Air Quality Modeling and Simulation - univ-reims.fr · POLYPHEMUS run, Forecast Emergency Center IRSN/CEREA Chernobyl 1986-04-26T13:03:00 1E-01 1E+00 1E+01 1E+02 D. QuØlo, M. Krysta,

Air Quality Modeling and SimulationA Few Issues for HPCN

Bruno Sportisse

CEREA, Joint Laboratory Ecole des Ponts/EDF R&DINRIA/ENPC CLIME project

TeraTech, 20 June 2007

B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 1 / 28

Introduction

Motivations

Atmospheric Chemical Composition• The atmosphere as a chemical reactor• Trace species: from ng m−3 to µg m−3

Applications• Risk assessment (NBC)• Photochemistry (ozone, nitrogen oxides, volatile organic compounds)• Transboundary pollution (heavy metals, acid rains)• Oxidizing power of the atmosphere and lifetime• Greenhouse gases and radiative effects• Stratospheric ozone (halogen compounds)• . . .

Model Uses• Process studies• Forecast (e.g. accidental release)• Impact studies


Introduction

Forecast and Risk Assessment

Chernobyl Accidental Release, 25 April-5 May 1986POLYPHEMUS run, Forecast Emergency Center IRSN/CEREA

Chernobyl

1986-04-26T13:03:00

1E-01 1E+00 1E+01 1E+02

D. Quélo, M. Krysta, M. Bocquet, O. Isnard, Y. Minier, and B. Sportisse. Validation of the POLYPHEMUS system: the ETEX,Chernobyl and Algeciras cases. Atmos. Env., 2007


Introduction

Impact Studies

POLYPHEMUS Run for the Impact of French Emission of Power Plantsfor the Year 2001 (NEC/CAFE Round)

Credit: Yelva Roustan (CEREA)

-10 -5 0 5 10 15 20

lon

35

40

45

50

55

lat

O3 mean

-0.40

-0.20

-0.10

-0.01

0.01

0.10

0.20

0.40

NOx

VOC

neutral

NOx limited

COV limited

favorable

Case A

Case B

unfavorable

favorable

increasing ozone


Introduction

Expertise for Disbenefit Effects and Dilemma

Three Case Studies• NOx disbenefit• Reduction of emitted mass versus increase of number for

secondary particles• Climate change versus air pollution (e.g.: impact of E85 flexfuel or

sulfate aerosols)


Introduction

Processes

VOC

chemistryaqueous

dust

gaseous chemistry

advection

condensation

evaporation

nucleation

coagulation

dry deposition

washout

VOC

NOx

H2SO4

turbulentdiffusion

particlesprimary

VOCNOx

VOC

SO2 NOx

Ozone

SVOC

HNO3

NH3

rainout

activation

TRAFFIC AGRICULTURE INDUSTRY BIOGENIC


Introduction

The Arms Race

• Air Quality Models (Chemistry-Transport Models) rely on subgridparameterizations.

• The resulting equations generate high-dimensional numericalissues.

• Both issues (modeling & numerics) are much more challenging foraerosol dynamics, based on more and more detailed models.

• Yet, even after having tackled these problems, models have to becarefully used, because of uncertainties. Ensemble modeling is onepossible answer.

• Coupling together observational data and numerical models iscarried out with data assimilation methods. Advanced issues arerelated to network design.

• For impact assessment, integrated modeling relies on look-uptables, to be computed with detailed models.


Parameterizations

Outline

1 Parameterizations

2 Numerics for CTM

3 Aerosol Modeling and Simulation

4 Towards Integrated Modeling

5 Uncertainty Propagation & Ensemble Forecast

6 Data Assimilation & Inverse Modeling


Parameterizations

Scales

Microphysics• Aerosols: 1 nm - 10 µm• Cloud droplets: 1 - 100 µm• Rain droplets: 0.01 - 0.1 mm

Numerics (Grid Cell)• Short-range (CFD): 1 - 10 m• Regional: 1 km• Continental: 10-100 km


Parameterizations

Scavenging of Radionuclides

Gas-Phase or Particle-BoundRadionuclides

• Detailed microphysicsversus tailoredparameterizations

• Uncertainties: rainparameters and sizedistribution -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.001 p )m�ni(dgol

-4.0

-3.5

-3.0

-2.5

-2.0

-1.5

-1.0

-0.5

0.0

0.5

01r

p)d,

D(E

gol

r mm1.0=D

r mm5.0=D

r mm1=D

r mm2=D

Size distribution of the aerosol collision efficiency

Towards Micro/Macro Models• Based on stochastic micro models


Parameterizations

Segregation Effect

Downdraft O3/Updraft NO• Rate of the titration reaction

NO+O3 →NO2:

ω = k 〈NO〉〈O3〉

1 +

〈

NO′

O′

3

〉

〈NO〉〈O3〉︸︷︷︸

Is

Closure Scheme• State-of-the-art in 3D models:

Is = 0 !• Towards Large Eddy Simulation ?

Reaction rate (DNS computation; credit: J.F. Vinuesa, JRC)


Numerics for CTM

Outline

1 Parameterizations

2 Numerics for CTM






Numerics for CTM

Time Integration of High-Dimensional Stiff Systems

Model Dimension (State Vector per Grid Cell)• Passive tracer: 1 tracer• Gas-phase: 50-100 surrogate species• Diphasic: 10-50 dissolved species• Aerosols: 20 species × 10 bins (size) × 1 family (internal mixing)

Wide Range of Timescales (Stiffness)

• From radical (τ = 10−10s) to inert species

Towards Highly Resolved Model• Next-generation mesoscale model: 1-3 km

grid• Unstructured meshes near sources ?


