Probabilistic Solar Power Forecasts Using a Large Ensemble

At last year’s annual meeting, we reported on solar power forecast uncertainty information from forecasts we provided for various probability of exceedances, probability bins, and measures of subhourly variability. We are extending this effort in a big way under a DOE-funded joint project, Solar Uncertainty Management and Mitigation for Exceptional Reliability in Grid Operations (SUMMER-GO), in which forecasts of the full probability distribution will be passed through decision support tools developed by the National Renewable Energy Laboratory (NREL) and integrated operations at the Electric Reliability Council of Texas (ERCOT). The tails of the distribution are particularly important because they represent the rare but high-risk scenarios which could be costly to electric grid managers if not mitigated against ahead of time. To better resolve the tails, a very large ensemble will be utilized. However, the available NWP sources which could be brought together have widely disparate temporal and spatial scales; the forecasts for these must be made congruent. Also, a well calibrated ensemble will need to account for such factors as the propensity for almost all current NWP models to be too sunny, partly due to mixing out low clouds beneath stable layers too quickly in the model forecast. This presentation will focus on our development of a sharp and calibrated multi-model ensemble for SUMMER-GO.