Model Output Statistics (MOS)
MOS (Model Output Statistics) is the technique of post-processing the output from numerical weather forecast models using statistics of local historical or actual weather measurements. The MOS service substantially improves the accuracy of forecast and historical data for temperature, humidity, wind speed and radiation (other variables can be specified on request).
Pure statistical models are excellent at “nowcasting” (forecasting short-term local weather) but are usually worthless beyond about six hours, because their accuracy is then lower than that of physical models. The MOS technique combines the model output and statistics, by using the complex numerical models based on the physics of the atmosphere to forecast large-scale weather patterns and then using regression equations in statistical post-processing to clarify surface weather details.
We can provide an estimate of the improvement with MOS, if you send us measurement data of the station(s) concerned from at least 1 year, obtained in or after 2004, preferably from 2008 onwards in .csv format. Measurements can have arbitrary time resolution (10 minutes, 15 minutes, 1h, 3h) as long as it is larger than 1 minute and smaller than 3 hours. The same time series can also change the interval during the measurement period (e.g. 5 months with 10 minute data followed by 7 months with 1h data is also fine). Additionally, we need the names (ID), the coordinates and (if possible) the altitude of the stations. Moreover, we should know, in which height (above ground) the sensors of the variables are located and which time zone is used for the measurement table (data).
STATIONID;latitude (decimal degrees);longitude (decimal degrees);altitude (m asl.)
TIME (UTC) YYYYMMDD hh:mm; wind speed (m/s)
This is an example for 10min data. Hourly data is also fine. Do not attempt to interpolate values in order to close data voids, as this will lead to a decrease in forecast quality. Mark the missing values as -999 or do not include these time steps at all in the file.
The accuracy is generally better than for a pure statistical model or for a pure numerical model (NWP) output.