where it matters.
Four layers, one pipeline: a machine-learned multimodel, our own regional model, AI corrections per location, and real-time nowcasting. Each layer feeds the next — and the whole thing feeds your applied models.
Most services refine a single global model — so their ceiling is that model’s accuracy. We go the other way: our machine learning studies the historical errors of every model in every meteorological situation and location, then blends them into one output that eliminates most biases. No single model is best everywhere; ours doesn’t have to be one.
The multimodel feeds our own regional models, which raise the resolution and account for local geography and 3D orography. Where station measurements exist, another AI layer corrects the output for the specifics of that exact place — and a professional meteorologist on duty can still step in.
Weather changes fast, so our outputs are continuously corrected against live measurements from stations and radars — most locations refresh every 10 minutes, minute-by-minute where it matters. The same engine then combines weather with your internal data to build applied models: demand, footfall, claims, production.
“No single weather model is best everywhere. One handles thunderstorms well, another winter inversions, a third night-time temperatures. The skill is knowing which to trust — for this place, this variable, this hour.”
Pick a few of your locations — we’ll run a back-test against the data source you use today and show you the difference, number by number.
Talk to our team