Return Pre Home Next



XUE Jishan

Chinese Academy of Meteorological Sciences


The recent progresses in the research and development of numerical weather prediction in China are reviewed in this paper. The most impressive achievements are the development of direct assimilation of satellite irradiances with a 3DVar data assimilation system and a non-hydrostatic model with semi-Lagrangian semi-implicit scheme. Progresses are also made in model physics and model application to precipitation and environmental forecasts. Some scientific issues of great importance for further development are discussed.

Key words: numerical weather prediction, data assimilation, dynamical core, model physics, precipitation environment


The researches on numerical weather prediction in China may be looked back to 1950s. Since mid 1980s, three big research projects focusing on global medium-range weather, tropical cyclone and heavy rain, and short-term climate prediction respectively were carried out consecutively. As the results of these research projects, the operational numerical prediction systems for global medium-range weather, tropical cyclones over western Pacific and rainfall over China were set up. The progresses in modeling technique have resulted obvious improvement of forecast skill in comparison with that a decade ago. For example, the time range of reliable global weather forecast (6 days) is almost double of the number (3-4 days) in earlier 1990s. The products of numerical models are the main basis of forecasters in operational forecast center when the forecasters prepare routine forecast to the public. However the current skill of numerical weather prediction is still not as satisfactory as people expect. The backward data assimilation technique not capable of handling huge amount of satellite data and parameterization schemes of model physics improper to the Eastern Asian monsoon area are two main problems causing degradation of numerical prediction in China. These two problems are closely related to the lack of advanced high-resolution model dynamical frame.

In late 1990s in order to enhance its observing capability, China started to set up its network of next generation weather Radar and initiated its new meteorological satellite project. These two projects imply that more and more remote sensing data will be available for the use of numerical prediction models. So the effective application of radar and satellite data in numerical weather prediction becomes urgent. At same time, China has built its capability of manufacturing high performance computers, so it is much easier for Chinese scientists to get computer resources to support researches on numerical modeling. The improvements in observational net work and computer resources bring forth a good opportunity to upgrade Chinese operational numerical weather prediction system and call for more research work in development of advanced models. In comparison to a decade ago when introducing model from abroad was thought as the most efficient way to upgrade Chinese model system, most Chinese scientist and policy makers agree that it is time to develop new model system based on efforts of Chinese scientist. In 2001, a key state research project: Research and Development of Chinese New Generation Numerical Weather Prediction System was launched. The goals of this 5 years' research project are to develop an advanced Global and Regional Assimilation and Prediction System (GRAPeS) which has the capability of improving routine forecast to meet the needs for high quality weather services. Earlier than GRAPeS a basic science research project: China Heavy Rain Experiment and Study has been carried out. This research project emphasizes the theoretical researches relevant to numerical prediction of heavy rain, so provides some theoretical basis applicable to GRAPeS. Implementations of these two projects and other projects supported by the National Natural Science Foundation or other agencies have caused advances in the researches of numerical weather prediction. The progresses in data assimilation, high-resolution model development, model physics and precipitation forecast will be given in following sections of this paper ended by a discussion about near future development. The progress in climate system model will be a special topic in this volume, so is not included in this paper. Some work completed recently has not been published, but they are of great importance in reflecting Chinese new progresses in researches on numerical weather prediction. They are cited in this paper without references listed with approval of the authors.




For a long period, sparseness of observational data in some key areas is a major difficulty encountered in the effort to improve the numerical weather prediction in China since most disastrous weather systems originate in the Tibetan Plateau or the Western Pacific where few conventional data are available. The development of advanced data assimilation scheme capable to assimilate remote sensing data is thought to be the first objective of GRAPeS project. In 2001, the scientist team of GRAPeS completed the scientific design and coding of a three dimensional variational assimilation system GRAPeS-3DVar (Xue 2002). It is set on grid system suitable to both global and regional data assimilation. A spatial filter based on spectral decomposition or a recursive filter is used as preconditioning to accelerate the convergence of iteration used in minimizing the cost function. The fast radiation transfer model and its tangential and ajoint (RTTOV) developed by ECMWF were then introduced into this 3DVar frame for direct assimilation of satellite irradiances. Results of preliminary experiments assimilating irradiances from satellite NOAA16 in microwave channels (AMSU-a and AMSU-b) show prominent improvement of forecast of typhoon track and intensity by introducing satellite information (Zhang et al. 2003). But more impressive is the inner structure of the typhoon revealed by the assimilation of AMSU data (see Fig.1 and Fig.2). Not only warm core and spiral belts of moisture, but also the asymmetric horizontal wind field and secondary vertical circulation is inferred. In addition to direct assimilation of satellite irradiances, some products derived from polar or geo-stationary satellites e.g. cloud drift wind derived from satellites FY-2 and GMS-5, Quikscat sea surface wind are also assimilated with GRAPeS-3DVar (Xue 2002). The combining use of irradiances in microwave channels from polar orbiting satellite and cloud drift winds also improve the analysis of large-scale environment of tropical storms and show positive impact on forecast of typhoon. The application of satellite-derived moisture is also studied and found to be useful in improving the forecast of rainfall (Meng et al. 2002, Zhou B. et al. 2002).



