Parameterizing Parameterizing convective convective
organizationorganizationa stab, in current a stab, in current
CAMCAM
and whyand whyBrian Mapes and Richard NealeBrian Mapes and Richard Neale
QuickTime™ and ampeg4 decompressor
are needed to see this picture.
Parameterizing Parameterizing convective convective
organizationorganizationa stab, in current a stab, in current
CAMCAM
and whyand whyBrian Mapes and Richard NealeBrian Mapes and Richard Neale
What did the movie show?
• (I claim, for current purposes): – Afternoon timing of rain set by an organization
delay time – This is many cu parcel turnover times– Aided by mtns in this case– More generally, aided by precipitation
• (via its evaporation - cold pools– a positive feedback
»When it rains, it pours
Outline
• Several more slides emphasizing concept and pervasiveness of ”organization”
• What this study isn’t about, and is
• Sensitivities and feedbacks
• WHAT WE DID TO CODE
• WHAT IT DID TO MODEL OUTPUTS
• Discussion type stuff and summary
A model view of diurnal development Khairoutdinov and Randall 2006
high-res simulations of shallow-deep transition (flat perioidc domain, specified diurnal surface flux)
Khairoutdinov and Randall 2006Organization (“upscale growth”) takes time
precip. evaporation is key
Khairoutdinov and Randall 2006
connected cloud
objects sorted by top height
true aspect ratio
connected cloud
objects sorted by top height
true aspect ratio
Mesoscale org. is ubiquitous in deep convectionCloudsat: an unbiased sample from the Asian monsoon
Mesoscale org. is ubiquitous in deep convectionCloudsat: an unbiased sample from the Asian monsoon
Mesoscale org is ubiquitous, II: (100s km, many hours scales)
Composite 10x10 deg 3-hourly evolution of IR, PW, 10m divergence around 1st appearance of cold
clouds (< 210K) on 0.5 deg grid
t=0t=0-12h-12h +9h+9h
Meso scale & lifetime clear even in this equal-weight composite
IRIR
PWPW
divdiv
Mapes Milliff Morzel in prep.
(strong rotation cases excluded)
A time domain view in tropical shipborne Doppler radar
data lag regression of vertical mass
flux w.r.t. surface rain
Mapes and Lin 2005 MWR
1000
800
600
400
200
p (mb)
Generic view of organized convection
shallow -> deep -> stratiform
shallowcu
deepcb
humid, cloudy heatedupdrafts
dry,rainy
downdraftcold conv. outflow
stratiform rain anvil
---(hours, days, even weeks)---->
So What? asks a Climo-Globo-Dynamician
So What? asks a Climo-Globo-Dynamician
Mean state implications Spatial patterns of convxn (tropical biases) Transient variability
– diurnal timing over land– PDF and nonlinear impacts
like ground hydrology
– MJO and other tropical waves apparently impacts ENSO
Mean state implications Spatial patterns of convxn (tropical biases) Transient variability
– diurnal timing over land– PDF and nonlinear impacts
like ground hydrology
– MJO and other tropical waves apparently impacts ENSO
Issues in GCM precipitationDai 2006: ”Precipitation Characteristics in Eighteen Coupled Climate Models”
Issues in GCM precipitationDai 2006: ”Precipitation Characteristics in Eighteen Coupled Climate Models”
...unrealistic double-ITCZ pattern over the tropical Pacific ...models fail to capture ... large intraseasonal variations ...too much convective (over 95% of total) and too little stratiform
precipitation* ...underestimate the contribution and frequency for heavy (>20 mm
day−1) and overestimate them for light (<10 mm day−1) precipitation...rains too frequently...
...Intensity... in storm tracks off eastern coasts... too weak... ...warm-season convection starts too early [in the day]...
– *is TRMM conv/strat = convection.F90 / stratiform.F90 ?? focus on the real process issues: profile, location, timing...
...unrealistic double-ITCZ pattern over the tropical Pacific ...models fail to capture ... large intraseasonal variations ...too much convective (over 95% of total) and too little stratiform
precipitation* ...underestimate the contribution and frequency for heavy (>20 mm
day−1) and overestimate them for light (<10 mm day−1) precipitation...rains too frequently...
