COSMO Sibiu 2013 Matthias Raschendorfer Some Challenges related to physical parameterizations Current WG3a-Activity towards solving the problems 1-st Part: Turbulence and Convection 2-nd Part: Microphysics and Radiation (U. Blahak)
COSMO Sibiu 2013Matthias Raschendorfer
Some Challenges related to physical parameterizations
Current WG3a-Activity towards solving the problems
1-st Part: Turbulence and Convection
2-nd Part: Microphysics and Radiation (U. Blahak)
The filtered model equations contain local and GS parameterizations:
p,Qp,ˆQkˆkˆˆˆ iiiiiiiiiit
vvv
SGS flux density roughness layer modification of transport
SGS source term including form drag
GS source termGS flux density
COSMO Sibiu 2013Matthias Raschendorfer
molecular flux density
Momentum:
(Enthalpy) ~Temperature:
water phases:
pressure gradient + gravity + Coriolis force
pressure work + cloud microphysics + radiation
cloud microphysics
functions in various covariance terms: of scalar variables
( + dissipation )
to be closed by restricting assumptions
Whole SGS variability needs to separated into classes to which specific closure assumptions can be applied
Turbulence: isotropic, normal distributed, only one characteristic length scale at each grid point, forced by shear and buoyancy closure by truncated 2-nd order equations
Circulation: non isotropic, arbitrarily skewed and coherent structures of several length scales, supplied by various pressure forces: closure adapted to process (e.g. by mass flux equations)Convection; kata- and anabatic flows; wakes; horizontal shear eddies; braking gravity waves
with turbulent and circulation contribution
Principal Problems:
COSMO Sibiu 2013Matthias Raschendorfer
Separation Turbulence Circulations
Parameterizations of source terms
Parameterizations of SGS processes
integrated in
Cloud-microphysics
Radiation transport
• Moist turbulence using statistical saturation adjustment, but
Turbulent contributions to phase change terms not yet considered in GS budgets
Non equilibrium processes (icing and precipitation) not included
• Convection scheme treats micro-physics including precipitation, but
Not complementary with GS microphysics
Radiation is not included
STICinteraction
Local parameterizations:
GS parameterizations:
• Radiation parameterization considers cloud properties, but Precipitating hydrometeors (snow) are not yet included SGS variability of cloud properties not properly considered
Missing interaction to be included :
• TKE scheme contains interaction terms, but
Some interaction terms are crude estimates and related circulations don’t even have their own contribution to transport (mixing) of 1-st order variables as well as TKE
• Mass flux convection scheme (seems not to be dispensible): Does not yet contain any dependency from turbulence Convective mixing of TKE not yet considered Is not separated against resolved convection (grey-zone, double counting) Is not able to give estimates of volume fractions of convective subdomains
Overlap of turbulent and convective contributions to microphysical source terms can’t be treated properly (no consistent description of cloud processes)
UTCS?
COSMO Sibiu 2013Matthias Raschendorfer
Some specific challenges:
• Some simplifying approximations are no longer valid due to increased or variable horizontal resolution:
3D-extensions: tilted columns; horizontal diffusion; transport of 2-nd order moments (TKE)Grey-zone; scale adaptive convection
Neglect of horizontal gradients compared to vertical ones, allowing single column solutions
Neglect of up- and downdraft fraction and mean vertical wind speed in convection parameterization (completely unresolved convection)
Roughness layer due to land use is only a small part of the lowest model layer, allowing to treat it in the SAT scheme only
vertically resolved roughness layer: additional form drag; smaller roughness length, modification of turbulent length scale
• More consistent and complete parameterizations: Avoiding numerical artefacts and instabilities Avoiding contradictory, artificial or unnecessary approximations Removing problems with diurnal cycle, stable boundary layer, low
level stratus; SAT Consolidation /merging of independent development
• Application of parameterizations in COSMO and ICON: common physics library: generation of clear interfaces; multi parameterization ensemble; modularization; cleaning up of NAMELISTs; adaptations for surface tiles; outstanding documentations
autom. parameter estim.; PP CALMO; statistical hyper-parameterization or post-processing
ongoing improvement; finishing PP UTCS; PT ConSAT and followers
