Air Quality Modeling
7.5 Air Quality Model Design and Simulation
7.5.2 Prepare for Modeling
7.5.2.1 Defining the Purpose of the Model
The first step in model development is to define the scientific, regulatory, or computational problem of interest. Example scientific topics include determining the effects of (1) aerosols on global climate, (2) aerosols on stratospheric ozone reduction, (3) aerosols on radiative absorption by clouds, (4) moisture and temperature fields on severe weather patterns, (5) soil moisture on wind patterns, (6) oceans on carbon dioxide loadings, (7) carbon dioxide buildup on global circulation patterns and climate, (8) clouds on climate, (9) clouds on tropospheric ozone concentrations, (10) urban pollution on regional and global gas loadings, (11) urban pollution buildup on local meteorology, and (12) aircraft emissions on ozone concentrations.
Some topics of interest to regulators and scientists include estimating the effects of (1) pollutant source controls on human exposure, (2) chlorofluorocarbon emissions regulations on stratospheric ozone, and (3) carbon dioxide emissions regulations on global warming.
7.5.2.2 Determining Scales and Dimension of Interest
The second step in model development is to determine the spatial and temporal scale dimension of interest according to the classification of models in figure 7.5.2-1 (as seen earlier in this chapter].
7.5.2.3 Selecting Processes
An important step in model development is to select the physical, chemical, and dynamical equations for the model and the best available tools to solve the equations. An ideal model includes every conceivable process, each simulated with the most accurate solver. Because computer speed and memory are limited, either the number of processes simulated or the accuracy of individual solutions must be limited.
Six major groups of processes simulated in atmospheric models are (1) meteorological, (2) transport, (3) cloud, (4) radiative, (5) gas, and (6) aerosol processes. When a model is developed, it is necessary to decide whether one or more of the groups can be excluded from the simulation or replaced by observations.
7.5.2.4 Selecting Variables
Once processes in a model are selected, variables must be chosen. If the model treats meteorology prognostically, some of the variables required include air temperature, air pressure, air density, horizontal velocity, vertical velocity, geopotential, water-vapor mass mixing ratio, liquid-water mass mixing ratio, and/or ice mass mixing ratio. If the model includes trace gases, prognostic variables include the concentration of each gas. If the model includes aerosols, prognostic variables include the particle number concentration and component volume concentration in each size bin. If radiative calculations are performed, prognostic variables may include heating rates. Many intermediate variables are also stored, but not permanently, in a model. Photolysis coefficients, extinction coefficients, particle growth rates, gas dry-deposition velocities, entrainment rates, and pressure-gradient forces are variables that are stored temporarily.
When model variables, especially gas and aerosol variables, are selected, computer central-memory limits must be considered.
7.5.2.5 Setting Time Steps and Intervals
Selecting model time steps and interval is the next step in model development. It depends on desired accuracy, computer-time requirements, and stability requirements. For a 5-km x 5-km horizontal grid, a typical time step for explicit dynamical calculations is 5 s. For a global grid of 5 x 5 , it is 300 s. For chemistry, the time step is variable with some solution methods and fixed with others. Generally, the longer the fixed time step, the less accurate the solution and the less computer time required.
Other processes, such as cloud, aerosol, radiative, and transport processes, use fixed or variable time steps. Aerosol chemical processes require small, variable time steps to maintain accuracy. Some physical processes, such as coagulation, are slow enough to allow long time steps, such as 300-900 s or more.
Time intervals between operator-split processes must be selected. The time interval may be chosen as the longest time step of any single process in a model. If chemistry requires a variable time step but transport requires a 300-s time step, a natural time interval is 300 s. Different groups of processes use different intervals. While the interval between gas chemistry and transport may be 300 s, that between gas chemistry and aerosol physics may be 900 s.
7.5.2.6 Setting Initial Conditions
One way to set meteorological conditions when initial data are absent is to run a preliminary model simulation that ends at the time of the start of the simulation of interest. The preliminary simulation also requires initial values that may not be correct. At the end of the presimulation, fields for temperature, pressure, velocity, and other parameters are available for every grid cell. Whether such values are accurate is open to question. Initializing a model with known data is the best way to reduce uncertainty.
7.5.2.7 Setting Boundary Conditions
In most atmospheric models, surface boundary conditions are needed. At the model top and surface, vertical velocities are usually set to zero, and variables are not transported through the boundaries by mean vertical velocities. At the surface, heat and moisture fluxes from the ground and gas/aerosol emission fluxes enter the model. Dry deposition and sedimentation remove gases and aerosols from the bottom layer.
In global models, the west-east grid generally wraps around on itself; thus, lateral boundary conditions are not needed. At the poles in a spherical-coordinate global model, south-north velocities may be set to zero so that all material travels around polar singularities. In regional models, lateral horizontal boundary conditions are needed. Boundary conditions for gases and particles often require concentrations from outside the boundary.
In many models, gas mixing ratios outside a lateral boundary are held constant, and such mixing ratios are allowed to advect into the model domain. In reality, gas mixing ratios outside a boundary vary during the day and night due to chemistry. Thus, solving time-dependent chemical equations for gas concentrations in a virtual row or column outside a lateral boundary improves estimates of gas fluxes into the model domain.
