Overview and Meteorological Validation of the Wind Integration National Dataset toolkit
Regional wind integration studies in the United States require detailed wind power output data at many locations to perform simulations of how the power system will operate under highpenetration scenarios. The wind data sets that serve as inputs to these studies must realistically reflect the ramping characteristics, spatial and temporal correlations, and capacity factors of the simulated wind plants, as well as be time-synchronized with available load profiles. The Wind Integration National
... gration National Dataset (WIND) Toolkit described in this report fulfills these requirements, and constitutes a state-of-the-art national wind resource data set covering the contiguous United States from 2007 to 2013 for use in a variety of next-generation wind integration analyses and wind power planning. The toolkit is a wind resource data set, wind forecast data set, and wind power production and forecast data set derived from the Weather Research and Forecasting (WRF) numerical weather prediction model. WIND Toolkit data are available online for over 116,000 land-based and 10,000 offshore sites representing existing and potential wind facilities. The WIND Toolkit wind resource data was generated on a 2-kilometer (km) by 2-km grid with a 20-meter (m) resolution from the ground to 160 m above ground, and includes meteorological and power data every 5 minutes. A state-of-the-art forecast data set was also created on a 6-km grid at 1-hour, 4-hour, 6-hour, and day-ahead forecast horizons using industry best practices. During this process, a team of developers focused on mimicking state-of-the art forecast accuracy. The power data were created using data from actual and hypothetical wind farms for 126,000 land-based and offshore wind power production sites. Barometric pressure, wind speed and direction (at 100 m above ground level), relative humidity, temperature, and air density data is available via an online interface. The conversion from wind to power included wind speed adjustment for wakes with an empirical function, application of power curves using different power curves for offshore and class 1-3 wind sites, and statistical adjustment to power. We used methods that respect the spatio-temporal correlations of typical forecast errors at all delivered horizons. We further applied statistical models at each site for horizons of <= 6h and created probabilistic forecasts using nonparametric error quantiles. Therefore, each power forecast contains a deterministic best-estimate value and the P10/P90 probability of exceedance values. Through this work, the research team discovered that creating and storing many terabytes of multiyear wind resource output data is challenging. As a result, we used parallel asynchronous I/ O (parallel-netcdf combined with WRF Quilt-I/O) to keep pace with the continuous generation of output data resulting from very high spatial and temporal resolutions for a large geographical area (continental United States). This document describes the selections of WRF settings to optimize the model output for wind turbine arrays and includes lessons learned from past projects. In addition, this document shows a comparison of observational data from six tall towers and three buoys with data from the WIND Toolkit, and thus serves as a validation report of the toolkit's meteorological data set. In this context, "validation" is taken to mean: "confirming that the WIND Toolkit meets the needs of a wind integration study by accurately capturing local 6 This report is available at no cost from the National Renewable Energy Laboratory (NREL) at www.nrel.gov/publications. wind conditions and their variability over a wide range of time scales." As described in this report, the WIND Toolkit meteorological data set appears to be an accurate description of the climate and weather at each of the sites that were investigated. Results also indicated that there is no obvious relationship between model performance and terrain. Based on previous experience, the data set is likely to be appropriate for conducting grid integration studies.