5 Center for Atmospheric Sciences, Scripps Institution of Oceanography, La Jolla, California, USA
Submitted to Journal of Climate
LASG, Institute of Atmospheric Physics,
Chinese Academy of Sciences,
P. O. Box 9804,
Beijing 100029, China.
Fax : 86-10-8299-5172
The performance of an interim version of the NCAR Community Atmospheric Model, namely CAM3.5, in simulating the East Asian Summer Monsoon (EASM) is assessed by comparing model results against observations and reanalyses. Both the climate mean states and seasonal cycle of major EASM components are evaluated. Special attention is paid to the sensitivity of model performance to changes in the convection scheme. This is done by analyzing four CAM3.5 runs with identical dynamical core and physical packages, but different modifications to its convection scheme, i.e. the original Zhang-McFarlane (ZM) scheme, Neale and Richter’s (NZM) modification, Wu and Zhang’s (WZM) modification and Zhang’s (ZZM) modification. The results show that CAM3.5 can capture the major climate mean states and seasonal features of EASM circulation system, including reasonable simulations of the Tibetan High in the upper troposphere and the Western Pacific Subtropical High (WPSH) in the middle and lower troposphere. The main deficiencies of model performance are found in monsoon rainfall and the meridional monsoon cell. The models fail to reproduce the subtropical monsoon rain belt and the related meridional monsoon circulation. This is caused by the weak meridional land-sea thermal contrasts in the model which contribute to the weaker monsoon circulation, and then insufficient rainfall in both tropical and subtropical regions of EASM. The seasonal migration of rainfall, as well as the northward jump of WPSH from late spring to summer, is reasonably simulated, except that the northward jump of the rain belt still needs to be improved. The monsoon rainfall distribution and its seasonal variation are sensitive to the modification of the ZM convection scheme, but the monsoon circulation exhibits less sensitivity. Three runs using modified schemes generally improve the model performance in EASM compared to the control run. The tropical rainfall distribution and the rainfall centers over the monsoon trough are more reasonably reproduced in the NZM. The WZM and ZZM have more realistic subtropical rainfall in the Meiyu/Baiu/Changma front as the closure assumption in these two modifications are explicitly linked to the large scale temperature and moisture advection. No evidence supports the idea that the multi-run ensemble is better than individual run.
The Asian summer monsoon is one of the most energetic components of the earth’s climate system, and it plays an important role in global climate variability (e.g. Ding, 1994; Webster et al., 1998; Wang et al., 2001). Numerous studies have documented that the huge Asian summer monsoon system can be divided into two subsystems: the Indian summer monsoon (ISM) and the East Asian summer monsoon (EASM) systems, which are to a large extent independent of each other, and at the same time, interacting with each other (Zhu, 1934; Yeh et al., 1959; Zhu et al., 1986; Tao and Chen, 1987). The EASM is not an eastward extension of the ISM. Lying downstream of the world’s highest Tibetan Plateau and in between the Eurasian continent and the Pacific Ocean, EASM has its own unique features. The ISM is a tropical monsoon in which the low-level winds reverse from winter easterlies to summer westerlies, whereas the EASM is a hybrid type of tropical and subtropical monsoon (Zhu et al., 1986; Tao and Chen, 1987; Chen et al., 1991; Ding, 1994). The generally recognized EASM system consists of three main components, namely, the East Asian Meiyu/Baiu/Changma front (a major rain-bearing system in the subtropics and mid-latitude), the Western Pacific Subtropical High (WPSH), and the tropical western Pacific monsoon trough or the western Pacific Intertropical Convergence Zone (ITCZ). These three components are intimately coupled (Nitta, 1987; Huang and Wu, 1989; Liu et al., 2008).
Simulations of the Asian summer monsoon and its variability have proven to be one of the most challenging issues (Kang et al., 2002; Wang et al., 2005). Many general circulation model (GCM) results have shown that the basic distribution of monsoon circulation, such as the subtropical high, the equatorial jets, and the monsoon low, etc., can be reproduced reasonably, but their strengths and variations are hard to simulate well (e.g., Hoskins and Rodwell, 1995; Zhou and Li, 2002). The monsoon rainfall is even harder to simulate (Kang et al., 2002). A wide skill range exists among GCMs in simulating monsoon precipitation, which is largely attributed to the different sub-grid scale parameterization schemes and horizontal resolutions (Sperber et al., 1994). Great efforts have been devoted to the simulations of the Indian monsoon (e.g., Yang et al., 1996; Meehl and Arblaster, 1998; Loschnigg et al., 2003; Meehl et al., 2006), but less attention has been paid to the East Asian monsoon. Due to the complex topography and the sharp land-sea thermal contrast, the EASM is much more complex than the ISM, and the model performances for the EASM are rather limited (Liang and Wang, 1998; Yu et al., 2000; Zhou and Li, 2002).