• Improvement by non-physical extensions:
using advantage of different approaches
using direct impact of error estimates
Including non-deterministic aspects stochastic physics
Expensive calculations can be called less frequently (smooth evolution in time) or can even be avoided (single column solutions) Adaptive parameterizations
short term
longer term
interdisciplinary ; longer term
COSMO Sibiu 2013Matthias Raschendorfer
Current WG3a-Activity towards solving the problems:
1-st Part: GS Parameterizations
Matthias RaschendorferCOSMO
Work on turbulence and SAT:
Allowing for TKE-advection (U.Blahak), o Implemented in 4.18; technically working; can be implemented in current version shortly
Adding scale interaction terms (M.Raschendorfer), Production due to SSO-wakes, horizontal shear eddies and convection
o SSO-term: operational at DWDo Production by convection: needs to be verified, but model output for EDR-forecasto Horizontal shear term: tuning parameter only estimated, but still used for EDR-forecast
Reformulation of TKE scheme (including SAT) (M.Raschendorfer), Changing positive definite solution of prognostic TKE-equation Weakening numerical security limits and modularization: common SUBS for turbulence and SAT Diffusion of conserved variables Same implicit diffusion solver for 1-st order variables and TKE with options for better coupling
o implemented in private test-version; and ICON not yet verified; common version in work
3D-Smagorinsky scheme implemented (W.Langhans, ETHZ) o Implemented in private test version (already documented)o Horizontal shear production + horizontal diffusion can be activated as well using current TKE
scheme
Diagnostics of TKE-scheme in stable conditions (Ines Cerenzia) Analytical and experimental study
o Report just availabe
Sibiu 2013
COSMO Sibiu 2013Matthias Raschendorfer
Thermodynamic corrections:
Carrying adiabatic source terms in prognostic pressure equations (U.Blahak) o Is implemented and being tested
Former isobaric grid scale saturation adjustment changed to an isochoric process (U.Blahak) Adjustment generates now a pressure correction, is mass conserving and fits to ICON
o Implemented and tested: only small impact
UTCS/TKESV: (D. Mironov, E. Maschuskaya) TKESV extension; statistical cloud scheme now based on double Gaussian distribution
o Implemented in test version
Turb-i-Sim: (J. Schmidli, O. Fuhrer, …) Evaluation and improvement of COSMO turbulence over Alpine topography
o Project at ETH and MeteoSwiss, just started
Deardorff-restriction of turbulent master length scale (M.Raschendorfer)o Implemented since more than a year in current version, needs to be verified
Mixed water-ice phase for turbulence and statistical saturation adjustment (M.Raschendorfer)o Implemented in old test-version only; tested by E.Avgoustoglou
Matthias RaschendorferCOSMO
Work on microphysics:
Implementation of 2-moment scheme (A.Seifert) Runtime 60-100% increased! Only as an reference or for special purpose (COSMO-ART) Further work on hail-microphysics and optimization
o Adopted as an extra code to 4.25 and tested: slightly better over all verification
Prognostic treatment of melted water fraction within solid water parcels (A.Seifert)o Ready for testing in case of snowo Further work for graupel and hail planned only as an extension of the 2-moment scheme
Almost ready improvement of the 1-moment scheme (F.Rieper) Changing exponential distribution function to a more general gamma-function Implementation of an improved sedimentation formulation for snow and rain Some bug fixes
o All to gather implemented in current version and being tested
Running improvement of 1-moment scheme Consideration of homogeneous ice nucleation in cirrus clouds allowing higher oversaturation (C.Köhler) Improved simulation of super-cooled water to improve forecast of aircraft icing (F.Rieper)
Sibiu 2013
Matthias RaschendorferCOSMO
Work on radiation:
Using an improved aerosol climatology (J.Helmert)o Test runs performed: currently too transparent clouds
Slightly modifying cloud cover diagnostics for ice clouds in radiation scheme (A.Seifert)o Already in current code
Considering precipitating hydrometeors in radiation calculation (U.Blahak) In particular slowly falling snow should be considered
o Work just started
Adaptive sampling of grid points used for radiation calculation (V.Venema, Uni Bonn) Running radiation only once for all grid points with similar properties related to radiation
o Promising, only research version prepared
Monte Carlo spectral integration (MPI Hamburg; B. Ritter) Varying stochastically the absorption coefficients of a reduced number of spectral bands
o Promising, only research version prepared
Sibiu 2013
Separated TKE equation (including scale interaction sources):
buoyancy production
eddy-dissipationrate (EDR)
0labil:neutral:stabil: 0
00
time tendency of TKE
transport(advection + diffusion)
shear production by sub grid scale circulations
0
v
shear production by the mean flow
0
v
Matthias Raschendorfer COSMO Lugarno 2012DWD
Additional Shear -Production of TKE by:
SSO wakesHorizontal shear eddiesVertical convective currents
Formal scale separation automatically produces interaction between GS parameterizations
of turbulence and circulations
More physically based mixing even for stable stratification
• Missing link; • Computationally extremely cheap; • clear impact
Matthias Raschendorfer COSMO Lugarno 2012DWD
pot. temperature [K]Wind speed [m/s]
referenceincluding horizontal shear – and SSO-production
including horizontal shear –, SSO- and convective production
mountain ridge
COSMO-US: cross section across frontal line and Appalachian mountains
Consequences of scale interaction terms and general model improvment:
Matthias RaschendorferDWD
More physical based TKE and mixing in the stable BL
- Is already beneficial for CAT-forecast needed for aviation (s. previous reports)
- Should be beneficial also for near surface SBL.
- Previous artificial security measures needs to be adopted!
First candidate: the minimal diffusion coefficient
- Previous value: tkv[h,m]min = 1.0 m2/s (same for scalars and momentum)
- Seems to dissolve BL clouds much to early now (and was presumably always a bit too large)
- Previous attempts to decrease it has not been successful
- After lots of general numerical improvement of the model and the introduction of at least the SSO-source term,
a further attempt has now been tried
- New value: tkv[h,m]min = 0.4 m2/s
CUS 2013
Computationally extremely cheap; large impact in particular for T_2m_Min (SK=-13.33 for a 2-month exp.) !!
Theta/[°C]
Vel/[m/s]
cloud-water-content/[Kg/Kg]:
time-height cutRoutine Experiment
all values are area averages
Diagnostics of PBL parameterization in stable conditions
Considerations about the stable PBL parameterizations in COSMO (operational setting in Arpa-SIMC) evidenced through a case study in the Po Valley
TKE forcings sum increased so that Ri never exceeds Ric
Very stable conditions not well described and led to less stable cases
Increase of turbulence in stable
conditions
Ines Cerenzia: ISAC-CNR, Arpa-SIMC
First test: reduction of TKMmin and TKHmin from 11 to 1010-2-2 m2/s
Diagnostics of PBL parameterization in stable conditions
Diff. Coeff fall below 1 m2/s in some periods in stable conditions
Increased amplitude of the oscillations in turbulence-related variables
CAUSE?
Effect on 2m Temperature
Ines Cerenzia: ISAC-CNR, Arpa-SIMC
Removal of the oscillations by setting pat_len=0
Diagnostics of PBL parameterization in stable conditions
Ines Cerenzia: ISAC-CNR, Arpa-SIMC
Neglect the triple term in TKE eq. due to pressure-velocity correlation
Effect on the whole PBL to be further investigated
Modelling Scalar Skewness: an Attempt to Improve the Representation of Clouds and Mixing Using a Double-Gaussian Based Statistical Cloud Scheme
Dmitrii Mironov 1, Ekaterina Machulskaya1, Ann Kristin Naumann2,
Axel Seifert 1,2, and Juan Pedro Mellado2
1) German Weather Service, Offenbach am Main, Germany
2) Max Plank Institute for Meteorology, Hamburg, Germany
Naumann et al. (2013) developed a statistical cloud scheme based on a 3-moment double-Gaussian PDF of linearized saturation deficit (s); the scheme requires mean, variance, and skewness of s as input
Transport equation for the skewness Ss of s is developed, and closure assumption for the third-order and fourth-order s-velocity correlations are formulated that account for high-skewness cloud regimes (e.g. cumuli)
The Ss equation is coupled to the TKE-Scalar Variance mixing scheme (see Machulskaya and Mironov 2013, COSMO Newsletter No. 13) and to the 3-moment double-Gaussian cloud scheme
The new scheme is tested against LES data (Heinze 2013) through single-column simulation of shallow cumuli (BOMEX test case); first results look promising
A statistical cloud scheme (statistical saturation adjustment) based on a pure Gaussian PDF is part of the current (separated) TKE scheme
In terms of scale separation, cloud processes due to non-Gaussian processes are due to circulations treated in different schemes (e.g. mass flux scheme for “shallow convection”).
This approach tries to treat these processes within a turbulence framework:
TKESV + New Cloud Scheme: Cloud Fraction and Cloud Water
BOMEX shallow cumulus test case (http://www.knmi.nl/~siebesma/BLCWG/#case5) .Profiles are computed by means of averaging over last 3 hours of integration (hours 4 through 6). LES data are from Heinze (2013).
Comprehensive testing of the new scheme (stratus and stratocumulus regimes, etc.)
Consideration of numerical issues
Implementation into COSMO an ICON
Further development of the scheme, e.g. consideration of the effect of microphysical processes on the scalar variance and skewness (in co-operation with the HErZ-CC team)
Outlook
COSMO Sibiu 2013Matthias Raschendorfer
Non physical parameterization complements:
• Trying to improve physical parameterizations using model error estimates:
“Verification with feed back” should help to ask the following questions:
1. How good is a certain model?
Meaningful error measure not unique but dependent on most important output for customers
Important information by comparison with other models
2. What is wrong with the model?
Errors need to be related to specific inaccurately simulated processes
Conditional verification by VERSUS or other specific diagnostic tools helps to limit the possible sources of errors
Arbitrary conditions in terms of model output, measurements and external parameters
Most selective conditions need to be found in cooperation with modelers
Detection of most crucial parts in physical parameterizations and testing alternative formulations
3. How can we improve the model systematically?
Statistical post-processing
o Offline application of regression functions generating user optimized products (T_2m, EDR, …)
Automatic parameter optimization (related to PP CALMO)
o Successive parameter estimation for specific processes separated by component testing
o Statistical hyper-parameterization (needed, if parameters are dependent on internal model state)
o Online expression of each physical parameter by a regression function in some model variables
Stochastic extensions (needed, if the model output still doesn’t describe the reality satisfactory)
o Non-deterministic contributions to initial conditions, boundary values, numerics or parameterizations
classical verification
model diagnostics
data assimilation
ensemble prediction; probabilistic forecast; error estimate
o introduction of additional statistical moments by simulation of stochastic processes Stochastic physics ?
Model input
boundary values t,M
0t,M
global constants
initial values
global parameter p
local (external) parameter Me
Model output
prognostic variables t,M
t,Mimplicit diagnostics
Explicite diagnostics t,M
Model integration
cdiagnosticmodel calculation
Observations T,L,,
Superobservation T,L,, Supercalculation T,L,,
averagingaveraging
compare
Additional parameters for explicit diagnostics
Md
assimilation
Principal of the parameterization complements:
Trying to improve physical parameterizations by systematic parameter tuning:
COSMO Sibiu 2013Matthias Raschendorfer
parametertuning
Providing a list of parameter sub sets containing as few as possible parameters, related to specific conditions and a verification quantity that can be compared with measurements and that is sensitive only to those parameters in the sub set in case of the applicability of that condition.
by minimizing the model error of a verification quantity
conditional sampling
regression coefficients k
Hyper- parameterization
decreasing stochastic complementincreasing
T,L,,
Stochastic variations of parameterizations:
might even decrease current stochastic complement
Stochastic variations of model input:
should decrease expectation of stochastic complement
• Stochastic variation of tendencies
• stochastic properties of SGS surface tiles or convective cells
Stochastic PhysicsMotivations
•to improve the model stochastically if it is not possible to do it deterministically•to estimate the background (model) error for the data assimilations purposes•to provide the users with an estimation of the forecast reliability and uncertainty
Possible steps
•to determine the entire model error and to approximate it with a random process with the same time and space correlations•to go further into the determination of different types of the model error•to develop a more consistent approach: noise structure should not be arbitrary, but should be determined by the governing equations