RAMS - Regional Atmospheric Modeling System

Transcription

1 RAMS - Regional Atmospheric Modeling Sstem Department of Atmospheric Science, Colorado State Universit and ASTeR Division of Mission Research Corporation Rosie Polkinghorne ATOC Meteorological Mesoscale Modeling 4/25/08

2 Non-hdrostatic, compressible equations = u K u K u K fv u w u v u u t u m m m ' π = v K v K v K fu v w v v v u t v m m m ' π = w K w K w K g w w w v w u t w m m m v 0 ' ' π rad il il h il h il h il il il il t K K K w v u t = = r K r K r K r w r v r u t r n h n h n h n n n n = w v u c R t v ' ρ ρ ρ ρ π π Equations of motion: Thermodnamic equation: Water species miing ratio continuit equation: Mass continuit equation:

3 Ener function R / C 0 p v p π = C ( p / p ) = C T / p The pressure perturbation term does not occur in the gravit term Removes the need to compute the densit perturbation - one less equation The vertical gradient of π is much less than that of p, introducing less error when using finite difference techniques

4 2-D or 3-D A few details Staggered Arakawa-C grid - all thermodnamic and moisture variables are defined at the same point, with the velocit components u, v, and w defined at 1/2 Δ, 1/2 Δ, and 1/2 Δ, respectivel Vertical grid spacing can be stretched Unlimited number of nested grids, grids can be added/removed during the simulation, grids can be moved during the simulation No minimum horiontal or vertical resolution No lower sie limit on a limited-area model domain, ma be run on a global domain

5 Initialiation Horiontall homogeneous from a single sounding RAMS/ISAN package Horiontal wind components, pressure, and relative humidit from the gridded global analsis and an available rawinsonde data are interpolated verticall to isentropic levels at a user-specified resolution. The Barnes objective analsis scheme is applied on the isentropic surfaces The isentropic dataset is then transferred to the model grid using overlapping polnomial interpolation

6 Advantages to the RAMS/ISAN package An objective analsis better approimates the interstation variabilit of the atmospheric fields since snoptic flow is roughl adiabatic Isentropes tend to be packed in frontal areas, providing enhanced resolution along discontinuities Short-wavelength features in Cartesian coordinates are transformed in the isentropic sstem into longerwavelength features that can be more accuratel analed objectivel with much less smoothing than with other coordinate sstems

7 Disadvantages to the RAMS/ISAN package The vertical resolution decreases as the atmospheric stabilit increases Isentropes frequentl intersect the ground a hbrid vertical coordinate is included, a miture of isentropic and terrain following coordinates

8 Solution Technique Velocit components and Ener function marched forward with leapfrog time differencing All scalar quantities other than the Ener function use forward-upstream time differencing Time-splitting is used to step forward the acoustic terms

9 Leapfrog phase and amplitude errors - Amplitude advection Δ Δ Δ Δ Phase Δ Δ Δ Δ Data from Pielke (2001)

10 Forward amplitude errors - advection From Tremback et al (1987)

11 Forward phase errors - advection 6th order scheme seems to have the best balance between accurac and efficienc The cost of a 6th order scheme is not prohibitive and errors are significantl smaller than lower-order schemes From Tremback et al (1987)

12 Effective Resolution Based on the phase and amplitude errors for both the forward and leapfrog schemes, the effective resolution of RAMS appears to be about 8Δ If a 6th-order forward scheme were used, the effective resolution could be increased to 4Δ

13 Coordinate Sstem Horiontal: Cartesian Polar stereographic Vertical: Terrain-following σ

14 Terrain-Following σ The top of the model domain is eactl flat and the bottom follows the terrain * = * = * = H g H g

15 Lateral Boundar Condition Radiative condition u t = ( u c) u Large-scale nudging Cclic boundaries

16 Radiative Condition Orlanski (1976) u c = t u Klemp and Lill (1978) - Orlanski phase velocities averaged in the vertical, then the average velocit applied to the whole column Klemp and Wilhelmson (1978) - phase velocit specified for a tpical gravit wave

