Suitability of Reanalysis Data for Wind Plant Revenue Estimation

Reanalysis datasets such as MERRA-2 and ERA5 are often used as long-term reference datasets in wind resource assessment. Reanalysis data are available at consistent spatial and temporal intervals across the globe and extend back in time by several decades. In order to use reanalysis data to estimate the expected long-term wind resource at a particular site, modeled data are correlated to on-site observations and corrected in a method known as measure-correlate-predict (MCP). However, even when MCP methods are applied to correct modeled plant generation at the monthly or annual level, significant errors can remain at the hourly level. Since power prices fluctuate on an hourly or sub-hourly basis and are often negatively correlated to wind generation, minor errors in capturing a wind farm’s diurnal production profile can result in large errors in the estimation of plant revenue.

In this presentation, observations from meteorological towers across the U.S. are used to identify time periods of systematic reanalysis wind speed bias at the hourly level. Observed and modeled wind speeds are passed through a theoretical wind plant power curve and paired with hourly power prices to demonstrate the effects of these biases on a wind plant’s merchant revenue estimation. Results indicate that reanalysis wind speeds consistently have significant errors during transitional periods around sunrise and sunset. These errors are particularly impactful during the fall and winter, when the period of largest reanalysis bias coincides with the secondary price peak around sunset. In addition, reanalysis datasets are unable to capture the diurnal variability of seabreeze effects, which leads to a significant underestimation of revenue for coastal projects.