Advancing Solar Irradiance Nowcasts on Long Island: Blending WRF-Solar with Observations

Solar power is increasingly being generated by a mix of distributed and utility-scale photovoltaic (PV) installations. The growing market penetration of PV has consequences for system load and grid stability, so forecasting and planning for generation is becoming increasingly important, including on nowcasting time scales of minutes to hours ahead. A key component of forecasting power generation is to have a good forecast of solar irradiance. Numerical weather prediction models like WRF-Solar®, which are specially tuned for solar power forecasts, have proven to be a valuable tool in producing such irradiance forecasts.

The Electric Power Research Institute (EPRI), working with utility partners in New York and other researchers and stakeholders, has developed a large collaborative effort focused on deploying and demonstrating advanced solar forecasting techniques in New York state. As part of this project, Brookhaven National Laboratory (BNL) has developed a system of high-definition (HD) sky imagers and software to support 0–30-minute nowcasts of cloud motion. The first set of cameras are deployed near BNL on Long Island, with eventual deployment planned for other sites across New York.

In this phase of the project we at the National Center for Atmospheric Research (NCAR) have developed a 1-year dataset of 6-h reforecasts with WRF-Solar over New York at 3-km grid spacing. Reforecasts are initialized several times per day to cover the daytime hours statewide and year-round. To improve upon the raw WRF-Solar irradiance forecast, we test various machine learning techniques, including random forest, gradient boosted regression, k-nearest neighbors, and Cubist, to blend meteorological and solar observations at BNL with WRF-Solar output. One year of model and observation data allows for testing of training period length, as well as assessing performance across multiple seasons. In future phases, the modeling and blending will be scaled up to a statewide and real-time system.