Atmospheric convection affects large-scale circulation and wave disturbances through the release of latent heat of condensation and vertical transport of heat, moisture and momentum. The large-scale forcing in turn influences and modulates the development and organization of convection and clouds (Wu et al., 2007). The coupling of convective processes with the large-scale dynamics is crucial for modeling the global distribution of precipitation (Zhang, 2005). Since rainfall in the EASM system results from a mixture of convection and stratiform clouds, and because there are planetary, synoptic and mesoscale systems in the EASM rain belts, the simulation of EASM rainfall and circulation depends strongly on the convection scheme (Huang et al., 2001).
The Community Atmosphere Model (CAM), as well as its previous version Community Climate Model (CCM), have been widely used for climate research (e.g., Ghan et al., 1996; Zhang, 2003; Collier, and Bowman, 2004; Zhou and Yu, 2004; Deser and Phillips, 2006; Hack et al., 2006). The model has also been used for monsoon studies (e.g. Hoerling et al., 1990; Yu et al., 2000; Liu et al., 2002). Meehl and Arblaster (1998) compared the aspects of the Asian–Australian monsoon system associated with El Nino–Southern Oscillation in global coupled Climate System Model (CSM) and its atmospheric component CCM3. They showed that the CSM captured most major features of the monsoon system in terms of mean climatology, interannual variability, and connections to the tropical Pacific, with largest discrepancies between the CSM, observations, and the CCM3 simulation in the equatorial eastern Indian Ocean and near the Philippines. Meehl et al. (2006) also examined the simulations of regional monsoon regimes in CAM3 and its coupled CCSM3. They noted that the major monsoon features were well represented in all simulations, but the details of the regional simulations were model dependent. Some aspects of the monsoon simulations, particularly in Asia, were improved in the coupled model compared to the SST-forced simulations. Nevertheless, the above analyses still concentrated on the tropical monsoon, i.e. the south Asian monsoon, and little attention was given to the EASM, especially its subtropical components. EASM has proven to be a good test-bed for climate models over large and complex topography (e.g., Zhou and Li, 2002). A comprehensive evaluation on the performance of CAM3.5 in EASM simulation will potentially reveal strengths and weaknesses of the model and may also serve as a useful reference for future model improvement.
The CAM model is recently improved by modifying the Zhang and McFarlane (1995, hereafter ZM) convection scheme and the representation of other physical processes. The purpose of this paper is to evaluate the performance of CAM3.5 in simulating the EASM. We focus on the EASM climatology and seasonal variation. We want to address the following questions: 1) What are the strengths and weaknesses of the NCAR CAM3.5 model in EASM simulation? 2) What are the influences of the modifications to the convection scheme on EASM simulations? Our results show that the model can capture the main characteristics of EASM circulation, but the simulation of Meiyu/Baiu/Changma rain belt and the meridional monsoon cell still needs to be improved. The precipitation is sensitive to changes in convection scheme, whereas the monsoon circulation is less so. The multi-run ensemble is not always superior to the individual simulation.
The rest of the paper is organized as follows. Section 2 provides a description of the model and three sets of modifications to the ZM convection schemes, as well as the observational and reanalysis datasets used in this study. The EASM climate mean state is depicted in section 3, while section 4 addresses seasonal and intraseasonal monsoon variation. Section 5 follows with conclusions and a brief discussion.
