Air Quality Modeling and Simulation A Few Issues for HPCN Bruno Sportisse CEREA, Joint Laboratory Ecole des Ponts/EDF R&D INRIA/ENPC CLIME project TeraTech, 20 June 2007 B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 1 / 28
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 2 / 28
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 3 / 28
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 4 / 28
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)
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 5 / 28
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 6 / 28
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.
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 7 / 28
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 8 / 28
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 9 / 28
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 10 / 28
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)
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 11 / 28
Numerics for CTM
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 12 / 28
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 ?
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 13 / 28
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)
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 14 / 28
Aerosol Modeling and Simulation
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 15 / 28
Aerosol Modeling and Simulation
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 16 / 28
Aerosol Modeling and Simulation
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 17 / 28
Towards Integrated Modeling
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 18 / 28
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)
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 19 / 28
In Models We Trust
In Models We Trust
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 20 / 28
Uncertainty Propagation & Ensemble Forecast
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 21 / 28
Uncertainty Propagation & Ensemble Forecast
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 22 / 28
Uncertainty Propagation & Ensemble Forecast
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)
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 23 / 28
Data Assimilation & Inverse Modeling
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 24 / 28
Data Assimilation & Inverse Modeling
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 25 / 28
Data Assimilation & Inverse Modeling
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 26 / 28
Data Assimilation & Inverse Modeling
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
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 27 / 28
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)
B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 28 / 28
IntroductionParameterizationsNumerics for CTMAerosol Modeling and SimulationTowards Integrated ModelingIn Models We TrustUncertainty Propagation & Ensemble ForecastData Assimilation & Inverse ModelingConclusion