Offshore Wind Farm Annual Energy Production Sensitivity To Model Assumptions

Offshore wind farm planning and layout design is critically contingent on accurate wind resource estimates and assessment of likely energy loss to wind turbine and whole wind farm wakes. Mesoscale models such as the Weather Research and Forecasting (WRF) model can represent the spatial variability in the flow field and thus the resource and operating conditions and can also employ simplified models of wind turbine aerodynamics to represent wind turbine wakes that can be propagated over many tens of kilometers. Engineering and other microscale models can more accurately represent the aerodynamics and wake generation processes and merging/dissipation of wakes from individual wind turbines but typically do not capture either the flow-field spatial variability or the whole-wind farm wake effects. Further, simple engineering models underestimate wake losses in large offshore wind farms because they do not consider atmospheric processes that start to limit the downward transfer of momentum into the wind farm. Thus a ‘scale’ gap exists which makes modeling of large wind farms or effective planning of lease areas very challenging. Scientific objectives of this research are to quantify uncertainties arising from the utilization of different datasets or parameterizations in the modeling process and to assess whether the limitations of the engineering models in terms of the flow-field spatial variability can be addressed.

The focus of the work is the US east coast Massachusetts/Rhode Island (MA/RI) group of eight lease areas that covers a total area of 3675 km² and the US West coast Humboldt lease area off California that covers 536 km². A number of uncertainties in modeling of Annual Energy Production (AEP) are explored here in the context of these lease areas using the PyWake platform. In these simulations it is assumed that in each lease area wind turbines are deployed on a uniform east-west north-south grid with a 1.85 km spacing. For a given lease area, the assumption is used that the (irregular) shape is filled with as many turbines as possible starting at the northeast corner and adding turbines column by column. For the Humboldt lease area this method gives 151 wind turbines while for MA/RI it is over 1000. Further the International Energy Agency 15 MW reference wind turbine that has a hub-height of 150 m and rotor diameter of 240 m is employed in all simulations.

• The first source of uncertainty that is explored is the source of the wind climate at the center of the lease area expressed in terms of the sector-wise Weibull A and k parameters. Here we sample across five different sources of the wind climate; (i) ERA5 hourly wind speed and direction at 100-m AGL computed over the 40-year period 1979-2018 and corrected to the wind turbine hub-height. (ii) 10-minute wind speed and direction at wind turbine hub-height from short-term deployments of buoy-mounted conically scanning lidars. (iii) climatologically reconstructed hub-height wind speed and direction derived from (ii) using a National Data Center buoy measurements (at 4-m height) for 1982-2022 and (iv) 10-minute output of wind speed and direction from simulations with the Weather Research and Forecasting (WRF) model.
• The second source of uncertainty that is explored is the wake model. Here we sample across six different wake models; (i) Niels Otto Jensen (NOJ), (ii) Fuga, (iii) FugaBlockage, (iv) TurboGaus, (v) GCL and (vi) BastankhGaussian. These models differ in terms of the complexity of the treatment of the individual wake expansion, the treatment of wake merging and whether or not they include losses due to the wind farm blockage effect.
• The third source of uncertainty is the spatial variability of flow conditions and the wind resource at 150 m AGL across the lease areas as sampled using WRF output at a grid spacing of 2 km by 2 km.

As an example of the first two uncertainties, for the Humboldt lease area:

• In the absence of wake losses the modeled AEP derived using the long-term (20 year) WRF simulation output is 11,459 GWh/year. However, using a wind climate derived from the ERA5 reanalysis yields an AEP estimate that is 8.9% lower. Similarly using the NCDC buoy long-term (1982-2022) dataset that has been corrected to hub-height, AEP is similar to that derived using the ERA5 reanalysis, 7.1% higher than using WRF.
• Using the six different wake models and the WRF wind resource yields wake losses that range from 3.7 to 14.1%. Using the climate-corrected lidar buoy data yields wake losses of 4.4 to 16.5%. Using ERA5 yields values of 4.9 to 17.7%.

This presentation will describe the robust framework that can be used to quantify uncertainty in AEP projections from offshore wind farms and to highlight and prioritize areas for investment to reduce that uncertainty.