Because errors in the initial conditions of model simulations are present and the model parameterization schemes are insufficiently representative of meso- and micro-scale physical processes in reality, the forecast of heavy rainfall in the warm sector is less accurate than that of the front-associated rainfall. It has been demonstrated that numerical simulations of a meso-scale convective system in the warm season is quite sensitive to the initial atmospheric conditions and the model physical parameterization schemes adopted (Luo et al. 2010; Wu et al. 2013). Compared with the single forecast, ensemble forecasts can provide further information on the uncertainty (Toth and Kalnay 1993, Tracton and Kalnay 1993), accuracy and predictability of the numerical models. This project will use data collected during the field experiment (2013-2014) and, along with other historical observations/data, to improve the model initial conditions and model physical processes, to carry out ensemble experiments, and to examine the model capability. The mesoscale models that will participate in the SCMREX NWP studies are listed in Table 9.
(1) Data Assimilation
The quality of the initial field is important for the short-term numerical weather prediction (NWP). It is a also the key to improving heavy rainfall forecasts during the early summer rainy season in South China and understanding corresponding physical mechanisms. A good initial field can also serve as the foundation stone for the construction of representative ensemble forecasts and their perturbation. Data assimilation has been used to reduce initial condition error. Applications of different methodologies in data assimilations of weather systems have been proposed and well developed, ranging from single conventional measurement to sorts of nonconventional measurements and from the single time observation to multi-times, such as the evaluation from the simple interpolation, successive correlation (Barnes，1964), optimal interpolation (OI) (Bratseth，1986) and three dimensional variation (3DVAR) (Lorenc，1981；Parrish and Derber，1992；Anderson et al.，1998) to ensemble Kalman filter (EnKF) (EnKF，Houtekamer et al.，1996； Hamill and Snyder，2000) and four dimensional variation (4DVAR) (Lewis and Derber 1985；Courtier and Talagrand 1990). More and more measurements are better assimilated due to the continuous advance in data assimilation methodology. Nowadays, various studies based on different measurements take account of the impact on analysis and forecast, and propose a lot of assessment methods (Purser and Huang 1993；Cardinali et al. 2004；Chapnik et al. 2004). The advanced methodologies help reveal how the assimilated measurements affect the model forecast via the data assimilation system, and what roles the measurements and background field in the analysis and forecast play. Based on these findings, an improved method of assimilation system can be proposed.
This project will use various measurements collected by different observation platforms during the field campaign to generate analysis datasets through data assimilation. A number of assimilation methods, such as 3DVAR, 4DVAR, EnKF and the hybrid method (Hamill and Snyder，2000), will be adopted. The datasets created by our data assimilation systems will be open to researchers of interests. Moreover, impacts aroused by various measurements including those from the COSMIC (http://www.cosmic.ucar.edu/) on forecast during the early summer rainy season in South China will be studied.
(2) Improvement of Model Physical Schemes
The model error is another major contributor to an inadequate numerical weather prediction. The triggering mechanisms for warm sector rainstorm generally involve complicated boundary layer processes, the surface topography and different properties between land and sea, the interaction among multi-scale systems, and sophisticated cloud-precipitation microphysical processes. A warm sector rainstorm is rarely well forecasted by numerical models, implying caveats in the physical processes among models. Since most key parameters of the applied physical processes in mesoscale numerical models are primarily based on some empirical values of similar weather backgrounds obtained in foreign areas. The representativeness of these empirical values is of major concern and may lead to model forecast errors associated with the peculiar monsoon rainstorm in China, especially in the warm sector scenario.
The project will take full advantage of the data collected during the field experiment to compare with model simulations, the microphysical fields and boundary layer profiles in particular. The integration/combination of the historical observation data, and the TRMM and CloudSat/CALIPSO satellite active measurements that reflect the cloud-precipitation characteristics, will also be utilized to evaluate/improve the parameterization schemes in mesoscale model by a variety of strategies (for instance: updating and improving the values of the key parameters in the model physical schemes via data assimilation based on the EnKF parameter estimation). Here, we focus on improving the boundary layer processes and the cloud-precipitation microphysical processes to reduce the model error and then improve the forecasting capability of warm sector rainstorm before and after the outbreak of summer monsoon rainfall.
Errors from model dynamics, physics schemes and initial conditions create uncertainties in model simulations. While a single simulation or prediction cannot provide information on uncertainty, ensemble prediction serves to provide the statistical state of a collection of possible results, which can help improve the forecast quality and extend valid prediction periods. The ensemble members created for one single model are generally approached with two methods, either by adding perturbations to initial condition, including random Monte Carlo (Leith 1974), Lagged Average Forecast, Singular Vector (Molteni et al. 1996, Ono et al. 2010), Breeding of Growing Modes (Toth and Kalnay 1993) and etc., or by selecting different model physical schemes and adopting different values of key parameters. Moreover, ensemble members across a couple of models can be used to provide the so-called “super ensemble forecasts”.
Making good use of initial conditions and physics perturbation methods (by changing physical schemes or parameters), we will carry out MEP experiments with model horizontal grid spacing less than 10 km for heavy rainfall in the first rainy season in South China. Super ensemble prediction will also be constructed based on individual model MEP experiments at all institutions in the present project, so as to evaluate the performance of MEP of heavy rainfall in South China.
Based on current operational system, CMA will set up a specified verification system for a careful assessment of all participating MEPSs. The verification of meso-scale ensemble forecast systems will be carried out against the shareable observational datasets in Southern China. Variables to be examined and verified are those of highest concerns in meso-scale forecasts, such as the surface and low-level troposphere parameters, including surface precipitation, 2-m temperature, 2-m dew point temperature, 2-m relative humidity, sea level pressure, 850-hPa temperature, 850 relative humidity, etc. Two main types of methods will be used to verify the ensemble forecasts conducted: one is the probability density function (PDF), and the other is the probability forecast exceeding some specific thresholds (e.g., equitable threat score, ETS).
(4) Computing Power and Facility
The State Key Laboratory of Severe Weather of CAMS will provide an IBM x3850X5 Sever for conducting the MEP experiments with the EnKF-WRF (citation TBA). This sever has 8 CPU processors with 80 core Intel Xeon 10C E7-8850 130W 2.00GHz/24MB, 128GB DDR3 RDIMM memory, SAS disk 1.2 T capable of expending up to 12 T.
Guangzhou Institute of Tropical and Marine Meteorology (GITMM) will run the GRAPES (citation TBA) use the high-performance computing systems at the Guangdong Meteorological Bureau, Dongguan City Meteorological Bureau, and GITMM, respectively, with the peak floating-point operations per second (FLOPS) of 500 T, 1.0944 T, and 2.816 T.
Chinese Academy of
46 Zhonguancun South Avenue,Beijing,100081,P.R.China