The 3D domain must be discretized, usually equally spaced on the ground, with variable height in the vertical direction. Recent models use a “terrain-following” vertical coordinate.

The boundary conditions are often unknown (or, at least, highly uncertain), so we try to put them as far as possible so that their influence on the interesting domain in minimal. However, solving with the same accuracy a very large domain is useless (and costly in terms of computation). The nesting procedure is thus: -Compute the situation of a large domain with coarse discretization -Use the computed values as boundary conditions for the inner domain more accurate discretization.

WRF (Weather Research and Forecast)/CHIMERE grid domains on Western Europe, France, and Eastern France, with a resolution of 45, 15, 3 km. Nesting is critical when using the model for decision making: which are the measures, hence the emissions, in the outer domains?

The “real-life” mechanism (chemical reactions take place at the same time of transport and dispersion) cannot be reproduced on a digital computer, hence THE SPLITTING SCHEME C n+1 (z 1,z 2,z 3 )= A xy A z A c A c A z A xy C n (z 1,z 2,z 3 ) where C n (z 1,z 2,z 3 ) is the pollutant concentration at step n A xy is the horizontal transport and diffusion module A z is the vertical transport and diffusion module A c is the chemical module Each module requires special simulation processors, e.g. to conserve the pollutant mass and avoid negative concentrations of any component.

The discretization of the equations may be obtained in several different ways. Ex. where n indicates the time index and k the spatial index along the coordinate z 1. The first approximation is called a “forward” difference (implicit), while the second is a “central” difference (again implicit). Similar approximations may be written for all the other terms.

The discretization method must be stable (i.e. prevent the indefinite accumulation of errors). Ex. only transport along only a component is discretized as: which is a very simple linear discrete-time system. So, for asymptotic stability, the eigenvalue must be smaller than one. Thus: which is again the Courant-Friedrichs-Lewy condition. This is relevant for air quality models since v may be high.

Contains of the most severe non-linearities, also the stiffest sub-system (several orders of magnitude of reaction time scales) The general form of chemical kinetics for two substances A, and B, that react to form a new substance C is: The problem is thus the determination of the coefficient k 1, k 2 and the integration of the complete system. The “schemes” differ for the number of equations considered and the chemical species involved.

The chemical scheme proposed by Carter, chemical species and operators -5 constant (O 2, H 2 O, CH 4, M, h ) -4 only produced (CO 2, - N, H 2, - C) -45 active:  2 alkanes (ALK1, ALK2)  3 alkynes (OLE1, OLE2, OLE3) + Isoprene  2 aromatics (ARO1, ARO2)  38 esplicit - 23 stationary (i.e. always in equilibrium given the speed of the reactions) 184 chemical reactions - 20 photochemical chemical Lumped molecules M: a synthetic component to account for incomplete description h  ultraviolet radiation

First ‐ order rate constants are in units of sec ‐ 1 and second ‐ order rate constants are in units of cm 3 molecule ‐ 1 sec ‐ 1.

The chemical reactions take few hours, so that the peak ozone concentrations are far from precursors sources. Tropospheric O 3 is a “secondary” pollutant (there are no ozone emissions) due to precursors.  NOx mainly due to road transports (76%) and domestic heating (21%)  VOC - such as CO, CH 4 mainly due to solvent use and industrial plants (44%) and road transport (49%), but also to agriculture. Ozone precursor emissions = 1.22 NOx CO CH 4 + NMVOC (conventional TOFP – Tropospheric Ozone-Forming Potentials)

The ozone formation follows the reaction: O 2 + O + M  O 3 + M (M being a molecule of O 2 or N 2 or others that absorbs the reaction energy) The availability of oxygens atoms depends on two reactions O 2 + h O + O caused by ultraviolet radiation with wave length  nm NO 2 + h NO + O caused by ultraviolet radiation with  nm Hence ozone formation is stronger in the mountains (higher radiation). Three ways of measuring ozone: - the daily max 8-hours running mean -AOT40 sum of differences between hourly concentrations (8am-8pm) and 80 mg/m 3 ×h (= 40 ppb) in a year (or another period) (vegetation) -SOMO35 sum of means over 35 ppb (daily maximum 8-hour) in a year (humans)