The CAM3.5, developed by the National Center for Atmospheric Research (NCAR) in collaboration with the climate modeling community, is the recently improved version of the state-of-the-art Atmospheric General Circulation Model (AGCM) and serves as an interim version to improve the model physics for the next generation CAM4. It is a global primitive equation model with 26 vertical levels. The integration used finite-volume dynamical core (Lin, 2004) and the horizontal resolution is about 2.5° longitude by 1.9° latitude. This version is closely related to its previous version CAM3 and includes changes to convection and cloud processes, land model and chemistry modules (Oleson et al., 2008; Stockli et al., 2008; Neale and Mapes, 2008; Neale et al., 2008; Richter and Rasch, 2008). The calculation of cloud fraction is updated and new hydrology, surface datasets, canopy integration are introduced in the land model. The other physical processes are the same as those of CAM3, which can be found in Collins et al. (2006). The CAM3 uses the standard Zhang-McFarlane (ZM, Zhang and McFarlane, 1995) convection parameterization scheme, which is a mass flux scheme inspired by the convective parameterization of Arakawa and Schubert (1974). An updraft ensemble of entraining convective plumes, all having the same mass flux at cloud base, relaxes the atmosphere toward a threshold value of convective available potential energy (CAPE). Simulations are not sensitive to the threshold used. In-cloud saturated downdrafts commence at the level of minimum saturated moist static energy. Detrainment of ascending plumes also begins at the level of minimum saturated moist static energy. Therefore, only ascending plumes that can penetrate through the conditionally unstable lower troposphere are present in the ensemble. In an effort that led to the development of CAM3.5, three revised versions of the Zhang-McFarlane convection scheme were tested. One was the version revised by Neale and Richter (hereafter NZM), in which the calculation of CAPE in the ZM scheme was modified to include the effect of lateral entrainment dilution (Neale and Mapes, 2008). In addition, convective momentum transport parameterized by Gregory et al. (1997) was included (Richter and Rasch, 2008). The second was the free tropospheric quasi-equilibrium version (hereafter ZZM) of the ZM scheme proposed by Zhang (2002) and used in Zhang and Mu (2005). In this version, in addition to changing the CAPE-based closure in the original ZM scheme to quasi-equilibrium-based, a relative humidity threshold of 80% at the parcel lifting level was imposed for convection. The third one is the Wu and Zhang version (hereafter WZM), which uses the ZZM version, but with the relative humidity threshold replaced by a threshold in large-scale CAPE forcing, based on cloud-resolving model simulation estimates (Wu et al., 2003, 2007; Zhang and Wu, 2003). Furthermore, as in the NZM version, convective momentum transport was also included, but using the more sophisticated parameterization of Zhang and Cho (1991). In this study, four AMIP-type integrations, forced by observed SSTs, are performed and compared by using the three revised versions as well as the original ZM scheme. They will be referred to as the NZM, WZM, ZZM and control runs, respectively. Table 1 shows a brief summary of each experiment. All the experiments are integrated more than 20 years and the data of 1980-1999 are used in this study.
In this paper, East Asia refers to the area between 90°E and 140°E and between 0°N and 45°N, including the monsoon region over the East Asia and western northern Pacific (Wang and Lin, 2002; Ding and Chan, 2005). To evaluate the model, the following reanalysis/observation datasets are used: 1) geopotential heights, zonal and meridional winds, air temperature, specific humidity from ERA40 reanalysis during 1980-1999 on a 2.5° grid (Uppala et al., 2005); and 2) CPC Merged Analysis of Precipitation (CMAP) based on blended satellite and in situ measurements during 1980-1999 on a 2.5° grid (Xie and Arkin, 1997).
Climatological mean pattern
Precipitation is one of the most important variables to measure monsoon activities. The 20-year mean JJA precipitation over East Asia is shown in Fig. 1. From the observation (Fig. 1a), two major rain belts can be seen. One is the tropical monsoon trough located between 10-20°N, as the reflection of intense rainfall corresponding to the ITCZ in tropical western Pacific. The other, known as the subtropical Meiyu/Baiu/Changma front, exhibits a zonally elongated structure extending from East China to northwestern Pacific around 30°N, with heavy rainfall in the lower reaches of the Yangtze River in China, South Korea and southwestern Japan. These two rain belts are closely related to each other, with a relatively dry region dominated by the WPSH between them. Overall, all simulations can reproduce a spatial pattern similar to that of the CMAP precipitation, although there are certain differences. Before examining the results from each modification of the scheme, the ensemble mean precipitation is calculated by averaging the results of four runs. The mean precipitation has a spatial pattern similar to that of CMAP, but the biases are also obvious, especially over the western Pacific (Fig. 1b). The largest deficiency is the underestimation of precipitation over the subtropical rain belt. The observed rain belt extends from the east flank of the Tibetan Plateau to the mid-Pacific, while the simulated rain band is located along the landward side of the eastern coast of Asian continent. The rainfall centers associated with Meiyu/Baiu/Changma front over the middle and lower reaches of the Yangtze River valley, South Korea and southwestern Japan are missing in the model. In the tropical region, the model simulates much weaker precipitation over the South China Sea and the subtropical western Pacific. The model produces excessive precipitation over the Tibetan Plateau and its eastern flank. This may be a systematic bias in NCAR atmospheric models, e.g. Yu et al. (2000) found that the NCAR CCM3 simulated unrealistically strong precipitation on the eastern periphery of the Tibetan Plateau. They attribute the heavy rainfall to the unrealistically strong surface sensible heating over the southeast and northeast of Tibetan Plateau that favors the forming of a powerful subtropical anticyclone over the eastern China.