17 Klemp and Wilhelmson (1978) A value of c=30 m/s was tested, along with c=10 m/s and c=50 m/s Results compared against a reference simulation where the model domain is increased b a factor of 4 The concluded that the simulation with c=30 m/s compares favorabl with the reference simulation, although a smaller value of c might be a little better This boundar condition is used in m eperiments with c=20 m/s From Klemp and Wilhelmson (1978)

18 Klemp and Wilhelmson vs Orlanski Tripoli and Cotton (1982) tested the sensitivit of a 2-D simulation of a thunderstorm to these two boundar conditions For that case, the Orlanski boundar conditions are superior to Klemp and Wilhelmson These results agree with the eperiments done b Clark (1979)

19 Top boundar condition Rigid lid Rigid lid with a high-viscosit laer aloft to damp gravit waves, b nudging to large-scale analsis or initial conditions

20 Inputs: Outputs: Surface boundar - LEAF-2 soil/vegetation tpe shortwave radiative transfer water transfer heat transfer longwave radiative transfer From Walko et al (2000)

21 Subgrid Miing Parameteriation Smagorinsk (1963) deformation-k closure scheme Deardorff level 2.5 K - Edd viscosit is a function of prognostic TKE (Deardorff, 1980) Mellor-Yamada level ensemble-averaged TKE (Mellor and Yamada, 1982)

22 Mellor-Yamada level 2.5 Inputs: wind potential temperature turbulent kinetic energ Outputs: subgrid-scale flues

23 Cumulus Parameteriation Modified Kuo - convection acts to consume the convective instabilit generated b larger scales Inputs: environmental potential temperature vertical velocit total water Outputs: cloud top, LCL profiles of heating and moistening convective tendenc terms for potential and total water

24 Radiation Parameteriation Mahrer and Pielke (1977) long/shortwave model - no cloud processes Chen and Cotton (1983) long/shortwave model - cloud processes considering all condensate as liquid Harrington (1997) long/shortwave model - twostream scheme interacts with liquid and ice hdrometeor sie-spectra

25 Harrington (1997) long/shortwave model Transfer equations are solved for three model gases - H 2 O, CO 2, and O 3 Optical properties of water droplets are computed used Loren-Mie theor, while the theories of Mitchell and Arnott (1994) and Mitchell et al (1996) are used to compute the optical properties of non-spherical ice crstals Raleigh scatter and absorption are treated as independent of wavelength across a given band Scattering properties are calculated based on an assumed gamma distribution

26 Harrington (1997) long/shortwave model Inputs: Outputs: model gas concentrations (H 2 O, CO 2, and O 3 ) hdrometeor miing ratios and sie distribution long/shortwave radiative flues

27 Stable precipitation parameteriation No cloud Condensation onl Single-moment bulk scheme (Walko et al, 1995) - prognose miing ratio Two-moment bulk scheme (Meers et al, 1997) - prognose miing ratio and number concentration

28 Two-moment bulk scheme (Meers et al, 1997) Water is categoried as vapor, cloud droplets, rain, pristine ice, snow, hail, aggregates, and graupel Hdrometeors in each categor are assumed to conform to a prescribed gamma distribution The conservation equation for hdrometeors includes advective and turbulent transport b the resolved and subgrid velocities in the model, sources and sinks which consist of all tpes of conversion from one categor to another, and local losses and gains due to gravitational sedimentation

29 Two-moment bulk scheme (Meers et al, 1997) Inputs: atmospheric pressure atmospheric temperature hdrometeor miing ratio hdrometeor number concentration Outputs: hdrometeor miing ratio hdrometeor number concentration

30 Computer aspects Parallel processing on distributed and shared memor platforms using Message Passing Interface (MPI) Originall run on mainframe computers such as CRAY and CYBER machines Now can be run on UNIX, LINUX, MS Windows 95/98/NT/2000 Currentl run on Lightning, an NCAR supercomputer running LINUX