Experiment Of WRF Short-range Precipitation Ensemble Forecasts In Southeastern China

Predicting the short range weather has been a major difficulty in numeric weather prediction. With the numeric models, physics parameterization schemes and bias correction techniques becoming fairly advanced and sophiscated, improving the forecast abilities of the single model is more expensive than before. Ensemble forecast appears to be a good choice in improving the short-range numeric prediction, it can also provide a of forecast products which tradition single deterministic forecast can not offer, and support the modern society’s needs in a better way. In China, the research of short-range ensemble forecasting is relatively late, and the key techniques such as multi-physics and multi-IC/LBC (intial condition/lateral boundary condition) need to be tested with more numeric experiments for southeastern China. In this study, an experimental short-range ensemble forecast system for southeastern China is constructed using WRF (Weather Research and Forecast) model, and carried out short-range ensemble forecast experiments using multi-physics and multi-IC/LBC methods.The first part of the research conducted an experiment using multi-physics method, the short-range ensemble forecast system used NCEP (National Centers for Environment Prediction) GFS (Global Forecast System) as its initial condition/lateral boundary condition and started the forecast cycle at May9th,2010. The results from May9th to June24th,2010are verified against upper-level data from NCEP GDAS (Global Data Assimilation System) and surface observation from more than1000stations in the forecast domain. The verification emphasized on figuring out the impact of how different physics parameterization schemes and the ensemble mean of all the members affect the forecast of meteorological elements. The result showed that: WRF model performed fairly well in forecasting the meteorological fields in southeastern China; For the whole forecast period (60h), the ensemble forecast system had certain skill in precipitation forecasting, judging from the performance of model members and ensemble mean; Meteorological elements at different pressure levels and precipitation at different thresholds were different in sensitivity to physical parameterization schemes and forecasting performance; For most meteorological fields, ensemble mean performed better than model members. A discussion of the possible optimized combination of physics processes used for multi-physics method based on the short-range ensemble forecast system is also done. The main conclusions are:an ensemble forecast system with more members would produce relatively better forecasts than an ensemble forecast system with less members, in this case, the whole ensemble forecast system which contains4types of physics processes produced the best forecasts. For surface temperature, multi-land surface model and multi-planet boundary layer parameterization schme are most efficient, while multi-land surface model and multi-cumulus parameterization scheme are best for sea level pressure forecast; For heavy precipitation and upper-level meteorogical elements, multi-cumulus parameterization scheme is the most efficient, followed by multi-planet boundary layer parameterization scheme, multi-land surface model and multi-microphysics scheme.Initial conditions and lateral boundary conditions are also very important to short-range ensemble forecast. The second part of the study used both multi-IC/LBC and multi-physics in the experiment. The forecasts were driven by the GFS, JMA (Japan Meteorological Agency) and T639global forecast data, started from June7, and last to July20,2012. An analysis on how different global forecast data and physics parameterization schemes affect the forecast of meteorological elements is done, how the ensemble mean of all the members improve the forecast qualities is examined. The results showed that:Ensemble mean reduced the uncertainties of the forecast and extended the valid time of the forecast, it also produced better forecast quality than the average quality of the single members. The upper-level elements are more sensitive to initial condition/lateral boundary condition relative to physics parameterization schemes. Among the three global forecast data, forecasts driven by JMA were worse than those driven by the other two global forecast data, forecasts driven by GFS were a little better than those driven by T639; Surface elements are sensitive to both initial condition/lateral boundary condition and model physics, like the upper-level elements, forecasts driven by JMA were worse than those driven by GFS or JMA, and the other global forecast data gave the almost same forecasts; Members using BMJ (Betts-Miller-Janjic) cumulus parameterization scheme produced better sea level pressure and surface temperature forecasts than those using KF (Kain-Fritsch) cumulus parameterization scheme, while for12-24hour accumulated precipitation, KF cumulus parameterization scheme is relative better than BMJ parameterization schme; WSM3microphysics scheme produced the best sea level pressure and surface temperature among all the microphysics schemes; For precipitation, moderate precipitation and heavy precipitation were sensitive to both initial condition/lateral boundary condition and model physics, torrential precipitation were not sensitive to microphysics schemes; Generally speaking, forecasts driven by GFS were best, closely followed by those driven by T639, and forecasts driven by JMA were worst, the KF cumulus parameterization scheme is slightly better than BMJ cumulus parameterization schme, Thompson scheme performed slightly better among the microphysics schemes.