NO.4 ZHANG Chaolin, CHEN Min, KUO Ying-Hwa, FAN Shuiyong and ZHONG Jiqin 389

Transcription

1 NO.4 ZHANG Chaolin, CHEN Min, KUO Ying-Hwa, FAN Shuiyong and ZHONG Jiqin 389 Numerical Assessing Experiments on the Individual Component Impact of the Meteorological Observation Network on the July 2000 Torrential Rain in Beijing ZHANG Chaolin 1 ( ), CHEN Min 1 ( ), KUO Ying-Hwa 2 ( ), FAN Shuiyong 1 ( ), and ZHONG Jiqin 1 ( ) 1 Institute of Urban Meteorology, China Meteorological Administration, Beijing , China 2 National Center for Atmospheric Research, P. O. Box 3000, Boulder, CO, , USA (Received October 30, 2006) ABSTRACT In an effort to assess the impact of the individual component of meteorological observations (ground-based GPS precipitable water vapor, automatic and conventional meteorological observations) on the torrential rain event in 4-5 July 2000 in Beijing (with the 24-h accumulated precipitation reaching 240 mm), 24-h observation system experiments are conducted numerically by using the MM5/WRF 3DVAR system and the nonhydrostatic MM5 model. Results indicate that, because the non-conventional GPS observations are directly assimilated into the initial analyses by 3DVAR system, better initial fields and 24-h simulation for the severe precipitation event are achieved than those under the MM5/Litter R objective analysis scheme. Further analysis also shows that the individual component of meteorological observation network plays their special positive role in the improvement of initial field analysis and forecasting skills. 3DVAR scheme with or without radiosonde and pilot observation has the most significant influence on numerical simulation, and automatic and conventional surface meteorological observations rank second. After acquiring the supplement information from the other meteorological observations, the ground-based GPS precipitable water vapor data can more obviously reflect initial field assimilation and precipitation forecast. By incorporating the groundbased GPS precipitable water vapor data into the 3DVAR analyses at the initial time, the threat scores (TS) with thresholds of 1, 5, 10, and 20 mm are increased by 1%-8% for 6- and 24-h accumulated precipitation observations, respectively. This work gives one helpful example that assesses the impact of individual component of the existing meteorological observation network on the high influence weather event using 3DVAR numerical system. Key words: three-dimensional variational data assimilation, global positioning system (GPS), severe rainfall, observation system experiment, numerical weather prediction (NWP) 1. Introduction The numerical variational assimilation system is superior to those traditional objective analysis methods, as it can easily incorporate meteorological observations directly without any presupposed retrieval. With the rapid development of nonconventional observational techniques, such as meteorological satellite and radar, which have been the indispensable tools and provided sufficient information for well understanding and precisely forecasting the atmospheric activity, variational assimilation is becoming the predominant method for the initial-field analysis of developing numerical weather prediction (NWP) system internationally. In the 1990s, the USA and the developed countries in Europe took the lead in operational NWP application of the three-dimensional variation assimilation (3DVAR) technique (see Lorenc et al., 2000). Recently, ECMWF, France, and Japan have practised 4DVAR system in the global and regional weather forecast operation. And now, Denmark and USA are also developing their operational 4DVAR system. China has carried out a series of research and development in the field of NWP variational assimilation and achieved some remarkable progress, and it provides a scientific basis and technical support for Supported by project of the Ministry of Science and Technology under Nos. 2005DIB3J098, 2003DFB00011 and 2002BA904B05, project of the Beijing New Star under No. H , projects of Beijing Municipal Science Technology Commission under Nos. H and H , and project of GPS application of Beijing Meteorological Bureau. Corresponding author: c lzhang@yahoo.com or clzhang@ium.cn.

