Abstracts
To eliminate the fast gravitational waves of great amplitude, which are not observed in the real atmosphere, the initial fields for numerical schemes of atmosphere forecasting and modeling systems are usually adjusted dynamically by applying balance relations. In this study we consider different forms of the balance equations and for each of them we detect the nonelliptic regions in the gridded atmosphere data of the Southern Hemisphere. The performed analysis reveals the geographical, vertical and zonally averaged distributions of nonelliptic regions with the most concentration in the tropical zone. The area of these regions is essentially smaller and less intensive for more complete and physically justified balance relations. The obtained results confirm the Kasaharas assumption that ellipticity conditions are violated in the actual atmospheric fields essentially due to approximations made under deriving the balance equations.
balance equations; initialization methods; nonelliptic regions
Para eliminar as ondas gravitacionais rápidas de grande amplitude, as quais não se observam na atmosfera real, os campos iniciais para esquemas numéricos de sistemas de modelagem e previsão atmosférica são usualmente ajustados dinamicamente aplicando as relações de balanço. Neste estudo consideramos formas diferentes de equações de balanço e para cada uma dessas detectamos as regiões não elípticas nos dados atmosféricos do Hemisfério Sul. A analise realizada mostra a distribuição geográfica, vertical e média zonal de regiões não elípticas com maior concentração na zona tropical. A área dessas regiões é essencialmente reduzida e a intensidade é visivelmente menor para as relações de balanço mais completas e fisicamente justificáveis. Os resultados obtidos confirmam a suposição de Kasahara de que as condições de ellipticidade são violadas nos campos atmosféricos reais essensialmente devido às aproximações feitas em dedução das equações de balanço.
equações de balanço; métodos de inicialização; regiões não elípticas
ARTIGOS
On nonelliptic regions and solvability of balance equations for atmosphere dynamics
Sobre regiões não elípticas e solvabilidade de equações de balanço para dinâmica da atmosfera
Andrei Bourchtein; Ludmila Bourchtein
Institute of Physics and Mathematics, Pelotas State University, Brazil Correspondence address: Rua Anchieta 4715 bloco K, ap.304, PelotasRS 96015420, Brazil Email: burstein@terra.com.br
ABSTRACT
To eliminate the fast gravitational waves of great amplitude, which are not observed in the real atmosphere, the initial fields for numerical schemes of atmosphere forecasting and modeling systems are usually adjusted dynamically by applying balance relations. In this study we consider different forms of the balance equations and for each of them we detect the nonelliptic regions in the gridded atmosphere data of the Southern Hemisphere. The performed analysis reveals the geographical, vertical and zonally averaged distributions of nonelliptic regions with the most concentration in the tropical zone. The area of these regions is essentially smaller and less intensive for more complete and physically justified balance relations. The obtained results confirm the Kasaharas assumption that ellipticity conditions are violated in the actual atmospheric fields essentially due to approximations made under deriving the balance equations.
Keywords: balance equations, initialization methods, nonelliptic regions
RESUMO
Para eliminar as ondas gravitacionais rápidas de grande amplitude, as quais não se observam na atmosfera real, os campos iniciais para esquemas numéricos de sistemas de modelagem e previsão atmosférica são usualmente ajustados dinamicamente aplicando as relações de balanço. Neste estudo consideramos formas diferentes de equações de balanço e para cada uma dessas detectamos as regiões não elípticas nos dados atmosféricos do Hemisfério Sul. A analise realizada mostra a distribuição geográfica, vertical e média zonal de regiões não elípticas com maior concentração na zona tropical. A área dessas regiões é essencialmente reduzida e a intensidade é visivelmente menor para as relações de balanço mais completas e fisicamente justificáveis. Os resultados obtidos confirmam a suposição de Kasahara de que as condições de ellipticidade são violadas nos campos atmosféricos reais essensialmente devido às aproximações feitas em dedução das equações de balanço.
Palavraschave: equações de balanço, métodos de inicialização, regiões não elípticas
1. INTRODUCTION
Numerical weather prediction, which is the core activity of atmospheric research and operational centers, consists basically of computation of solution to a set of partial differential equations expressing the conservation laws of mass, momentum and energy for compressible continuum medium in the noninertial system related to a rotating sphere. The chosen differential model is solved numerically as initial value (or initialboundary value) problem, requiring the definition of initial data. Data assimilation schemes supply these initial conditions, but they may not be well dynamically adjusted, which means that fast oscillations of great amplitude, which are not observed in the real atmosphere, are generated at the initial stages of the numerical solution. These oscillations may contaminate physically meaning solution up to some days of forecast depending on the mechanisms of physical and computational diffusion included in the model. The process of adjusting the initial data to the prediction model to ensure small amplitudes of the fast waves is called initialization.
