Modeling Clouds and Climate: A computational challenge

Modeling Clouds and Climate:

A computational challenge

Stephan de Roode

Clouds, Climate & Air Quality

Multi-Scale Physics (MSP), Faculty of Applied Sciences

with contributions from Harm Jonker (MSP) and Pier Siebesma

(KNMI,MSP)

Length scales in the atmosphere

Landsat 60 km 65km

Large Eddy Simulation 10km

~mm ~100m~1m-100m

Earth 107 m

Courtesy Harm Jonker

Cloud dynamics

10 m 100 m 1 km 10 km 100 km 1000 km 10000 km

turbulence Cumulus

clouds

Cumulonimbus

clouds

Mesoscale

Convective systems

Extratropical

Cyclones

Planetary

waves

Large Eddy Simulation (LES) Model

Cloud System Resolving Model (CSRM)

Numerical Weather Prediction (NWP) Model

Global Climate Model

The Zoo of Atmospheric Models

DNS

mm

Cloud microphysics

Rain and Radiation

~mm

~1m-100m

aircraft observations during ASTEX, Duynkerke et al., 1999

drizzle drops

Observed cloud droplet spectrum

cloud water

1 minute course on cloud thermodynamics

Adiabatic plume (does not mix with its environment)

Conservation of energy

€

cpTenthalpy{

+ gzgravitational

potentialenergy

{ = sdry staticenergy

{ = cst

1 minute course on cloud thermodynamics

Adiabatic plume (does not mix with its environment)

Conservation of energy

€

cpTenthalpy{

+ gzgravitational

potentialenergy

{ = sdry staticenergy

{ = cst

s

he

igh

t

temperature T

€

∂T∂z

= -gcp

≈ -10 K km-1

Rising plume

1 minute course on cloud thermodynamics

Adiabatic plume (does not mix with its environment)

Conservation of energy

Conservation of water

€

cpTenthalpy{

+ gzgravitational

potentialenergy

{ = sdry staticenergy

{ = cst

€

qvap

watervapor

{ = qtot

total watercontent

{ = cst

s

he

igh

t

qtot temperature T

€

∂T∂z

= -gcp

≈ -10 K km-1

qsaturation

Rising plume

1 minute course on cloud thermodynamics

Adiabatic clouds (clouds that do not mix with their

environment)

Conservation of energy

Conservation of water

€

cpTenthalpy{

+ gzgravitational

potentialenergy

{ − Lvqliq

condensation/evaporation

1 2 3 = sliq

liquidstatic energy

{ = cst

€

qvap

watervapor

{ + qliq

liquidwater

{ = qtot

total watercontent

{ = cst

sliq

he

igh

t

qtot

qsaturation

qliq temperature T€

∂T∂z

= -gcp

≈ -10 K km-1

€

∂T∂z

= -gcp

+Lv

cp

∂qliq

∂z≈ -5 K km-1

Cloud droplet size (condensational growth only)

€

∂qliq

∂z=α

€

qliq = ρ liqNdroplet43πRdroplet

3

qliq €

Rdroplet =3αz

4πρ liqNdroplet

3

Condensation too small droplet sizes for rain (Rrain > 100 m)

Rain forms by droplet collisions gravity and in-cloud turbulence

Collision efficiency laboratory experiments and by Direct Numerical Simulation

0 10 20 30 400

1000

2000

3000

4000

Cloud droplet radius ( )m

Ndroplet

= 400 cm-3

pollutant continental air

Ndroplet

= 40 cm-3

clean marine air

More rain in the weekend?

0 10 20 30 400

1000

2000

3000

4000

Cloud droplet radius ( )m

Ndroplet

= 400 cm-3

pollutant continental air

Ndroplet

= 40 cm-3

clean marine air

Mon-Friday Sat-Sunday

More rain in the weekend?

0 10 20 30 400

1000

2000

3000

4000

Cloud droplet radius ( )m

Ndroplet

= 400 cm-3

pollutant continental air

Ndroplet

= 40 cm-3

clean marine air

Mon-Friday

Sat-Sunday?

Fewer but larger droplets lead to more a more efficient formation of rain. Some investigations suggests a weak correlation between day of the week and precipitation, other ones do not.

"weekdays"

"weekend"

Sat-Sunday

Droplet concentration and Radiation:

"Indirect" aerosol effect

Cloud albedo (reflectivity) depends oncross sectional area A of cloud dropletshaving a concentration N

€

Apolluted

Aclean

=N polluted

Nclean

⎛

⎝ ⎜

⎞

⎠ ⎟

1 / 3

> 1

Feedback effects in a changing climate

Dufresne & Bony, Journal of Climate 2008

Radiative effects only

Water vapor feedback

Surface albedo feedback

Cloud feedback

Ensemble forecast with the ECMWF model:

50 simulations with perturbed initial conditions

http://www.knmi.nl/exp/pluim/vijftiendaagse/index.html

Edward Lorenz(1917-2008)

Assess uncertainty in global temperature change due to

uncertainties in parameterization coefficients/switches

Murphy et al. 2004, Nature

Uncertainty in cloud lateral mixing is identified as a major

contributor to the large spread in the PDF

Murphy et al. 2004, Nature

current PhD project:LES of deep convection

(Steef Boing)

Siebesma & Holtslag ‘96

The playground for cloud physicists: Hadley circulation

deep convection shallow cumulus stratocumulus

Atlantic Stratocumulus to cumulus Transition EXperiment

(ASTEX)

LES, 1995 LES, 1999

64x64x60 grid pointssimulation time: 3 hoursruns were done on a CRAY supercomputer

2010: run full Lagrangian transition (40 hours) on 256x256x128 grid points

De Roode and Duynkerke, 1997

EU Cloud Intercomparison,

Process Study and

Evaluation Project

(EUCLIPSE)

Future

Sea water temperature: T+T

enhanced surface evaporation

Present

Sea water temperature: T

Positive Feedback?

Entrainment drying dominates moisture

tendency

Negative Feedback?

Entrainment in a water tank (Harm Jonker's laboratory)Convection driven by a salinity flux at the surfaceFinding: considerable less top entrainment than in LES models

atmosphere tank (heat) tank (salt)

Reynolds number Re=108 Re=103 Re=103

Prandtl number Pr=1 Pr=10 Pr=1000

computationally expensive

Why different entrainment rates?

izw*Re

Pr

Site Architecture Max nr cores used Grid

SARA IBM Power 6 1024 1024 x 1024 x 768

CINECA IBM BCX/5120 2048 2048 x 2048 x 1024

LRZ SGI Altix 4700 3072 1536 x 1536 x 768

Juelich Bluegene 32,768 3072 x 3072 x 1536

DEISA: Distributed European Infrastructure for Supercomputing Applications

resource allocation: 1.9M cpu-hr

€

N x = N y = 2048, Nz = 1024, N procs = 2048

Re = 30,000 Pr = 1

(potential) Temperature animation

Animation of the temperature (Harm Jonker)

Prandtl-number:

Re number must be really large before fluid-properties can be neglected

The importance of large computations (Harm Jonker)

Top

ent

rain

men

t ef

ficie

ncy

A

range LES and observationsatmosphere

Outlook

Large Eddy Simulation of clouds

+ Large domains and fine grid resolution

+ Long simulations (diurnal cycle, equilibrium solutions)

+ Exploration of parameter space and its effect on cloud transitions

(surface temperature, inversion strength, subsidence etc.)

+ Rate of turbulent mixing across cloud interfaces

(entrainment/detrainment in shallow and deep convection)

Postprocessing

- giant data sets are produced