Downscaling global reanalyses with WRF for wind energy resource assessment

Meteorological reanalysis datasets are increasingly being used, in conjunction with mesoscale models such as WRF, to provide a long-term (several decades) view of the potential wind energy resource at a site that is under consideration for wind development. These multi-decadal 3D gridded global datasets of basic meteorological variables can be downscaled with WRF to a few km spatial resolution, to provide long-term, high spatial resolution fields of hub-height wind speed at any prospective wind farm on the globe. This presentation will give an overview of the considerations and strategies used in downscaling reanalysis data sets for wind energy resource assessment. In addition, in recent years, several leading operational weather forecast centers have developed new reanalysis datasets that offer potential improvement over the work-horse NCEP/ NCAR Reanalysis Project (NNRP) dataset that was developed in the mid-1990s, and has been used widely in the renewable energy industry. Improvements in new reanalysis datasets include better representation of physical processes and finer grid resolution in the underlying global model, as well as more sophisticated data assimilation techniques. However, it has yet to be confirmed whether the improvements in the production methods of the new reanalysis datasets translate into improved performance for use in wind resource assessment studies. This study will directly address this question, by comparing the performance of three newer reanalysis datasets to that of NNRP: the Climate Forecast System Reanalysis (CFSR) developed at NCEP; the Modern-Era Retrospective analysis for Research and Applications (MERRA) developed at the National Aeronautics and Space Administration (NASA); and the ECMWF Interim Reanalysis (ERA-Interim) developed at the European Centre for Medium Range Weather Forecasts (ECMWF). These comparisons will examine the performance of hub-height wind speeds pulled directly from the raw datasets themselves, as well as the performances of downscaled versions of the datasets using WRF.