Air Quality Modeling and Simulation

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

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 1986 POLYPHEMUS run, Forecast Emergency Center IRSN/CEREA 1986-04-26T13:03:00 Chernobyl 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 Plants for the Year 2001 (NEC/CAFE Round) Credit: Yelva Roustan (CEREA) NOx O3 mean 0.40 favorable Case A 55 0.20 unfavorable COV limited 0.10 increasing ozone 50 0.01 lat 45-0.01 NOx limited 40-0.10 neutral Case B -0.20 favorable 35-10 -5 0 5 10 15 20 lon -0.40 VOC B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 4 / 28

Introduction Expertise for Disbenefit Effects and Dilemma Three Case Studies NO x 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

Processes Introduction advection gaseous chemistry VOC Ozone NOx SVOC nucleation condensation activation aqueous chemistry H2SO4 evaporation rainout HNO3 coagulation turbulent diffusion VOC NOx primary particles NH3 SO2 NOx VOC dust washout VOC dry deposition TRAFFIC AGRICULTURE INDUSTRY BIOGENIC B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 6 / 28

The Arms Race Introduction Air Quality Models (Chemistry-Transport Models) rely on subgrid parameterizations. The resulting equations generate high-dimensional numerical issues. Both issues (modeling & numerics) are much more challenging for aerosol dynamics, based on more and more detailed models. Yet, even after having tackled these problems, models have to be carefully used, because of uncertainties. Ensemble modeling is one possible answer. Coupling together observational data and numerical models is carried out with data assimilation methods. Advanced issues are related to network design. For impact assessment, integrated modeling relies on look-up tables, to be computed with detailed models. B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 7 / 28

Outline Parameterizations 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-Bound Radionuclides Detailed microphysics versus tailored parameterizations Uncertainties: rain parameters and size distribution log 1 0 E(D,d r p ) 0.5 0.0-0.5-1.0-1.5-2.0-2.5-3.0-3.5 D r=0.1 D r=0.5 D r=1 D r=2-4.0-3.0-2.5-2.0-1.5-1.0-0.5 0.0 0.5 1.0 dp (in log1 m) 0 mm mm mm mm 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

Segregation Effect Parameterizations Downdraft O 3 /Updraft NO Rate of the titration reaction NO+O 3 NO 2 : NO O ω = k NO O 3 1 + 3 NO O 3 } {{ } I s Closure Scheme State-of-the-art in 3D models: I s = 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

Outline Numerics for CTM 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 10 s) 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 a passive tracer (mass consistency error) Numerics: discrepancies in the wind fields for ρ and c Convective episodes 70 65 60 55 50 45 40 35-10 0 10 20 30 40 50 60 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

Outline Aerosol Modeling and Simulation 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 Dynamics Aerosol Modeling and Simulation primary gas emission primary particle emission heterogeneous reactions condensation coalescence evaporation homogeneous reactions nucleation coagulation coagulation hygroscopic activation activation and scavenging dry deposition wet scavenging gas molecules D < 0.01 0.01 < D < 0.1 0.1 < D < 1 1 < D < 10 aerosols aerosols aerosols aerosols cloud droplets B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 16 / 28

Aerosol Modeling and Simulation Towards External Mixing of Fractal Particles Spherical case Internal mixing with a carbon core External mixing Internal mixing with a carbon surface Soot Mixing state and radiative properties Geometrical configuration of soot B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 17 / 28

Outline Towards Integrated Modeling 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: min e F impact F CTM F economic 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

Outline Uncertainty Propagation & Ensemble Forecast 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 Uncertainties Cloud attenuation ±30% Dry deposition (O 3 and NO 2 ) ±30% Boundary Conditions (O 3 ) ±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 Models E = {M m ( )} m Ensemble mean: EM( ) = 1 M m ( ) E M E Super-ensemble: ELS( ) = m α m M m ( ) with weights α m to forecast on the basis of past observations Ensemble & relative uncertainties (POLYPHEMUS run for ozone) B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 23 / 28

Outline Data Assimilation & Inverse Modeling 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

Background Data Assimilation & Inverse Modeling Data assimilation for ozone (Credit: Lin Wu/CLIME/CEREA) 140 120 reference observation optimal interpolation enkf 4dvar rrsqrt Montandon Monitoring Networks Terrestrial sensors Satellite data Key Features Variational and sequential methods Inverse modeling of emissions High-dimensional systems Second-order sensitivity Concentration 100 80 60 40 20 0 10 20 30 40 50 Observation numbers 9 11 May 8 12 May 13 May 7 14 May 15 May 6 11-15 May (Mean) 11-15 Mai (Simulated) 5 4 3 2 1 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 time (hours) 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) See movie B. Sportisse Air Quality Modeling and Simulation TeraTech, 20 June 2007 26 / 28

Data Assimilation & Inverse Modeling alpha=2 alpha=0.5 alpha=1 Reseau geometrique Network Design IRSN Descartes monitoring network design over France to monitor potential radionuclides releases accidents (Credit: Marc Bocquet, CEREA) 20000 simulated releases 54 52 50 48 46 44 42 40-10 -5 0 5 10 15 54 52 50 48 46 44 42 40-10 -5 0 5 10 15 54 52 50 48 46 44 42 40-10 -5 0 5 10 15 54 52 50 48 46 44 42 40-10 -5 0 5 10 15 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