• Asselin, R. A., 1972: Frequency filter for time integrations. Mon. Wea. Rev., 100 , 487490.

  • Bénard, P., 2003: Stability of semi-implicit and iterative centered-implicit time discretizations for various equation systems used in NWP. Mon. Wea. Rev., 131 , 24792491.

    • Search Google Scholar
    • Export Citation
  • Benoit, R., M. Desgagné, P. Pellerin, Y. Chartier, and S. Desjardins, 1997: The Canadian MC2: A semi-Lagrangian, semi-implicit wideband atmospheric model suited for finescale process studies and simulation. Mon. Wea. Rev., 125 , 23822415.

    • Search Google Scholar
    • Export Citation
  • Benoit, R., and Coauthors, 2002a: The real-time ultrafinescale forecast support during the special observing period of the MAP. Bull. Amer. Meteor. Soc., 83 , 85109.

    • Search Google Scholar
    • Export Citation
  • Benoit, R., C. Girard, M. Desgagné, S. Chamberland, and W. Yu, 2002b: Finescale orography and the MC2 dynamics kernel. Preprints, 10th Conf. on Mountain Meteorology, Park City, UT, Amer. Meteor. Soc., 219–222.

  • Caya, D., and R. Laprise, 1999: A semi-implicit semi-Lagrangian regional climate model: The Canadian RCM. Mon. Wea. Rev., 127 , 341362.

    • Search Google Scholar
    • Export Citation
  • Charney, J. G., and N. A. Phillips, 1953: Numerical integration of the quasi-geostrophic equations for barotropic and simple baroclinic flows. J. Meteor., 10 , 7199.

    • Search Google Scholar
    • Export Citation
  • Denis, B., 1990: Introduction de la topographie dans un modèle atmosphérique non-hydrostatique. M.S. thesis, UQAM, 121 pp. [Available from R. Laprise, Département de physique, UQAM, P.O. Box 8888, Stn “Downtown,” Montréal, Québec H3C 3P8, Canada.].

  • Denis, B., J. Côté, and R. Laprise, 2002: Spectra decomposition of two-dimensional atmospheric fields on limited-area domains using the discrete cosine transform (DCT). Mon. Wea. Rev., 130 , 18121829.

    • Search Google Scholar
    • Export Citation
  • Eckart, C., 1960: Hydrodynamics of Oceans and Atmospheres. Pergamon, 290 pp.

  • Gal-Chen, T., and R. C. Somerville, 1975: On the use of a coordinate transformation for the solution of Navier–Stokes equations. J. Comput. Phys., 17 , 209228.

    • Search Google Scholar
    • Export Citation
  • Girard, C., and M. Desgagné, 2004: An unstable semi-implicit scheme made stable. Preprints, 38th CMOS Congress, Edmonton, AB, Canada, Canadian Meteorological and Oceanographic Society, 120.

  • Héreil, P., and R. Laprise, 1996: Sensitivity of internal gravity waves solutions to the time step of a semi-implicit semi-Lagrangian nonhydrostatic model. Mon. Wea. Rev., 124 , 972999.

    • Search Google Scholar
    • Export Citation
  • Kasahara, A., 1974: Various vertical coordinate systems used for numerical weather prediction. Mon. Wea. Rev., 102 , 507522.

  • Klemp, J. B., W. C. Skamarock, and O. Fuhrer, 2003: Numerical consistency of finite differencing in terrain-following coordinates. Mon. Wea. Rev., 131 , 12291239.

    • Search Google Scholar
    • Export Citation
  • Kwizak, M., and A. Robert, 1971: A semi-implicit scheme for grid point atmospheric models of the primitive equations. Mon. Wea. Rev., 99 , 3236.

    • Search Google Scholar
    • Export Citation
  • Laprise, R., and C. Girard, 1990: A spectral general circulation model using piecewise-constant finite-element representation on a hybrid vertical coordinate system. J. Climate, 3 , 3252.

    • Search Google Scholar
    • Export Citation
  • Laprise, R., D. Caya, G. Bergeron, and M. Giguère, 1997: The formulation of the André Robert MC2 (Mesoscale Compressible Community) model. Numerical Methods in Atmospheric and Oceanic Modeling Atmosphere-Ocean: The Andre J. Robert Memorial Volume, C. A. Lin, R. Laprise, and H. Ritchie, Eds., Canadian Meteorological and Oceanographic Society/NRC Research Press, 195–220.

    • Search Google Scholar
    • Export Citation
  • Pinty, J-P., R. Benoit, E. Richard, and R. Laprise, 1995: Simple tests of a semi-implicit semi-Lagrangian model in 2D mountain wave problems. Mon. Wea. Rev., 123 , 30423058.

    • Search Google Scholar
    • Export Citation
  • Robert, A., 1966: The integration of a low order spectral form of the primitive meteorological equations. J. Meteor. Soc. Japan, 44 , 237244.

    • Search Google Scholar
    • Export Citation
  • Robert, A., 1969: The integration of a spectral model of the atmosphere by the implicit method. Proc. WMO/IUGG Symp. on NWP, Tokyo, Japan, Japan Meteorological Agency, 19–24.

  • Robert, A., 1981: A stable numerical integration scheme for the primitive meteorological equations. Atmos.–Ocean, 19 , 3546.

  • Robert, A., 1982: A semi-Lagrangian and semi-implicit numerical integration scheme for the primitive meteorological equations. J. Meteor. Soc. Japan, 60 , 319325.

    • Search Google Scholar
    • Export Citation
  • Robert, A., J. Henderson, and C. Turnbull, 1972: An implicit time integration scheme for baroclinic models of the atmosphere. Mon. Wea. Rev., 100 , 329335.

    • Search Google Scholar
    • Export Citation
  • Robert, A., T. L. Yee, and H. Ritchie, 1985: A semi-Lagrangian and semi-implicit numerical integration scheme for multilevel atmospheric models. Mon. Wea. Rev., 113 , 388394.

    • Search Google Scholar
    • Export Citation
  • Schär, C., D. Leuenberger, O. Fuhrer, D. Lüthi, and C. Girard, 2002: A new terrain-following vertical coordinate formulation for high-resolution numerical weather prediction models. Mon. Wea. Rev., 130 , 24592480.

    • Search Google Scholar
    • Export Citation
  • Simmons, A. J., B. J. Hoskins, and D. M. Burridge, 1978: Stability of the semi-implicit method of time integration. Mon. Wea. Rev., 106 , 405412.

    • Search Google Scholar
    • Export Citation
  • Tanguay, M., A. Robert, and R. Laprise, 1990: A semi-implicit semi-Lagrangian fully compressible regional forecast model. Mon. Wea. Rev., 118 , 19701980.

    • Search Google Scholar
    • Export Citation
  • Tanguay, M., E. Yakimiw, H. Ritchie, and A. Robert, 1992: Advantages of spatial averaging in semi-implicit semi-Lagrangian schemes. Mon. Wea. Rev., 120 , 113123.

