Application of Machine Learning Towards the Calibration and Forecasting of Marine Renewable Resources

The rising climate crisis, largely driven by fossil fuel dependence, urgently necessitates a global transition to renewable energy sources. However, the successful development and safe operation of offshore energy infrastructure, and their subsequent integration into the power grid, critically depend on accurate and reliable assessments and forecasts of renewable energy sources. Errors in these assessments can lead to significant engineering and economic risks, including structural failures, operational inefficiencies, and underestimated extreme event impacts, thereby endangering offshore structures and personnel. While the European Reanalysis Fifth Generation (ERA5) climate dataset is comprehensive and widely utilised for such evaluations, its accuracy of crucial metocean parameters, especially in complex environments like enclosed and semi-enclosed seas (e.g., the Mediterranean), requires substantial refinement. This limitation poses a direct challenge to ensuring the reliability and safety of offshore infrastructure design and operation. This thesis directly addresses this critical gap by presenting a multi-faceted approach to significantly enhance the utility, accuracy, and reliability of publicly available climate datasets, specifically for safe and efficient renewable energy development. Addressing these limitations, we make several key contributions, leveraging statistical and advanced machine learning methodologies to improve and forecast marine renewable energy resources, specifically wave and wind. First, we develop a statistical approach to calibrate concurrent and collocated ERA5 wave data, achieving closer alignment with groundtruth wave buoy observations. Second, to address the inherent nonlinearity and stochastic nature of wave resources, machine learning algorithms are employed to map ERA5 data to observational wave buoy data. This enables comprehensive calibration and integration of spectral wave parameters across the entire Mediterranean Sea, yielding significantly more reliable information crucial for engineering design, oceanographic research, and the precise estimation of extreme events, even extending calibrated data to areas without buoy observations. Third, we explore advanced deep neural network approaches for wind speed forecasting within potential offshore floating wind farm development areas suitable for floating wind installations. Fourth, we examine the parameterisation of air-sea momentum and energy exchange, essential for understanding and predicting metocean variables. Specifically, we focus on the wind friction velocity, and the sea surface drag coefficient, both fundamental for accurately assessing and modelling wind