Fig.1.  Vertical–zonal cross section of temperature deviation from the zonal mean, assimilation of NOAA – 16 AMSU data, Typhoon RAMMASUN 18 UTC July 2 2002 (Zhang H., 2003)



Fig.2.  Vertical cross section of mean tangential wind speed, assimilation of NOAA-16 AMSU data, typhoon RAMMASUN 18UTC July 2,2002  (Zhang H. 2003).

The assimilation of radar data is another issue attracting attention of a number of scientists. By use of radar observation data collected in two field experiments: South China Heavy Rain Experiment in 1998 and China Heavy Rain Experiment and Study in 2001-2002, the retrieval and assimilation of Doppler weather radar data were studied. The research work includes wind retrieval from single and dual Doppler radar observation, assimilation of moisture profile using precipitation estimated from radar observation. Some simple but practical algorithms were developed. One example of them is the horizontal wind field estimation based on tacking the movement of clouds. The assimilation of these data shows positive impact on improving typhoon forecast.

The techniques to assimilate the vertical profile or column integration of moisture from the Global Positioning System (GPS) data or other satellite-borne sensors are also studied. The key point is the vertical distribution of water content. Different approaches, from the simplest statistical model to the sophisticated varitional approach consistent with model parameterization of cloud and precipitation, are introduced.  There are evidences showing that the impact of using these moisture information on heavy rain forecast is obvious. However the results are case-dependent, so more experiments should be done before conclusion of the effectiveness of these approaches is reached.

Progress is also achieved in the research on four-dimensional variational assimilation. Based on a regional numerical weather prediction model developed in the Institute of Atmospheric Physics (AREM), Mu et al. developed an experimental system of four-dimensional variational assimilation. Much attention is paid to the so-called ‘on-off' problem encountered in developing the adjoint of some physical parameterization scheme, such as convection.



One goal of GRAPeS is to meet the needs in numerical predictions of different scales with one common dynamic core. The highest spatial resolution should not lower than 1km. For this purpose, the model dynamic core should be optional either non-hydrostatic or hydrostatic, and either global or limited area. Along this line, a new high-resolution dynamic model has been set up. It is a non-hydrostatic semi-implicit semi-Lagrange model based on the fully compressible atmospheric equations. The model employs latitude-longitude grid points with Arakawa-C staggering horizontally. In the vertical, terrain-following height coordinate and Charney-Philips staggering are used.  A vector interpolation is introduced in order to keep the accuracy of the calculation of departure point in semi-Lagrange scheme. The vital issue in the deployment of semi-implicit scheme for non-hydrostatic model is the algorithm solving the three-dimensional Helmholtz equation of pressure variable which is sensitive to vertical boundary conditions related to the formulation of vertical acceleration. A three-dimensional reference atmosphere in hydrostatic balance is induced to alleviate difficulties caused by numerical treatment of topography. A series of ideal trials were carried out. The model correctly simulates the evolution of geostrophic balanced flows and other hypothetical flow patterns. The simulation experiments with an isolated bell-shaped mountain are employing to verify the model's ability to reproducing the well-studied mountain waves.

Some researchers devoted to improving the long-term integration with semi-Lagrange scheme. Chen et al. proposed application of non-interpolating semi-Lagrange scheme and developed a energy-conserving semi-Lagrange scheme by introducing energy-interpolation (Chen J. et al. 2000, 2001). The semi-Lagrange or semi-implicit scheme is also used in modeling convective clouds (Lou et al. 2002, Chen Z. et al. 2001).