Aerosol concentrations outside a lateral boundary are difficult to estimated and are usually held constant. In reality, such values are affected by time-dependent physical and chemical processes, such as coagulation, growth, and aqueous chemistry. Time-dependent equations for the change in aerosol concentration can be written for a virtual row or column outside a lateral boundary.
7.5.2.8 Input Data
Input data, such as topographical, land-use, soil moisture, emissions, chemical-rata, absorption cross-section, and activity-coefficient data, are needed to run a model. Topographical data are necessary for calculating surface geopotential and graphing model predictions. Land-use data may be used to estimate average grid-cell values of the surface roughness length for momentum, soil specific heat, soil density, soil porosity, surface albedo, and leaf area index.
Soil moisture (soil liquid water) data are needed for calculating surface temperatures in a soil model. Soil moisture varies with relative humidity, air temperature, soil temperature, soil porosity, and soil specific heat. Ideal sources of foil moisture data for model initialization are measurements from remote sensing instruments, which are generally housed on an aircraft or satellite.
Emissions data are required to simulate gas or particle pollution buildup. Emissions data must be accurate to properly simulate urban air pollution, since emissions make up the bulk of the source gases and particles entering an urban atmosphere. Urban emissions inventories are often prepared by state and local agencies. Global emissions rates are more uncertain because taking into account every gas or aerosol source in each grid cell of a global model requires a significant effort.
Chemical-rate coefficient data are essential for simulating gas, aqueous, or reversible chemical reactions. Such data include temperature-and/or pressure-dependent uni-, bi-, and the molecular rate coefficients for gas-or aqueous-phase chemical reactions. Rate-coefficient data also include equilibrium-coefficient data.
Absorption cross-section data are important for estimating photolysis coefficients of gases and aqueous species and for calculating extinction coefficients due to gas absorption.
Activity-coefficient data are necessary for properly simulating reversible chemistry in concentrated aerosols.
7.5.2.9 Simulations
After a model has been developed and data have been obtained, simulations can be run. When a simulation is first started, it usually does not run to completion, because bugs still exist in the program. Debugging can take hours to weeks, de-pending on the number and severity of bugs and on the debugging experience of the programmer. Nevertheless, bugs are usually ironed out, and a baseline simulation can be performed.
During a baseline simulation, predictions and statistical comparisons to data should be gathered, stored, and/or printed out. The primary purpose of model development is to study a scientific or regulatory issue, and the baseline simulation should be designed with this study in mind.
The first step in model development is to define the scientific, regulatory, or computational problem of interest. Example scientific topics include determining the effects of (1) aerosols on global climate, (2) aerosols on stratospheric ozone reduction, (3) aerosols on radiative absorption by clouds, (4) moisture and temperature fields on severe weather patterns, (5) soil moisture on wind patterns, (6) oceans on carbon dioxide loadings, (7) carbon dioxide buildup on global circulation patterns and climate, (8) clouds on climate, (9) clouds on tropospheric ozone concentrations, (10) urban pollution on regional and global gas loadings, (11) urban pollution buildup on local meteorology, and (12) aircraft emissions on ozone concentrations.
Some topics of interest to regulators and scientists include estimating the effects of (1) pollutant source controls on human exposure, (2) chlorofluorocarbon emissions regulations on stratospheric ozone, and (3) carbon dioxide emissions regulations on global warming.
7.5.2.2 Determining Scales and Dimension of Interest
The second step in model development is to determine the spatial and temporal scale dimension of interest according to the classification of models in figure 7.5.2-1 (as seen earlier in this chapter].
7.5.2.3 Selecting Processes
An important step in model development is to select the physical, chemical, and dynamical equations for the model and the best available tools to solve the equations. An ideal model includes every conceivable process, each simulated with the most accurate solver. Because computer speed and memory are limited, either the number of processes simulated or the accuracy of individual solutions must be limited.
Six major groups of processes simulated in atmospheric models are (1) meteorological, (2) transport, (3) cloud, (4) radiative, (5) gas, and (6) aerosol processes. When a model is developed, it is necessary to decide whether one or more of the groups can be excluded from the simulation or replaced by observations.
7.5.2.4 Selecting Variables
Once processes in a model are selected, variables must be chosen. If the model treats meteorology prognostically, some of the variables required include air temperature, air pressure, air density, horizontal velocity, vertical velocity, geopotential, water-vapor mass mixing ratio, liquid-water mass mixing ratio, and/or ice mass mixing ratio. If the model includes trace gases, prognostic variables include the concentration of each gas. If the model includes aerosols, prognostic variables include the particle number concentration and component volume concentration in each size bin. If radiative calculations are performed, prognostic variables may include heating rates. Many intermediate variables are also stored, but not permanently, in a model. Photolysis coefficients, extinction coefficients, particle growth rates, gas dry-deposition velocities, entrainment rates, and pressure-gradient forces are variables that are stored temporarily.