    • Search Google Scholar
    • Export Citation
  • Thomas, S. J., A. V. Malevsky, M. Desgagné, R. Benoit, P. Pellerin, and M. Valin, 1997: Massively parallel implementation of the mesoscale compressible community model. Parallel Comput., 23 , 21432160.

    • Search Google Scholar
    • Export Citation
  • Thomas, S. J., C. Girard, R. Benoit, M. Desgagné, and P. Pellerin, 1998: A new adiabatic kernel for the MC2 model. Atmos.–Ocean, 36 , 241270.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    A variant of the Charney–Phillips–Robert vertical discretization grid.

  • View in gallery

    Simulations of idealized flow past topography: (a) Eulerian scheme with realistic time step (Co = 0.12), (b) inconsistent semi-Lagrangian scheme with very small time step (Co = 0.006), (c) inconsistent semi-Lagrangian scheme with Courant number equal to one (Co = 1.0), and (d) consistent semi-Lagrangian scheme with realistic time step (Co = 0.12).

  • View in gallery

    (a) Topography field, contours every 250 m; (b) gchen, inconsistent SL scheme in original terrain-following coordinate; (c) sleve, inconsistent SL scheme in SLEVE coordinate; and (d) slagw, consistent SL scheme in SLEVE coordinate. (b)–(d) show a 6-h forecast valid 1800 UTC 19 Sep 1999, made with MC2 at a resolution of 2 km, vertical motion field at 10 km, and contours every 50 cm s−1.

  • View in gallery

    A 6-h forecast valid 1800 UTC 19 Sep 1999, made with MC2 at a resolution of 2 km: (a), (c), (e) vertical motion fields at 5 km over a subdomain, and (b), (d), (f) 15-km-high cross sections along arrow, with contours every m s−1. (a), (b) gchen: inconsistent semi-Lagrangian scheme in original coordinate; (c), (d) sleve: inconsistent semi-Lagrangian scheme in SLEVE coordinate; and (e), (f) slagw: consistent semi-Lagrangian scheme in SLEVE coordinate.

  • View in gallery

    A 6-h forecast valid 1800 UTC 19 Sep 1999, made with MC2 at a resolution of 2 km: Spectra of (left) horizontal kinetic energy and (right) temperature variance at three levels in the atmosphere (5, 10, and 15 km). The spectra were calculated from reduced grids of 201 × 201 points taken in the middle of the domain.

  • View in gallery

    Merlin IV flight 26 track segment (flight level: 160–4880 m above sea level), with time labels (hh:mm multiplied by 10, e.g., 171.00 = 17:10 = 17h 10min) every other minute overlaid on the MC2 vertical velocity at 5000 m above sea level (slagw run). Gray shade interval is 1 m s−1.

  • View in gallery

    Comparison between Merlin IV averaged in situ vertical velocity measurements (gray line) and MC2 results interpolated to the flight track (heavy lines): (a) slagw run and (b) galchen run.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 98 51 3
PDF Downloads 44 25 0

Finescale Topography and the MC2 Dynamics Kernel

View More View Less
  • 1 Recherche en Prévision Numérique, Meteorological Service of Canada, Dorval, Quebec, Canada
Full access

Abstract

The Canadian Mesoscale Compressible Community (MC2) model provided daily forecasts across the Alps at 3-km resolution during the Mesoscale Alpine Programme (MAP) field phase of 1999. Among the results of this endeavor, some have had an immediate impact on MC2 itself as it increasingly became evident that the model was spuriously too sensitive to finescale orographic forcing. The model solves the Euler equations of motion using a semi-implicit semi-Lagrangian scheme in an oblique terrain-following coordinate. To improve model behavior, typical approaches were tried at first. These included a generalization of the coordinate transformation to make the terrain influence decay much more quickly with height as well as the introduction of nonisothermal basic states to diminish the amplitude of numerical truncation errors. The concept of piecewise-constant finite elements was invoked to reduce coding arbitrariness. But it was later pointed out that the problem was very specific and due to a numerical inconsistency. The true height of model grid points is fixed and known in height-based coordinates. Nevertheless, it was discovered that for this semi-Lagrangian scheme to be consistent, the departure height is an unknown that must be obtained in the same manner as the other unknowns.

Corresponding author address: Claude Girard, Recherche en Prévision Numérique, 2121 Trans-Canada Highway, Room 500, Dorval QC H9P 1J3, Canada. Email: claude.girard@ec.gc.ca

Abstract

The Canadian Mesoscale Compressible Community (MC2) model provided daily forecasts across the Alps at 3-km resolution during the Mesoscale Alpine Programme (MAP) field phase of 1999. Among the results of this endeavor, some have had an immediate impact on MC2 itself as it increasingly became evident that the model was spuriously too sensitive to finescale orographic forcing. The model solves the Euler equations of motion using a semi-implicit semi-Lagrangian scheme in an oblique terrain-following coordinate. To improve model behavior, typical approaches were tried at first. These included a generalization of the coordinate transformation to make the terrain influence decay much more quickly with height as well as the introduction of nonisothermal basic states to diminish the amplitude of numerical truncation errors. The concept of piecewise-constant finite elements was invoked to reduce coding arbitrariness. But it was later pointed out that the problem was very specific and due to a numerical inconsistency. The true height of model grid points is fixed and known in height-based coordinates. Nevertheless, it was discovered that for this semi-Lagrangian scheme to be consistent, the departure height is an unknown that must be obtained in the same manner as the other unknowns.

Corresponding author address: Claude Girard, Recherche en Prévision Numérique, 2121 Trans-Canada Highway, Room 500, Dorval QC H9P 1J3, Canada. Email: claude.girard@ec.gc.ca

1. Introduction

The main feature of the Mesoscale Compressible Community (MC2) model is its semi-implicit semi-Lagrangian (SISL) numerical time integration scheme developed by the late Andre Robert (Robert 1969; Kwizak and Robert 1971; Robert et al. 1972 for the SI part and Robert 1981, 1982; Robert et al. 1985; Tanguay et al. 1990 for the SL part). It includes spatial averaging of trajectory-endpoint values of the nonlinear terms evaluated at central time in addition to an off-centering of all space–time trajectory averages for the linear terms (Tanguay et al. 1992), both elements being used to reduce gravity wave noise believed connected mostly with stationary forcing, in particular but not exclusively orographic forcing, and occurring at large Courant numbers. The model behavior with respect to orographic forcing was extensively studied by Pinty et al. (1995), showing that SISL gave good results without using off-centering on all kinds of stationary nonlinear as well as linear orographic forcing provided appropriate time resolutions were used. The consequences of using large time steps, with and without off centering, were examined by Héreil and Laprise (1996).