Cloud and precipitation may be what Chinese scientists are concerned mostly. More attentions are now paid to explicit scheme due to the increase of model resolution. Hu et al. developed a new dual-parameter mixed-phase microphysical scheme with 5 species of water substances using a quasi-implicit calculation method (Lou et al., 2002). The scheme includes water contents and number concentrations of cloud, ice crystal, rain and graupel, totally 11 cloud variables and 31 microphysical processes. The processes of deposition, freezing, ice nucleation, autoconversion of cloud to rain or ice to snow, snow to graupel are improved comparing to other explicit scheme currently used in mesoscale models. This scheme was introduced to mesoscale model MM5 with prognostic variables added. The results of simulation of tropical storm with this new scheme are compared with three explicit schemes in MM5. Some micro-physical variables seem well simulated with the new scheme, but the intensity of the storm is not as good as the original explicit schemes. The scheme is also used to simulate the heavy rain events in the South China and the Eastern China. The results indicate that graupel, snow and ice are important in cold areas of clouds and most rain-drops result from the melting of graupel and snow particles.

Efforts are also made to the parameterization of boundary layer processes. Zhou et al. design a new boundary layer model based on Mellor-Yamada's second order turbulent closure method Level 4 (Zhou et al. 2002). The model is coupled with MM5 to simulate a rain-storm happening in the South China in June 1998. Comparing with the MM5 boundary layer scheme, the new model better reproduces the major weather system and rainfall pattern. Specifically, the new model captures major features of turbulence in boundary layer, which removed the un-reasonable rainfall in the western area of the South China existing in simulations with other schemes.

Using surface observation data in the Tibetan Plateau, Xu et al. verified the flux of sensible heat derived by a new boundary layer model based on the theory of multi-scale turbulence. Then the boundary layer model is coupled with the mesoscale atmospheric model MM5. The comparison with observation indicates that the multi-scale turbulence theory well describes the vertical heat transport process in the real atmosphere and is more feasible than the classic similarity theory. Not only it is as accurate as classic similarity theory in computation of gradient transport, but it also gives reasonable result for anti-gradient transport. The simulations of heavy rain events in both South China and the Central China indicate that in the case of complex terrain and surface categories the forecast of the detail of both the location and intensity of heavy rain with parameterization based on the multi-scale turbulence theory is better than that with MRF scheme and Blackadar scheme. The simulation indicates also that heating by land surface makes contribution to the formation of low level jet and mesoscale systems. Accordingly, in cases of complex terrain and land surface, the parameterization of boundary layer based on the theory of multi-scale turbulence has the potential to improve the prediction of heavy rain.



Improvement of numerical prediction of precipitation is always a main objective of the researches in numerical weather prediction. The project of China Heavy Rain Experiment and Study developed an experimental NWP system focusing on the heavy rain along Yangtze River in the Meiyu season (Yu et al., 2002). The core of the system is a regional model AREM developed by Yu et al. The prototype of AREM is the dynamic frame of a regional eta-coordinate model REM having been developed by Yu since 1989 aiming at dealing with the significant effects of topography on the precipitation in China (Yu et al. 1989). REM has been popularly used to the summer precipitation predictions and the heavy rain studies since 1992. In the upgrade of AREM (advanced REM) many attentions are paid to the standardization and modularization of model codes and introduction of advanced model physics. The model physical processes consist of large-scale condensation, modified Betts-Miller convective adjustment, bulk aerodynamic surface flux and non-local diffusion planetary boundary layer parameterization. To understand the capability of AREM in modeling rainfall in China, hindcast experiments are firstly conducted using historic data of June and July in 1998 and 2001. The threat scores (TS) for light rain (1mm/d), moderate rain (10mm/d), heavy rain (25mm/d) and torrential rain (50mm/d) are 0.57, 0.32, 0.18 and 0.76 respectively in 1998, and even higher in 2001. So the performance of AREM is encouraging in these two years. In 2002 the model was tested for real time forecast. The TS skills along the Yangtze River for 1mm/d, 10mm/d, 25mm/d and 50mm/d are 0.47, 0.29, 0.17, and 0.07 respectively, showing the stable performance of the model. The model is also used for studying the mechanism of heavy rain forming along the Meiyu front. Some case studies with AREM show that the model can capture the structures and evolutions of the rain systems. For example, the model reveals the mechanism of formation and evolution of mesoscale vortex in the east periphery of the Tibetan Plateau causing torrential rainfall over Sichuan Basin and down-stream areas. The careful analysis of model integration shows that when mesoscale eddy moves from the top of Tibetan Plateau to the upper level of Sichuan Basin and overlaps to the southwest vortex, the coupled vortex could reach to the upper troposphere. When the system grows mature, the ascending motion is strong in the center of the vortex and the large latent heating in the middle troposphere enhances the deep convection. Heavy rain is mainly caused by deepened deep convective cloud. This experiment helps the understanding on the mechanism of torrential rain.