The results for different modifications of the convection scheme are shown in Figs. 1c-1f. They all show similar biases to those in the ensemble average, e.g., the landward distribution rather than zonally elongated subtropical rain belt, the missing of rainfall centers associated with the Meiyu/Baiu/Changma front, the underestimated precipitation over the South China Sea and Philippines Sea, and the excessive rainfall over the Tibetan Plateau and its eastern periphery. The landward distribution of rainfall is obvious in these runs, leading to the heavy rainfall shift northward over the continent. However, regional details are different among the four runs. In the control run, the tropical rainfall over the western Pacific is greatly underestimated, i.e., the precipitation amount over the South China Sea and the Philippines Sea is less than 6 mm day-1, which is only about half of that in the observation. The rainfall over the central eastern China (110-120°E, 30-40°N) is overestimated, but that over the southeast coastal regions is underestimated. The NZM run reproduces the two major rain belts in EASM, but the amplitude is weaker than the observation, especially over the Meiyu/Baiu/Changma rain belt. Similar results are also evident in the WZM and ZZM runs, but they have more realistic Meiyu/Baiu/Changma rain belts compared to the NZM run. The rainfall centers over the South Korea and southwestern Japan are only evident in the WZM run, except with a weaker intensity. The tropical rain belt is separated into two parts in the ZZM run, as the rainfall over the South China Sea is dramatically decreased. The simulation of EASM precipitation is generally improved in three modifications compared to the ZM scheme, particularly in the tropical region. None of the simulations realistically reproduces the observed Meiyu/Baiu/Changma rain band in the region from East China to the mid-Pacific, but the WZM and ZZM runs show better results than the NZM run. The simulation of the Meiyu/Baiu/Changma rain band seems to be a universal problem in current state-of-the-art AGCMs (Kang et al. 2002). The Meiyu/Baiu/Changma front is a rain band of thousands of kilometers long across East Asia and the northwestern Pacific, and is closely related to the planetary scale circulation. Precipitation along the Meiyu/Baiu/Changma front is usually caused by convection and clouds organized into eastward moving mesoscale convective systems (Chen et al., 1991). Because WZM and the ZZM are explicitly related to the large scale forcing, they do better than the NZM over the subtropical rain belt.
To quantitatively evaluate the model’s performance in simulating the geographical distribution of monsoon precipitation, a Taylor diagram (Taylor, 2001) is employed to compare the ability of different runs in East Asia (Fig.2). Before the comparison, model results are interpolated to the spatial grid of the observation. The rainfall distributions over the tropical (5-15°N) and subtropical (25-35°N) regions are also compared. The standard deviations of the simulations are smaller than those of the observation except for the subtropical rainfall in the ZZM run, suggesting that the spatial variance in the model is smaller than the observation. In comparison with the control run, the three revised schemes improve the monsoon rainfall simulations. This improvement is evident in both tropical and subtropical rainfalls. The NZM run simulates the EASM rainfall more realistically in its tropical rain belt, but weakly in the subtropical rainfall. The subtropical rainfall simulated by WZM is the most reasonable, which may be related to its closure assumption. The performance of multi-run ensemble is generally better than most individual schemes, indicating that the multi-run ensemble can overcome some biases in the individual run. However, the ensemble result is not always the best, which may be caused by the high resemblance among the four versions of the ZM scheme compared here.
EASM horizontal circulation
Monsoon rainfall distribution is closely related to monsoon circulation. In the upper level (100 hPa), the most outstanding feature of EASM is the huge anticyclonic circulation (the so-called Tibetan High) centered over the southern edge of the Tibetan Plateau, with the axis of the anticyclone along 30°N, westerly jet to the north along 40°N and the Tropical Easterly Jet (TEJ) to the south of 25°N (Fig. 3a). This anticyclone provides a favorable divergent field for the underlying convective activity along the Meiyu/Baiu/Changma front. The multi-run ensemble reasonably reproduces the structure of the Tibetan High except with a weaker strength (Fig.3b), indicating a weaker heating from the Tibetan Plateau. Contrary to the ensemble mean, the intensity of the High simulated in the control run is stronger than the reanalysis, and the center of the High shifts westward (Fig.3c). The Tibetan High simulated by three revised versions of the ZM scheme (Figs.3d-f) shows similar patterns to the ensemble mean, suggesting that the upper level circulation is less sensitive to changes in the convection scheme.