Numerics for CTM

Towards on-line Coupling

Many Motivations• Physics: conservation of

homogeneous mixing ratio for apassive tracer (mass consistencyerror)

• Numerics: discrepancies in thewind fields for ρ and c

• Convective episodes

-10 0 10 20 30 40 50 6035

40

45

50

55

60

65

70

-50

-30

-10

-5

5

10

30

50

Relative difference for the Chernobyl release (fitted w)


Aerosol Modeling and Simulation

Outline

1 Parameterizations

2 Numerics for CTM







Aerosol Dynamics

aerosols aerosols aerosolsaerosolsgas molecules cloud droplets

primary gas emission

coalescence

homogeneous reactions

heterogeneousreactions

0.01 < D < 0.1D < 0.01 0.1 < D < 1 1 < D < 10

primary particle emission

dry deposition

scavengingwet

evaporation

hygroscopic activationcoagulationcoagulationnucleation

condensation

activation and scavenging



Towards External Mixing of Fractal Particles

Internal mixing with a carbon core

Internal mixing with a carbon surface

External mixing

Mixing state and radiative properties

Spherical case

Soot

Geometrical configuration of soot


Towards Integrated Modeling

Outline

1 Parameterizations

2 Numerics for CTM






Towards Integrated Modeling

Model Reduction

Arms Race versus Robustness• Impact studies over many (meteorological) years (Long Range

Transport Air Pollution/Clean Air For Europe)• “Integrated” modeling:

mine

Fimpact ◦ FCTM ◦ Feconomic activity(e)

where e stands for emissions• 4D distributed systems with a few observations versus

low-dimensional models

Many strategies• Source-Receptor matrices (2500 × 5 × 5 × 5 runs of one

meteorological year)• Look-up tables (HDMR, chaos expansion)


In Models We Trust

In Models We Trust


Uncertainty Propagation & Ensemble Forecast

Outline

1 Parameterizations

2 Numerics for CTM







Uncertainties in CTM

Major Uncertainties• Input data: emissions, met. data• Parameterizations and physics• Numerics• Bugs

Data UncertaintiesCloud attenuation ±30%Dry deposition (O3 and NO2) ±30%Boundary Conditions (O3) ±20%Anthropogenic emissions ±50%Biogenic emissions ±100%Photolytic rate ±30%

In Models We Trust: the Overtuning Issue• Too few observational data (chemical, vertical, time)• Key target: ozone peak (impact study versus forecast)

Some strategies• Sensitivity analysis• Monte Carlo simulations on the basis of Probability Density Functions (PDF)• Ensemble meteorological forecasts• Multi-configuration/multi-model runs



Ensemble Forecast

Ensemble (Set) of ModelsE = {Mm(·)}m

• Ensemble mean:

EM(·) =1|E|

∑

M∈E

Mm(·)

• Super-ensemble:

ELS(·) =∑

mαmMm(·)

with weights αm to forecaston the basis of pastobservations

Ensemble & relative uncertainties (POLYPHEMUS run for ozone)


Data Assimilation & Inverse Modeling

Outline

1 Parameterizations

2 Numerics for CTM







Background

Monitoring Networks• Terrestrial sensors• Satellite data

Key Features• Variational and sequential methods• Inverse modeling of emissions• High-dimensional systems• Second-order sensitivity

Data assimilation for ozone (Credit: Lin Wu/CLIME/CEREA)

0 10 20 30 40 50

Observation numbers

20

40

60

80

100

120

140

Con

cen

trati

on

Montandon

referenceobservationoptimal interpolationenkf4dvarrrsqrt

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23time (hours)

0

1

2

3

4

5

6

7

8

9

11 May12 May13 May14 May15 May11-15 May (Mean)11-15 Mai (Simulated)

Inverse modeling of NOx emissions



Forecast and Risk Assessment

Source Localization and Operational Forecast of a Release• Pre-operational case• Maximum Entropy technique• POLYPHEMUS run, credit: Marc Bocquet (CEREA)

Seemovie



Network Design

IRSN Descartes monitoring network design over France• to monitor potential radionuclides releases accidents (Credit:

Marc Bocquet, CEREA)• '20000 simulated releases

-10 -5 0 5 10 15

40

42

44

46

48

50

52

54

alpha=2

-10 -5 0 5 10 15

40

42

44

46

48

50

52

54

alpha=1

-10 -5 0 5 10 15

40

42

44

46

48

50

52

54

alpha=0.5

-10 -5 0 5 10 15

40

42

44

46

48

50

52

54

Reseau geometrique


Conclusion

Many Challenging Issues for HPCN

An Increasing Spatial Resolution• Towards 1-kilometer grid• Parameterization, adaptive unstructured meshes ?

An Increasing Chemical Resolution• From surrogate species to chemical species• Secondary Organic Aerosol, external mixing, . . .

An Increasing Complexity: Coupling Models and Scales• Towards multi-media integrated modeling• From off-line coupling to on-line coupling

From Deterministic to Probabilistic Models• CTM are not deterministic models.• From all-in-one models to a new generation of modeling systems (ensemble

modeling)


IntroductionParameterizationsNumerics for CTMAerosol Modeling and SimulationTowards Integrated ModelingIn Models We TrustUncertainty Propagation & Ensemble ForecastData Assimilation & Inverse ModelingConclusion

Air Quality Modeling and Simulation - univ-reims.fr · POLYPHEMUS run, Forecast Emergency Center IRSN/CEREA Chernobyl 1986-04-26T13:03:00 1E-01 1E+00 1E+01 1E+02 D. QuØlo, M. Krysta,

Documents