A Data-Driven Approach to Offshore Wind Forecasting in the Celtic Sea

Abstract Accurate weather forecasts are critical for various industries - including offshore wind - which, has a significant part to play in realizing global net zero energy goals. Traditionally, such forecasts have been made using physics-based numerical weather prediction techniques, however, recently, machine learning models, trained on historical data, have shown promise in learning patterns not always represented by discretized physical equations and therefore have the potential to enhance the accuracy and efficiency of forecasts produced. This paper applies a low-cost machine learning-based framework (MaLCOM) to offshore wind forecasting in the Celtic Sea. It uses an attention-based long short-term memory (LSTM) recurrent neural network (trained on in-situ observations) to learn temporal patterns coupled with a random forest-based spatial nowcast model (trained on the ERA5 reanalysis) for complete spatiotemporal prediction. Winds derived from wave spectra measured by coastal buoys are integrated, showing the performance of the framework even with imperfect data. Validation with independent observations from floating lidar units in 2023 confirms the framework’s potential for regional wind prediction. This work extends previous MaLCOM-based ocean condition predictions to offshore wind forecasting, showcasing new methods for enhancing the value of discrete metocean measurements and improving real-time decision-making for offshore planning.