Research project P4S/251/MACH-PP (Research action P4S)
What’s postprocessing ?
Postprocessing plays an important role in meteorology by significantly improving the accuracy, reliability, and usability of weather forecasts. These techniques are essential for reducing systematic biases and errors inherent in numerical weather prediction (NWP) models. They transform raw model outputs into more precise and actionable weather information by integrating multiple data sources to compensate for the models limitations. By quantifying forecast uncertainty and refining predictions, postprocessing enables meteorologists to provide more reliable and locally relevant forecasts that better match observed weather patterns. These methods are particularly valuable for extreme weather events, and in adapting predictions to specific applications such as renewable energy, aviation, and agriculture. As weather prediction technologies continue to advance, postprocessing remains a critical tool in bridging the gap between complex computational models and the practical need for accurate, localised, and actionable weather forecasts.
The project
For more than 20 years, the Royal Meteorological Institute of Belgium (RMI) has been at the forefront of research for the efficient postprocessing of weather forecasts using statistical methods. However, this was only recently transposed to its operational system, with the postprocessing of the European Centre for Medium-Range Weather Forecasts (ECMWF) medium-range forecasts. The challenges for RMIB in this domain are numerous and the room for improvement is enormous. The MACHPP project aims to push this effort to the next level, by bringing Machine Learning methods into the processes.
The project will first continue the development of statistical postprocessing tools for operational implementation RMI, in particular for all surface parameters used by the weather office, with the aim of obtaining gridded and station corrections. The tools will range from classical techniques to Machine Learning-based methods and will be applied to the forecasts of ECMWF.
The project will also investigate the postprocessing of the in-house prediction model Alaro, using all of the data sources available, and relying on new Machine Learning techniques. The benefits of its application would be to improve the local weather forecasts, which are particularly valuable in the case of extreme precipitation events.
Specific meteorological variables such as solar radiation and wind speed at wind turbine heights to improve downstream applications in the renewable energy sector will also be the target of these improvements. This includes the improvement of operational services such as the Storm Forecast Tool for the Belgian Offshore Zone developed by RMI for Elia, and new applications such as wind power ramp forecasting and prediction of Dunkelflautes events (little or no wind and solar irradiation for long periods of time).
