After two years of development, we have released the meteoblue Learning Multi-model (mLM). The mLM uses over 25 different forecast models as well as observations from over 50000 stations to compute a most likely consensus forecast.

Currently, Europe is experiencing an unusual heat wave. These situations are difficult to forecast, because statistical methods are not be properly trained to such extreme events. However, the mLM is doing an remarkable job even in this situation.
We routinely test our global forecasts against more than 50000 weather stations and the maps below indicate if our forecasts were too warm (red) or too cold (blue); light green and yellow are good forecasts. In the top left of the images, we have summarized the performance with mean and RMS errors and with counts at how many locations the forecasts had errors smaller than 1.5, 2 or 3°C at the given forecast hour, shown on the top right of the image. A good forecast is generally defined as being within 2°C errors and mLM mastered this task for almost 80% of the locations 3 days ahead. Current forecast methods (MOS) achieved only 53%.

The verification shows that the heatwave on 26 July 2018 was forecasted one to four days ahead.

Forecast issued on current day (day 1 forecast)

Forecast issued a day earlier (day 2 forecast)

Forecast issued 2 days earlier (day 3 forecast)

Forecast issued 3 days earlier (day 4 forecast)

And of course, mLM works globally!

With this heat, temperatures in cities and sunny places may even rise higher than official measurements. Let us know your opinion.