The ozone creation and destruction cycle NO 2 + h NO + O O 2 + O + M  O 3 + M O 3 + NO  NO 2 + O 2 is much faster that other atmospheric reactions. This means that the correspondent equation: dNO 2 /dt= k o3 O 3 NO – k NO2 NO 2 may be considered always in equilibrium, i.e. dNO 2 /dt=0. O 3 = k NO2 NO 2 / (k o3 NO) in equilibrium conditions (i.e. always). Thus ozone increases with nitrogen dioxide (diesel emissions) even if most anthropogenic emissions are NO. It is a catalytic cycle, since NO 2 is essential, but is not consumed

The NOx cycle results in relatively low ozone levels because, although ozone is formed, it is destroyed in reacting with NO 2. Adding VOC, allows NO 2 to be regenerated without destroying ozone, bypassing O 3 + NO  NO 2 + O 2 OH radicals (also generated by various reactions among pollutants in the atmosphere) convert some VOC to peroxy radicals, which then regenerate NO 2 as follows: VOC–OO + NO  NO 2 + VOC–O where the two oxygen atoms ("OO") are the peroxide group attached to a VOC. Ozone formation thus depends on the ratio of VOC to NOx (VOC/NOx).

At high VOC/NOx ratios, ozone formation is controlled by the amount of NOx available, and the last reaction is the main route to regenerate NO 2 from NO. Under this "NOx-limited" situation, decreasing NOx reduces ozone, decreasing VOC has little or no effect. At low VOC/NOx, ozone formation is limited by the amount of VOC available. In addition, NO 2 competes with VOC to react with OH radicals, slowing the rate at which VOC is converted to peroxy radicals, and thereby slowing the rate of reaction. Under this "VOC-limited" condition, reducing VOC reduces ozone, reducing NOx increases ozone. This because NOx reductions slow down the rate of ozone destruction through the NOx cycle, and speed up the rate of NO 2 regeneration.

 High ozone ground levels concentrations observed since the 70’s in USA and Europe  The process takes place only at summer temperatures (over 30° C) In Lombardy increasing ozone trends claim for effective reduction policies

We want to design effective ozone reduction policies for Lombardy region solving a multi-objective optimisation problem… ozone pollution reduction [% max] 0% 30% 60% 100% 20%40%60%80%100% reduction costs [% max] ? CLE MFR MFR

 Select an ozone indicator (max 8h average) = Air Quality Index  Simulate different scenarios through an eulerian photochemical model, CALGRID (highly complex, time consuming)  Train a surrogate model to represent CALGRID outputs  Evaluate precursors reduction costs  Select the decision variables (spatially uniform precursors reduction rates in each emission sector)  Solve the multi-objective planning problem, modelling ozone dynamics through the surrogate model

Point wise meteo data Land use Air pollution model CALGRID Hourly 3 D concentration fields Initial and boundary concentrations Meteorological preprocessing Chemical speciation and preprocessing Topography Emissions (model) Hourly wind fields Hourly emission values

It requires on each cell:  Orography  Hourly wind field  Hourly emissions Lombardia – wind field

It is not possible to measure emissions from all of the individual sources (e.g. passenger cars, domestic heating, etc.) In practice, atmospheric emissions are estimated on the basis of measurements made at selected or representative samples of the (main) sources and source types. The basic model for an emission estimate is the product of (at least) two variables, for example: an emission measurement (rarely available) over a period of time and the number of such periods emissions occurred in the required estimation period Or (more frequently) an activity statistic and a typical average emission factor for the activity. ♠

E = A*fe*(1-re*AR) Emission of a given pollutant in a given scenario Energy used by a certain activity (in a nominal year) (Unabated) Emission factor Efficiency of abatement technology DECISIONS Applications rate Emission estimates are computed though inventories (i.e. databases) containing, for each pollutant: - Location of the source -Direct measurements (rarely available) -Emission factors -A measure of the activity -Operating conditions -Details on the measurement procedure/instruments or the estimation EMISSION INVENTORIES

EMISSION INVENTORIES - CORINAIR CORINAIR 1990 Inventory recognises 11 main source sectors (as agreed with EMEP): 1.Public power, cogeneration and district heating plants 2.Commercial, institutional and residential combustion plants 3.Industrial combustion 4.Production processes 5.Extraction and distribution of fossil fuels 6.Solvent use 7.Road transport 8.Other mobile sources and machinery 9.Waste treatment and disposal 10.Agriculture 11.Nature. They are provided on large point sources on an individual basis and on other smaller or more diffuse on an area basis, usually by administrative boundary at the county, department level (NUTS level 3). CORINAIR is a European standardization project (since 1985) aiming to provide a complete, consistent and transparent air pollutant emission inventory for Europe within a reasonable time scale.

