Eastern Wind Dataset
Here you will find information about the Eastern Wind Dataset including the methodology used to develop the accuracy of the data, site selection, power output, and forecasts.
This dataset was originally created for the Eastern Wind Integration and Transmission Study. These data are modeled data and not actual measured data. Learn more about the datasets including the similarities and differences between the Eastern and Western datasets and the differences from the NREL state wind maps.
Methodology
AWS-Truewind created the Eastern Dataset with oversight and assistance from NREL. The Eastern Dataset consists of three years (2004-2006) of 10-minute wind speed and plant output values for 1,326 simulated wind plants, as well as next-day, six-hour, and four-hour forecasts for each wind plant.
After testing various mesoscale models, AWS-Truewind settled on the MASS v.6.8 model for the generation of the wind data grid for the Eastern Wind Integration and Transmission Study. The model is initialized with input from the NCEP/NCAR Global Reanalysis data set, and assimilates both surface and rawindsonde data. The model used a nested grid scheme, with the final output at a resolution of 2km.
Wind plant locations were determined by using a proprietary AWS-Truewind wind speed map of the study area along with 10 years of speed distributions previously computed by AWS-Truewind. The wind resource at each cell was computed by combining these data sets with a composite turbine power curve. Wind plants were then created by combining cells with close proximity and similar wind characteristics (subject to environmental, topographic and other exclusions). Size and state-by-state distribution of wind plants were controlled to give a diverse selection of plants. (For example, it was necessary to reduce the minimum capacity factor threshold in order to get enough plants in Connecticut, Rhode Island, New Jersey and Delaware.)
The wind speed and power output time series for each wind plant were computed by combining the mesoscale grid output with the composite turbine power curves, the list of cells contained in each plant, and other data. Adjustments were made for model biases, wake losses, the impact of gusts, availability and other factors.
AWS-Truewind has completed a final report describing the overall project detailing the modeling and inputs.
Site Selection
Land-based sites are actually simulated wind plants, composed of many nearby grid points that have similar wind characteristics. Site sizes range from 5 to 160 km2, and maximum power output ranges from 100 to 1435 MW. (Data from the individual grid points and the shape and layout of the sites are NOT available.) The site screening process excluded areas of open water, wetlands, parks, areas of steep slope, and non-public federal lands. Airports and developed areas were also excluded, along with a buffer zone around each such area.
The final site list contains 1326 sites totaling 580 GW. The bulk of the sites fall between 100 MW and 600 MW in size. A smaller number (150) of "megasites" with rated capacities exceeding 1000 MW were also chosen. All of these are in the Great Plains. A separate screening with a lower capacity factor threshold was done for Connecticut, Rhode Island, New Jersey, and Delaware, with 30 sites selected in these states.
Offshore sites were chosen from a 2-km grid, where each grid point represented 20 MW of offshore wind capacity. Selected grid points were at least 8 km from shore and in water no deeper than 30 m. A total of 4948 sites in the Atlantic Ocean and 4 of the 5 Great Lakes were selected.
Wind Plant Output
Wind plant output time-series values were computed by adding up the contributions of each grid cell in the site. Three composite power curves (computed by averaging 2 or 3 power curves from commercial turbines) were available, and the choice of power curve was based on the average wind speed at the site. Adjustments were made for model biases, wake losses, wind gusts, turbine and plant availability, and other factors.
Forecasts
AWS Truewind produced hourly forecasts for three different time horizons: next-day, six-hour, and four-hour. Each set of forecasts was synthesized by running a statistical forecast synthesis tool written by AWS Truewind called SynForecast. This tool uses actual forecasts and observed plant output to develop a set of transition probabilities, which are then applied stepping forward in time from a random starting point in a process known as a Markov chain.