2 390 ACTA METEOROLOGICA SINICA VOL.20 practising variation assimilation into Chinese NWP systems. However, there is still a significant lagdifference in numerical variational assimilation applications, especially in operation, between China and other advanced countries in the world. The variational assimilation technique can be used to assess the impact of individual component of the existing observation network via Observation System Experiments (OSEs). Great achievement in this field has been made abroad, which provides valuable information for the atmospheric observation network design and construction (Guo et al., 2000). In recent years, through numerical experiments and based on the numerical analysis of vertical wind distribution characteristic sensitivity to high impact weather evolution, Zhang and Wang (2002) assessed the impact of the planning wind profiler observation network in Beijing areas of China using the nonhydrostatic mesoscale weather model MM5V2 (PSU/NCAR). However, little research has been conducted to investigate the impact of individual component of the existing observation network on severe weather with the variational assimilation technique. The observation network around Beijing will be updated for supporting the art-of-state weather service in 2008 Olympic Games. It is imperative to conduct scientific research to provide scientific suggestions on how to construct or update observation networks. In this paper, focusing on the most severe precipitation event occurring in the summer of 2000 in Beijing, we investigated the individual component impact of the meteorological observation network on this torrential rain event by MM5/WRF 3DVAR system and MM5 mesoscale nonhydrostatic model, and achieved some valuable results for the future observation network construction in the Beijing area. 2. Brief introduction of the 3DVAR system and numerical schemes 2.1 MM5/WRF 3DVAR system The MM5/WRF 3DVAR system used here is the latest official version released in June 2003 (V1.3.0, UCAR), which has been remarkably updated compared to the earlier version (see Barker et al., 2003, 2004). It not only can be widely used in meteorological scientific research, but also plays an important role in practical operational applications. In China, Zhang et al. (2002) have conducted variational data assimilation analysis of the Wuhan heavy rainfall on 20 July 1998, using an earlier version of the MM5 3DVAR 1.0 system. To support the research and operational applications in Beijing, we have conducted the following work on the MM5/WRF 3DVAR system: (1) We developed local observation preprocessing module for groundbased GPS precipitable water content, radiosonde, upper-air wind, surface conventional, and automatic weather station. This allows us to incorporate all the above-mentioned meteorological data in Beijing and its surrounding areas into the MM5/WRF 3DVAR system. (2) Using the daily 24- and 12-h forecast outputs from MM5V3.6 that are valid at the same time covering the period of July 2000, we estimated the climatic background error covariance matrix via the NMC-method (see Parrish and Derber, 1992). Further updating of the initial lateral conditions for the MM5 model is carried out using 3DVAR analyses. (3) According to the climatic characteristics and observations in Beijing and its surrounding areas, we adjust the horizontal length scale of the background field errors, which is calculated by the NMC-method, to more reasonable values. 2.2 Model used Numerical experiments are carried out with the MM5 (V3.6, PSU/UCAR) nonhydrostatic mesoscale model (see Dudhia et al., 1993, 2003). The horizontal resolution is 10 km with 81 grids in the west-east direction and 67 grids latitudinally, centered at (40 N, 116 E). In the vertical direction the terrain-following coordinates, σ-coordinates, are used with 23 vertical levels. Numerical integrations are advanced with the time-step of 30 s. The physics schemes include Dudhia simple ice explicit microphysics and Grell cumulus parameterization, MRF planetary boundary layer parameterizations, Dudhia cloud radiation, and fivelayer soil scheme. 2.3 Experimental design Case Driven by small-scale convective activities augmented by the larger-scale weaker convective system

3 NO.4 ZHANG Chaolin, CHEN Min, KUO Ying-Hwa, FAN Shuiyong and ZHONG Jiqin 391 and local topography effects (see Zhang et al., 2005), a torrential rainfall weather event occurred from 00:00 GMT 4 to 00:00 GMT 5 July 2000, in which the 24-h accumulated precipitation of the automatic weather station exceeds 240 mm (we simply referred to it as the July 2000 torrential rain). Statistics show that it is the heaviest rainfall event of the summer of 2000 in Beijing Distribution of the individual component of atmospheric observation network in Beijing and its surrounding areas Figures 1a-1d indicate the distribution of the existing conventional radiosonde, pilot, automatic and conventional surface observations, and the GPS-MET networks within the model domain, respectively. And Figs.1a and 1b obviously show that the radiosonde and pilot observations are fewer, but they are wellproportioned distribution, at intervals of about km within the model domain. In observational viewpoints, they have a reasonable horizontal distribution for well monitoring the evolution of the upper atmospheric system on large-scale and mesoscale. Figure 1c shows that on the land the automatic and conventional surface weather observations are about km station-spacing, with the high density of these surface observations, especially within the Beijing area, which are very useful for monitoring the genesis/developing/dissipating evolution of meso- and micro-scale weather systems. Apparently, the Beijing s ground-based GPS-MET experimental network (see Fig.1d), consisting of 10 GPS integrated precipitable water vapor (IPW) remote-sensing stations with km station-spacing, is helpful for capturing IPW evolution over the Beijing area Experimental scheme design According to the existing observation network distribution in Beijing and its surrounding areas as shown in Fig.1 and the discussion in Section 2.3.2, seven Fig.1. Distribution of the individual component of atmospheric observation network within the simulation domain: (a) conventional radiosonde, (b) upper-air pilot, (c) automatic and conventional surface observations, and (d) ground-based GPS-MET.