Point sources The sources to be provided as point sources are: Power plant with thermal input capacity >= 300MW Refineries Sulphur acid plant Nitric acid plant Integrated iron/steel with production capacity > 3 Mt/y Paper pulp plant with production capacity > 100 kt/y Large vehicle paint plant with production capacity > vehicles/yr Airports with > LTO cycles/y Other plant emitting >= 1000 t/y SO 2, NOx or VOC or >= t/yr CO 2 Available for all European countries Using a standardized methodology and classification (SNAP - Standardized Nomenclature for Air Pollutants) Adopting a transparent approach by the provision, within the inventory, of activity statistics/data and emission factors (or details of emission measurements where available) used to calculate emissions and through the supply of full references to the sources of these data.

CORINAIR developed a complete three level nomenclature (macrosector – sector – activity). More recently, also the “fuel” has been added. Pollutant covered are: Sulphur dioxide (SO 2 ) Nitrogen oxides (NOx) Non-methane volatile organic compounds (NMVOC) ammonia Carbon monoxide Methane Nitrous oxide Carbon dioxide The number of sources to be considered as point sources is presently of several thousands. Activities and pollutants To serve as input to air quality models, all this information must be spatialized (i.e. redistributed on the model gridded domain). This is obtained by using more detailed proxy variables (ex. population distribution, local car fleets,…)

EMISSION INVENTORIES Ex. INEMAR ( /Inemar/HomeLombardia) Ex. GAINS (

WIND FIELDS A preprocessor (ex. CALMET) is needed to generate wind speed and direction in each cell of the domain with the required frequency (e.g. every hour). This must be done interpolating local measurements and taking into account: the different rugosity of each cell, its orography, and satisfying the mass conservation principle.

Simulation of a heavy pollution episode in June ‘96 The threshold level for human health is 110  g/m 3 for the 8h moving average

- 35% NOx - 35% VOC & NOx - 35% VOC + 35% VOC & NOx Hypothetical precursor reduction scenarios and consequent O3 changes in a given time interval The mountain part of the region is insensitive to both NOx and VOC reductions on the whole domain; thus, we can focus on the plain part (VOC - limited)

 Receptor (4km * 4km) : a given cell in the gridded domain (ozone indicator evaluation)  Emissions (12km * 12 km): cells in the square centered in the receptor (emission patterns, initial concentration conditions)  A polynomial LOCAL surrogate model fitted to CALGRID output on a given time interval T. We develop a simple model which interprets the air quality index in a cell as a function of nearby emissions 4 km

NOx emissions VOC emissions Air quality index Non linear terms a ij, b ij, c ij, d ij, and e ij are parameters to be estimated. The air quality index is assumed to be a polynomial function of nearby emissions

Model parameters can be estimated: On all the cells of the domain (a single surrogate model) except for borders On a sub-region of the entire domain On the individual cells The resulting surrogate model can be validated: Against a time interval not used in the fitting process Against a number of cells (10-20% randomly selected) not used in the fitting process

CORINAIR emission categories subdivide emissions in 11 macrosectors (EMEP/CORINAIR,1999): 1.public power, cogeneration and district heating plants; 2.commercial, institutional and residential combustion plants; 3.industrial combustion; 4.production processes; 5.extraction and distribution of fossil fuels; 6.solvent use; 7.road transport; 8.other mobile sources and machinery; 9.waste treatment and disposal; 10.agriculture; 11.nature. Each macrosector is subdivided in sectors and these in activities (e.g. EURO 6 diesel cars cc in sector diesel cars, macrosector 7)

 The reduction plan requires to select VOC reduction rates for:  solvent use (current emission 470 ton/day)  road transport(408 ton/day)  waste treatment (110 kg/day)  fossil fuel distribution (50 ton/day)  production without combustion (23 ton/day) each represents an activity sector s  We thus assume reduction factors r s (one value for all the emissions of sector s in t he region) as decision variables. This means a proportional decrease of the “emission surface” of each sector.

 r s : reduction for sector s: R s maximum feasible  E ij s : VOC emission on cell (i,j) for sector s  c s : reduction costs function for sector s  I i,j : ozone indicator on cell (i,j) decision variables Two-objective NL optimization problem with only 5 decision variables and 5 constraints (plus non- negativity)

Noticeable improvements in ozone levels can be reached with moderate investments… …provided they are targeted to relevant sectors, such as road transport and industrial solvents.