Meteorological Analysis of Large Power-Error Events in the Wind Forecast Improvement Project (Wfip) Data Set

Electric grid and utility operators rely on numerical weather prediction model forecasts of power production from wind plants. Of greatest concern are instances when the models either greatly over-forecast or greatly under-forecast the amount of power that a wind plant or group of wind plants will produce. We use one year of observation and model data from the Wind Forecast Improvement Project (WFIP) to analyze the characteristics of large power error events. WFIP was a joint DOE, NOAA, and private sector field campaign whose goal was to improve wind forecasts for the wind energy industry. Observations were collected during WFIP for a 12 month period in 2011-2012 in two study areas within the United States, one in the Northern Great Plains, and one in west Texas.

Large error events are defined on the basis of the 3 hour running mean of the difference between forecast and observed power, aggregated within each of the two WFIP study areas. Using 6 hour forecasts from the 3 km resolution NOAA/High Resolution Rapid Refresh (HRRR) model, and observed pseudo-power derived from anemometers on 127 tall tower sites in the two WFIP study areas, we find 27 events in the North Study Area with aggregate error greater than 20% of capacity, and 31 events in the South Study Area with aggregate error greater than 30% of capacity over the WFIP year-long field program. These events are categorized according to season, meteorological condition, being an over- or under-forecast, and length of the event. We also compare statistics to large errors calculated using the 13km resolution NOAA/Rapid Refresh model (RAP), and to errors calculated for individual tower locations (instead of the aggregated power).