aerosol mixing

18
Evaluating the Role of Aerosol Mixing State in Cloud Droplet Nucleation using a new activation parameterization Daniel Rothenberg and Chien Wang Massachusetts Institute of Technology Department of Earth, Atmospheric, and Planetary Sciences Program in Atmospheres, Oceans, and Climate December 11, 2013 EAPS Department of Earth, Atmospheric, and Planetary Sciences Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 1 / 12

Upload: steveconti

Post on 22-Dec-2015

26 views

Category:

Documents


1 download

DESCRIPTION

Aersols make the air smell nice

TRANSCRIPT

Evaluating the Role of Aerosol Mixing Statein Cloud Droplet Nucleation

using a new activation parameterization

Daniel Rothenberg and Chien Wang

Massachusetts Institute of TechnologyDepartment of Earth, Atmospheric, and Planetary Sciences

Program in Atmospheres, Oceans, and Climate

December 11, 2013

EAPSDepartment ofEarth, Atmospheric,and Planetary Sciences

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 1 / 12

The Mixing State Problem

How do you represent mixtures of aerosols in GCMs/CRMs?

Uniform, homogeneous compositionparticles

Different populations of chemicallyhomogeneous particles

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 2 / 12

The Mixing State Problem

Real aerosol populations arechemically and physicallyheterogeneous

Different particles with varyingoptical and microphysicalproperties

Important to capture thisdiversity in order to resolveanthropogenic aerosoleffects on clouds/climate!

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 3 / 12

The Mixing State Problem

Real aerosol populations arechemically and physicallyheterogeneous

Different particles with varyingoptical and microphysicalproperties

Important to capture thisdiversity in order to resolveanthropogenic aerosoleffects on clouds/climate!

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 3 / 12

Example Effect: Droplet Nucelation

Put aerosols in an updraft...adiabatic cooling

↓supersaturated environment

↓condensational growth of

aerosol/droplets↓

bifurcate aerosol into cloud droplets(”activation”) and haze

Critical factor - Smax (function oftemperature, updraft speed, aerosolproperties)

How does the aerosol mixing statecontribute to potential dropletactivation?

Droplet nucleation simulated with detailed parcel model

cloud droplets

haze

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 4 / 12

Error in Droplet Nucleation due toInternal Mixing Assumption

10−4 10−3 10−2 10−1 100

rp (µm)

0

5000

10000

15000

20000

25000

30000

35000

40000

45000

nN

(rp)

(cm

−3)

Original - internal mixture

10−4 10−3 10−2 10−1 100

rp (µm)

Decomposed

mixed - internalsulfate - externalcarbon - external

α

γ (�)

Internal mixture ofcarbon/sulfate →decompose into spectrumof mixtures preservingnumber and mass of eachspecies

internal 0.2 0.4 0.6 0.8 external

alpha

sulfate

0.2

0.4

0.6

0.8

carbon

gam

ma

Nucleated Droplet Count Error(internal - external)

-250

-200

-150

-100

-50

0

50

100

150

200

250 Error in predicted dropletnumber from 1 m/s updraft,explicitly computed withdetailed parcel model

+100% error when mostlycarbon - important fordownwind of intense biomassburning/industrial emissions?

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 5 / 12

MARC = a Multimode, 2-Moment, and Mixing-state-resolving !Model of Aerosols for Research of Climate!

Log-normal distribution, 2 prognostic moments (Q, N) + BIM & OIM, prescribed σ !

(Kim et al., JGR, 2008)

gaseous oxidation

aqueous oxidation

Condensation

Evap-resuspense

Aging (surface preparation)

Nucleation

Growth Coagulation)

Radiation

Clouds

Meteorology

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 6 / 12

Droplet Nucleation / Activation Parameterization

Smax - the “Activation Equation”

From parcel theory, can derive (Ghan et al, 2011)

αV

γ=

4πρwρa

GSmax

Smax∫0

r2(tact) + 2G

tmax∫tact

Sdt

1/2

dN

dScdSc

Need assumptions,1 aerosol modes have bulk properties (e.g. hygroscopicity)2 instantaneous particle growth in equilibrium with relative humidity3 activation instantly happens when particle sees critical S (Kohler Theory)

Basic equation underlying parameterizations used in GCMs/CRMs to predictdroplet nucleation

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 7 / 12

Droplet Nucleation Errors in CESM+MARC

180°W 120°W 60°W 0° 60°E 120°E 180°E

60°S

30°S

30°N

60°N

Error, predicted droplet nucleation (ARG - explicit)1/cm3, Avg: -36.8 (-597.9 − -0.3)

240

160

80

0

80

160

240

180°W 120°W 60°W 0° 60°E 120°E 180°E

60°S

30°S

30°N

60°N

Error, predicted droplet nucleation (FN - explicit)1/cm3, Avg: -5.2 (-360.7 − 26.1)

240

160

80

0

80

160

240

Severe underprediction inareas with high carbonaceousaerosol loading

Exactly in regions mostimportant for anthropogenicaerosol effects

Need to better parameterizemixing state / competitioneffects on droplet nucleation

Parameterizations:ARG - Abdul-Razzak and Ghan, 2000FN - Fountoukis and Nenes, 2005explicit - numerical parcel model