The first comprehensive description of the model in terrain-following coordinates was given in Laprise et al. (1997). The model was later specialized to two major types of application: mesoscale forecasting (Benoit et al. 1997) and regional climate simulations (Caya and Laprise 1999). Substantial modifications were brought about in the forecast version by Thomas et al. (1997) for efficiency on various computer architectures and by Thomas et al. (1998, hereafter T98) to further improve (i) the efficiency of the SL scheme, (ii) the stability of the SI scheme over steep terrain, and (iii) the open lateral boundary conditions. In the fall of 1999, the model was used to provide daily forecasts across the European Alps at 3-km resolution during the entire special observation period (SOP; 7 September–15 November 1999) of the Mesoscale Alpine Programme (MAP). An overview of its performance during MAP is given in Benoit et al. (2002a).

In spite of the relatively good performance of the model on real as well as canonical cases, a problem handling finescale topography was becoming increasingly evident. From the beginning (Denis 1990), the model had been showing rather high sensitivity to orographic forcing. In consequence, topography used by the model was smoothed more than what is customarily done. A partial solution was to use a modified vertical coordinate as proposed by Schär et al. (2002), hereafter SLFLG). Another one was to use a modified basic state (Benoit et al. 2002b). A major step in finding a true solution to the problem was achieved when it was realized that not every model presented that much sensitivity. And recently its cause was diagnosed by Klemp et al. (2003), hereafter KSF) to be a numerical inconsistency between certain terms involving SL advection and other parts of the code.

In section 4, we unveil what we feel is the simple but true solution to this numerical inconsistency problem, and we show the obvious and immediate impacts of the correction. But first we give an up-to-date account of the model developments mentioned above, all of them mostly motivated by this search for a solution to the high spurious sensitivity of the model to finescale topography. In section 2, we review the model equations and describe the transformations of the SISL scheme to admit nonisothermal basic states and of the coordinate system to allow for more general oblique vertical coordinates. In section 3, we show how the concept of piecewise-constant finite elements is used to formalize spatial discretization and thus reduce code arbitrariness. We conclude in section 5.

2. Model equations

a. Basic features

In addition to the SISL scheme, the model has a number of other worthy characteristics. The first characteristic is a thermodynamic simplification. In effect, the equations of motion usually involved three main thermodynamic variables—temperature, pressure, and density—related diagnostically by an equation of state. Completely eliminating one of these variables simplifies the dynamic system while restraining the freedom of choice for the final set of equations. Here, density is eliminated with advantage.

The second characteristic involves a change of thermodynamic variables. With the equations written in terms of deviations, linear and nonlinear terms are easily distinguished. Previously, the basic state from which deviations could be obtained was specialized to the isothermal case and the new variables were simply taken as the perturbations of temperature and pressure. With general hydrostatic basic states, we find that the SISL scheme is more conveniently implemented using a generalization of the perturbation variables, a meaningful one. The variables namely become buoyancy and generalized pressure and the fundamental relevant parameters, the speed of sound and buoyancy frequency, are emphasized.

In the “horizontal,” that is, in the direction perpendicular to gravity, the equations are written in a so-called invariant form (Robert et al. 1972) readily admitting, with the specification of a single scaling parameter S, a choice of orthogonal coordinates: Cartesian (S = 1), spherical, and even cylindrical (rotating annulus experiments) as well as all conformal mappings (stereographic, Mercator, . . .) of the spherical earth. This feature of the model is, however, irrelevant for the problem at hand. It will not be further discussed. The equations will in fact be presented in absolute Cartesian coordinates (omitting the Coriolis force).

In the “vertical,” a terrain-following oblique coordinate of the height variety is used. The metric has now been generalized to greatly increase the flexibility in the choice of vertical coordinate definition. In particular, we may use the smooth level vertical (SLEVE) coordinate of SLFLG.

Space discretization is done using finite differences with variables distributed on a set of staggered grids particularly well suited for deriving the elliptic-type numerical equation that characterizes the SI scheme. Since the scheme is absolutely stable in principle, a balance can easily be achieved between space and time truncation errors by adjusting the time step. But even after the grid structure and local order of accuracy of finite differencing have been decided, there still remains some degree of arbitrariness for the final code. Such arbitrariness has now practically been eliminated through the application of the concept of piecewise-constant finite elements. The method can be used with advantage to construct conservative schemes on staggered grids in particular (see Laprise and Girard 1990 for further details). We cannot however claim that the resulting code is anything other than a particular set of at best second-order accurate, finite-difference equations.

Horizontal resolution is taken to be uniform in the chosen coordinate (e.g., in spherical coordinates it is uniform in latitude and longitude while in Mercator projection it is uniform in map coordinates). Vertical resolution may be varied at will using a stretching transformation (Laprise et al. 1997). It is then again uniform in the final indicial coordinate k, though this last transformation will be omitted from the following description.

b. Basic dynamic equations

We consider a fluid exclusively composed of well-mixed perfect gases (dry air). In an absolute reference frame, the basic equations for its evolution in time are
i1520-0493-133-6-1463-e1a
i1520-0493-133-6-1463-e1b
i1520-0493-133-6-1463-e1c
i1520-0493-133-6-1463-e2
Pressure p, density ρ, temperature T, and velocity υ = (u, υ, w) characterize the fluid; d/dt = ∂ /∂t + υ. ∇ is the material derivative and Φ the geopotential field, ∇Φ = gk; f = (fu, fυ, fw) refers to all nonconservative forces and Q to all heat sources; the gas constant R is the difference between the heat capacities at constant pressure cp and constant volume cυ.
Contrary to the other equations that are independent prognostic equations, the equation of state is diagnostic. This suggests that one thermodynamic variable is practically superfluous and can be eliminated. Let us then proceed to eliminate density from (1) using (2). We get
i1520-0493-133-6-1463-e3a
i1520-0493-133-6-1463-e3b
i1520-0493-133-6-1463-e3c
with κ = R/cp. In meteorological applications the various phases of water are of course considered. This implies the presence of appropriate continuity equations for each added substance along with modifications of (3) to take into account in particular the buoyancy effects of moisture and suspended hydrometeors.

c. Change of thermodynamic variables and separation of terms

It is always possible to subtract from the above equations a hydrostatic basic state. The step of identifying and separating advection and linear terms from the rest of nonlinear and source terms is a necessary one for the application of a SISL scheme. Here, we show how the scheme has been modified to allow for nonisothermal basic states leading to a set of terms linked with variable coefficients instead of constant ones. We found that redefining the thermodynamic variables greatly helped in clarifying the end result.

