6.6 Assessing the Fidelity of the North American Wind Climate and Impacts of Wind Farms Using High-Resolution Modeling

Wednesday, 10 January 2018: 2:45 PM
Room 16AB (ACC) (Austin, Texas)
T. J. Shepherd, Cornell Univ., Ithaca, NY; and R. J. Barthelmie and S. C. Pryor

The installed wind energy capacity of the US stands at 84 GW (July 2017) and thus wind energy contributes over 6% of the nations’ electricity. This fraction is expected to continue to rise rapidly to exceed 20% by 2030, requiring quadrupling of installed capacity to reach 300 GW. The cost of wind energy is competitive with conventional electricity generation but the cost per kWh is determined by the wind resource and the uncertainty of its prediction in both the long and the short term. Accurate quantification of the wind resource and annual energy production presents a significant challenge to current numerical models, and even small uncertainties in modeled wind speeds can significantly impact the cost of investment capital. Numerical models have also been used to investigate the potential of large wind farms’ inadvertent modification of local and regional climate. The results are highly dependent on study assumptions and direct observations have indicated small to negligible meteorological impacts.

The aims of this research are therefore to assess how skillful numerical weather prediction (NWP) models are as a research tool for simulating the wind climate, and to address whether, using realistic scenarios, there is inadvertent weather modification on local and regional climate scales as a result of wind farm development in the US. Specific research questions addressed in this presentation are:

  1. How well does the Weather Research and Forecasting (WRF) model simulate aspects of the wind climate of relevance to the wind energy industry?
  2. What impacts, if any, do wind farms have on the surrounding local atmosphere immediately downstream of major wind deployments (i.e. adjacent grid cells) and on regional climate scales?
  3. Can these impacts be characterized temporally in terms of diurnal and/or seasonal effects?

A series of simulations using WRF v3.8.1 for the entire calendar year of 2008 have been completed. The year 2008 was selected for an initial suite of simulations as it was a good representative year for near neutral climate conditions. Indeed, after a moderate La-Nina in the first half of the year, ENSO-neutral conditions developed by July-August. In the continental US, air temperature was only 0.1 C above the twentieth-century (1901 – 2000) mean.

The WRF model domain in this study encompasses the eastern two-thirds of North America from 100° W. Following a full simulation without any wind turbines, the simulations were repeated using the Fitch wind farm parameterization and explicit 2014 wind turbine locations and rated capacity. The 20 most-frequently used wind turbine types were parameterized explicitly using the manufacturers’ power and thrust curves while the remainder used an average power curve scaled to the rated capacity for each turbine.

The outer model domain resolution is 12 km with 320 × 320 cells, with a nested 4 km domain of 676 × 676 cells. This nested domain resolution captures the wind climate for most of the eastern US at high resolution, and is also appropriate for operation of the Fitch wind farm parameterization. There are 41 vertical levels up to a model top of 50 hPa, 18 of these levels are in the first 1 km of the atmosphere, to suitably capture the planetary boundary layer (PBL). The lateral boundary conditions are updated 6 hourly, with ERA-Interim data, and Real Time Global (RTG) SST analyses provide initial conditions. The key physics settings include the Eta microphysics scheme, rapid radiative transfer scheme for longwave radiation, and Dudhia for shortwave, Revised Monin-Obhukov similarity scheme for the surface layer physics, the Noah land surface model, Mellor-Yamada-Nakanishi-Niino PBL scheme, and the Kain-Fritsch cumulus parameterization. Convection is explicitly resolved in the 4 km domain.

The fidelity of the simulated wind climate has been primarily assessed against 5-minute wind speeds at 10-m a.g.l. from over 500 sonic anemometers deployed as part of the National Weather Service Automated Surface Observing System (ASOS). The following statistical metrics are computed for wind speeds in each grid cell containing an ASOS station:

  1. RMSD (root mean squared difference between wind speeds from WRF versus those from the in situ measurements).
  2. MAD (mean absolute difference)
  3. Mean bias.
  4. Temporal correlation coefficient (r)

Initial results imply relatively high model skill (low RMSD, bias and MAD and high r) except in the western portion of the domain (Central Great Plains).

To evaluate the impact of current wind farm deployments, we compare the no-wind turbine simulation directly against the simulation including wind turbines using statistical metrics 1-3. We analyze the 4 km domain 10-m wind speed, 2-m air temperature, 2-m specific humidity, turbulent kinetic energy (TKE) budget, near hub-height wind speed, PBL height, annual precipitation total, and soil moisture. For each 4 × 4 km grid cell we compute the performance metrics where the no-wind turbine simulation is set as the ‘target’, to effectively allow direct assessment of the impacts of wind farms on the local and regional climate i.e. according to WRF with the Fitch parameterization enabled, this is the impact of wind turbines on near-surface climate.

Results of further simulations of 2008 with doubled installed wind turbine capacity (to 62 GW in the domain), and with quadrupled wind turbine installed capacity (125 GW) will also be reported, along with a decadal (2001 – 2010) simulation of the current climate, and a simulation of a future climate (2070 – 2080) designed to quantify the potential for reduction/enhancement of the wind energy resource under climate change.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner