Sensitivity to convective parameterization in regional climate models

1
Sensitivity to convective parameterization in
regional climate models

• Raymond W. Arritt
• Iowa State University, Ames, Iowa USA

2
Acknowledgments

• Zhiwei Yang
• PIRCS organizing team William J. Gutowski, Jr.,
  Eugene S. Takle, Zaitao Pan
• PIRCS Participants
• funding from NOAA, EPRI, NSF

3
Overview

• Survey of convective parameterizations
• Sensitivity to specification of closure
  parameters in the RegCM2 implementation of the
  Grell scheme
• Sensitivity to the choice of cumulus
  parameterization in regional climate simulations
  using MM5

4
Survey of some commonly used convective
parameterizations in regional models

• Kuo-Anthes
• RegCM2, RAMS, MM5
• Kain-Fritsch
• MM5, RAMS (being implemented)
• Grell
• RegCM2, MM5
• Betts-Miller
• Eta, MM5

5
Survey of cumulus parameterization methods

• History and variants
• Mode of action
• What is the fundamental assumption linking the
  grid scale and cumulus scale?
• Cloud model, trigger, etc.

6
Kuo-Anthes scheme

• Originally developed by Kuo (1965) with
  refinements by Anthes (1974)
• Mode of action
• assume convection is caused by moisture
  convergence (this is wrong!)
• moisture convergence into a column is partitioned
  between column moistening and precipitation
• thermodynamic profiles are relaxed toward a moist
  adiabat over a time scale t

7
Partitioning of moisture convergence in the Kuo
scheme
column moistening b moisture convergence
precipitation (1-b) moisture convergence
Anthes parameter b varies (inversely) with
column relative humidity
moisture convergence
8
Grell scheme

• Simplification of the Arakawa and Schubert (1974)
  scheme
• there is only a single dominant cloud type
  instead of a spectrum of cloud types
• Mode of action
• convective instability is produced by the large
  scale (grid scale)
• convective instability is dissipated by the small
  scale (cumulus scale) on a time scale t
• there is a quasi-equilibrium between generation
  and dissipation of instability

9
Grell scheme

• Lifting depth trigger
• vertical distance between the lifted condensation
  level and the level of free convection becomes
  smaller than some threshold depth Dp
• default Dp 150 mb in RegCM2 and default Dp 50
  mb in MM5

LFC
Dp
LCL
10
Kain-Fritsch scheme

• Refinement of the approach by Fritsch and
  Chappell (1980, J. Atmos. Sci.)
• the only scheme originally developed for
  mid-latitude mesoscale convective systems
• Mode of action Instantaneous convective
  instability (CAPE) is consumed during a time
  scale t
• makes no assumptions about relation between
  grid-scale destabilization rate and
  convective-scale stabilization rate

11
Kain-Fritsch scheme

• Trigger Parcel at its lifted condensation level
  can reach its level of free convection
• a parcel must overcome negative buoyancy between
  LCL and LFC
• a temperature perturbation is added that depends
  on the grid-scale vertical velocity
• Detailed and flexible cloud model
• updrafts and downdrafts, ice phase
• entrainment and detrainment using a buoyancy
  sorting function

12
Entrainment and detrainment in the Kain-Fritsch
scheme
mix cloud and environmental parcels, then
evaluate buoyancy
positively buoyant parcels are entrained
negatively buoyant parcels are detrained
13
Betts-Miller scheme

• Based mainly on tropical maritime observations,
  e.g., GATE
• variant Betts-Miller-Janjic used in the Eta model
• Mode of action When convective instability is
  released, grid-scale profiles of T and q are
  relaxed toward equilibrium profiles
• equilibrium profiles are slightly unstable below
  freezing level
• basic version of the scheme has different
  equilibrium profiles for land and water this can
  cause problems (see Berbery 2001)

14
Questions

• Within a given cumulus parameterization scheme,
  how sensitive are results to specification of the
  closure parameters?
• Within a given regional climate model, how
  sensitive are results to the choice of cumulus
  parameterization scheme?

15
Sensitivity to closure parameters

• Perform an ensemble of simulations each using a
  different value for a closure parameter or
  parameters
• must truly be an adjustable parameter e.g.,
  dont vary gravitational acceleration or specific
  heat
• parameter value should be reasonable e.g.,
  convective time scale can't be too long
• Here in the Grell scheme of RegCM2, vary
• Dp (lifting depth threshold for trigger)
• t (time scale for release of convective
  instability)

16
Closure parameter ensemble matrix
Dp
t
17
Test cases

• Two strongly contrasting cases over the same
  domain
• drought over north-central U.S. (15 May - 15 July
  1988)
• flood over north-central U.S. (1 June - 31 July
  1993)
• output archived at 6-hour intervals
• initial and boundary conditions from NCEP/NCAR
  Reanalysis

18
Verification measures

• Root-mean-square error
• compute RMSE at each grid point in the target
  region (north-central U.S. flood area) and
  average
• Number of days that each parameter combination
  was within the 5 best (lowest RMSE) of the 25
  combinations
• attempts to show consistency with which the
  parameter combinations perform

19
Flood case RMS precipitation error (mm) over the
north-central U.S.
20
Drought case RMS precipitation error (mm) over
the north-central U.S.
21
Flood case number of days for which each
ensemble member was among the 5 members with
lowest RMSE
22
Drought case number of days for which each
ensemble member was among the 5 members with
lowest RMSE
23
Variability with different convective schemes A
mixed-physics ensemble

• How much variability can be attributed to
  differences in physical parameterizations?
• Perform a number of simulations each using
  different cloud parameterizations
• convective parameterization Kuo, Kain-Fritsch,
  Betts-Miller, Grell
• shallow convection on or off

24
Mixed-physics ensemble
Mean
Spread
25
Multi-model ensemble (PIRCS-1B)
Mean
Spread
26
Area-averaged precipitation in the north-central
U.S.
Mixed Physics
Multi-Model (PIRCS 1B)
27
Preliminary findings

• Results can be sensitive to choice of closure
  parameters
• best value of closure parameter varies depending
  on the situation it is not realistic to expect
  a single best value
• Use of different cumulus parameterizations
  produced about as much variability as use of
  completely different models
• Beware of statements such as MM5 (RAMS, RegCM2
  etc.) has been verified... without stating the
  exact configuration!
• There may be potential for this variability to
  aid in generating ensemble forecasts it is
  easier to run one model with different
  parameterizations than to run a suite of
  different codes

28
(No Transcript)