Calibration of MetCoOp-based ensemble forecasts for aggregated wind power in Finland

The amount of installed wind power capacity has grown in Finland over the last 10 years. According to statistics from Renewables Finland, there were 2,002 wind turbines installed with a total capacity of 9,433 MW by the end of 2025. Electricity statistics from Finnish Energy show that wind power amounted to 26% of the total Finnish electricity production during 2025, which makes wind power an important energy source. Since wind power production by nature is volatile due to natural variations of wind, and has a big share of total electricity production, it is important to have both accurate and reliable forecasts for wind power. 

The wind power forecast model at the Finnish Meteorological Institute is based on forecast data from the MetCoOp ensemble prediction system (MEPS). MEPS is based on the limited area model Harmonie-AROME and covers Northern Europe with a horizontal resolution of 2.5 km. Every hour a five-member ensemble is run with a maximum forecast lead time of 66h. The output of MEPS is used as the input to the wind power model. The wind power model further considers the locations of the wind turbines and information about their heights and power curves when computing an aggregated wind power forecast. The wind power forecast model is computed for all ensemble members and for all forecast lead times. After this, a 30-member lagged ensemble is formed by collecting the forecast output from five members over a six-hour time window.  

In this research, we study ensemble wind power forecasts for aggregated wind power in Finland and how to calibrate and post-process wind power forecasts with the aim of having accurate and reliable probabilistic forecasts. We use statistical post-processing tools, such as ensemble model output statistics (EMOS), and study how to adapt it to a lagged 30-member ensemble. As observations we use statistics on wind power production provided by Finnish Energy and Finland’s transmission system operator Fingrid. A total of two years of data is used in this study. We will present the results from our on-going study on probabilistic wind power forecasts for Finland.