In addition to development and improvement of model system, ensemble approach is found to be a way having the potential to improve the heavy rain forecast. Studies show that ensemble prediction based on model perturbation is more effective in depicting the uncertainty of numerical prediction of heavy rain events in comparison to that based on the perturbation of initial conditions. However Chen's further study (Chen et al. 2003) shows that perturbation of initial conditions is also effective if proper model elements are perturbed. Most heavy rain events in China occur in the large-scale backgrounds of weak baroclinity. The perturbation algorithms for initial conditions used by most operational centers are based on the theory of baroclinic instability, so not able to capture the diversity of predictions of heavy rain due to the uncertainty in the initial conditions. An approach to disturb the initial fields relevant to the meso-scale gravity wave based on the model perturbation is proposed in their work. A hybrid method of perturbations in initial condition, in model physics and in boundary condition is also tested in Chen's work and gives result better than single perturbation.



As the resolution of numerical weather prediction models increases, the model outputs are more useful in driving models of environmental elements. On the basis of Shao's dust storm model (Shao 2000), an integrated experimental numerical prediction system for dust storm has been developed in the National Meteorological Center (NMC/CMA). This system comprises an atmospheric model, a land surface model, a wind erosion scheme, a transport and deposition scheme and a geographic information database. This system is nested with operational global model T213. The system captured the main features of severe dust storm events in the spring of 2002 in its pre-operational trials.

Efforts are also made to the development of prediction model for urban pollution. A high-resolution model of concentration of pollutants in Beijing city is developed in the Chinese Academy of Meteorological Sciences (CAMS). The atmospheric model is based on the meso-scale non-hydrostatic model MM5 with atmospheric chemistry processes and transportation of pollutants included. The resolution of the model is as high as 500 meters, so small-scale terrain and main features of the city land surface and building blocks are resolved in some extent in the model. The model is expected applicative in environmental management of Beijing city.

Wang et al. use regional climate model (Wang et al. 2002) to drive high-resolution distributed hydrology-soil-vegetation model (DHSVM). They improved the parameterization schemes for land surface and vegetation and evapotranspiration process in the model. The model is used to study the hydrological process in north China and the local response to the climate change. The model succeeded in simulating water cycle in the period 1979-1991, proving its usefulness in solving water resource issues.



Stimulated by the requirements for the high quality products of numerical weather prediction and the rapid progress in observing system and computer power in China, a key state research and development project was launched in China. This 5-year project, known as GRAPeS, aims at developing Chinese new generation data assimilation and prediction system for global medium- range and regional mesoscale weather prediction. In the past 2 years, the scientist teams of GRAPeS project have completed the development of a non-hydrostatic model dynamic core and some main physical processes. A three-dimensional variational data assimilation system capable of assimilating both conventional and satellite irradiances data is also set up. Progresses are also made on numeric algorithms, model physics, coupling of atmospheric model with land surface, hydrological or pollutant transport models and ensemble forecasting. The achievements have set up a solid foundation for the new generation numerical system in China. However they are still preliminary from the point of view of long-term goal of the development of numerical prediction system. In the near future, some scientific issues relevant to numerical weather prediction must be emphasized.

(1) Assimilation of radar and other high-resolution remote sensing data. Few studies dealt with the assimilation of radar observation up to now even though more and more data become available. The assimilation of radar data may be more difficult than satellite irradiances due to more uncertainties in dealing with the reflection of radar beams by rain and cloud drops. The incompleteness of the radar observation and inconsistency in the spatial resolution with other observational platforms and background field are also big challenges. More studies on the physical basis of assimilation of radar may be necessary before effective operational scheme is set up.

(2) Numeric algorithms for high-resolution model. The fully compressible high-resolution model retains the three dimensional acoustic waves which cause new difficulties in numerical computations. For instance, the semi-Lagrangian scheme commonly used in current models may be unstable in these cases. For the same reason, the treatment of model terrain seems more difficult in the high-resolution model. Some schemes of treating terrain used in hydrostatic models are not effective any more in non-hydrostatic model. The inaccurate computation of dynamic and thermodynamic variable around the small-scale mountains may cause more serious problems and destroy the model integration ultimately due to the three-dimensional propagation of the acoustic waves. The development of new terrain scheme is necessary for accurate prediction of mountain-induced rain.