4 392 ACTA METEOROLOGICA SINICA VOL.20 Table 1. The scheme illustration of numerical experiments Experiment Analysis method for Background field scheme Kind and number Total station number within initial field (in parentheses) the model domain of observation networks Using the daily 24- and 12-h conventional radiosonde forecast outputs from MM5V3.6 (6), pilot(4), automatic that are valid at the same time and conventional surface 328 covering the period of July 2000, observations (308), and Exp.1 MM5/WRF 3DVAR the climatic background error ground-based GPS-IPW covariance matrix for 3DVAR of stations(10) MM5 is estimated via the NMC-method. And the initial lateral boundary conditions of the MM5v3.6 model are generated and updated every 12 h according to the forecasts provided by the global spectral model T106 (NMC, Beijing, China) The lateral boundary conditions conventional radiosonde of the MM5v3.6 model are (6), pilot(4), automatic Exp.2 Cressman s objective generated and updated every 12 h and conventional surface 318 analysis of MM5 according to the forecasts observations (308) Litter R package provided by the global spectral model T106 (NMC, Beijing, China) Exp.3 same as Exp.1 same as Exp.1 same as Exp Exp.4 same as Exp.1 same as Exp.1 conventional radiosonde 10 (6), pilot(4) Exp.5 same as Exp.1 same as Exp.1 automatic and conventional 308 surface observations (308) Exp.6 same as Exp.1 same as Exp.1 ground-based GPS-IPW 10 stations(10) Exp.7 same as Exp.1 same as Exp.2 non 0 numerical OSEs listed in Table 1 were designed to assess the impacts of the individual component of the existing meteorological observation networks around Beijing (conventional radiosonde, pilot, automatic and conventional surface observations, and ground-based GPS-IPW observations) on the July 2000 heavy rainfall. For each numerical experiment of Exp.1- Exp.7, the MM5 model was utilized to simulate this heavy rain process. Twenty-four-hour numerical integration was carried out with MM5/WRF 3DVAR s or the MM5 Litter R s objective analysis results as the initial fields. By comparing and analyzing the initial, forecasting and objective verification results between different numerical schemes as illustrated in Exp.1-Exp.7, the positive impacts of the individual component of the existing meteorological observation networks in Beijing and its surrounding areas on the July 2000 torrential rain process were studied. Here, the following main points should be noted: (1) For those experimental schemes with the MM5 Litter R s objective analysis as the initial fields, observations from conventional radiosonde, pilot, automatic and conventional surface observations within the model domain, as well as real-time data of 12- county surface stations in the Beijing areas have been included. (2) For those experimental schemes with the MM5/WRF 3DVAR s analysis as the initial fields, the 10 ground-based GPS-IPW observations of the Beijing s atmospheric water vapor experiment in the summer of 2000 also could be assimilated besides other data. (3) Each numerical experiment is integrated for 24 h, its background fields at the initial time are introduced from the 12-h forecasting outputs of the global spectral model T106 (NMC, Beijing, China). And the

5 NO.4 ZHANG Chaolin, CHEN Min, KUO Ying-Hwa, FAN Shuiyong and ZHONG Jiqin 393 lateral boundary conditions of the MM5v3.6 model are generated and updated every 12 h according to the 12-, 24-, and 36-h forecasting results provided by the global spectral model T106. For any of those numerical experiments with 3DVAR analysis result as the initial condition, its initial lateral boundary condition is further updated with 3DVAR s initial analysis. (4) Objective verification is conducted for prognostic elements of each experiment s simulation. Therein, precipitation observations for verification consist of the automatic and conventional surface stations within the simulating domain, and other real observations are based on the objective analysis of conventional surface observation and upper radiosondes, in which the global spectrum model T106 s analysis fields at corresponding time, operationally provided by National Meteorological Center (Beijing, China), are used as the background information. 3. Analyses and discussions 3.1 Impacts of observation networks on the precipitation forecast skill The threat scores (TS) of forecasting station precipitation of Exp.1-Exp.7 are listed in Table 2, which obviously show that the maximum TS for 24-h precipitation reaches 0.8 or more under many of the experiments at threshold of 1 mm, and the minimum is also high up to 0.54 (Exp.6 for threshold of 20 mm). Almost all of the TS values of the 24-h simulated precipitation of experiments are over 0.7 for the thresholds of 1, 5, 10, and 20 mm. The above model results indicate that the amount, location, and starting time of precipitation are simulated very well for the Beijing July 2000 torrential rain by using MM5 3DVAR system and MM5V3.6 model. By comparing Exp.1 with Exp.2, most of threat scores of Exp.1 are higher than those of Exp.2, except that those in the period of with thresholds of 1 and 5 mm, and the 24-h verifications with 1-mm thresholds are a little lower than TS of Exp.2. Therefore, it suggests that MM5/WRF 3DVAR is a credible and feasible system that can be used to generate reasonable initial NWP conditions, and the precipitation forecasting results better than that of the traditional MM5 Litter R objective analysis method could be achieved after it assimilated 10 GPS-IPW data over the Beijing area directly. Comparing the threat scores for the 6-h Table 2. Precipitation threat scores (TS) of numerical experiments vs observations Threshold Prognostic element Valid period Exp.1 Exp.2 Exp.3 Exp.4 Exp.5 Exp.6 Exp.7 for verification (MMDDHH) mm 6 h AP h AP mm 6 h AP h AP mm 24 h AP mm 24 h AP Note: AP denotes accumulated precipitation. precipitation between Exp.1 and Exp.3, most of the threat scores of Exp.1 are 1%-8% higher than those of Exp.3, except TS for the threshold of 5 mm during the two periods of and (TS of 0.55 and 0.3 in Exp.1, 0.56 and 0.31 in Exp.3, respectively). Moreover, the TS difference for the 24- h forecast between them is also quite similar to those mentioned above. The above precipitation TS comparisons among Exp.1, Exp.2, and Exp.3 show coincidently that the accuracy of short-term precipitation

6 394 ACTA METEOROLOGICA SINICA VOL.20 forecast is improved under Exp.1, because with Beijing s GPS-IPW 3DVAR assimilation its higher quality initial fields are achieved compared to those based on the objective analyses (Exp.2) and those only assimilated conventional upper and surface observations (Exp.3). Comparing the TS values of Exp.4, Exp.5, Exp.6, and Exp.7 listed in Table 2, Exp.5, in general, has the best scores, Exp.4 is next, and then is Exp.7. Except for a few thresholds and periods, the TSs of Exp.6 are a little higher than those of Exp.7, and most scores of Exp.6 are comparative or even lower than the values of Exp.7. Due to the fact that Exp.7 represents the information contribution of background fields to this precipitation simulation, all of the results discussed above indicate that the automatic and conventional surface weather observations have the most important effects on the improvement of assimilation quality of initial fields, which could result from the fact that their data (up to 318 in the simulating domain, referred to Table 1 and Fig.1) are much more abundant than the others. However, it should be noted that the impact from only 10 atmospheric three-dimensional detection stations (sounding and pilot) on the forecasting improvement has ranked second. From this point, one can find that to enrich three-dimensional atmospheric information by developing upper atmospheric detection instruments has important significance for the short-term NWP improvement. In general, the TSs of Exp.6 are lower than those of Exp.7, suggesting that without synthesizing other kinds of conventional upper and surface observations the individual role of GPS-IPW assimilation only can be summarized as the negative or neutral kind on this precipitation simulation. Combined with the previous analyses of precipitation forecasting TS about Exp.1 and Exp.3, the following conclusion could be drawn: In order to improve assimilation/analysis quality of initial fields and increase precipitation skills of short-term NWP forecasting, the ground-based GPS-IPW data assimilation should be conducted, together with additional predominance-supplemented information from other meteorological observation networks. Good precipitation TS skills of Exp.7 further show that the medium-term global spectral model T106 (NMC, Beijing, China) has a higher-level operational forecasting skill and has provided the good background and lateral information for this simulation. 3.2 Impacts of observation networks on forecasting skill of other key prognostic elements The root mean square (RMS) errors are calculated for the 0-, 12-, and 24-h prognostic elements of wind, temperature and moisture fields on the model s grids at 300-, 500-, 700-, and 850-hPa levels for Exp.1- Exp.7, respectively (time and space variation of those RMS values listed in Table 3, omitted here). According to Table 3, at the initial time (0-h forecasting) the forecasting wind, temperature, and moisture at the high troposphere (300 hpa), mid-troposphere (500 hpa), and low troposphere (700 and 850 hpa) of Exp.2 have the least RMS scores generally among all the seven numerical experiments, and Exp.7 has the largest RMS values. Furthermore, among the rest experiments the RMS scores of Exp.4 are closest to those of Exp.2. Taking the RMS score of the initial moisture field at 850 hpa, for example, the corresponding RMS scores of Exp.1-Exp.7 are 1.1, 1.0, 1.1, 1.0, 1.5, 1.4, and 1.6 g kg 1, respectively. On the one hand, the above RMS results could be mainly due to the fact that the observation data of Exp.2, which only passed the rough data quality check, are regarded as the denotation of the real true atmospheric conditions and are fully credible according to the MM5 Litter R objective analysis method while analyzing the initial fields of model, i.e., the representative and instrumental errors of these data are not concerned. And then the initial analysis under Exp.2 is closest to true atmospheric conditions as defined in the verification method (referring to verification design described in Section 2.3.2). On the other hand, the Exp.7 s verification results also reflect the fact that it just takes the background information into consideration, and ignores the conventional observation data regarded as totally reliable in our verification method. Additionally, the RMS scores of Exp.4 are the closest to

7 NO.4 ZHANG Chaolin, CHEN Min, KUO Ying-Hwa, FAN Shuiyong and ZHONG Jiqin 395 those of Exp.2, suggesting the importance of threedimensional information of the upper-level atmospheric monitoring. The RMS values under other experiments directly reflect the integrated contributions from the background and different observation information in the 3DVAR system. The little difference of RMS errors of 12- and 24-h forecasts among Exp.1- Exp.7 is not evident, meaning that using the method of MM5/WRF 3DVAR at the initial analysis does not cause abnormal increase of short-range forecast errors, and it would be feasible and reliable in practical NWP application. Of the NWP prognostic variable verifications the above results also reasonably reflect the theoretic disagreement upon the true atmospheric state representation between variational assimilation and objective analysis methods, i.e., using the traditional objective analyses as the true states of the atmosphere is inconsistent with the theoretic foundation of variation method, in which observational errors are always regarded as one objective and important factor on data analysis. Therefore more objective and effective methods for NWP verification should be developed in numerical applications of data variational assimilation in the future. 3.3 Simulation on the July 2000 torrential rain Figure 2 shows the distribution of 24-h accumulated precipitation of observation and simulation in Exp1. Figure 2a shows that the observed rainfall is distributed in a northeastern-southwestern belt in the simulation domain during 00 GMT 4 to 00 GMT 5 July 2000, with two heavy rain centers, one in Beijing with the maximum of 24-h precipitation exceeding 240 mm (Mentougou Automatic Weather Station, Beijing), the other near the southern edge of the simulation domain with the maximum of 24-h precipitation being 113 mm (Xingtai Conventional Weather Station, Hebei Province). Statistics show that it is the heaviest rainfall in the summer of 2000 in Beijing. By comparing Fig.2a with Fig.2b, it can be seen that both the rainfall distribution and the two heavy rainfall centers are simulated successfully by Exp.1. The simulated 24- h rainfall maximum in southwestern Beijing reaches 285 mm. The amount and location are nearly identical to the actual observations (e.g., the maximum of 240 mm). Furthermore, the rainstorm in Xingtai, Hebei Province of China, is also well simulated near the southern edge of the domain, except that the maximum of 24-h rainfall reaches mm, which is much larger than the observed amount of 138 mm. It is notable that there is only conventional surface and upper observations in simulation domain except Beijing areas where a high-resolution AWS network is working. The rainfall observation in Xingtai surely proves the occurrence of the heavy rainfall, but may not be representative of the maximum rainfall nearby. Whatever, it is certain that the simulation of rainfall in Xingtai areas is successful. From the distribution of rainfall respectively from the explicit scheme and Grell cumulus parameterization scheme of Exp.1 (figure omitted ), it can be found that the simulated distribution and the intensity of the heavy rainfall over Beijing areas mostly result from the simple explicit ice scheme (with 10 km horizontal resolution). The Grell cumulus parameterization scheme only plays a minor role in the simulated rainfall. On the contrary, the Grell cumulus parameterization is responsible for most of the local heavy rainfall in Xingtai and its surrounding areas, showing that the rainfall there mainly results from the synoptic scale convective system. The Beijing radar observation and retrieved TBB data from Japanese GMS5 satellite images (figure omitted) indicate that the above numerical simulations are in good agreement with the observations. Namely, the July 2000 torrential rain over Beijing is mainly driven by small-scale convective activities, which is augmented by the larger-scale weaker convective system but the precipitation near Xingtai obviously results mainly from mesoscale convective activities. 3.4 Data assimilation influence on the initial fields of the numerical model On the initial moisture field Figure 3a shows the precipitable water vapor at initial time in the simulation of Exp.1. It can be seen from Fig.3a that an extensive south-northoriented Ω-shaped moisture tongue is located over Beijing, in front of the mountains. The windward side of the mountains exhibits a strong southwest-northeast-