...Intensity... in storm tracks off eastern coasts... too weak... ...warm-season convection starts too early [in the day]...
– *is TRMM conv/strat = convection.F90 / stratiform.F90 ?? focus on the real process issues: profile, location, timing...
Might “organization” help?Might “organization” help?
shallowcu
deepcb
stratiform rain anvil
developmentsensitivities &
feedbacks
continuationpast mean-state stabilization
late stageimpacts
What this talk ISN’T about, 1What this talk ISN’T about, 1 We are NOT attempting to parameterize the
impacts of mesoscale anvils (stratiform precipitation) as another category of moist vertical eddy– a la Donner (1993, 2001)
{other authors literally compare TRMM “stratiform” rain to model “large-scale” rain, presuming that this is the job of the cloud scheme, NOT convection scheme}
– philosophical debate needed? elsewhere.
We are NOT attempting to parameterize the impacts of mesoscale anvils (stratiform precipitation) as another category of moist vertical eddy– a la Donner (1993, 2001)
{other authors literally compare TRMM “stratiform” rain to model “large-scale” rain, presuming that this is the job of the cloud scheme, NOT convection scheme}
– philosophical debate needed? elsewhere.
We are NOT attempting to parameterize the impacts of two-dimensionality or steadiness (“organization” as used in some contexts), e.g. on vertical momentum flux
We are NOT attempting to parameterize the impacts of two-dimensionality or steadiness (“organization” as used in some contexts), e.g. on vertical momentum flux
What this talk ISN’T about, 2What this talk ISN’T about, 2
What this talk IS aboutWhat this talk IS about Organization here means sub-grid variations of
(mainly) thermodynamic variables, correlated* with convective updraft occurrence
Organization here means sub-grid variations of (mainly) thermodynamic variables, correlated* with convective updraft occurrence
“organized” by
precipitation
“random”
*this makes it sound too unlikely -- convection is a highly systematically self-selection process for the most buoyant parcels - “special” parcels are inevitable, common, essential even if the weather doesn’t look “organized” (2D, etc.)
Considering convection-LS interaction
in param’z’n framework
Q1
local column impacts
Q1(T tend.)
Q2(q tend.)
Q3(u tend.)
z
z
z
SC DC ST
PBLmean
e
lower trop. q
upper trop. q
T in lower trop.
inversions
T upper
shearaerosolother e.g. “dyn”?
PBLsubgridvar.
(SST)
sensitivities of convection(? unknown ?)
Sensitivity problems (CCM3-CAM2,3)
deep convection closed on undilute parcel CAPEsensitivities of convection
(? unknown ?)
PBLmean
e
lower trop. q
upper trop. q
lower trop T
inversions
deep T
shearaerosolother e.g. “dyn”?
PBLsubgridvar.
(SST)
PBLmean
e
deep layer mean T
(SST)
sensitivities of CAPE
q aloft
Consequences rediscovered repeatedly over past several years in CCSM comm...
Sensitivity determines local feedbacks
sensitivities
Q1
impacts
Q1
Q2
Q3
z
z
z
SC DC ST
PBLmean
e
deep layer mean T q aloft
negative
negative
Rain & downdraft feedbacks: not all negative!
sensitivities
Q1
impacts
Q1
Q2
Q3
z
z
z
SC DC ST
PBL mean e
PBLsubgridvar.
lower trop. q
upper trop. q
lower trop. T
inversions
deep T
shearaerosol
other
negative
POSITIVE
(Updrafts don’t sample a (Updrafts don’t sample a homogeneous mean state homogeneous mean state
every 20 minutes with downdraft every 20 minutes with downdraft outflows blended into PBL)outflows blended into PBL)
Sensitivity update (CAM3.5)deep convection closed on dilute
parcel CAPE
PBLmean
e
deep layer mean T
sensitivities of dCAPE
q aloft
Better variability, incl. ENSO improvement (Neale et al. in prep.)
PBLmean
e
deep layer mean T
sensitivities of CAPE
q aloft
Q3Also impacts update - Richter
Sensitivity update (CAM3.5)deep convection closed on dilute
parcel CAPE
PBLmean
e
deep layer mean T
sensitivities of dCAPE
q aloft
• latent heat of freezing also added, to compensate loss of mean CAPE
• convection now clusters more in moister parts of space-time (increasing variability)
• still, negative feedbacks locally (deep conv. chills PBL, dries, heats)
What we did for summer vacationWhat we did for summer vacation Add “org” as local postive feedbacks Add “org” as local postive feedbacks
ordinary PBL T’ (Ok-ARM)
org (tied to recent (3h) rain
Convective closure entrains moister air: (capped at saturation)
params chosen so org ~1 for rainrate of P = 3mm/d
Didn’t do enough to make its impact clear: try a bigger hammer!Perturb initial parcel T:
Compensating for downdraft chilling...Compensating for downdraft chilling...
mean heating from ZM convection scheme in single column test
DASH: tests doubling alfa, a downdraft mass flux valve (to its documentation value, then 2x)
DOT: doubling Ke, a “stratiform” rain evaporation parameter, then 4x
mean heating from ZM convection scheme in single column test
DASH: tests doubling alfa, a downdraft mass flux valve (to its documentation value, then 2x)
DOT: doubling Ke, a “stratiform” rain evaporation parameter, then 4x several
K/d in month mean
Single column model testsSingle column model tests
When it rains, it pours... (increased variance)
When it rains, it pours... (increased variance)
diurnal delaydiurnal delay
Diurnal delay as expected (tau = 3h)
at least in SCM
Diurnal delay as expected (tau = 3h)
at least in SCM
What does it do? Mean stateWhat does it do? Mean state
Mean state can be more stable (warmer aloft) since convection is happening in org-enhanced areas
Mean state can be more stable (warmer aloft) since convection is happening in org-enhanced areas
control
diff
Mean stateMean state
Warmer tropics, higher Z300 in tropics,...
Warmer tropics, higher Z300 in tropics,...
Mean Mean
...westerly jet stream changes
...westerly jet stream changes
Drier tooDrier tooMean Mean
Drier, since deep convection is occurring in special org-enhanced places and buffered from entrainment
Drier, since deep convection is occurring in special org-enhanced places and buffered from entrainment
Mean Mean
Drier, and less cloudy -
except in stratus regions (due to enhanced stability?)
Drier, and less cloudy -
except in stratus regions (due to enhanced stability?)
Variability Variability
When it rains, it pours
When it rains, it pours
Variability Variability
Where it rains, it pours (& the converse)
Where it rains, it pours (& the converse)
Focus on Asian
monsoon
Focus on Asian
monsoonHotter in the desert(2m T)
Time variations
Time variations 10 days in
July
Where it rains it pours
(“noise?”)
10 days in July
Where it rains it pours
(“noise?”)
QuickTime™ and aCinepak decompressor
are needed to see this picture.
PDF viewpointPDF viewpointreference CAM with ORG
Rather extreme, but maybe Rather extreme, but maybe a step in a useful a step in a useful
direction?direction?
Improvements requiredImprovements required
org should be able to move– advect w/ low level wind? spread, upshear enhancement, etc.?
(more than we could do here, in parallel code...no neighbors!)
other sources besides precip– precip evaporation, really– subgrid geography?– deformation (gradient tightening)?
impacts on convection deciders should be calibrated– e.g. downdrafts don’t really heat inflow; they just don’t cool it
Ideally, seek consistency w/ PBL and cloud scheme subgrid assumptions - but dist. tails are key to convxn
Resolution dependent (to make whole system more resolution independent)?
org should be able to move– advect w/ low level wind? spread, upshear enhancement, etc.?
(more than we could do here, in parallel code...no neighbors!)
other sources besides precip– precip evaporation, really– subgrid geography?– deformation (gradient tightening)?
impacts on convection deciders should be calibrated– e.g. downdrafts don’t really heat inflow; they just don’t cool it
Ideally, seek consistency w/ PBL and cloud scheme subgrid assumptions - but dist. tails are key to convxn
Resolution dependent (to make whole system more resolution independent)?
Beyond CAM-lineage constraintsBeyond CAM-lineage constraints I wish it governed shallow-deep transition, not just
the strength of deep convection– Park-Bretherton unified PBL-convx suite?
We still need to get late-stage impacts right– top-heavy Q1 profile, i.e. impacts observed during
stratiform precipitation “meso” subroutine in convect.f ? water passed to stratiform.f ?
should we care ? really governed by bottom-heavy Q1 elsewhere?
– (rain in cu)
I wish it governed shallow-deep transition, not just the strength of deep convection– Park-Bretherton unified PBL-convx suite?
We still need to get late-stage impacts right– top-heavy Q1 profile, i.e. impacts observed during
stratiform precipitation “meso” subroutine in convect.f ? water passed to stratiform.f ?
should we care ? really governed by bottom-heavy Q1 elsewhere?
– (rain in cu)
Too heuristic? Too heuristic? Literalists will want org to be a quantity that can
be objectively measured (e.g. in CRMs), not just tuned for impact.
Internal inconsistencies with other subgrid schemes pinch over time – but subgrid dist. & overlap assumptions devised for
area (radiation) may not be good for the small but important (systematically self-selecting) buoyant “tail” parcels, especially a key subset (deciders) driving development...
Literalists will want org to be a quantity that can be objectively measured (e.g. in CRMs), not just tuned for impact.
Internal inconsistencies with other subgrid schemes pinch over time – but subgrid dist. & overlap assumptions devised for
area (radiation) may not be good for the small but important (systematically self-selecting) buoyant “tail” parcels, especially a key subset (deciders) driving development...
SummarySummary ORG variable used to enhance local positive
feedbacks (opposing some excessive local negative feedbacks)
Effect: when it rains, it pours– implication: rest of atm more stable
Bigger dynamic range, delayed diurnal development, more variability, rain persists past marginal stability (discharge-recharge), strong dev. sensitivity to moisture yet without mean-state unstable (cold) biases
some of these seem potentially desirable
ORG variable used to enhance local positive feedbacks (opposing some excessive local negative feedbacks)
Effect: when it rains, it pours– implication: rest of atm more stable
Bigger dynamic range, delayed diurnal development, more variability, rain persists past marginal stability (discharge-recharge), strong dev. sensitivity to moisture yet without mean-state unstable (cold) biases
some of these seem potentially desirable
Convection: a 2-scale circulation
time
Time mean
Deformation radius R
Gravity wave speed c
Cloud C
Convection in a low-res gridtime
Time mean
R
Grid ScaleC<G<R
G< G/2c
G/c
2G/c
3G/c
c c
C
Convection’s tendencies in model
process & scale categoriesAdiabatic transport
Phase changes
G-sized area averages
Advection.fAdvection.f ConvectionConvection.f.f
Deviations therefrom
ConvectioConvection.fn.f
(No net (No net effect on effect on resolved resolved scales)scales)
Example: evaporation and downdraftsExample: evaporation and downdrafts
mean heating from ZM convection scheme in single column test
DASH: tests doubling alfa, a downdraft mass flux valve (to its documentation value, then 2x)
DOT: doubling Ke, a “stratiform” rain evaporation parameter, then 4x
mean heating from ZM convection scheme in single column test
DASH: tests doubling alfa, a downdraft mass flux valve (to its documentation value, then 2x)
DOT: doubling Ke, a “stratiform” rain evaporation parameter, then 4x
Effects of minimum in ensemble
Arakawa & Schubert (1974): an ensemble of plumes with different entrainment rates i
• Low- and high- clouds compete for PBL moisture
• Competition decided on Aintegral of b over height of b>0 layer
• Lowest- cloud very deep & buoyant: dominates
• Separate shallow scheme may have to be used (or can tune by specifying critical work function for each cloud type)
Parameterization priorities, 5-10y
Plain old Q1 and Q2 are still big Q’s
e.g. model MJO Q1 profilesMJO heating (anomaly) profile
Lin et al. 2004
Point scale (5’ vertically pointing cloud radar vs. gauge
rain, EPIC 2001):
cudyn.