We start by introducing perturbations T ′ = TT∗, (lnp)′ = ln (p/p∗), whereby T∗(z) and p∗(z) are related hydrostatically [∂lnp∗/∂z = −g/RT∗]. Relevant parameters are the speed of sound c2 = cp/cυ(RT∗) and the buoyancy frequency N2 = g(βA + γA) with βA = ∂lnT∗/∂z and γA = g/cpT∗. We proceed to separate terms, leaving to the left-hand sides of the equations total derivatives and linear terms and sending to the right-hand sides nonlinear terms to join source terms:
i1520-0493-133-6-1463-e4a
i1520-0493-133-6-1463-e4b
i1520-0493-133-6-1463-e4c
Finally we define buoyancy b = gT′/T∗, generalized pressure P = R T∗(lnp)′, and generalized buoyancy B = b-γAP to obtain
i1520-0493-133-6-1463-e5a
i1520-0493-133-6-1463-e5b
i1520-0493-133-6-1463-e5c
with the right-hand sides explicitly given by
i1520-0493-133-6-1463-e6a
i1520-0493-133-6-1463-e6b
i1520-0493-133-6-1463-e6c
The left-hand side of system (5) is quite general because it only contains material derivatives plus linear forcing terms associated with acoustic and gravity wave oscillations. The equations are therefore independent of the choice of original dependent variables. We could have started for example with equations written in terms of potential temperature θ and the Exner function π and find B = ′/θ* and P = cp θ*π′. Moreover the equations are independent of the type of fluid (see, e.g., Eckart 1960).

d. Generalized vertical coordinate

With a terrain-following Z coordinate, specification of surface boundary condition becomes trivial: WS = 0. The simplest formulation is given by the following linear relationship (Gal-Chen and Somerville 1975):
i1520-0493-133-6-1463-e7
Here zS(x, y) represents the terrain heights, that is, the value of z corresponding to the bottom of the model, zT corresponding to the top, and H being a normalizing constant. For a model with a rigid top, zT = constant and we may choose H = zT.
Our code has been generalized to allow for more complex relations; for example, we may use the relation proposed by SLFLG:
i1520-0493-133-6-1463-e8
in which not only the effect of terrain decays much more quickly with height (exponentially rather than linearly) but also in which the decay rate is made dependent on the horizontal scale of the relief. In (8), there are two topography fields—zS1 and zS2—containing relief contributions of different scales (zS1 + zS2 = zS), each associated with its own vertical decay parameter, sS1 or sS2.
The transformation of the equations from orthogonal (x, y, z) to oblique terrain-following (x, y, Z) coordinates, in which the wind vector is not being transformed, is quite straightforward in particular because only one coordinate is involved in the transformation. The wind components u, υ, and w are simply but correctly treated as three independent scalars. Indeed it is sufficient to transform the spatial derivatives through a standard application of the chain rule (Kasahara 1974). With partial space derivatives of z named as follows,
i1520-0493-133-6-1463-e9
we have the following transformation laws:
i1520-0493-133-6-1463-e10a
i1520-0493-133-6-1463-e10b
i1520-0493-133-6-1463-e10c
Fully expanded, the resulting equations (left-hand sides only) are
i1520-0493-133-6-1463-e11a
i1520-0493-133-6-1463-e11b
i1520-0493-133-6-1463-e11c
i1520-0493-133-6-1463-e11d
i1520-0493-133-6-1463-e11e
The material derivative may be written in terms of the new coordinates as follows:
i1520-0493-133-6-1463-e12a
Comparing (12a) above with
i1520-0493-133-6-1463-e12b
we find, using transformation laws (10), that generalized vertical velocity W is given by
i1520-0493-133-6-1463-e13
Divergence, which takes the form
i1520-0493-133-6-1463-e14a
may also be written as follows:
i1520-0493-133-6-1463-e14b
Going from one form to the other implies obvious relations between metric coefficients,
i1520-0493-133-6-1463-e15
in addition to the definition (13) of W. The second form (14b), being a flux form, is a better choice for discretization. It however involves W, which is not a prognostic variable of the model. Discretizing (14b) therefore means discretizing (13) as well.

3. Space–time discretization

a. Semi-implicit semi-Lagrangian time discretization

Paradoxically, the details of the SISL time-integration scheme are not relevant to the problem at hand, except perhaps for the fact that Lagrange cubic polynomials are used for the required interpolations. We therefore only give a brief summary. Representing the set of five dependent variables by the vector ΨT = (u, υ, w, B, P/c2), linear terms by L, and right-hand sides by RT = (Ru, Rυ, Rw, RB, RP), system (11) is written
i1520-0493-133-6-1463-e16
Except for the addition of a weak Robert time filter needed to damp the computational mode inherent to the three-time-level scheme (Asselin 1972), the SISL scheme is fully described by
i1520-0493-133-6-1463-e17
with the following space–time difference and average operators:
i1520-0493-133-6-1463-e18a
i1520-0493-133-6-1463-e18b
i1520-0493-133-6-1463-e18c
in which ε is an off-centering parameter. The variables now only appear at three relative time levels t − Δt, t, and t + Δt identified by upper indices (, o, +) and two relative space points: the arrival point (x, y, Z), which will also be a model grid point, and the departure (upwind) point (x − 2α, y − 2β, Z − 2γ), where α = umΔt, β = vmΔt, and γ = WmΔt are the displacements along model coordinate axes evaluated at the trajectory midpoint (xα, yβ, Zγ). Full details of the procedure used to obtain α, β, and γ are given in section 3 of T98. If χ is a term to be evaluated at one of these space–time points, then
i1520-0493-133-6-1463-e19a
i1520-0493-133-6-1463-e19b
i1520-0493-133-6-1463-e19c
i1520-0493-133-6-1463-e19d
Expanding (17) using (18) leads to
i1520-0493-133-6-1463-e20
For convenience, we have written ▵t± for (1 ± ε)▵t. Note again that the linear terms contain variable coefficients, not only Go, G1, and G2, which were previously introduced in T98 and have become fully generalized here, but also N2∗ and c2∗, which, being made functions of true height z(x, y, Z), must take on a different value at every grid point in oblique coordinates. In practice these coefficients are conveniently precalculated and stored as three-dimensional fields, increasing the memory requirements of the model but leaving the computing costs unchanged.

The option of nonisothermal basic states has contributed (Benoit et al. 2002b) to the development of a satisfactory solution to the problem at hand. It has however its own drawbacks; numerical stability may be compromised (Simmons et al. 1978) for example. And, since it has little relevance to the main result of this paper, we will not discuss it further. All simulations presented here having been done using an isothermal basic state with T∗ = 273 K.

b. Staggered space discretization

The placement of grid points where dependent variables are defined is an important element of spatial discretization. In the horizontal, we use an Arakawa C type grid. In the vertical, we use a variant of the Charney–Phillips (1953), Robert (1966), vertical grid displayed in Fig. 1. Basically, N layers are considered. At their center we locate u, υ, and P (full levels). At the N − 1 interfaces (half levels) we locate B and w. This accounts for the 5N − 2 clearly internal variables and associated prognostic equations. Top and bottom boundary conditions on w account for another couple of variables and equations sufficient for closing the system when thermally isolated. In general though extra boundary conditions on B are required, and often this cannot be done directly. Here fully prognostic equations are introduced for B in the half layers touching the boundaries and possibly involving boundary-induced turbulent fluxes.