8 396 ACTA METEOROLOGICA SINICA VOL.20 oriented moisture gradient with precipitable water vapor highly reaching 6 cm. These results indicate that the initial moisture field distribution is very favorable for the genesis/development of heavy rainfall. The above initial moisture field of model indicates that the moisture characteristics, plentiful condition, and advantageous advection, previous to the precipitation occurrence, could be reasonably captured while all kinds of meteorological observations have been adopted with the 3DVAR technique. Figure 3b illustrates the initial IPW analysis difference between Exp.1 and Exp.3. It presents the further influence from the nonconventional GPS-IPW data assimilation, after introducing conventional meteorological observations and the background fields, on the improvement of the atmospheric moisture fields at the model s initial step. As clearly shown in Fig.3b there are extensive positive (negative) difference areas of the model s initial IPW fields between Exp.1 and Exp.2. And the maximum (minimum) difference reaches +5.6 (-3.3) mm, the former (latter) central value is located over the southwest/east area to Beijing, respectively. Moreover, a density transition zone between them is just along a southeast-northwest axis, and its center is just over Beijing. It clearly indicates that GPS-IPW data have an important effect on the water vapor analysis of the model s atmosphere. Introducing GPS-IPW data into initial analysis by 3DVAR scheme, both the pattern and intensity of model s IPW are much greatly influenced. And the assimilated initial IPW data contribute to the increase (decrease) of the atmospheric integrated precipitable water vapor in the southwest (northeast) areas of Beijing, and then help strengthen the southwest-northeastward vapor gradient variation over the southwest areas of Beijing, especially Mentougou District. Obviously, this effect of ground-based GPS-IPW data on the initial atmospheric moisture fields is helpful to enhancing the advantageous condition for the future evolution of the heavy rain of Beijing under Exp.1, and then forms the advantageous water vapor distribution pattern as shown in Fig.3a. Notice that 10 GPS-IPW remotesensing stations are within Beijing and its adjoining regions (see Fig.1d), and that the range of positive (negative) difference shown in Fig.3b is more extensive, suggesting that the GPS-IPW observations exert extensive impacts because of the general balance mechanism restriction among all analysis variables from the cost function of the 3DVAR scheme, and the dynamical/thermodynamical physical conditionality in the model. For the local area over the GPS-IPW network (Beijing and its adjoining regions), the major role of GPS-IPW assimilation is to strengthen reasonably the moisture gradient over Beijing City, so that the precipitation simulation under Exp.1 is better than that under Exp.3. Figure 3c indicates the initial analysis difference distribution of the IPW field of Exp.6 minus Exp.7, which reflects the assimilation impact of only introducing the non-conventional ground-based GPS-IPW observations on the basis of background fields. By contrasting Fig.3c with Fig.3b, it can be seen that the horizontal distribution characteristics of the positive and negative difference areas are similar, further proving the above conclusion of the ground-based GPS- IPW data assimilation effects. However, by comparing the distribution and central intensity between Exp.1 minus Exp.3 and Exp.6 minus Exp.7, it should be noticed that under Exp.1 minus Exp.3 (see Fig.3c) the negative (positive) difference area in the northeast (southwest) to Beijing is larger (smaller), and the central intensity is stronger (weaker) than those of Exp.6 minus Exp.7 (see Fig.3b). This fact shows that the restriction among ground-based GPS water vapor monitoring, conventional meteorological observations and background fields is an interactive and complex process in the 3DVAR system. Each of the individual component of meteorological observation network makes its special contribution to the generation of the initial moisture field of the numerical model. The initial assimilations of the upper and surface conventional meteorological observations also play an advantageous restriction role in the analysis improvement of the moisture fields. The direct and indirect physical balance restricted among different kinds of observations makes the analysis results better reflect the real conditions of the atmosphere in the variational assimilation analysis. And it is also the reason why under

