Enhancing Offshore Wind Farm Met-Ocean Data Accessibility: A Machine Learning Approach With Satellite-Derived Wave Measurements in the Celtic Sea

Marine operations are a significant expense for offshore wind farms, representing up to one third of total project costs. An improved understanding of the variation of met-ocean conditions across a wind farm site offers the potential to reduce weather downtime and associated costs. This work employs a machine learning approach utilising a surrogate wave model trained on the relationship between the wave conditions at discrete measurement locations to wave conditions across the entire model domain. The surrogate model can then be run with real-time data inputs from the discrete measurement locations to provide a spatial dataset for waves, without the high computational power needed to run the physics-based wave model itself. This new method enhances the accessibility of met-ocean data to allow more informed decision making for the installation, operation, and maintenance of offshore wind farms. The approach has already proven successful with fixed measurement buoys, and work is ongoing to adapt the modelling framework to use satellite-derived wave data as an input. With freely available global coverage, satellite data is a useful complementary data source to wave buoy data. Several Earth Observation satellite missions host radar altimeters that report significant wave height along the satellite’s ground track. The first step towards utilising radar altimeter data with the machine learning framework is assessing the impact of using only significant wave height data as measurement inputs. This paper compares the model outputs from running the model with wave height, period, and direction data, and with wave height data only. The results show that running the model with wave height data only produces a small reduction in the accuracy of output wave predictions in coastal areas.