(multi-cellular)
anvil dyn. µ-physics…
…includingstratiform
rain
Synoptic scale (6h 1000-km humidity vs.
budget rain, COARE)
days
Problems with weak entrainment in closure
– Deep convection dries the lower troposphere, but feels no feedback (brake): dry bias in mean state
– Moisture storage/xport/discharge mechanisms weak:
too little variability: P too much like E– All sens. in PBL: premature diurnal rainfall over land– Over warm SST, but dry aloft, scheme too active
(“double ITCZ” in SE Pac ?)
Entrainment and large-scale convective weather variability
(aqua-planet, rad.-conv. equilibrium,
prescribed “radiation” (cooling), uniform warm SST)
Low entrainment rate .125/HPBL
Virginie Lorant Ph.D. 2001
High entrainment rate .185/HPBL
Longitude (repeated)
Tim
e
Longitude (repeated)
Rain Rain
Entrainment: a Good Thing?• Entrainment restrains too-easy deep
convectionBUT…• makes top height of convection too low
– (in realistic stratification)
• ---> unstable sounding bias – (in climate)
» e.g. cold tropical upper troposphere
• Unwanted trade-off: variability Unwanted trade-off: variability (encouraged by entrainment constraint) (encouraged by entrainment constraint) vs. mean state (better with less dilution)vs. mean state (better with less dilution)
A solution: entrainment, but possibly of non-average air
• All plumes entrain strongly. First clouds entrain clear air, so are very sensitive to q, but are typically shallow.
• Later convection may entrain air pre-moistened by prior clouds, becoming deeper.
• This gives deep clouds an indirect q dependence, and opens up questions of cloud-field organization.
1
3
2
Discretize mixing by generationsrather than by entrainment rate
1 1 12
A minimal model: 4 levels
Trade layer
Middle trop.
Upper trop.
humidity density
Two (or 3) vertical modes indynamics
Trimodal clouddistribution: cu, cg, cb(good # for microphysics/precip. knobs)
PBL
4 subcolumn typeseach w/area fraction & prior rain rate
env. RH
plume model
1
anorm
1
cloudbase area
inhibition
energy
New values ofL(c-e), div[w’X’],rain & downdrafts,for each of the 4prior subcolumn types (which mayproduce any of
{cu, cg, cb} at thepresent time step)
4-level toy model summary
PBL
LT
MT
UT
b1 b2 b4b3
PBL
LT
MT
UT
w1 w2 w4w3
PBL
LT
MT
UT
PBL
LT
MT
UT
pure env cu cg cb 1 2 3 4
cb4
cg3
cu2
RAINPROD
1 1
RH=100%
random trigger E: random trigger E: TKETKE
OrgE: amt OrgE: amt T;T;by type by type rainrain
cu
cu
cg
cb
4-layer atm on warm beta-plane channel
Sensitivities: lifting
subsidence
control
T’ effect only (~50%)
unimodal LSD param. flawedunimodal LSD param. flawed
low-level lift, subsidence above
low-level subsidence, lift above
1. Lifting good for rain (duh)
2. Linear for ~1K stimuli
3. T’ ~half of total (q’ rest)
4. Low levels dominate5. Strong initial
sensitivity, but vertical structure changes stretch LSD param. too far
6. Shear of 10 m/s has no significant effect on rain
Moisture sensitivity
(courtesy Tetsuya Takemi)
5 6 8
568
568
Need to map & calibrate sensitivities
Example: Convecting CRM’s Q1 response to observed T’, q’
T
q
lower half of T’
upper half of T’
lower q’
upper q’
perturb
COARE sounding regressionsjust beforerain
Response not totally deterministic, even with
128 x 128 km domain.Stochastic aspect to convection must
be recognized.
Top heaviness of deep conv. heating
Lack of stratiform processes per se, or of cumulus showers?
“representing organization” - start with, say,
prognostic plume entrainment?GCM
Deep convectionheating in GCM
Lee Kang Mapes 2001
20N-20S cooling
Deep convectionheating
obs
Earth
Mapes 2000
Example of successive entraining
plume buoyancies (equatorial
EPAC)
qsat(T,p)q(p)
very dry (no deep convection)
undilute
12 3
4