Ensemble modeling of coronal mass ejections (CMEs) provides a probabilistic forecast of CME arrival time which includes an estimation of arrival time uncertainty from the spread and distribution of predictions and forecast confidence in the likelihood of CME arrival. The real-time ensemble modeling of CME propagation uses the Wang-Sheeley-Arge (WSA)-ENLIL+Cone model installed at the Community Coordinated Modeling Center (CCMC) and executed in real-time at the CCMC/Space Weather Research Center. The current implementation of this ensemble modeling method evaluates the sensitivity of WSA-ENLIL+Cone model simulations of CME propagation to initial CME parameters. We discuss the results of real-time ensemble simulations for a total of 35 CME events which occurred between January 2013 -July 2014. For the 17 events where the CME was predicted to arrive at Earth, the mean absolute arrival time prediction error was 12.3 hours, which is comparable to the errors reported in other studies. For predictions of CME arrival at Earth the correct rejection rate is 62%, the false-alarm rate is 38%, the correct alarm ratio is 77%, and false alarm ratio is 23%. The arrival time was within the range of the ensemble arrival predictions for 8 out of 17 events. The Brier Score for CME arrival predictions is 0.15 (where a score of 0 on a range of 0 to 1 is a perfect forecast), which indicates that on average, the predicted probability, or likelihood, of CME arrival is fairly accurate. The reliability of ensemble CME arrival predictions is heavily dependent on the initial distribution of CME input parameters (e.g. speed, direction, and width), particularly the median and spread. Preliminary analysis of the probabilistic forecasts suggests undervariability, indicating that these ensembles do not sample a wide enough spread in CME input parameters. Prediction errors can also arise from ambient model parameters, the accuracy of the solar wind background derived from coronal maps, or other model limitations. Finally, predictions of the K P geomagnetic index differ from observed values by less than one for 11 out of 17 of the ensembles and K P prediction errors computed from the mean predicted K P show a mean absolute error of 1.3.

Modeling is an important tool in understanding physical processes in the space weather. Model performance studies evaluate the quality of model operation by comparing its output to a measurable parameter of interest. In this paper we studied the performance of the combination of the halo coronal mass ejection (CME) analytical cone model and ENLIL three‐dimensional MHD heliosphere model. We examined the CME arrival time and magnitude of impact at 1 AU for different geoeffective events, including the October 2003 Halloween Storm and the 14 December 2006 storm CMEs. The results of the simulation are compared with the ACE satellite observations. The comparison of the simulation results with the observations demonstrates that ENLIL cone model performs better compared to reference mean velocity and empirical models.