(3) Model physics. The heavy rain events in China often occur in the background of weak baroclinity. However the parameterization schemes used in the models are based on the studies on either baroclinic mid- latitudes or typical tropical atmosphere. This discrepancy is thought to be one of the main reasons causing big errors in forecast of rainfall. More studies with data of field experiments are fundamental for improvement of forecast of rainfall.

(4) Coupling with and development of models beyond atmosphere. Many evidences show the impact of physical processes out of atmosphere on the short and medium range variability of weather. Coupling with land surface model is among the most important issues for future model development. The complexity of the characteristics of land surface in China makes the problem very complicated and difficult. The first step may be the development of coupled atmospheric and land surface models focusing on the time scale of several days. Since most studies on land surface or hydrological process models are motivated for the climate change, new researches are required.


Chen Jiabing, 2000: An application of the mass convection law to the spectral climate model of national climate center. Chinese Journal of Atmospheric Sciences, 24, 608-614

Chen Jiabing et al., 2001: A study on energy-conserving semi-Lagrangian scheme. Chinese Journal of Atmospheric Sciences, 25, 837-846

Chen Jing, et al., 2003: The impact of non-adiabatic physics parameterization schemes on meso-scale heavy rainfall short-range simulation. Acta Meteorologica Sinica, 61, 2

Chen Zeyin et al., 2000: An atmospheric model with semi-Lagrangian and semi-implicit Eulerian equations preliminary numerical tests on deep convection. Chinese Journal of Atmospheric Sciences, 24, 615-624

Kang Hongwen, 2000: A study on application of adaptive grids technique in numerical model, part 1: One-dimensional case test. Chinese Journal of Atmospheric Sciences, 24. 615-624 part 2: Two-dimensional problem, Chinese Journal of Atmospheric Sciences, 24, 749-760

Lou Xiaofeng, Hu Zhijin, Shi Yueqing, Wang Pengyun and Zhou Xiuji, 2002: The development of a new explicit microphysical scheme and simulations of hurricane and heavy rainfall. International Conference on Mesoscale Convective and Heavy Rainfall/snowfall in East Asia, Tokyo, Japan, 293-296

Meng Zhiyong, 2002: Tbb-nudging four-dimensional data assimilation method and simulation on heavy rain process in Wuhan on 20 July 1998. Chinese Journal of Atmospheric Sciences, 26, 663-676

Shao, Y., 2000: Physics and Modeling of Wind Erosion. Kluwer Acdemic Publication, pp393

Wang Jun, Chen Jiabing, 2000: An application of non-interpolating semi-Lagrangian scheme to one-dimensional Burgers equation and two dimensional shallow water equations. Chinese Journal of Atmospheric Sciences, 24, 493-508

Wang Shourong et al., 2002: Improvements of a distributed hydroloymodel and its climatological-hyreological off-line assimilation experiments. Acta Meteorologica Sinica, 60, 290-300

Wang Shourong, et al., 2002: Numerical simulation experiments by nesting Hydrology model DHSVM with regional climate model RegCM2/China.  Acta Meteorologica Sinica, 60, 421-427

Xue Jishan, 2002: Satellite data assimilation plan in CMA. Topic Chairman and Rapporteur Reports of the Fifth WMO International Workshop on Tropical Cyclones (IWTC-5), WMO/TD No.1136

Yu Rucong, 1989: Design of the limited area numerical weather prediction model with steep mountains. Chinese Journal of Atmospheric Sciences, 13(2)

Yu Rucong, Xu Youping, Cheng Rui and Li Kun, 2002: A regional numerical prediction model and its simulations on the heavy rainfalls along the Yangtze River. International Conference on Mesoscale Convective and Heavy Rainfall/snowfall in East Asia, Tokyo, Japan, 339-344

Zhang Hua et al., 2003: Satellite observation study for typhoon over Northwest Pacific. Unpublished

Zhou Bing et al., 2002: Numerical simulation of CMWDA with impacting on torrential rain forecast. Acta Meteorologica  Sinica, 60, 208-317

Zhou Xuiji, Zhao Ping and Sun Jian, 2002: A new PBL model with second-order turbulent closure and its implement into MM5. International Conference on Mesoscale Convective and Heavy Rainfall/snowfall in East Asia, Tokyo, Japan, 297-301

Return Pre Home Next