Simulating wind power from local to Continental scale: A data and Model comparison for Brazil

Studying the possibilities of integration of higher shares of renewables into the energy production matrix requires long time series of renewable power generation in order to assess diurnal, seasonal and inter-annual fluctuations in power generation. Timeseries should possibly also be highly resolved spatially to account for resources in different locations.

Reanalysis data are a freely accessible source for climate data which can be used to simulate time series of several decades of renewable power generation – but do these time series match with actual power generation?

In our new paper in Energy, we have simulated and validated Brazilian wind power generation from MERRA-2 on the spatial scale of windparks, states, subsystems and the whole country with a turbine specific power curve model from Ryberg et al.

As the spatial resolution of MERRA-2 data is rather coarse with about 50 km, different approaches were tested to increase the spatial resolution. Horizontal interpolation methods had low influence on model quality. A further attempt to downscale the data was using the Global Wind Atlas (i.e. adapting the mean wind speed to the GWA) and comparing it to another approach using locally measured wind speeds. Both data sources helped increase the simulation quality (lower RMSE) in many cases, however on wind park level data quality of locally measured wind speeds may produce outliers. Also temporal bias correction was attempted using the measured wind speeds, but it had only little impact on the results, as it was hardly applied due to limits in data availability.

In general results looked good: For Brazil and state level correlations > 90% were achieved with some exceptions only. For single windparks correlations are lower (range of 60%-80%). The knowledge that the Global Wind Atlas can compete with local measurements in terms of improving model quality opens opportunities for developing a global model.

Not shown in the paper, but also relevant: In an original approach a standard turbine model was used to convert wind speeds into power which was updated to Ryberg et al.'s turbine specific models. This helped increase correlation (slightly) and decrease RMSE (slightly, in most cases) compared to assuming one generic turbine model (Vestas V80, Enercon E82 or Siemens SWT2.3). See figure below.

The paper is online behind a paywall, but the accepted manuscript can be found on arxiv. The corresponding code is available here and the resulting data sets here.