Forecast performance assessment of a kinematic and a magnetohydrodynamic solar wind model

Performance evaluation of operational models guides forecasters on use of their products, focuses model developers in making improvements, and informs other modelers considering use of the output for forcing data. Two operational solar wind models, the Enlil magnetohydrodynamic model, and the Wang‐Sheeley‐Arge kinematic model, were executed daily in 2007–2011 from solar photosphere magnetograms compiled from the Global Oscillation Network Group telescope system. The original (uncorrected) magnetic field specification and a zero point‐corrected (ZPC) version were used as inner boundary conditions (IBCs) in separate 7 day forecast executions. Forecasts of solar wind radial speed ( V sw ) and interplanetary magnetic field (IMF) polarity were compared with observations from the Advanced Composition Explorer satellite. Forecast verification metrics were computed by forecast day, year, and uncorrected or corrected IBCs. High speed events (HSEs) and IMF polarity changes (IPCs) predicted and observed were compared. Neither model exhibited a significant systematic error except in 2009, when both failed to represent the slow solar wind. Using the ZPC initial conditions resulted in smaller forecast‐observation differences in the years with greater V sw variance. This was due to in part to reduced variance in the V sw predictions from the ZPC IBCs. Differences were nil or worse in the other years. The time‐varying component of the forecast‐observation differences was smallest at forecast days 3 to 5, followed by a sharp rise. Impact of ZPCs on IMF polarity predictions was small. HSE prediction performance depended on detection algorithm used. Both models under predicted the number of forecast periods having IPCs.