The long history of balance relations aimed to adjust the initial data may be traced back to the famous nonlinear balance equation by Charney (1955). A review of various initialization procedures, including nonlinear balance and omega equations, developed up to mid 70s, is given by Bengtsson (1975). The current approaches to initial adjustment include nonlinear normal mode initialization (NMI) introduced by Machenhauer (1977) and Baer and Tribbia (1977), boundary derivative method (BDM) presented first by Browning et al. (1980) and digital filter technique proposed by Lynch et al. (e.g., Lynch and Huang 1992). One of the most effective versions of the NMI is the vertical (or implicit) normal mode initialization (Bourke and McGregor 1983, Temperton 1988, Fillion and Roch 1992), which is equivalent to BDM approach (Kasahara 1982, Bijlsma and Hafkenscheid 1986, McGregor and Bourke 1988).
In his seminal paper, Daley (1981) presented the basic concepts of initialization and formulated a series of problems whose solution may improve understanding the principal properties of initialization equations. In this study we investigate one of these issues: the nonellipticity of the balance diagnostic relations under fixed pressure field, socalled pressure (geopotential) constrained initialization. In the last years the initialization procedure was dropped in some atmospheric centers due to increased quality of observational network and objective analysis. Even having this tendency, the solution of the stated problem is important on its own because it could clarify a nature of the balance involving atmospheric fields.
The first studies of the ellipticity conditions for balance relations were made by Charney (1955) and Houghton (1968) in the case of nonlinear balance equation on the fplane and on the sphere. Since the last equation is the particular case of the MongeAmper equation (Charney 1955, Kasahara 1982), these studies essentially were the applications of the welldeveloped theory of MongeAmpere equation.
The first theoretical study on the nonellipticity of simplified NMI/BDM equations was presented by Tribbia (1981), who constructed theoretical example demonstrating that a certain restriction on meteorological fields must be satisfied in order to obtain a solution of the initialization system with fixed geopotential. He used the model of isolated barotropic vorticity on the fplane and obtained that this restriction is close to ellipticity condition of the nonlinear balance equation. The violation of ellipticity condition can lead to the divergence of the iterative method of solving the NMI/BDM equations when the height constrained initialization is required. This problem was first reported by Daley (1978) when applying Machenhauer iteration procedure to the shallow water equations. The speculations about the reasons for this problem centered on two possibilities: the shortcomings of applied iterative algorithms and the mathematical inconsistency of the boundary value problem due to existence of nonelliptic regions in the real atmospheric data (Daley 1981, Tribbia 1981, Errico 1983, Rasch 1985).
Among different studies on nonelliptic regions in the isobaric height fields we should note the papers by Kasahara (1982), Paegle et al. (1983) and Knox (1997), and discussion of the respective issues by Daley (1991). As it was pointed out by Kasahara (1982), in the past the occurrence of these regions used to be considered as a result of observational inaccuracies even though the changes made to recover ellipticity criterion sometimes exceeded probable data errors, specially at higher levels. Probably, Kasahara (1982) was the first who stated in an explicit way that the ellipticity condition is a mathematical constraint on atmospheric fields, which can produce nonelliptic regions, and, consequently, impossibility of the required balance, simply because the assumptions made in deriving the balance relations could be not totally satisfied in real atmosphere. Therefore, one of the points of the different studies (e.g., Kasahara 1982, Paegle et al. 1983, Randel 1987 and Knox 1997) is to adjust ellipticity conditions by including the terms neglected in nonlinear balance relations. The new conditions, called realizability conditions, have essentially reduced the area of nonelliptic regions supporting the Kasaharas supposition. However this approach is based on evaluation of the contribution of different terms of the primitive divergence equation for possibly recovering the ellipticity of regions rather than on the consistent system of balance relations. The only considered balance equation was the nonlinear balance equation on the fplane or on the sphere.
In the recent years some new and more complex mathematical criterions of ellipticity have been obtained for NMI/BDM equations, which are much more general balance system based on more accurate and reliable assumptions than nonlinear balance equation (Bourchtein 2002, Bourchtein 2006). In this way, many terms neglected in the derivation of the nonlinear balance equation have been recovered in NMI/BDM equations. Therefore, one can expect that respective ellipticity conditions should be more soft and related nonelliptic regions should be more scarce in order to confirm the Kasaharas statement.
In the present study we compare nonelliptic regions related to nonlinear balance equation (on the fplane and sphere) and NMI/BDM equations for the shallow water model. In section 2 we present the NMI/BDM equations for the shallow water model and give a brief exposition of the respective ellipticity conditions. The results of the evaluation of nonelliptic regions in South Hemisphere for different balance relations are presented in section 3 followed by concluding remarks in section 4.
2. BALANCE EQUATIONS AND ELLIPTICITY CONDITIONS
In local Cartesian coordinates x, y the classic nonlinear balance equation on a tangent plane has the form (Charney 1955)
where Y is the streamfunction, F is the geopotential, is a chosen value of the Coriolis parameter f, and is the Laplace operator. Considered as equation for the streamfunction with a given geopotential field, it is a special case of MongeAmpere equation. If this equation is to be solved on bounded domain D with imposed values of the streamfucntion on the boundary ¶D, then the problem is well posed only if the equation is of elliptic type. It requires the ellipticity condition to be satisfied, which has the following form for equation (1):
Similarly, using spherical coordinates l (longitude) and j (latitude), the nonlinear balance equation assumes the form (Houghton 1968)
where a is the Earths radius, f = 2W sin j, b = 2W cos j/ a, W is the angular velocity of the Earths rotation and u and v are the longitudinal and meridional components, respectively, of nondivergent wind, that is,
Again, the solution of the boundary value problem for (3) with respect to the streamfunction requires the ellipticity condition, which can be written as follows (Houghton 1968):
The NMI/BDM systems have much more complex structure and contain a set of equations. For the shallow water equations on a sphere, the system contains two equations, which can be expressed in longitudelatitude coordinates (l, j) as follows (Browning et al. 1980, Bourke and McGregor 1983, Temperton 1988):
where
for any vector function (U, V) and any scalar function h. Here u and v are the (full) physical components of velocity, is a mean geopotential height and Q_{u}, Q_{v}, Q_{F} contain all the nonlinear and variable coefficient terms of the shallow water equations, that is,
The system (6)(7) contains three unknown functions u, v and F, so it admits different closure conditions. The following natural versions of these conditions are frequently considered (Daley 1981, Daley 1991):
i.e., initialization with unchanged slow mode p (frequently called unconstrained initialization), unchanged streamfunction Y(streamfunction constrained initialization) or unchanged geopotential F (geopotential constrained initialization). Function p is the potential vorticity of the linearized barotropic equations on the fplane. The nonlinear system of partial differential equations (6)(7) with one of the closure conditions (10), (11) or (12) forms wellposed boundary value problem if it is elliptic. The ellipticity is guaranteed if its homogeneous form (characteristic determinant) is definite, that is, it does not change sign in the domain D of the problem.
For each of the closure conditions (10)(12) the ellipticity criterion of the respective differential problem have been derived in Bourchtein (2002) and Bourchtein (2006). In particular, it was shown that the closures (10) and (11) generate the same ellipticity condition in the simple form
It means that the boundary value problem for NMI/BDM with unchanged streamfunction can be well posed if, and only if, the phase speed of gravitational waves c = is greater than the advective speed throughout the entire domain D. Of course, this condition is satisfied for the barotropic model equations of the atmosphere and for the first (fastest) vertical modes of a baroclinic model. However, the condition (13) is violated for very thin layers, which is the cause of the divergence of iterative algorithms applied to solve the initialization equations. Respectively, a similar behavior can be expected for slow internal modes of a baroclinic model and was observed in numerical experiments with different multilevel models reported in many papers (e.g., Daley 1981, Errico 1983, Temperton and Roch 1991).
If the geopotential constrained initialization is used, the ellipticity condition is much more complex and its approximate form in spherical coordinates can be written as follows (Bourchtein 2006):
In the next section we show that this condition can be violated in some points of the analysis data. Even though the area covered by points with the negative values of E_{3} is usually small in comparison with the total area of a chosen domain, it leads to mathematical inconsistency of the boundary value problem for considered differential equations. This inconsistency causes divergence of any iterative method applied to solve the boundary value problem. In this way, we confirm the Daley assumption that the reason behind the problem of divergence is of mathematical nature.
3. ANALYSIS OF DISTRIBUTION OF THE NONELLIPTIC REGIONS
In this section we apply the ellipticity criteria (2), (5) and (14) to investigate the occurrence of the respective nonelliptic regions in the gridded data of the NCEP (National Centers for Environmental Prediction) analysis for the Southern Hemisphere. The data for this study were taken from the global NCEP analysis available on a spatial grid with regular latitude/ longitude resolution of 1^{0} and 26 vertical pressure levels. The analysis was restricted to the data of the Southern Hemisphere at 850, 500 and 200 hPa pressure levels for 0000 GMT 05 November 2005. The meteorological elements used are the longitudinal and meridional velocity components u and v, and the geopotential F.
First, we compute the ellipticity measures E_{1} and E_{2} defined by (2) and (5) on three chosen pressure levels. The nondivergent velocity components in (5) were evaluated by formula (4) with the streamfunction found from Poissons equation Ñ^{2}y = z, where the Laplace operator is defined in (9) and z is the relative vorticity defined by the second formula in (8) with gridded data of the velocity components u and v. The obtained values of E_{2} are systematically slightly greater than E1, but the difference is too small and can be certainly neglected for this study. The charts of the distribution of E_{2} are shown on Figs.16 separately for each pressure surface and Eastern and Western Hemispheres. Contour intervals are 4 · 10^{8}s^{2}. To avoid the noisy maps, only nonelliptic regions are plotted. It can be seen the strong tendency in increasing the nonelliptic area toward higher levels. There is some relation between nonelliptic area location at different levels but it is not observed systematically. At each pressure surface the nonelliptic regions mostly appear in the tropics and subtropics, though there are some nonelliptic regions in the middle and even high latitudes as well. The geographical distribution of the nonelliptic regions in the tropics appears to be almost random. To give one example of the relation between measures E_{1} and E_{2} we also show the chart of E_{1} for 500 pressure surface, Western Hemisphere (Fig.7). As one can see the values of two measures are virtually identical for the purpose of our study. Therefore, hereafter we use only the measure E_{2}, which is theoretically more complete.
The following series of six charts (Figs.813) shows the elliptic measure E_{3} for corresponding surfaces and Hemispheres. Contour intervals are 2 · 10^{9}s^{2} and again only the nonelliptic regions are plotted. On all charts for E_{3} the nonelliptic regions cover significantly less area and have much less intensity in comparison with the measure E_{2}. The spatial distribution of the nonelliptic E_{3} areas seems to follow the pattern of the measure E_{2}: these are more concentrated in tropic and subtropic zone with rather chaotic geographical distribution and increased area and intensity at the 200 hPa pressure level.
As it was pointed out by different researchers (e.g., Kasahara 1982, Knox 1997), an area average of ellipticity measure is another important index for examining the nature of nonelliptic regions. The longitudinal averages of the measures E_{2} and E_{3} are presented in Figs.1419. The solid line is for E_{2} and the pointed for E_{3}. Evidently, the negative values of E_{3} are much more rare and have much smaller amplitudes when compared to E_{2}. Also, the negative values of E_{3} are confined to very narrow tropical zone and all of them are clustered near boarder line between negative and positive values.
If we compare the results for E_{3} with the respective results obtained for realizability conditions by Kasahara (1982) and Knox (1997), we can note a great similarity. Indeed, the use of ellipticity criterion E_{3} allows to recover ellipticity in the major part of the negative area of the measure E_{2} and to strongly decrease the remaining negative values bringing them to the boarder line. The average qualitative distribution of measure E_{3} exhibits the same principal characteristics as the realizability measures, namely, the negative area is confined to tropicsubtropic zone and it increases toward higher pressure levels. The main difference is that the effect of the compensation of negative ellipticity achieved in Kasahara (1982) and Knox (1997) by the inclusion in the realizability conditions of additional terms from the divergence equation, is obtained in our study by substituting the ellipticity criterion for more simple balance relation by another ellipticity criterion corresponding to more complex and justifiable NMI/BDM method. In this way, we substantiate the Kasaharas statement that ellipticity conditions can be violated in the actual atmospheric fields essentially due to approximations made under deriving the balance relations.
4. CONCLUSIONS
The nonlinear balance equation introduced by Charney, as well as its spherical generalization introduced by Houghton, and the NMI/BDM initialization equations were considered in this work. The ellipticity conditions associated with these balance relations were presented and the respective nonelliptic regions were found using the gridded data of Southern Hemisphere. These are practically all known until now ellipticity criterions of balance relations, because the wellknown Tribbias condition for spectrally reduced shallow water system virtually coincides with the Charney criterion of nonellipticity. The obtained results showed the existence of nonelliptic regions even for more complex balance systems of the NMI/BDM method. However, these regions have essentially smaller areas and intensity as compared with those for nonlinear balance equation. This result confirms the Kasaharas conclusion that the occurrence of nonelliptic regions is of physical nature and related to making simplifications in the derivation of the balance relations.
5. ACKNOWLEDGEMENTS
All the geographical charts were plotted with the use of free software GrADS (Grid Analysis and Display System). This research was supported by Brazilian science foundations CNPq and FAPERGS.
6. REFERENCES
Received April 2006
Accepted July 2006
 Baer, F.; Tribbia, J.J. On complete filtering of gravity modes through nonlinear initialization. Mon. Wea. Rev., v.105, p.15361539, 1977.
 Bengtsson, L. Fourdimensional assimilation of meteorological observations. WMO/ICSU Joint Organizing Committee, GARP Publ. Series, No.15, 1975, 76 p.
 Bijlsma, S.J.; Hafkenscheid, L.M. Initialization of a limited area model: a comparison between the nonlinear normal mode and bounded derivative methods. Mon. Wea. Rev., v.114, p.14451455, 1986.
 Bourchtein, A. Ellipticity of normal mode initialization equations. Appl. Math. Comput., v.133, p.193211, 2002.
 Bourchtein, A. Ellipticity conditions of the shallow water balance equations for atmospheric data. J. Atmos. Sci., v.63, p.15591566, 2006.
 Bourke, W.; McGregor, T. A nonlinear vertical mode initialization scheme for a limited area prediction model. Mon. Wea. Rev., v.111, p.22852297, 1983.
 Browning, G.L.; Kasahara, A.; Kreiss, H.O. Initialization of the primitive equations by the bounded derivative method. J. Atmos. Sci., v.37, p.14241436, 1980.
 Charney, J. The use of the primitive equations of motion in numerical prediction. Tellus, v.7, p.2226, 1955.
 Daley, R. Variational nonlinear normal mode initialization. Tellus, v.30, p.201218, 1978.
 Daley, R. Normal mode initialization. Rev. of Geoph. and Space Phys., v.19, p.450468, 1981.
 Daley, R. Atmospheric data analysis. Cambridge University Press, 1991, 457 p.
 Errico, R.M. Convergence properties of Machenhauers initialization scheme. Mon. Wea. Rev., v.111, p.22142223, 1983.
 Fillion, L.; Roch, M. Variational implicit normal mode initialization for a multilevel model. Mon. Wea. Rev., v.120, p.10511076, 1992.
 Houghton, D.D. Derivation of the elliptic condition for the balance equation in spherical coordinates. Generalized nonlinear balance criteria and inertial stability. J.Atmos. Sci., v.25, p.927928, 1968.
 Kasahara, A. Nonlinear normal mode initialization and the bounded derivative method. Rev. of Geoph. and Space Phys., v.20, p.385397, 1982.
 Kasahara, A. Signifi cance of nonelliptic regions in balanced flows of the tropical atmosphere. Mon. Wea. Rev., v.110, p.19561967, 1982.
 Knox, J.A. Generalized nonlinear balance criteria and inertial stability. J.Atmos. Sci., v.54, p.967985, 1997.
 Lynch, P.; Huang, X.Y. Initialization of the HIRLAM model using a digital filter. Mon. Wea. Rev., v.120, p.10191034, 1992.
 Machenhauer, B. On the dynamics of gravity oscillations in a shallow water model, with applications to normal mode initialization. Contrib. Atmos. Phys., v.50, p.253271, 1977.
 McGregor, T.; Bourke, W. A comparison of vertical mode and normal mode initialization. Mon. Wea. Rev., v.116, p.13201334, 1988.
 Paegle, J.; Paegle, J.N.; Dodd, G.C. On the occurence of atmospheric states that are nonelliptic for the balance equations. Mon. Wea. Rev., v.111, p.17091723, 1983.
 Randel, W.J. The evaluation of winds from geopotential height data in stratosphere. J. Atmos. Sci., v.44, p.30973120, 1987.
 Rasch, P.J. Developments in normal mode initialization. Part II. A new method and its comparison with currently used schemes. Mon. Wea. Rev., v.113, p.17531770, 1985.
 Temperton, C. Implicit normal mode initialization. Mon. Wea. Rev., v.116, p.10131031, 1988.
 Temperton, C.; Roch, M. Implicit normal mode initialization for an operational regional model. Mon. Wea. Rev., v.119, p.667677, 1991.
 Tribbia, J.J. Nonlinear normal mode balancing and the ellipticity condition. Mon. Wea. Rev., v.109, p.17511761, 1981.
Publication Dates

Publication in this collection
28 Jan 2008 
Date of issue
Dec 2007
History

Accepted
July 2006 
Received
Apr 2006