754

ACTA METEOROLOGICA SINICA

VOL.25

Comparison of Dierent Order Adams-Bashforth Methods in an Atmospheric General Circulation Model

) and ZHANG Bo (

National Meteorological Center, China Meteorological Administration, Beijing 100081

ABSTRACT The Asselin-Robert time lter used in the leapfrog scheme does degrade the accuracy of calculations. As an attractive alternative to leapfrog time dierencing, the second-order Adams-Bashforth method is not subject to time splitting instability and keeps excellent calculation accuracy. A second-order AdamsBashforth model has been developed, which represents better stability, excellent convergence and improved simulation of prognostic variables. Based on these results, the higher-order Adams-Bashforth methods are developed on the basis of NCAR (National Center for Atmospheric Research) CAM 3.1 (Community Atmosphere Model 3.1) and the characteristics of dynamical cores are analyzed in this paper. By using Lorenz nonlinear convective equations, the ltered leapfrog scheme shows an excellent pattern for eliminating 2t wave solutions after 20 steps but represents less computational solution accuracy. The fourth-order AdamsBashforth method is closely converged to the exact solution and provides a reference against which other methods may be compared. Thus, the Adams-Bashforth methods produce more accurate and convergent solution with dierencing order increasing. The Held-Suarez idealized test is carried out to demonstrate that all methods have similar climate states to the results of many other global models for long-term integration. Besides, higher-order methods perform better in mass conservation and exhibit improvement in simulating tropospheric westerly jets, which is likely equivalent to the advantages of increasing horizontal resolutions. Based on the idealized baroclinic wave test, a better capability of the higher-order method in maintaining simulation stability is convinced. Furthermore, after the baroclinic wave is triggered through overlaying the steady-state initial conditions with the zonal perturbation, the higher-order method has a better ability in the simulation of baroclinic wave perturbation. Key words: Adams-Bashforth method, Lorenz nonlinear convection equations, baroclinic wave test, amplitude accuracy Citation: Zhao Bin and Zhang Bo, 2011: Comparison of dierent order Adams-Bashforth methods in an atmospheric general circulation model. Acta Meteor. Sinica, 25(6), 754764, doi: 10.1007/ s13351-011-0606-6.

1. Introduction Since Bjerkness (1904) rst proposed the scientic concept of Numerical Weather Prediction (NWP), the atmospheric general circulation models (AGCMs) have acheived broad development and application in weather forecasting (Simmons and Temperton, 1997; Chen and Xue, 2004). The deciency in the treatment of the model dynamical framework and in the accuracy of numerical integration is always a major limiting factor for model development. Currently, the traditional Eulerian treatment of advection in the three-time-level (leapfrog) time differencing scheme is generally used in AGCMs. As an
Supported Corresponding

ecient method, the leapfrog scheme achieves secondorder accuracy with one function evaluation per time step. The leapfrog scheme neither articially damps linear oscillatory motion nor produces instability by amplifying the oscillations (Durran, 1991). However, these advantages are greatly diminished by the large phase-speed error of the leapfrog scheme. The AsselinRobert time lter (Asselin, 1972; Robert, 1966) is always used to reduce the eects of these disadvantages but it degrades the accuracy of the calculations (Feng et al., 2002; Ji et al., 1980, 2005). Recently, a modication on the Asselin-Robert time lter has been introduced (Williams, 2009) as a simple method to ensure the averaged-state conservation, achieving

by the China Meteorological Administration Special Fund for Numerical Prediction of GRAPES (2200504). author: [email protected]. c The Chinese Meteorological Society and Springer-Verlag Berlin Heidelberg 2011

NO.6

ZHAO Bin and ZHANG Bo

755

third-order amplitude accuracy. Yong (1968) and Pfeer et al. (1992) pointed out that two categories of methods can serve as substitutes of the leapfrog scheme. The rst category is the updating method with multistep, i.e., the rstorder Matsuno scheme (Matsuno, 1966), which performs two function evaluations per time step and heavily damps the solution. Although the second-order Runge-Kutta methods (including the huen method; Mesinger and Arakawa, 1976) can increase accuracy, it also requires two function evaluations per time step. The second one is the Adams-Bashforth method, the second-order of which is always used in some semiimplicit semi-Lagrangian models. The second-order Adams-Bashforth method has the same second-order accuracy as the leapfrog scheme, but it performs only one function evaluation per time step. Unlike the leapfrog scheme, it is not subject to time splitting instability and does not need time lter. Lilly (1965) concluded that the second-order Adams-Bashforth method exhibits better stability and satisfactory accuracy. Durran (1991) presented a detailed comparison of the third-order Adams-Bashforth method with the widely used Asselin-Robert ltered leapfrog scheme, and found that the third-order Adams-Bashforth method is more accurate than the leapfrog scheme, i.e., phase-speed errors are smaller and amplitude errors are considerably less than those generated by the Asselin-Robert time ltering. Zhao and Zhong (2009a, 2010) and Zhao et al. (2010) developed an atmospheric general circulation model with the second-order Adams-Bashforth method as an alternative method to the original leapfrog scheme in the NCAR (National Center for Atmospheric Research) CAM 3.0 (Community Atmosphere Model 3.0). The Held-Suarez idealized test and an idealized baroclinic wave test were carried out. Their results showed that the climate states of the second-order Adams-Bashforth model were similar to those of many other global models (Fox-Rabinovitz et al., 1997; Lin, 2004; Wan et al., 2008) with converged solutions for long-term integration as well as better simulation ability for short-term idealized integration. A long-term real case integration during 19801999 shows that due to the reduction of simulation er-

rors of prognostic variables, the second-order AdamsBashforth method also has better ability in simulation of the diagnostic variables. Based on the preliminary studies above, the focus of this paper is to analyze performance of the even higher-order Adams-Bashforth methods as alternatives to the leapfrog scheme. As one of the new generation AGCMs for studying weather and climate, the CAM 3.1 is released in April 2005 as an updated version of the NCAR CAM 3.0 (Collins et al., 2006; James et al., 2006). It employs the Eulerian spectral method with triangular truncation to solve numerically the governing equations of the atmosphere. Fully updated physical parameterizations improve the capability of the model in simulating the global climate and general circulation. The higher-order Adams-Bashforth methods are coded into the CAM 3.1 to keep higher-order accuracy for explicit non-linear dynamical calculations, and a Crank-Nicholson semi-implicit scheme is incorporated for the treatment of fast gravity modes. In this paper, the Lorenz nonlinear convective equations are adopted to analyze the eect of the leapfrog time lter as well as the computational accuracy of dierent alternative methods. The remainder of this paper is organized as follows. The Held-Suarez idealized test is carried out to examine the improvement of the simulation and the mass conservation attribute of dierent order time differencing schemes. An idealized baroclinic wave test is conducted to examine the ability of alternative methods in simulating baroclinic wave perturbation and maintaining the initial data stability. 2. The Adams-Bashforth methods The general second-order Adams-Bashforth approximation to the ordinary dierential equation d = F ( ) dt has the form:
1 n+1 n = F ( n+ 2 ), t 3 1 n+ 1 F ( 2 ) = F ( n ) F ( n1 ), 2 2 3 n+1 n 1 n = F ( ) F ( n1 ). t 2 2

(1)

(2)

756

ACTA METEOROLOGICA SINICA

VOL.25

Using the third-order Adams-Bashforth method within Eq. (1), it should be written as: n+1 n 23 16 = F ( n ) F ( n1 ) t 12 12 5 n2 + F ( ). 12

it becomes: 55 59 n+1 n = F ( n ) F ( n1 ) t 24 24 9 37 + F ( n2 ) F ( n3 ). 24 24

(4)

(3)

Using the fourth-order Adams-Bashforth method,

A summary of the properties of the AdamsBashforth methods is given in Table 1, in which the storage factor and amplitude and phase errors are

Table 1. Comparison of the storage factor, amplitude error, and phase error of dierent time dierencing schemes
Method Forward Second-order AB method Third-order AB method Fourth-order AB method Storage factor 2 3 4 5 Amplitude error 2 p4 1+ 4 3p4 1 8 13p6 1 24 1+ p2 Phase error p2 3 5p2 1+ 12 289p4 1+ 720 251p4 1 720 1

Here p= t, where t is time step and is a real constant (Durran, 1991).

provided. 3. Lorenz nonlinear convective equations Before the AGCM dynamical core is modied with the Adams-Bashforth methods, the simple Lorenz nonlinear convective equations are employed to analyze the eect of the leapfrog lter and the computational solutions with dierent time dierencing schemes. Theoretically, the leapfrog solution must be ltered in order to avoid spurious amplication of the computational mode. Zhao and Zhong (2009b) analyzed the negative inuence of the Asselin-Robert lter in an AGCM and found it decreases the simulation accuracy. Lorenz (1971) demonstrated that the equations governing the two-dimensional Rayleigh convection can be truncated to three components whose amplitudes are governed by the system of equations as following (Durran, 1991): X = aY aX, t Y = XZ + bX Y, t Z = XY cZ. t Within the equations, variables a and b are set to 12, c is 6, and the initial value (X0 , Y0 , Z0 )=(10,

10, 25). From the standpoint of predictability, the computational solutions will decay with time and there is no sensitive dependence on the initial conditions. The chaotic equations should be analyzed because the computational solutions exaggerate the dierence with various time dierencing schemes. The susceptibility and ltered instability of dierencing schemes will be examined. Test is performed with the leapfrog without time lter, the second- to fourth-order Adams-Bashforth methods, and the Asselin-Robert ltered leapfrog (with time lter = 0.06) for about two periods. Besides, we analyze the evolution of computational solutions with dierent time steps. The numerical solution for X of the ltered and non-ltered leapfrogs is plotted in Fig. 1. The numerical test is performed with time step of 0.015 and a total time steps of 100. It is shown that the computation mode decays during integration and the unltered leapfrog scheme rapidly generates 2t computational modes which inevitably induce time splitting instability during integration. As a treatment to this computational error, the Asselin-Robert lter is used in the leapfrog with lter of 0.06, and it is found that the ltered leapfrog solution presents an excellent pattern that has eliminated 2t waves after 20 steps. Various time steps as 0.015, 0.01, 0.005, and

NO.6

ZHAO Bin and ZHANG Bo

757

Fig. 1. Comparison of the leapfrogs without (solid line) and with (dashed line) time lter coecients in the Lorenz equations. The time step is 0.015.

0.0025 are compared in Fig.

2.

With time reso-

lution increasing, the fourth-order Adams-Bashforth method has stable solution at all resolutions and is converged to the exact solution (Figs. 2a-d). Thus, the fourth-order Adams-Bashforth method provides a reference against which the other schemes may be compared. The solution error of the second- and thirdorder Adams-Bashforth is much smaller than that of the ltered leapfrog scheme in Fig. 2a with time step of 0.015. The ltered leapfrog solution is essentially lagged behind about 1/10 period than that of the fourth-order Adams-Bashforth method. As for the other time dierencing schemes, the results identically indicate that the higher-order Adams-Bashforth method is closer to the exact solution, that is to say, the higher-order method produces more accurate solution. With the time resolution being increased from

Fig. 2. The numerical approximations to a nonchaotic solution of Lorenz equations with time dierencing schemes. The time step is (a) 0.015, (b) 0.01, (c) 0.005, and (d) 0.0025. 2OR-AB, 3OR-AB, and 4OR-AB denote second-, third-, and fourth-order Adams-Bashforth, respectively.

758

ACTA METEOROLOGICA SINICA

VOL.25

0.01 to 0.025 (Figs. 2b-d), the improvement of computational solution is signicant for the ltered leapfrog and the Adams-Bashforth methods. At resolution of 0.05, the second- and third-order Adams-Bashforth methods are close to the fourth-order one and the leapfrog solution obtains the equivalent accuracy at the resolution of 0.025. 4. The Held-Suarez test In this section, an idealized test proposed by Held and Suarez (1994) is used for climate-scale test. In designing the forcing and dissipation, we use a simple Newtonian relaxation of the temperature eld to a zonal symmetric state and the Rayleigh damping of low-level winds to represent boundary friction. Detailed specications are as follows: V = Kv ( )V , t T = KT (, )[T Teq (, p)], t Teq = max 200, [315 (T )y sin2 p p ()z log( )cos2 ]( ) , p0 p0 b KT = Ka + (Ks Ka )max[0, ]cos4 , 1 b b Kv = Kf max[ ], 1 b Kf = 1 day1 , b = 0.7, Ka = 1/40 day1 , Ks = 1/4 day1 , = (T )y = 60 K, ()z = 10 K, R 2 = . Cp 7

There are no moist processes, no topography, no land-sea contrast, and no seasonal or diurnal cycle during the integration. The intention is to investigate the accuracy of simulation and the stability of dynamical core under idealized forcing. Simulations are conducted at T85L26. All the tests are integrated for 1200 days and the rst 200 days are considered as spin-up time and are discarded. The results are obtained by averaging the last 1000 days and the time step is 600 s. Figure 3 shows the time-zonal mean zonal winds as a function of latitude and -coordinates for different order Adams-Bashforth methods. The overall

features such as the midlatitude tropospheric westerly jets located in both hemispheres at = 0.2 (approximately 200 hPa) with a maximum wind speed of more than 28 m s1 , the equatorial surface easterlies, and the tropospheric jets are well represented. Similar to other studies with Held-Suarez forcing, these patterns compare quite well with those obtained with Held and Suarez (1994) and other AGCMs (Wan et al., 2008; Fox-Rabinovitz et al., 1997; Lin, 2004); especially, the midlatitude jets in the upper troposphere are clearly visible. Comparison of the maximum tropospheric westerly jets in these simulations is important. Zhao and Zhong (2009a) demonstrated that the tropospheric jets will become stronger with horizontal resolution decreasing, inducing the equatorward extended strong westerly jets to join and cut o the zero line near the equator. Conversely, with the resolution increasing, the weakening tropospheric jets contract and shift poleward. Similar discrepancy in the tropospheric westerly jets between low- and high-resolution simulations is also found in the results produced by the lowerand higher-order Adams-Bashforth methods. The extended jets weaken and contract poleward with the time dierencing order increasing. It is likely that the advantage of the higher-order method is equivalent to that of the higher-resolution. The transient eddy temperature variance is an index of baroclinic wave activities. Figures 4 shows the vertical proles of zonal mean time-averaged transient eddy temperature variance for dierent order AdamsBashforth methods. Similar patterns appear, such as the maximum eddy temperature variance exhibits two pairs in midlatitudes, appearing in the lower troposphere and extending upward and poleward. The magnitude of the eddy temperature variance is sensitive to the simulation accuracy. Higher-order AdamsBashforth method shows stronger eddy temperature variance in midlatitudes. It represents the fact that, with time dierencing order increasing, wave activity is enhanced and the maximal eddy temperature variance increases from about 32 K2 (Fig. 4a) to more than 36 K2 (Fig. 4c).

NO.6

ZHAO Bin and ZHANG Bo

759

Fig. 3. Time-zonal mean zonal wind (m s1 ) produced by (a) second-order, (b) third-order, and (c) fourth-order Adams-Bashforth methods.

Time variation of the global-mean surface pressure may gauge the models performance on the mass conservation (Cheong, 2006). Theoretically, if the surface pressure instead of the logarithmic surface pressure is integrated in the surface pressure equation, the total mass should be conserved in the dynamical core as: d dt
2 0
2

ps cosdd = 0.

(5)

Because logarithmic surface pressure is always incorporated in traditional AGCMs in consideration of the noises caused by the hydrostatic equations aliasing problems, it is worth checking if the total mass conserves in the dynamical core that incorporates the logarithmic surface pressure as the prognostic massrelated variable to facilitate model integration. Further comparison of results in the last 1000 days (Fig. 5) is carried out. The time variation of the total mass anomaly is examined. It is shown that the total

mass anomalies produced by dierent methods have similar patterns, but with varying amplitudes and evolution trends. The second-order Adams-Bashforth method gives maximum anomaly amplitude between 25 kg m2 (approximately 2.5 hPa and 0.25% of total mass). The linear least square method is used to dene a trend line for the evolution of anomaly. The slope of trend is 5.7103 for the second-order Adams-Bashforth method, i.e., the total mass has an overall decreasing trend of 5.7 kg m2 for 1000 days. Moreover, the standard deviation is as high as 10.7 kg m2 (approximately 1.1 hPa and 0.11% of total mass). The maximum anomaly amplitude of the thirdorder Adams-Bashforth method is not so large as that of the second-order one. The third-order method presents a perturbation of 15 kg m2 (approximately 1.5 hPa and 0.15% of total mass) and the slope of trend is 2.4103 , which means an increasing trend of 2.4 kg m2 (0.024% of total mass) for 1000 days. The

760

ACTA METEOROLOGICA SINICA

VOL.25

Fig. 4. The zonal mean time-averaged transient eddy temperature variance (K2 ) produced by (a) second-order, (b) third-order, and (c) fourth-order Adams-Bashforth methods.

It is interesting to note that the maximum anomaly amplitude of the fourth-order method is less than 10 kg m2 (approximately 1.0 hPa and 0.1% of total mass). In contrast to the second- and thirdorder methods, the slope of the fourth-order method is just 1.1103 , meaning that the total mass only increases 1.1 kg m2 (0.011% of total mass) for 1000 days. The standard deviation is as small as 3.4 kg m2 . Therefore, a conclusion can be easily drawn that higher-order Adams-Bashforth method performs better in mass conservation.
Fig. 5. Time variation of mass anomaly with trend line of total mass (kg m2 ) for 2011200 days. (a) Second-order, (b) third-order, and (c) fourth-order Adams-Bashforth. Solid: anomaly line, dashed: trend line, and SD: standard deviation.

5. The baroclinic wave test An idealized baroclinic wave test in the Northern Hemisphere is carried out on synoptic scale. The idealized test is a deterministic initial value test for dry dynamical cores of AGCMs. It aims to analyze the evolution of baroclinic waves in the Northern Hemis-

standard deviation is 8.2 kg m2 , smaller than that of the second-order one by about 23.4%.

NO.6

ZHAO Bin and ZHANG Bo

761

phere over a 30-day period. The concept of the initial value test was developed by Hoskins and Simmons (1975). Jablonowski and Williamson (2006) further formed a new kind of idealized test scheme by adding a local perturbation in the initial condition. The test was designed to analyze the evolution and convergence of dynamical cores with overlaying zonal perturbation. The simulated baroclinic wave can also be analyzed by this test. The conguration and description of the deterministic initial data of the test have been reported in a previous paper (Zhao and Zhong, 2009b). 5.1 Steady state test Following Jablonowskis design, the steady state initial conditions are used in the experiments for a 30-day integration with T170L26 and the time step is chosen as 150 s. In this case, we dene l2 as error norm of surface pressure to test the stability of dierent time dierencing schemes in maintaining the initial data. 1 l2 (ps (t) ps (t0 )) = ( 4
2 0 2
2

steady state test. It is seen that the leapfrog scheme keeps the error norm about 2.2 hPa and manifests several unstable oscillations. The dierent order AdamsBashforth methods present similar l2 norm trends and the second-order one is as low as 0.7 hPa. Moreover, the fourth-order Adams-Bashforth method keeps l2 norm lower than the second-order by about 0.1 hPa during the 30-day simulation and shows a better stability. The third-order Adams-Bashforth method shows more superiority to the second-order one during the rst 10 days and then l2 norm increases slightly for the last 20 days. Furthermore, we select the longitudelatitude point (0 , 0 ) to test the l2 error norm and nd a similar evolution trend (Fig. 6b). Some other points are also calculated and similar patterns are revealed. Thus, it can be concluded that the higherorder Adams-Bashforth method will maintain the initial data more stable. 5.2 Perturbation state test The evolution of the baroclinic wave is examined by overlaying the steady state initial conditions with the zonal perturbation. Figure 7 compares the 850hPa relative vorticity eld between the third- and fourth-order Adams-Bashforth methods on day 9 with resolution T170L26. The overall vorticity patterns agree reasonably well. In particular, the maximum vorticity center with the value of 12.2105 s1 in Fig. 7b is 1.3105 s1 stronger than that in Fig. 7a (10.9105 s1 ).

2 1/ 2

ps (, , t) ps (, , t0 ) cosdd) (
i j [ps (i , j , t) i

ps (i , j , t0 )]2 j
j

)1/2 , (6)

where wj = |sin(j +1/2 ) sin(j 1/2 )|, ps is surface pressure, and j is Gaussian weight. Figure 6a shows the l2 norm distribution in the

Fig. 6. Distributions of the l2 norm with dierent time dierencing schemes in (a) the steady-state test and (b) the evolution of error between ps (0 , 0 ) and the initial data.

762

ACTA METEOROLOGICA SINICA

VOL.25

Fig. 7. Comparison of 850-hPa relative vorticity (105 s1 ) with T170L26 resolution on day 9 for (a) third-order and (b) fourth-order Adams-Bashforth methods. MV means maximum vorticity.

Fig. 8. Comparison of the lowest surface pressure change during (a) 015 days and (b) 915 days.

During the evolution of baroclinic waves, the lowest surface pressure can be regarded as one of the criterions to assess the ability of simulation. Figure 8a shows a comparison of the lowest surface pressure for three kinds of Adams-Bashforth methods against the leapfrog scheme during 015 days. Figure 8b presents a zoomed-in version of Fig. 8a. It is found that the results of the dierent time dierencing schemes are in complete agreement with those in the rst 8 days (Fig. 8a). After day 8 (Fig. 8b), slight deviations appear. It is clear that the surface pressure of the higher-order method is deeper (approximately 1.0 hPa) than that of the lower-order one, indicating that the higher-order Adams-Bashforth method has a better ability in simulating the baroclinic waves. 6. Summary Considering that the Asselin-Robert time lter degrades the second-order accuracy of the unadulterated leapfrog method, it can be a feasible way to im-

prove the performance of AGCMs by replacing the Asselin-Robert time lter with the Adams-Bashforth method. The second-order Adams-Bashforth method as an alternative to the ltered leapfrog scheme has been analyzed in our preliminary works (Zhao and Zhong, 2009a, 2010; Zhao et al., 2010) Based on these results, an analysis of the dierent order Adams-Bashforth methods is presented, along with a detailed comparison of the higher-order AdamsBashforth method with the widely used AsselinRobert ltered leapfrog scheme. Not surprisingly, the higher-order Adams-Bashforth method proves to be more ecient in avoiding the instability of time lter and it exhibits better accuracy than the leapfrog scheme. In this paper, the Lorenz nonlinear convective equations are used to compare the computation solution accuracy of four kinds of time dierencing schemes. Idealized dynamical tests of Held-Suarez and baroclinic wave tests demonstrate the improvement of the new dynamical cores. The results are as follows:

NO.6

ZHAO Bin and ZHANG Bo

763

1) The analysis of Lorenz nonlinear convective equations reveals that the unltered leapfrog scheme rapidly generates two computational modes which inevitably induce time splitting instability during the integration. The ltered leapfrog solution shows an excellent pattern for eliminating the two-wave solution after 20 steps, but it has a 1/10 period lag with the time resolution of 0.015. The fourth-order AdamsBashforth method is converged to the exact solution and provides a reference against which other methods may be compared. Hence, the Adams-Bashforth methods produce more accurate and convergent solutions with time dierencing order increasing. 2) All time dierencing schemes successfully retain the characteristics of the Held-Suarez test to the results of many other global models. However, the discrepancy in producing tropospheric westerly jets between the lower- and higher-order methods is likely equivalent to the results produced between low- and high-resolution. Moreover, considering the logarithmic surface pressure instead of surface pressure is always incorporated in traditional AGCMs, the comparison of total mass conservation in the last 1000 days is carried out. The higher-order method exhibits less anomaly uctuation, slight slope of trend, and smaller standard deviation, demonstrating that the higher-order Adams-Bashforth method performs better in mass conservation. 3) The synoptic-scale idealized test of baroclinic waves shows that the higher-order method can well maintain the initial data. After overlaying the steady state conditions with zonal perturbation, better simulation of baroclinic wave perturbation is generated by the higher-order Adams-Bashforth method than the lower-order one. The time dierencing schemes are now begin preliminarily put into operational application. There are many aspects demanding improvement. For example, with the accuracy arising, the numerical cost increases. In addition, the memory needed in the high-order scheme decreases the model eciency due to the limitation of computational resources although the related storage requirements are not particularly burdensome. Detailed comparison of the numerical

cost should be discussed in the future and a further discussion should be made on the relationship between the amplitude and phase-speed errors that develop when the schemes are used to integrate. Moreover, development of new dynamical cores with full physical parameterizations are needed so as to improve the models ability in climate simulation and prediction.

REFERENCES
Asselin, R. A., 1972: Frequency lter for time integrations. Mon. Wea. Rev., 100(6), 487490. Bjerkness, V., 1904: Das problem der wettervorhersage, betrachtet vom standpunkte der mechanik und der physic. Meteorol. Zeitschr, 21, 17. Chen Dehui and Xue Jishan, 2004: An overview on recent progresses of the operational numerical weather prediction models. Acta Meteor. Sinica, 62(5), 120 130. (in Chinese) Cheong, H. B., 2006: A dynamical core with double Fourier series: Comparison with the spherical harmonics method. Mon. Wea. Rev., 134(4), 1299 1315. Collins, W. D., and Coauthors, 2006: The formulation and atmospheric simulation of the Community Atmosphere Model Version 3 (CAM3). J. Climate, 19(11), 21442161. Durran, D. R., 1991: The third-order Adams-Bashforth method: An attractive alternative to leapfrog time dierencing. Mon. Wea. Rev., 119(3), 702720. Feng Guolin, Cao Hongxing, Gu Xiangqian, et al., 2002: A new scheme for improving the accuracy of numerical prediction-retrospective time integration scheme. J. Appl. Meteor. Sci., 23(2), 207217. (in Chinese) Fox-Rabinovitz, M. S., G. L. Stenchikov, M. J. Suarez, and L. L. Takacs, 1997: A nite-dierence GCM dynamical core with a variable-resolution stretched grid. Mon. Wea. Rev., 125(11), 29432968. Held, I. M., and M. J. Suarez, 1994: A proposal for the intercomparison of the dynamical cores of atmospheric general circulation models. Bull. Amer. Meteor. Soc., 75(10), 18251830. Hoskins, B. J., and A. J. Simmons, 1975: A multilayer spectral model and the semi-implicit method. Quart. J. Roy. Meteor. Soc., 101(429), 637655. Jablonowski, C., and D. L. Williamson, 2006: A baroclinic wave test case for dynamical cores of general

764

ACTA METEOROLOGICA SINICA

VOL.25

circulation models: Model intercomparisons. NCAR Technical Note NCAR/TN-469+STR, Boulder, CO, 89 pp. James, W. H., J. K. James, A. S. Phillips, J. Caron, and J. Yin, 2006: The dynamical simulation of the Community Atmosphere Model Version 3 (CAM3). J. Climate, 19(11), 21622183. Ji Liren, Xing Runan, and Chen Xingshan, 1980: Improving experiments of a three-level primitive equation model. Chinese J. Atmos. Sci., 14(4), 310318. (in Chinese) , Chen Jiabin, and Zhang Daomin, 2005: Review of some numerical aspects of the dynamic framework of NWP model. Chinese J. Atmos. Sci., 29(1), 120130. (in Chinese) Lilly, D. K., 1965: On the computational stability of numerical solutions of time-dependent non-linear geophysical uid dynamics problems. Mon. Wea. Rev., 93, 1126. Lin, S. J., 2004: A vertically lagrangian nite-volume dynamical core for global models. Mon. Wea. Rev., 132(10), 22932307. Lorenz, E. N., 1971: An N-cycle time dierencing scheme for stepwise numerical integration. Mon. Wea. Rev., 99, 644648. Matsuno, T., 1966: Numerical integrations of the primitive equations by a simulated backward dierence method. J. Meteor. Soc. Japan (Ser. 2), 44, 7684. Mesinger, F., and A. Arakawa, 1976: Numerical methods used in atmospheric models. GARP Publication Series, 17(1), 64 pp. Pfeer, R. L., I. M. Navon, and X. L. Zou, 1992: A comparison of the impact of two time-dierencing schemes on the NASA-GLAS climate model. Mon. Wea. Rev., 120(7), 13811393.

Robert, A., 1966: The integration of a low-order spectral form of the primitive meteorological equations. J. Meteor. Soc. Japan, 44(2), 237244. Simmons, A. J., and C. Temperton, 1997: Stability of a two-time-level semi-implicit integration scheme for gravity wave motion. Mon. Wea. Rev., 125(4), 600615. Wan, H., M. A. Giorgetta, and L. Bonaventura, 2008: Ensemble Held-Suarez test with a spectral transform model: variability, sensitivity, and convergence. Mon. Wea. Rev., 136(3), 10751092. Williams, P. D., 2009: A proposed modication to the Robert-Asselin time lter. Mon. Wea. Rev., 137(8), 25382546. Yong, J. A., 1968: Comparative properties of some time dierencing schemes for linear and nonlinear oscillations. Mon. Wea. Rev., 69, 357364. Zhao Bin and Zhong Qing, 2009a: The dynamical and climate test of an atmospheric general circulation model with the second-order Adams-Bashforth method. Acta Meteor. Sinica, 23(6), 738749. and , 2009b: An alternative method to leapfrog time dierencing and its applications in an atmospheric general circulation model. Climatic Environ. Res., 14(1), 2130. (in Chinese) and , 2010: The development of a nonhydrostatic global spectral model. Adv. Atmos. Sci., 27(6), 709717. , Zhang Bo, and Zhong Qing, 2010: Sensitivity analysis of second-order Adams-Bashforth method on vertical resolutions. Sci. Meteor. Sinica, 30(1), 814. (in Chinese)