When model variables, especially gas and aerosol variables, are selected, computer central-memory limits must be considered.
7.5.2.5 Setting Time Steps and Intervals
Selecting model time steps and interval is the next step in model development. It depends on desired accuracy, computer-time requirements, and stability requirements. For a 5-km x 5-km horizontal grid, a typical time step for explicit dynamical calculations is 5 s. For a global grid of 5 x 5 , it is 300 s. For chemistry, the time step is variable with some solution methods and fixed with others. Generally, the longer the fixed time step, the less accurate the solution and the less computer time required.
Other processes, such as cloud, aerosol, radiative, and transport processes, use fixed or variable time steps. Aerosol chemical processes require small, variable time steps to maintain accuracy. Some physical processes, such as coagulation, are slow enough to allow long time steps, such as 300-900 s or more.
Time intervals between operator-split processes must be selected. The time interval may be chosen as the longest time step of any single process in a model. If chemistry requires a variable time step but transport requires a 300-s time step, a natural time interval is 300 s. Different groups of processes use different intervals. While the interval between gas chemistry and transport may be 300 s, that between gas chemistry and aerosol physics may be 900 s.
7.5.2.6 Setting Initial Conditions
One way to set meteorological conditions when initial data are absent is to run a preliminary model simulation that ends at the time of the start of the simulation of interest. The preliminary simulation also requires initial values that may not be correct. At the end of the presimulation, fields for temperature, pressure, velocity, and other parameters are available for every grid cell. Whether such values are accurate is open to question. Initializing a model with known data is the best way to reduce uncertainty.
7.5.2.7 Setting Boundary Conditions
In most atmospheric models, surface boundary conditions are needed. At the model top and surface, vertical velocities are usually set to zero, and variables are not transported through the boundaries by mean vertical velocities. At the surface, heat and moisture fluxes from the ground and gas/aerosol emission fluxes enter the model. Dry deposition and sedimentation remove gases and aerosols from the bottom layer.
In global models, the west-east grid generally wraps around on itself; thus, lateral boundary conditions are not needed. At the poles in a spherical-coordinate global model, south-north velocities may be set to zero so that all material travels around polar singularities. In regional models, lateral horizontal boundary conditions are needed. Boundary conditions for gases and particles often require concentrations from outside the boundary.
In many models, gas mixing ratios outside a lateral boundary are held constant, and such mixing ratios are allowed to advect into the model domain. In reality, gas mixing ratios outside a boundary vary during the day and night due to chemistry. Thus, solving time-dependent chemical equations for gas concentrations in a virtual row or column outside a lateral boundary improves estimates of gas fluxes into the model domain.
Aerosol concentrations outside a lateral boundary are difficult to estimated and are usually held constant. In reality, such values are affected by time-dependent physical and chemical processes, such as coagulation, growth, and aqueous chemistry. Time-dependent equations for the change in aerosol concentration can be written for a virtual row or column outside a lateral boundary.
7.5.2.8 Input Data
Input data, such as topographical, land-use, soil moisture, emissions, chemical-rata, absorption cross-section, and activity-coefficient data, are needed to run a model. Topographical data are necessary for calculating surface geopotential and graphing model predictions. Land-use data may be used to estimate average grid-cell values of the surface roughness length for momentum, soil specific heat, soil density, soil porosity, surface albedo, and leaf area index.
Soil moisture (soil liquid water) data are needed for calculating surface temperatures in a soil model. Soil moisture varies with relative humidity, air temperature, soil temperature, soil porosity, and soil specific heat. Ideal sources of foil moisture data for model initialization are measurements from remote sensing instruments, which are generally housed on an aircraft or satellite.
Emissions data are required to simulate gas or particle pollution buildup. Emissions data must be accurate to properly simulate urban air pollution, since emissions make up the bulk of the source gases and particles entering an urban atmosphere. Urban emissions inventories are often prepared by state and local agencies. Global emissions rates are more uncertain because taking into account every gas or aerosol source in each grid cell of a global model requires a significant effort.
Chemical-rate coefficient data are essential for simulating gas, aqueous, or reversible chemical reactions. Such data include temperature-and/or pressure-dependent uni-, bi-, and the molecular rate coefficients for gas-or aqueous-phase chemical reactions. Rate-coefficient data also include equilibrium-coefficient data.
Absorption cross-section data are important for estimating photolysis coefficients of gases and aqueous species and for calculating extinction coefficients due to gas absorption.
Activity-coefficient data are necessary for properly simulating reversible chemistry in concentrated aerosols.
7.5.2.9 Simulations
After a model has been developed and data have been obtained, simulations can be run. When a simulation is first started, it usually does not run to completion, because bugs still exist in the program. Debugging can take hours to weeks, de-pending on the number and severity of bugs and on the debugging experience of the programmer. Nevertheless, bugs are usually ironed out, and a baseline simulation can be performed.
During a baseline simulation, predictions and statistical comparisons to data should be gathered, stored, and/or printed out. The primary purpose of model development is to study a scientific or regulatory issue, and the baseline simulation should be designed with this study in mind.
