Our models & technology

Forty models.
One forecast.

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.

40+ numerical models Extended station network & radar Updates every 10 min
Machine-learning combination of forecast models: AI adjustments for local conditions and real-time corrections leading to best-in-class accuracy.

More accurate than any single model

ML picks the best of 40+ models for each situation and place.

Hyperlocal

Regional models with detailed geography and 3D orography.

Always current

Nowcasting corrections every 10 minutes, minute-by-minute where it matters.

How it works

From global models to your doorstep

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.

01 · ML multimodel

One forecast from all the world’s 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.

  • 40+ global & regional models — GFS, ECMWF, UKMO, ICON and more
  • Error-aware blending — each model weighted by its past skill, per situation
  • Probabilities included — full distributions of every variable, not one number
Forecast pipeline: numerical models such as ECMWF, GFS and ICON enter model combination, data refinement and local scaling, producing local forecasts, reduced errors and API data.
02 · Regional model & AI corrections

Down to your street, your field, your track

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.

  • 90 m resolution — 3D orography cells and node interpolation
  • AI per-location correction — learned from local stations’ history
  • Human in the loop — meteorologists on duty, every day
High-resolution forecast grid laid over mountainous terrain with a lake, illustrating how the regional model captures orography.
03 · Nowcasting & applied models

Corrected every 10 minutes — then applied to your business

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.

  • Real-time corrections — from our extended station network, radars and partners
  • Applied indicators — demand, traffic, campaign efficiency, occupancy
  • Trained on your data — sales, claims or production history welcome
Forecasts refreshed every 10 minutes feeding applied models across industries: energy, insurance, retail, smart cities, agriculture, construction, road and rail.
How you get the data
Accuracy, measured
“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.”
Forecasts.cloud · the approach behind meteocentrum.cz and meteocentrum.de
10 min
refresh cycle for most locations
40+
numerical models blended by our ML
Denser coverage
our own stations on top of official networks
90 m
resolution with detailed 3D orography
Four base runs a day · continuous nowcasting · probabilities and distributions on request
Where it’s used

One engine, seven industries

Put our accuracy to the test.

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