Ensemble forecasting for wind power prediction in Northwestern China

By the end of 2011, China had a cumulative wind energy capacity of more than 62 GW, making China the largest wind power provider in the world. China plans to increase its wind energy capacity to 100 GW by 2015, to meet ~11.4% of its energy needs from non-fossil fuel sources. In order to effectively and safely integrate wind power into the China State Grid electric grid system, accurate prediction of wind power is essential. Numerical weather prediction (NWP) models are able to simulate the evolution of weather systems that generate winds, making them an indispensable tool for wind power prediction. Many of the Chinese wind farms are located in Northwestern China where weather observations are sparse and complex terrain dominates. In such situation, forecast from a single NWP model realization contains much uncertainty and can be very inaccurate. An ensemble of NWP forecasts that samples and propagates the uncertainties in the model and observation data presents exceptional benefit to providing more reliable forecast, and associated highly-needed forecast uncertainties.

In collaboration with the China Electric Power Research Institute (CEPRI), the Research Applications Laboratory (RAL) at NCAR has developed a multi-model, multi-physics, and multiple perturbations, rapid-cycling, real-time ensemble weather forecast system with a built-in continuous 4-dimensional data assimilation capability. The system is expected to provide real-time operational weather forecast in support of wind energy forecast for two major wind farm clusters in Northwestern China by early 2013. The ensemble system covers China and a large portion of the Asian landmass with a nested grid configuration down to a horizontal grid spacing of 2.7 km over the wind farm clusters. The ensemble model includes WRF-ARW, WRF-NMM and MM5 model cores, perturbations from different cumulus parameterizations, microphysics, planetary boundary layer schemes, perturbations from different initial/boundary conditions constructed using NCEP GFS, Canadian GEM, and Japanese GSM model forecasts, perturbations to observations and data assimilation parameters, and an ensemble Kalman filter based initial condition perturbation scheme. We will show results from error statistics of real-time ensemble forecast as well as case studies.