We find the dependent variables given at the following internal grid points: [u]i−1/2jk, [υ]ij−1/2k, [w]ijk−1/2, [B]ijk−1/2, [P]ijk. In the horizontal, variables having half-integer subscripts fall on frontiers of the domain in the direction concerned. Values of variables on these boundary points are given by boundary conditions. Details on how these are applied numerically are given in T98. At the top and bottom, vertical motion w is obtained diagnostically setting W = 0 in (13). At the top, this also means that wN+1/2 = 0 and therefore dw/dt = 0; PN+1/2 may then be obtained hydrostatically using BN+1/4. At the surface though, dw/dt does not generally vanish. But w1/2 is known diagnostically and therefore also implicitly dw/dt. This means that P1/2 may, indeed must, be obtained nonhydrostatically using dw/dt as diagnostically defined; P1/2 and PN+1/2 are required because extra thermodynamic equations are solved for B3/4 and BN+1/4 (and these are more accurately defined when interpreted to be located halfway in between).

c. Piecewise-constant finite elements

To formalize space discretization, we have introduced piecewise-constant finite elements, basis functions in terms of which dependent variables are defined and with respect to which errors are orthogonalized through integration. In all, six 1D elementary basis functions are required, for example,
i1520-0493-133-6-1463-e21a
i1520-0493-133-6-1463-e21b
in the x direction and similar ones in the other directions, to build four 3D basis functions (boxes),
i1520-0493-133-6-1463-e22a
i1520-0493-133-6-1463-e22b
i1520-0493-133-6-1463-e22c
i1520-0493-133-6-1463-e22d
to be used to represent the five dependent variables (Einstein’ summation convention used):
i1520-0493-133-6-1463-e23

d. Projection operator

Spatial discretization derives automatically from the application of a projection operator: each equation, multiplied by its appropriate basis function, is integrated over the whole domain, resulting in pure algebraic relations involving the expansion coefficients (gridpoint values). For any term or set of terms F, we may for example write formally,
i1520-0493-133-6-1463-e24
Here we have chosen the basis function appropriate for the vertical momentum equation. For a quantity in phase with the test function—for example, w—the projection acts as a simple selector of the appropriate coefficient:
i1520-0493-133-6-1463-e25
Note the passage from half-integer indices for the integral to integer indices (corresponding then to the indices used in the code) for the coefficient (gridpoint value). The square brackets accompanied by subindices are reserved to indicate such projections (integrals). For a quantity staggered with respect to the test function—P for example—the projection operates an average, which may be written in the following equivalent forms:
i1520-0493-133-6-1463-e26
For the derivative of a variable staggered with respect to the test function—for example, ∂P/∂Z—the projection gives a centered difference, which we may write equivalently as follows:
i1520-0493-133-6-1463-e27
In summary, the integrals always reduce to algebraic relations between coefficients. Most of the time, these integrals may be represented by combinations of means and differences involving the coefficients. A more complex example is given by the following nonlinear product taken from the right-hand side of the u momentum equation:
i1520-0493-133-6-1463-e28

e. Discretization in Cartesian z coordinates

Applied on (20) in Cartesian z coordinates, the method outlined above simply gives (dropping the + upper index):
i1520-0493-133-6-1463-e29a
i1520-0493-133-6-1463-e29b
i1520-0493-133-6-1463-e29c
i1520-0493-133-6-1463-e29d
i1520-0493-133-6-1463-e29e
The (˜) T∗ is used to indicate difference operators in z coordinates.

f. Discretization in oblique Z coordinates

In oblique Z coordinates, it would seem more appropriate to consider weighted projections (since dz = GodZ). Replacing F by GoF in the previous definition, we may write, for example,
i1520-0493-133-6-1463-e30
Analytically the Jacobian G0 is unique. Numerically, we consider four different piecewise-constant parameters, one for each of the four possible types of integrals,
i1520-0493-133-6-1463-e31
As thus defined, the parameters uncouple from the projected functions F no matter how both they and the functions are defined. The weighted projection (30) gives results no different from the simple one (24).
With these definitions for projectors and with the fundamental metric parameters given by
i1520-0493-133-6-1463-e32a
i1520-0493-133-6-1463-e32b
i1520-0493-133-6-1463-e32c
discrete divergence, which in Cartesian coordinates is given by
i1520-0493-133-6-1463-e33a
in oblique coordinates becomes
i1520-0493-133-6-1463-e33b
If discrete generalized vertical motion W is defined as follows:
i1520-0493-133-6-1463-e34
and noting that
i1520-0493-133-6-1463-e35
in complete analogy with differential relations (15), we obtain the following flux form:
i1520-0493-133-6-1463-e33c
This in a way guarantees mass conservation as far as calculation of divergence using (33b) is concerned. Note however that (33c) is not part of the model code (see T98).
To describe the discrete subsystem in oblique Z coordinates equivalent to the one in z coordinates (29), we simply replace the difference operators valid in z coordinates by their equivalent in oblique Z coordinates, that is,
i1520-0493-133-6-1463-e36a
i1520-0493-133-6-1463-e36b
i1520-0493-133-6-1463-e36c
as appearing in calculation of divergence (33b) for the continuity Eq. (29e) and
i1520-0493-133-6-1463-e37a
i1520-0493-133-6-1463-e37b
i1520-0493-133-6-1463-e37c
as appearing in calculation of pressure gradient terms in momentum Eqs. (29a)(29c). And this essentially takes care of space discretization since terms on the right-hand sides of the equations involve either the same operators or slightly more complicated versions of them for nonlinear terms. Compared to earlier versions of the model (Laprise et al. 1997; T98), the code is now more general, admitting nonisothermal basic states and more general oblique coordinate systems.

4. Problem and solution

a. Discovery of the problem

As mentioned in the introduction, the high sensitivity of MC2 to finescale orographic forcing was a fairly well known feature, hence the tendency to rather heavily smooth the topography used by the model. This could not be better illustrated than in Benoit et al. (2002b) where the topography of the MC2 at 3-km resolution, as prepared for and used during the MAP field phase, did not seem more detailed than that of the then operational Swiss model at 14 km. Thus use of smooth topography combined with the introduction of a sponge layer at high levels, sometimes also combined with augmented horizontal diffusion coefficient on momentum, all contributed to diminish the sensitivity and hide the problem.

The sensitivity is related to the following two things: (i) the nature of the oblique terrain-following coordinate system and (ii) the amplitude of the thermodynamic perturbations. This was documented by SLFLG using MC2 simulating 2D pure (except for a damping top layer) dynamic flow past idealized topography. We repeat here the description of the main experiment. The topography is given by a bell-shaped structure modified by small-scale (8Δx) periodic variations:
i1520-0493-133-6-1463-e38
where zo= 250 m, a = 5 km, and λ = 4 km. The upstream flow is characterized by uniform velocity uo = 10 m s−1, uniform buoyancy frequency N∗ = 0.01 s−1, surface temperature Ts = 288, and pressure ps = 105 Pa. The domain is Lx = 200 km and Lz = 19.5 km. The resolution is given by Δx = 500 m, Δz = 300 m, Δt = 6 s. The simulation is carried out until a quasi steady state is reached. SLFLG showed that “a dramatically distorted unphysical wave pattern at upper levels” produced using the Gal-Chen–Somerville coordinate (7) is (i) greatly diminished if the SLEVE coordinate (8) with scaling parameters sS1 = 5 km, sS2 = 2 km is used and (ii) eliminated if the perturbations are kept small (choosing N∗ = 0.01871 such that the mean and basic states of the simulated atmosphere become identical).

The idealized case with N∗ = 0.01 is generating perturbation quantities at high levels, which are rather large even for real situations. This suggested the possible benefits of introducing the option of nonisothermal basic states. This feature was indeed implemented (Benoit et al. 2002b) and essentially leads to perfect simulations when mean and basic states nearly coincide.

For real-case simulations, SLFLG showed greatly reduced numerical noise using the new SLEVE coordinate. Benoit et al. (2002b) were also able to reduce noise on real cases using nonisothermal basic states.

b. Nature of the problem

The presence of noise in models using oblique terrain-following coordinates is quite common, and causes for it may be varied. Here we are fortunate to have a canonical experiment emphasizing one specific problem. In fact, following the work of SLFLG, KSF succeeded in clearly relating the problem to a numerical inconsistency between the SL advection scheme and other parts of the code. They showed in particular that the problem was essentially resolved if less accurate quadratic interpolations, instead of the usual cubic ones, were used to obtain upstream values in just one of the equations, namely the thermodynamic equation. And we are able to verify that the problem vanishes when we replace the SL scheme by an Eulerian scheme (Fig. 2a), a result that we use here as our control. If any doubt persists, there is a classical test for numerical inconsistency that can be made, the convergence test: decrease the time step and see how the solution evolves. In fact, the time step chosen here is adequate for an Eulerian scheme but is already relatively small from the point of view of an SL scheme, and decreasing the time step by a factor of 20 (Fig. 2b) does not improve the SL solution. Hence the model is numerically inconsistent.

c. The solution

The first clue toward a true solution came to us when, instead of decreasing the time step, we increased it. The amplitude of the erroneous distortions then diminishes, with best result occurring for a Courant number Co = 1 (Fig. 2c). This canonical mountain wave is already quasi linear: decreasing zo by any factor gives the same patterns of vertical velocities but with amplitude smaller by the same factor. We nevertheless did linearize the model. In particular, the advection operator is linearized as follows:
i1520-0493-133-6-1463-e39
Here we must assume w∼0, not W∼0. From (13) we find
i1520-0493-133-6-1463-e40
With Co = 1 the SL scheme is not interpolating horizontally. The horizontal displacement is exactly 2Δx. The vertical displacement from grid point I − 2 to grid point i calculated at midpoint of the trajectory I − 1 is given by
i1520-0493-133-6-1463-e41
which is the exact solution. So (41) explains why the model is essentially correct at Co = 1. This may also explains why the model is incorrect away from Co = 1 and especially as Co → 0. In effect, what is really involved in that calculation is the advection of true height z on Z surfaces, which is too crudely performed with the Eulerian form of discrete Eq. (34). It introduces a numerical inconsistency that can easily be removed by a more accurate Lagrangian estimate. Replacing (13) by the equivalent,
i1520-0493-133-6-1463-e42
where the subscript Z indicates that the derivative is to be evaluated along constant-Z surfaces, we discretize (42) in Lagrangian fashion, calculating Wm at the midpoint of the trajectory as follows:
i1520-0493-133-6-1463-e43
where z = z(x, y, Z) is the actual height of the Z surface at the arrival grid point and zuZ = (x − 2α, y − 2β, Z), the height of the Z surface at the departure point obtained by interpolation. The meaning of (42) is that W, the vertical displacement in oblique coordinate, corresponds to w, the absolute vertical displacement, corrected by (dz/dt)Z, the absolute height displacement resulting from “horizontally” moving along a constant-Z surface. The numerical inconsistency problem is thus indeed resolved in theory (as acknowledged, indeed demonstrated, by KSF, who became aware of our solution) and experiment (see Fig. 2d). It is noticeable that the only place where Eulerian relation (34) was used explicitly in the model nowadays was exactly in the calculation of Lagrangian displacements. KSF suggested another solution, that is, to interpolate in Cartesian rather than in oblique space. This solution also avoids calculating vertical displacements based on (34), but it would not have been so obviously and easily implemented.

d. Impact of the solution

The expected impact of removing this numerical inconsistency on real 3D simulations is obviously a reduction in the level of noise associated with finescale orography. To document this, we show various fields resulting from three 6-h integrations made with the model over a region covering the Alps. The case is chosen from the MAP SOP and corresponds to the intensive observing period 2b (IOP-2b) event with initial time 1200 UTC 19 September 1999. The 2-km horizontal resolution used here is slightly higher than the 3-km resolution used during the MAP field phase (Benoit et al. 2002a). The horizontal grid contains 400 × 400 points and there are 44 levels. The basic state is kept isothermal with T∗ = 273 K. The physical effects package is essentially the same as that used then. This time, however, absolutely no smoothing is applied to the topography field, and the model is integrated without any horizontal diffusion and without a sponge layer aloft. For the first integration, gchen, the old method of calculating W and original Gal-Chen vertical coordinate are used. In the second integration, sleve, the SLEVE coordinate (sS1 = 10 km, sS2 = 6 km) is used. Finally the third integration, slagw, uses both the SLEVE coordinate and the new method of calculating W.

There is little need to emphasize the location and magnitude of the original noise problem (Figs. 3 and 4) as well as the impact of changing methods for calculating W and, to a smaller but significant degree, the advantage of using the SLEVE coordinate. As might be suspected the noise is greatly reduced on other fields as well. Shown in Fig. 5 are the spectra of horizontal kinetic energy and temperature variance at 5, 10, and 15 km, calculated from a reduced grid of 201 × 201 points taken from the middle of the domain using the technique described in Denis et al. (2002). We observe numerical noise reduction in every field and at all levels, in the smaller scales of course, but these include wavelengths even longer than 8–10 Δx. Not surprisingly, the precipitation field (not shown) is also affected.

e. Comparison with reality

We use the vertical velocity measured during the Merlin IV flight 26 of 19 September 1999 of the MAP field project to compare the model calculations with reality. The flight track, an Alpine transect over the upper Rhine valley, is presented in Fig. 6, with time labels, overlaid with the slagw solution for w. We focus on the portion of the flight that is level at 4880 m above sea level (flight level 160) and take the model solution at 5000 m as is, without height adjustment. Since the “raw” in situ values from the Merlin have a sampling rate of 1 Hz, which corresponds to a travel distance of ∼100 m, we apply a moving average of the N points surrounding the current data point, with N = 21; in that way, the model values, on a 2-km grid, can be better compared with the measurements. The model fields are cubically interpolated to the instantaneous 1-Hz aircraft positions, on which they form smoothly varying curves.

