New Scientific Paper: Improved Local Weather Forecasting Using Machine Learning

ISLANDER partner elmy has published the open access paper (awaiting peer-review) “Improved local weather forecasting using machine learning”.

The study focuses on enhancing weather forecasts for the German North Sea island of Borkum, targeting three key variables: air temperature, solar irradiance, and wind speed. These forecasts are essential for managing energy demand and supply in island systems.

Using over four years of historical data (2018–2022) from DWD, Météo-France, and ECMWF, the research compares numerical weather prediction (NWP) models with in-situ measurements. It introduces machine learning models that significantly outperform traditional forecasts:

• Wind speed: 35–50% improvement in mean absolute error (MAE)
• Air temperature: 12–30% MAE reduction
• Solar irradiance: 9–12% MAE reduction

The paper also evaluates dynamic weighting of third-party forecasts, which showed only marginal gains.

📄 Read the full publication: https://open-research-europe.ec.europa.eu/articles/5-271/v1