9 NO.4 ZHANG Chaolin, CHEN Min, KUO Ying-Hwa, FAN Shuiyong and ZHONG Jiqin 397

10 398 ACTA METEOROLOGICA SINICA VOL.20

11 NO.4 ZHANG Chaolin, CHEN Min, KUO Ying-Hwa, FAN Shuiyong and ZHONG Jiqin 399 Exp.1 the better precipitation-simulated TS are achieved in respect to the observational stations rather than under Exp.6, which is on the basis of Exp On the initial thermal and dynamical fields Figure 4 presents the initial difference distribution for horizontal wind and temperature fields at the sigma level. Although only additional 10 GPS- IPW data are further assimilated relative to Exp.2, from the initial thermal and dynamical difference fields of Exp.1-Exp.2 (see Fig.4a), Exp.1 is more helpful for forming the warm and humid (cold and dry) air-mass distribution in the southeast (northwest) areas in Beijing, and is more advantageous to forming a northeastsouthwestward wind shear zone over Beijing in the lower atmosphere. Obviously, those differences are helpful for the simulation of the heavy rainfall in the upcoming 24 hours, which shows that the improvement of precipitation forecast in Exp.1 not only lies in the fact that the GPS-IPW data assimilated in initial time provide more useful information, as discussed in the previous section, on atmospheric water vapor structure and its distribution in the early period of this rainfall, but also the other fact that with the 3DVAR assimilation method Exp.1 integrally considers the interaction among observation, background, model and their errors, especially instead of simply taking the observation as the true state of atmosphere. Therefore, compared to Exp.2, in which initial fields are analyzed by the MM5 Litter R scheme presuming the observations directly as the true atmospheric conditions, the GPS-IPW monitoring data assimilation of Exp.1 also makes significant improvements in the analysis quality of the thermal and dynamic fields at the initial assimilation step, and then increases the level of precipitation simulation. Figure 4b shows the possible impacts of numerical experiments with or without the GPS-IPW 3DVAR assimilation on the atmospheric thermal and dynamical conditions. It can be found that the difference in Fig.4b is not obvious, and even could be negligible compared with those of Exp.1 minus Exp.2. It correctly shows the fact that the GPS- IPW data assimilation should mainly affect the water vapor analysis rather than the atmospheric wind and temperature fields, because water vapor information is analyzed as a passive and independent of other analysis variables in the present MM5/WRF 3DVAR system. Figures 4c and 4d give the difference distribution in the lower troposphere from Exp.4 minus Exp.7 (Fig.4c) and Exp.5 minus Exp.7 (Fig.4d), and respectively show the roles of the upper-layer and surface conventional meteorological observations in the model s initial fields. Figure 4c indicates the data assimilation influences of conventional upper-layer observational networks on the precipitation forecast improvement, which mainly results from strengthening the lower-layer warm-advection transportation of the east wind component over Beijing areas. Meanwhile, the major roles of surface meteorological observations are to strengthen the advection confluence from the colder north (warmer south) wind component in the western (eastern) areas to Beijing at low levels (see Fig.4d). Moreover, the analysis differences of the temperature field in Fig.4c are obviously smaller than those in Fig.4d. It further suggests that in this precipitation event, the surface data assimilation is more helpful for objectively describing the true temperature information in the lower-layer atmosphere relative to the upper-layer meteorological observations. This reveals, to a certain extent, the reason why better TS of simulated precipitation are obtained with conventional surface observation assimilation, rather than with acquired conventional upper-layer meteorological data. 4. Conclusions By using the MM5/WRF 3DVAR system and the nonhydrostatic MM5 model, the meteorological data in Beijing and its surrounding areas, such as the ground-based GPS precipitable water vapor, radiosonde, upper-wind pilot reports, conventional surface observations, and automatic weather stations, were successfully assimilated at the initial step. And then 24-h numerical application experiments were also carried out to investigate the impact of the individual component of meteorological observations on the July 2000 torrential rain. The main conclusions can be drawn as follows: (1) With the MM5/WRF 3DVAR system more