In Fig. 7, we compare individually the slagw and galchen solutions with the smoothed in situ w measurements. Visually from Fig. 7, it is clear that the slagw solution is quite closer to the measurements than the galchen solution. Quantitatively, the correlation coefficients, for the flight-time period 1645–1710, are 0.26 and 0.02 respectively; as a scale of reference, the correlation coefficient between the “raw” and smoothed measurements is 0.85, and that between the two model solutions is 0.20.

Although still far from ideal, the solution presented here for the w problem in MC2 constitutes a major improvement in accuracy.

5. Summary and conclusions

SISL schemes are very efficient schemes for the numerical integration of the Euler equations of motion of the atmosphere. In principle these schemes are absolutely stable such that accuracy becomes the sole factor in determining time steps. In terrain-following coordinates though, high and steep topography has remained a challenge for both the SI and SL components. But after the work of Héreil and Laprise (1996) and T98, MC2 appeared to be well prepared for the task of high-resolution forecasting over steep mountains like the Alps.

Although sensitivity to finescale topography had been observed, it was not thought to be a serious problem. The work by SLFLG, to which we participated, was initially seen as a valuable contribution to finescale noise reduction. This remains true: the computational cost of running with the SLEVE coordinate is negligible. Furthermore the equations in the generalized coordinate, described in section 2, look simpler, and the code itself, described in section 3, is definitely simpler than the precedent (T98), especially after application of the concept of piecewise-constant finite elements.

The idealized quasi-linear case studied by SLFLG had another particularity besides the finescale structure of the relief, namely a mean state quite far from the typical isothermal basic state used in the model. A choice of basic state closer to the actual one was bound to reduce numerical errors, in a similar manner that SLEVE reduces the adverse influence of finescale terrain features on model variables aloft. This motivated us to modify the SISL scheme accordingly. But as mentioned in section 3, we did not fully implement this option of using a nonisothermal basic state for three reasons. First, as we now know, the real problem is elsewhere; second, as is well known (Simmons et al. 1978), stability of the SI scheme may then be compromised; third, after Bénard’s (2003) evaluation of our SI scheme showing that in the absence of both a time filter and off-centering the scheme was actually absolutely unstable (and in fact without off-centering the model is known to blow up sometimes), we have developed (Girard and Desgagné 2004) a more stable SI scheme using what amounts essentially to a locally defined time-dependent basic state, a more accurate scheme that does not require off-centering and will be the subject of a separate study.

The true nature—numerical inconsistency—of a noise problem has now been revealed and a solution has been found. The results presented in section 4 lead us to conclude that absolute height, although a known fixed quantity in height-based coordinates, must nevertheless be transported in a Lagrangian manner to obtain a departure height consistent with departure values of the other prognostic variables.

REFERENCES

  • Asselin, R. A., 1972: Frequency filter for time integrations. Mon. Wea. Rev., 100 , 487490.

  • Bénard, P., 2003: Stability of semi-implicit and iterative centered-implicit time discretizations for various equation systems used in NWP. Mon. Wea. Rev., 131 , 24792491.

    • Search Google Scholar
    • Export Citation
  • Benoit, R., M. Desgagné, P. Pellerin, Y. Chartier, and S. Desjardins, 1997: The Canadian MC2: A semi-Lagrangian, semi-implicit wideband atmospheric model suited for finescale process studies and simulation. Mon. Wea. Rev., 125 , 23822415.

    • Search Google Scholar
    • Export Citation
  • Benoit, R., and Coauthors, 2002a: The real-time ultrafinescale forecast support during the special observing period of the MAP. Bull. Amer. Meteor. Soc., 83 , 85109.

    • Search Google Scholar
    • Export Citation
  • Benoit, R., C. Girard, M. Desgagné, S. Chamberland, and W. Yu, 2002b: Finescale orography and the MC2 dynamics kernel. Preprints, 10th Conf. on Mountain Meteorology, Park City, UT, Amer. Meteor. Soc., 219–222.

  • Caya, D., and R. Laprise, 1999: A semi-implicit semi-Lagrangian regional climate model: The Canadian RCM. Mon. Wea. Rev., 127 , 341362.

    • Search Google Scholar
    • Export Citation
  • Charney, J. G., and N. A. Phillips, 1953: Numerical integration of the quasi-geostrophic equations for barotropic and simple baroclinic flows. J. Meteor., 10 , 7199.

    • Search Google Scholar
    • Export Citation
  • Denis, B., 1990: Introduction de la topographie dans un modèle atmosphérique non-hydrostatique. M.S. thesis, UQAM, 121 pp. [Available from R. Laprise, Département de physique, UQAM, P.O. Box 8888, Stn “Downtown,” Montréal, Québec H3C 3P8, Canada.].

  • Denis, B., J. Côté, and R. Laprise, 2002: Spectra decomposition of two-dimensional atmospheric fields on limited-area domains using the discrete cosine transform (DCT). Mon. Wea. Rev., 130 , 18121829.

    • Search Google Scholar
    • Export Citation
  • Eckart, C., 1960: Hydrodynamics of Oceans and Atmospheres. Pergamon, 290 pp.

  • Gal-Chen, T., and R. C. Somerville, 1975: On the use of a coordinate transformation for the solution of Navier–Stokes equations. J. Comput. Phys., 17 , 209228.

    • Search Google Scholar
    • Export Citation
  • Girard, C., and M. Desgagné, 2004: An unstable semi-implicit scheme made stable. Preprints, 38th CMOS Congress, Edmonton, AB, Canada, Canadian Meteorological and Oceanographic Society, 120.

  • Héreil, P., and R. Laprise, 1996: Sensitivity of internal gravity waves solutions to the time step of a semi-implicit semi-Lagrangian nonhydrostatic model. Mon. Wea. Rev., 124 , 972999.

    • Search Google Scholar
    • Export Citation
  • Kasahara, A., 1974: Various vertical coordinate systems used for numerical weather prediction. Mon. Wea. Rev., 102 , 507522.

  • Klemp, J. B., W. C. Skamarock, and O. Fuhrer, 2003: Numerical consistency of finite differencing in terrain-following coordinates. Mon. Wea. Rev., 131 , 12291239.

    • Search Google Scholar
    • Export Citation
  • Kwizak, M., and A. Robert, 1971: A semi-implicit scheme for grid point atmospheric models of the primitive equations. Mon. Wea. Rev., 99 , 3236.

