A mesoscale ensemble prediction system (MEPS) was designed and its application for mid-latitude cyclones was explored in this paper. The uncertainties existing in the initial data, model physics, and lateral boundary condition were considered and incorporated into the MEPS. Three separate ensemble subsets were created with respect to the three uncertainties, respectively. Three mid-latitude cyclones that occurred over the United States were chosen for implementation of the MEPS using the Penn State-NCAR fifth-generation Mesoscale Model (MM5), which was adopted as a very useful tool in short-range ensemble forecasting.

The perturbations used to represent the initial condition errors in this study were generated by three breeding methods, which are the Bred Monte-Carlo Perturbations method (MC), the Breeding of Growing Modes method (BGM) and the Perturbed Observations method (PO). The uncertainties existing in model physics were explored through the use of multiple model physical parameterizations (MPEF), while the use of bred perturbations in the outermost domain generated the lateral boundary ensemble subset (LBEF). Each ensemble subset consisted of 20 48-h forecasting members except for MPEF, which had 48 members.

Through the examination of the ensemble prediction results such as verifying cyclone central pressure, cyclone track, etc., combined with statistical analysis, we gained a good understanding of the ability of the MEPS for mid-latitude cyclone prediction. Questions addressed concerned the relative merits and role of each ensemble subset in the MEPS. The ensemble that can provide sufficient information to contain realistic atmospheric states will lead us to examine whether an ensemble forecast could provide insight into the potential forecast skill and improve the accuracy of subsequent probabilistic forecasts for mid-latitude cyclones.