A virtual weather station provides the most accurate weather data for any location worldwide on an hourly basis without any delay. It therefore combines multiple data sources in real time with sophisticated state of the art technologies:
Machine learning: Site specific combination of different weather data sources
Downscaling: Elevation correction to the actual position
Nowcasting: Real time correction with observations and measurements
The data from the closest measuring stations or observations are interpolated respecting microclimatic conditions (downscaling) and, if necessary, supplemented by simulation or forecast data.
The exact composition of the values depends on the location of the selected point (in relation to the weather stations):
If one or more nearby and representative weather stations provide data, these are used. The greater the distance, the more simulation data is used. If no measurement data is available in the vicinity, forecast values are used (multi-model).
Furthermore, for variables like clouds, precipitation and solar radiation, the recent observation from satellite images and radar networks are implemented to achieve the highest accuracy.
Therefore, the accuracy depends on the availability of measurement and observation data (like reanalysis datasets like ERA5), which is best in Europe and North America. Our quality verification shows that mLM forecast already achieve higher accuracy than ERA5, which is further reduced by nowcasting (especially for precipitation and solar radiation).
To set up a virtual weather station, the forecast package API is called every hour and the most recent data is stored as a time series, which requires a special caching license for long term storage.