To quantify the model performance, the intensity of Tibetan High (defined as the averaged geopotential height over 60-120°E, 20-40°N), the zonal wind speed of westerly jet (defined as the zonal wind speed averaged over 60-120°E, 40-50°N) and the TEJ (defined as the zonal wind speed averaged over 60-120°E, 10-20°N) are listed in Table 2. In the control run, the intensity of the Tibetan High and the westerly jet are stronger than the reanalysis, while the TEJ is slightly weaker. The intensities of the High, westerly jet and TEJ simulated by the three revised versions of the ZM scheme are weaker than the reanalysis. The weak High indicates a weak divergence in the upper troposphere. The weak TEJ in the simulation is consistent with the weak rainfall over the tropical monsoon trough, as the TEJ is closely linked to monsoon rainfall in Asia through the meridional vertical circulation (Zeng and Guo, 1982).
In the middle and lower troposphere, WPSH greatly influences the climate of EASM. The position, shape and strength of WPSH dominate the large-scale quasi-stationary frontal zones and associated rain band in East Asia (Tao and Chen, 1987; Ding, 1994; Zhou and Yu, 2005). Here we use 500-hPa geopotential height to measure the WPSH, which is widely used in previous studies (e.g., Yu et al., 2000; Liu et al., 2002; Zhou and Li, 2002). In the ERA40 reanalysis, a strong anticyclone dominates the subtropical western Pacific and a weak trough appears over the northeastern China (Fig.4a). Three indices, which respectively represent WPSH intensity (IS), westward extension (IW) and northern edge (IN), are calculated to make an objective comparison between the reanalysis and the model. IS is defined as the regional average of the grids with geopotential height greater than 5860 gpm over the region 100-140°E and 10-40°N. To compare the westward extension of WPSH in reanalysis and model simulations, each of the 500-hPa geopotential height values over the selected region are subtracted by IS. IW is then defined as the longitude of the westward extension of zero gpm contours of the subtracted field (Zhou et al., 2008). The WPSH ridge, defined as and (Li and Chou, 1998), is constructed before calculating IN. Afterward, IN is defined as the latitude of WPSH ridge position. The values of three indices are listed in Table 3.
The WPSH in the multi-run ensemble resembles the reanalysis (Fig. 4b), except that it extends westward about 20°and the ridge shifts northward more than 5° (Table 3). The westward extension of WPSH, along with the downward motion, is consistent with the underestimated precipitation over the subtropical regions of East Asia. The WPSH can also accelerate the southwestern monsoon flows to reach relatively higher latitudes of East Asia. The northward shift of the WPSH ridge results in the southerlies deeply penetrating into the northern China. This corresponds to the excessive precipitation in North China but deficient Meiyu/Baiu/Changma rainfall along the Yangtze River (Fig. 1). The mid-latitude westerly in the model seems to be more flat. No trough is evident in this zone. The WPSH simulated by the control run is much stronger than the reanalysis, and the subtropical Eurasian continent is dominated by the anticyclone (Fig.4c). Although the simulations are improved in three revised versions of ZM scheme (Figs.4d-4f), the northward shift of the WPSH still exist, especially in the ZZM version (Table 3). The westward extension is obvious in all the three revised schemes, with the WZM run slightly better. Generally, the stronger WPSH is associated with a westward extension and northward shift. According to Rodwell and Hoskins (2001), the westward extension can be attributed to the intensified rainfall over the ISM region (Fig. 1) in term of the Sverdrup vorticity balance. This bias of WPSH then causes the drier bias of rainfall over the western Pacific, which is dominated by the ridge.
The water vapor transport is crucial to monsoon rainfall, and closely resembles the large-scale monsoon circulation in the lower troposphere (Zhou and Yu, 2005). From Fig.5a, three main branches of water vapor transport to East Asia can be found: a strong transport by southwesterly from ISM, a moderate transport by southeasterly from the western Pacific, and a weak one linked to the cross-equator flow straddling 105°E-150°E. The southwesterly transport in all three simulations is weaker than the reanalysis (Figs. 5b- 5f) which lead to the weak rainfall centers over the Bay of Bengal and South China Sea. The simulated moisture transport turns northwestward in the north of the Bay of Bengal around 20° N and this leads to a weaker contribution of the southwesterly monsoon flow to the water vapor transport over the EASM. The southeasterly transport of the water vapor from the western Pacific shifts westward and extends northward as a result of the biases in WPSH simulation. This deficiency is related to the weak rainfall associated with the subtropical Meiyu/Baiu/Changma front and the relatively strong rainfall over the mid-latitude continent north of 45°N. Therefore, the bias of water vapor transport is consistent with that of precipitation shown above.