MET Nordic dataset: post-processing of model output near-surface fields

MET Nordic is a dataset created by the Norwegian Meteorological Institute (MET Norway), providing  near-surface variables for Scandinavia, Finland, and the Baltic countries with a resolution of 1 km. This dataset is available through two distinct production streams: MET Nordic RT designed for real-time data provision to applications, and Met Nordic LTC aimed at supporting applications that require consistent long-term data.The dataset includes the following near-surface variables: temperature at two metres, precipitation, sea-level air pressure, relative humidity, wind speed and direction, solar global radiation, long-wave downwelling radiation, and cloud area fraction.

MET Nordic RT provides updated products every hour with a 20 minute delay, and it has an archive that goes back to 2012. The dataset consists of post-processed products that (a) describe the current and past weather (analyses), and (b) give our best estimate of the weather in the future (forecasts). The products integrate output from MetCoOp Ensemble Prediction System (MEPS) as well as measurements from various observational sources, including crowdsourced weather stations. These products are deterministic, that is they contain only a single realisation of the weather. The forecast product forms the basis for the forecasts on Yr (https://www.yr.no). Both analyses and forecasts are freely available for download in NetCDF format.

For temperature and precipitation, the model output is combined with unconventional observations, such as data from citizen weather stations. Their inclusion shows a clear improvement to the accuracy of short-term temperature forecasts, especially in areas where existing professional stations are sparse. In this study, we will summarise the results obtained with the post-processing and we will share the main lessons learned, which can also be useful for systems that want to use these observations for data assimilation.

The MET Nordic LTC is currently in an experimental phase and undergoing significant modifications, more than the RT stream. The primary goal of LTC is to extend the temporal coverage of the variables provided by RT, possibly back to 1961, with an emphasis on achieving time consistency—a feature not prioritised in the RT dataset. The approach involves applying post-processing techniques similar to those used for RT but with notable distinctions. Firstly, we utilise a reanalysis dataset, such as the 3-km Norwegian Reanalysis (NORA3),  rather than outputs from numerical weather predictions. Secondly, we are developing methods to incorporate observational data in a way that maintains the time consistency of the dataset. This may necessitate using only a selection of high-quality observations that are available over extended periods.