    • Search Google Scholar
    • Export Citation
  • Laprise, R., and C. Girard, 1990: A spectral general circulation model using piecewise-constant finite-element representation on a hybrid vertical coordinate system. J. Climate, 3 , 3252.

    • Search Google Scholar
    • Export Citation
  • Laprise, R., D. Caya, G. Bergeron, and M. Giguère, 1997: The formulation of the André Robert MC2 (Mesoscale Compressible Community) model. Numerical Methods in Atmospheric and Oceanic Modeling Atmosphere-Ocean: The Andre J. Robert Memorial Volume, C. A. Lin, R. Laprise, and H. Ritchie, Eds., Canadian Meteorological and Oceanographic Society/NRC Research Press, 195–220.

    • Search Google Scholar
    • Export Citation
  • Pinty, J-P., R. Benoit, E. Richard, and R. Laprise, 1995: Simple tests of a semi-implicit semi-Lagrangian model in 2D mountain wave problems. Mon. Wea. Rev., 123 , 30423058.

    • Search Google Scholar
    • Export Citation
  • Robert, A., 1966: The integration of a low order spectral form of the primitive meteorological equations. J. Meteor. Soc. Japan, 44 , 237244.

    • Search Google Scholar
    • Export Citation
  • Robert, A., 1969: The integration of a spectral model of the atmosphere by the implicit method. Proc. WMO/IUGG Symp. on NWP, Tokyo, Japan, Japan Meteorological Agency, 19–24.

  • Robert, A., 1981: A stable numerical integration scheme for the primitive meteorological equations. Atmos.–Ocean, 19 , 3546.

  • Robert, A., 1982: A semi-Lagrangian and semi-implicit numerical integration scheme for the primitive meteorological equations. J. Meteor. Soc. Japan, 60 , 319325.

    • Search Google Scholar
    • Export Citation
  • Robert, A., J. Henderson, and C. Turnbull, 1972: An implicit time integration scheme for baroclinic models of the atmosphere. Mon. Wea. Rev., 100 , 329335.

    • Search Google Scholar
    • Export Citation
  • Robert, A., T. L. Yee, and H. Ritchie, 1985: A semi-Lagrangian and semi-implicit numerical integration scheme for multilevel atmospheric models. Mon. Wea. Rev., 113 , 388394.

    • Search Google Scholar
    • Export Citation
  • Schär, C., D. Leuenberger, O. Fuhrer, D. Lüthi, and C. Girard, 2002: A new terrain-following vertical coordinate formulation for high-resolution numerical weather prediction models. Mon. Wea. Rev., 130 , 24592480.

    • Search Google Scholar
    • Export Citation
  • Simmons, A. J., B. J. Hoskins, and D. M. Burridge, 1978: Stability of the semi-implicit method of time integration. Mon. Wea. Rev., 106 , 405412.

    • Search Google Scholar
    • Export Citation
  • Tanguay, M., A. Robert, and R. Laprise, 1990: A semi-implicit semi-Lagrangian fully compressible regional forecast model. Mon. Wea. Rev., 118 , 19701980.

    • Search Google Scholar
    • Export Citation
  • Tanguay, M., E. Yakimiw, H. Ritchie, and A. Robert, 1992: Advantages of spatial averaging in semi-implicit semi-Lagrangian schemes. Mon. Wea. Rev., 120 , 113123.

    • Search Google Scholar
    • Export Citation
  • Thomas, S. J., A. V. Malevsky, M. Desgagné, R. Benoit, P. Pellerin, and M. Valin, 1997: Massively parallel implementation of the mesoscale compressible community model. Parallel Comput., 23 , 21432160.

    • Search Google Scholar
    • Export Citation
  • Thomas, S. J., C. Girard, R. Benoit, M. Desgagné, and P. Pellerin, 1998: A new adiabatic kernel for the MC2 model. Atmos.–Ocean, 36 , 241270.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

A variant of the Charney–Phillips–Robert vertical discretization grid.

Citation: Monthly Weather Review 133, 6; 10.1175/MWR2931.1

Fig. 2.
Fig. 2.

Simulations of idealized flow past topography: (a) Eulerian scheme with realistic time step (Co = 0.12), (b) inconsistent semi-Lagrangian scheme with very small time step (Co = 0.006), (c) inconsistent semi-Lagrangian scheme with Courant number equal to one (Co = 1.0), and (d) consistent semi-Lagrangian scheme with realistic time step (Co = 0.12).

Citation: Monthly Weather Review 133, 6; 10.1175/MWR2931.1

Fig. 3.
Fig. 3.

(a) Topography field, contours every 250 m; (b) gchen, inconsistent SL scheme in original terrain-following coordinate; (c) sleve, inconsistent SL scheme in SLEVE coordinate; and (d) slagw, consistent SL scheme in SLEVE coordinate. (b)–(d) show a 6-h forecast valid 1800 UTC 19 Sep 1999, made with MC2 at a resolution of 2 km, vertical motion field at 10 km, and contours every 50 cm s−1.

Citation: Monthly Weather Review 133, 6; 10.1175/MWR2931.1

Fig. 4.
Fig. 4.

A 6-h forecast valid 1800 UTC 19 Sep 1999, made with MC2 at a resolution of 2 km: (a), (c), (e) vertical motion fields at 5 km over a subdomain, and (b), (d), (f) 15-km-high cross sections along arrow, with contours every m s−1. (a), (b) gchen: inconsistent semi-Lagrangian scheme in original coordinate; (c), (d) sleve: inconsistent semi-Lagrangian scheme in SLEVE coordinate; and (e), (f) slagw: consistent semi-Lagrangian scheme in SLEVE coordinate.

Citation: Monthly Weather Review 133, 6; 10.1175/MWR2931.1

Fig. 5.
Fig. 5.

A 6-h forecast valid 1800 UTC 19 Sep 1999, made with MC2 at a resolution of 2 km: Spectra of (left) horizontal kinetic energy and (right) temperature variance at three levels in the atmosphere (5, 10, and 15 km). The spectra were calculated from reduced grids of 201 × 201 points taken in the middle of the domain.

Citation: Monthly Weather Review 133, 6; 10.1175/MWR2931.1

Fig. 6.
Fig. 6.

Merlin IV flight 26 track segment (flight level: 160–4880 m above sea level), with time labels (hh:mm multiplied by 10, e.g., 171.00 = 17:10 = 17h 10min) every other minute overlaid on the MC2 vertical velocity at 5000 m above sea level (slagw run). Gray shade interval is 1 m s−1.

Citation: Monthly Weather Review 133, 6; 10.1175/MWR2931.1

Fig. 7.
Fig. 7.

Comparison between Merlin IV averaged in situ vertical velocity measurements (gray line) and MC2 results interpolated to the flight track (heavy lines): (a) slagw run and (b) galchen run.

Citation: Monthly Weather Review 133, 6; 10.1175/MWR2931.1

Save