12 400 ACTA METEOROLOGICA SINICA VOL.20 nonconventional atmospheric motoring data, such as ground-based GPS-IPW observations, can be directly incorporated in the assimilation step. Therefore it can be widely used in meteorological numerical analysis and NWP applications. Compared to the objective analysis method this 3DVAR system provides higher quality initial fields for MM5, and then improves the accuracy of short-term precipitation forecast because its initial analysis fields cover more abundant and reasonable moisture, thermo- and dynamical information of the monitored atmosphere. (2) Further analysis also shows that each of the individual component of meteorological observation network (the nonconventional ground-based GPS precipitable water vapor, automatic and conventional meteorological observations) makes its special contributions to the improvement of initial field analysis and forecasting skills. The ground-based GPS-IPW data provide useful moisture information closely related with strong convective weather events. This study shows that the ground-based GPS-IPW data obviously play positive roles in the improvement of the initial field assimilation and precipitation forecast accuracy after the supplement information was ingested from the other meteorological observations. (3) There exits a theoretic disagreement upon the true-state atmospheric representation between variational assimilation and objective analysis methods. In general, the traditional NWP verification, which is using the traditional objective analyses as the true states of the atmosphere, is inconsistent with the theoretic foundation of variation method, in which observational errors always are regarded as one objective and important factor on data analysis. In the future, more objective and effective methods for the NWP verification should be developed in numerical applications of data variational assimilation. (4) This research only focuses on one severe rain case, which has given a clue to assess the impact of individual components of the existing meteorological observation network on high influence weather events or climate changes with 3DVAR numerical system. And further investigation into it is still necessary in the future. Acknowledgments. We would like to thank Dr. Dale Barker (NCAR, USA), Prof. YongRun Guo (NCAR, USA), Prof. Wei Huang (NCAR, USA), Dr. Hans Xiang-Yu Huang (Denmark Meteorological Institute) for valuable technical guidance to 3DVAR applications. Engineer Chu Yanli (IUM, China), Senior Engineer Zhang Jingjiang (IUM, China), and Senior Engineer Wang Jingli (IUM, China) provided assistance in processing the ground-based GPS-IPW data. Prof. Mao Jietai (Peking University, China), Associate Prof. Li Chengcai (Peking University, China), Prof. Fang Zongyi (NSMC, China), and Prof. Xia Qing (NSMC, China), and Dr. Cao Yunchang (NSMC, China) took part in the 2000 GPS observational experiment. REFERENCES Barker, D. M., W. Huang, Y. -R. Guo, and A. Bourgeois, 2003: A Three-Dimensional Variational (3DVAR) Data Assimilation System for Use with MM5. Available from MM5 3DVAR Web-site Barker, D. M., Y. R. Guo, W. Huang, A. J. Bourgeois, and Q. N. Xiao, 2004: A three-dimensional variational data assimilation system for MM5: Implementation and initial results. Mon. Wea. Rev., 132, Dudhia, J., 1993: A nonhydrostatic version of the Penn State/NCAR mesoscale model: Validation tests and simulation of an Atlantic cyclone and cold front. Mon. Wea. Rev., 121, Dudhia, J., D. Gill, K. Manning, W. Wang, C. Bruyere, et al., 2003: PSU/NCAR Mesoscale Modeling System, Tutorial Class Notes and User s Guide: MM5 Modeling System Version 3. Guo Y. R., Y. H. Kuo, J. Dudhia, D. Parsons, and C. Rocken, 2000: Four-dimensional variational data assimilation of heterogeneous mesoscale observations for a strong convective case. Mon. Wea. Rev., 128, Lorenc, A. C., S. P. Ballard, R. S. Bell, et al., 2000: The Meteorological Office global three-dimensional variational data assimilation scheme. Quart. J. Roy. Meteor. Soc., 126, Parrish, D. F., and J. C. Derber, 1992: The National Meteorological Center s spectral statistical

13 NO.4 ZHANG Chaolin, CHEN Min, KUO Ying-Hwa, FAN Shuiyong and ZHONG Jiqin 401 interpolation analysis system. Mon. Wea. Rev., 120, Zhang Chaolin, Ji Chongping, Kuo Ying-Hwa, et al.,2005: Numerical simulation of topography effects on the July 2000 severe rainfall in Beijing. Progress in Natural Sciences, 15(9), Zhang Chaolin and Wang Yingchun, 2002: Numerical study on weather influences of location design of wind profiler network for Beijing Area. Acta Meteorologica Sinica, 60(6), (in Chinese) Zhang Xin, Wang Bin, Ji Zhongzheng, et al., 2002: Data assimilation research of 3DVAR on July 1998 heavy rainfall in Wuhan. Progress in Natural Sciences, 12(2), (in Chinese)