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 8 / 12

Polynomial Chaos Expansion of Parcel Model

Polynomial emulator of full-complexity model

Computationally-cheap (produce/run), accurate distribution of modeled response

Detailed Parcel Model

Updraft speed, temperature, pressure,

aerosol properties

Smax

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 9 / 12

Polynomial Chaos Expansion of Parcel Model

Polynomial emulator of full-complexity model

Computationally-cheap (produce/run), accurate distribution of modeled response

Detailed Parcel Model

Updraft speed, temperature, pressure,

aerosol properties

Smax

Polynomial Chaos Expansion

Produce sets of model runs based on PDFs of input parameters

Save response function(s)

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 9 / 12

Polynomial Chaos Expansion of Parcel Model

Polynomial emulator of full-complexity model

Computationally-cheap (produce/run), accurate distribution of modeled response

Detailed Parcel Model

Updraft speed, temperature, pressure,

aerosol properties

Smax

Polynomial Chaos Expansion

Produce sets of model runs based on PDFs of input parameters

Save response function(s)

Parcel Model Emulator

Numerical quadrature to compute coefficients of orthogonal basis functions

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 9 / 12

Emulation Results - Single Mode Aerosol

101 102 103 104

Droplet Concentration, cm−3 (Detailed Parcel Model)

101

102

103

104

Dro

ple

t C

on

cen

trati

on

, cm

−3 (

Poly

nom

ial

Ch

aos)

SM1

SM2

SM3

SM4

SM5

10-3 10-2

Supersaturation (Detailed Parcel Model)

10-3

10-2

Su

pers

atu

rati

on

(P

oly

nom

ial

Ch

aos) SM1

SM2

SM3

SM4

SM5

10-2 10-1 100 101

Updraft Speed (m/s)

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Su

pers

atu

rati

on

(%

)

ARG

FN

parcel

pce

TOP: Emulator well-calibrated exceptfor large number concentrations ofsmall particles (SM5)

LEFT: Reproduces non-linear responsein Smax due to important variables

Next step - extend to multiple aerosolmodes

Aerosols (SMi) from Nenes and Seinfeld, 2003

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 10 / 12

Emulation Results - Single Mode Aerosol

101 102 103 104

Droplet Concentration, cm−3 (Detailed Parcel Model)

101

102

103

104

Dro

ple

t C

on

cen

trati

on

, cm

−3 (

Poly

nom

ial

Ch

aos)

SM1

SM2

SM3

SM4

SM5

10-3 10-2

Supersaturation (Detailed Parcel Model)

10-3

10-2

Su

pers

atu

rati

on

(P

oly

nom

ial

Ch

aos) SM1

SM2

SM3

SM4

SM5

10-2 10-1 100 101

Updraft Speed (m/s)

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Su

pers

atu

rati

on

(%

)

ARG

FN

parcel

pce

TOP: Emulator well-calibrated exceptfor large number concentrations ofsmall particles (SM5)

LEFT: Reproduces non-linear responsein Smax due to important variables

Next step - extend to multiple aerosolmodes

Aerosols (SMi) from Nenes and Seinfeld, 2003

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 10 / 12

Conclusions/Summary

Resolving some degree of the “mixing state” of heterogeneous aerosol calculationswill change the potential for droplet nucleation and calculated cloud droplet burden.

Existing parameterizations for global models may not be well-calibrated for thecomplexity and diversity of aerosol populations predicted from mixing-state resolvingmodels.

Biases in physics calculations due to complex mixing state must be addressed toaccurately simulate anthropogneic aerosol effects on clouds and climate.

Polynomial Chaos and other advanced statistical techniques could help produceefficient activation parameterizations specifically tuned for applications in GCMs andCRMs.

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 11 / 12

Conclusions/Summary

Resolving some degree of the “mixing state” of heterogeneous aerosol calculationswill change the potential for droplet nucleation and calculated cloud droplet burden.

Existing parameterizations for global models may not be well-calibrated for thecomplexity and diversity of aerosol populations predicted from mixing-state resolvingmodels.

Biases in physics calculations due to complex mixing state must be addressed toaccurately simulate anthropogneic aerosol effects on clouds and climate.

Polynomial Chaos and other advanced statistical techniques could help produceefficient activation parameterizations specifically tuned for applications in GCMs andCRMs.

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 11 / 12

Conclusions/Summary

Resolving some degree of the “mixing state” of heterogeneous aerosol calculationswill change the potential for droplet nucleation and calculated cloud droplet burden.

Existing parameterizations for global models may not be well-calibrated for thecomplexity and diversity of aerosol populations predicted from mixing-state resolvingmodels.

Biases in physics calculations due to complex mixing state must be addressed toaccurately simulate anthropogneic aerosol effects on clouds and climate.

Polynomial Chaos and other advanced statistical techniques could help produceefficient activation parameterizations specifically tuned for applications in GCMs andCRMs.

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 11 / 12

Acknowledgments

This material is based upon work supported by the National Science FoundationGraduate Research fellowship under NSF Grant No. 1122374

We would also like to thank Steve Ghan (PNNL) for providing a reference parcel modeland code for his activation scheme; Rotem Bar-Or and Alex Avramov (MIT) for helpingrun the CESM+MARC and for helpful discussion; Dan Czizco, Ron Prinn, and PaulO’Gorman (MIT) for feedback and comments while preparing portions of this work forthe MIT PAOC General Examination.

Coupled CESM+MARC runs performed using the NCAR Yellowstone supercomputer.

Rothenberg/Wang (MIT) AGU Fall 2013 December 11, 2013 12 / 12