Skip to main content

meteoblue Learning Multimodel (mLM) Forecast


For operators of weather stations, meteoblue offers station specific forecast, which are optimized on the measurements of the specific weather station. To reach the highest accuracy level, the forecast needs to be adapted to the station specific conditions, using its measurement data. Therefore, the measurements must be transferred to meteoblue at least four times a day, to integrate the measurements into the meteoblue forecast pipeline via nowcasting. meteoblue uses the measurements only for internal purposes to respect the property rights of the owner of the data, which will be specified in an individual business agreement. Based on the measurements of the last months, our proprietary artificial intelligence finds the best model combination for each weather station, adapted to the microclimatic situation and according to the actual weather conditions.

Weather station measurements should be provided to meteoblue through an API access via http call. The data of all stations should be delivered in only one file in a readable text format (e.g. TXT, CSV, JSON) with a unique StationID for each station.

Adding new stations obviously shouldn't change the file format. In consequence, once the export is set up successfully, it should be very safe to scale up to more stations. The mLM still delivers forecast data if your measurement data delivery fails for 2-3 days.

For measurement data delivery from our customers station data, meteoblue proposes the following procedure and format. meteoblue will download the file from HTTP or FTP server or make an API-call to your system.

Ideally, the most current observations from all stations are submitted in one file every hour. More time steps can be included, e.g. the last 3 hours of observations from all available stations. This is generally a good strategy as the data transfer becomes more robust to compensate for missing data. Double submission of the same values does not cause any problems.

You can check your file on our website here.

The preferred data format is a simple CSV file:


  • StationID, where possible, shall only contain numbers and capital letters and have a length of approximately 8 signs
  • The number of decimals of the float can vary arbitrary for any single entry
  • All stations should be submitted in one file or able to be accessed with one call
  • The order of stations and time step can be arbitrary. It is not necessary to group all time steps from the same station. It is not necessary to have any order in the time steps.
  • Missing data values have to be set to -999.0
  • The columns in the csv file should be named exactly like described in samples below:


MYID3423,2020-12-31 23:58,45.33,-107.234234,432.43,22.8,12.4,320.3,55.5,-999.0,0.02
MYSTATION223,2020-12-30 22:18,25.33,17.234234,432.43,22.8,13.4,320.3,55.5,-999.0,-999.0
MYID3423,2020-12-31 22:18,45.33,-107.234234,432.43,-999,15.4,320.3,55.5,-999.0,-999.0

STATIONIDstringUnique identifier of the station, this should never change
TIMEstringUTC-TIME YYYY-MM-DD hh:mmSpecifying when the observation was taken
LATfloat° (WGS84 / GPS coordinates)Latitude of the station in decimal degrees, negative values for South (range -90 to 90)
LONfloat° (WGS84 / GPS coordinates)Longitude of the station in decimal degrees, negative values for West (range -180 to 180)
HEIGHTfloatm aslAltitude above sea level of the station, this is not the measurement height above ground
TEMPERATUREfloat°CAir temperature
WINDSPEEDfloatm/sWind Speed
WINDDIRECTIONfloat°Meteorological standard convention: 0° = wind coming from North; 90° = wind blowing from the east; 180 = wind coming from South
RELATIVEHUMIDITYfloat%relative humidity
DEWPOINTfloat°Cdewpoint temperature
PRECIPITATIONfloatmmprecipitation within the last hour