Meso-scale Ensemble Forecasts on Physical Perturbation Using a Non-hydrostatic Model

The past decade has seen increasing interest in ensemble methods for operational numerical weather prediction. Ensemble forecasting is motivated by the recognition that numerical predictions always contain both initial condition uncertainties and numerical model uncertainties. It is traditionally desirable to use an ensemble method focused mainly on uncertainties in the initial conditions for the medium range forecasts. Encouraged by the success of global medium-range ensemble forecast, investigations are made in examining the short-range ensemble forecast (SREF) with meso-scale models. It is found that the ensemble approach could also improve short-range weather forecasts, especially the forecasting of quantitative precipitation, meso-scale convective systems. However, it is still challenging for meso-scale ensemble forecast due to the difficulty to generate a well spreading perturbation which triggers meso-scale and convective processes in a short time.Motivated by the previous studies on meso-scale ensemble forecasts, a meso-scale ensemble forecast system based on a limited area non-hydrostatic meso-scale NWP model (GRAPES-Meso) is carried out, to construct meso-scale ensemble, especially considering the perturbation of the physical sensitive factors and the initial conditions. As a preliminary step, the objective of the study is to understand the impact of model physics uncertainties on predicting extreme precipitation events using meso-scale ensemble forecast. A flash flood case on July 10, 2004 in Beijing is particularly chosen for the study of a 36 h meso-scale ensemble forecast. The "7.10" flash flood causes a high impact on traffic and open air activities because of the failure of the forecast. In particular, the uncertainties in convective parameterization schemes are focused on. The members of SREF are constructed by specifying the closure assumptions, triggering parameters and precipitation efficiency. Meanwhile meso-scale ensemble forecast is constructed using multi-physical parameterizations and the uncertainty of initial condition is added using the Monte Carlo method.The verification results show some characteristics of the meso-scale system could be captured by GRAPES and meso-scale ensemble is feasible to improve on the forecast site and the forecast intensify of the precipitation. Different ensemble experiment has different result, even in the same experiment, members differ with each other, and it means there is an appropriate spread among members. In spread analysis, from the surface to the middle-level of troposphere there is a maximum value area in Beijing, i.e., this area has large uncertainties. Meanwhile it is found that it's difficult to construct ensemble system only considering model uncertainties, if initial condition uncertainties are added, the results are preliminary but encouraging, it is helpful to capture more meso-scale information and construct more effective meso-scale ensemble system.