The meteoblue learning multimodel (mLM) is a new technique of post-processing the output from numerical weather forecast models using actual weather measurements. The mLM reads actual weather measurement data and selects the best simulation model for making a forecast.
For temperature, the model accuracy of the mLM was tested for one year before introduction in August 2018.
The operational model accuracy of the mLM was verified for a period of two months (September and October 2018) on more than 30'000 meteorological stations worldwide. This analysis shows a model accuracy of 1.2 K for the 24-hour ahead (24h) forecast and a model accuracy of 2.0 K for the 6 day forecast. The model accuracy of the mLM 24h forecast is significantly better than established standards:
- 0.8 K better than using 'stand-alone' numerical weather forecast models (24h forecast).
- 0.3 K better than models simulations using MOS .
- 0.3 K better than the reanalysis model ERA5 (which uses measurements for model correction).
We could demonstrate that the 6-day air temperature forecast of the mLM is as good as the 24h forecast of 'stand-alone' numerical weather forecast models. This improvement corresponds to the average improvement achieved by weather forecast every 10 years over the past decades.
Validation results for the other variables where mLM is implemented already will be added.
For further information, please refer to the technical documentation.