Mesoscale Models
On the mesoscale range, the inputs to the models
are the representation/parameterization of the
topography and the natural features of the surface
necessary for computations down to the smallest
scale L M /R M (which typically is of order
1/3–1 km). This is therefore the scale over which
the surface conditions are averaged, including
ground surface elevation z s (x, y), roughness length
z 0 (x, y), surface heat flux F θ (x, y, t), and surface
temperature θ s (x, y, t) (which is usually derived in
operational mesoscale models by coupling the
atmospheric model to a thermal model for ground
temperature, allowing for radiation from/to the
earth’s surface). In fully computational models,
FCMs, the boundary conditions at the edges of
the domain (and, in some models, above the
domain) have to be specified and are usually
derived from larger scale regional or global
numerical models. These are usually updated at
regular intervals, for example, every 3–12 h, with
observational data taken at all levels in the atmosphere. This process of “data assimilation” is also
beginning to be applied directly to mesoscale
models, for example, in urban areas with many
measurement sites.
A critical feature of any FCM, especially in the
boundary layer and when the flow is influenced by
mountains, buildings, surface heating etc., is the
representation of the effects on residual
(or computed) scales greater than L M or R M of
the turbulence at the smaller space and timescales.
Some models, such as HOTMAC [41], contain
quite complex sub-models of the turbulence statistics (with extra equations for the turbulent
kinetic energy and turbulent dissipation rate
(k-ε)), while others, for example, WRF [42],
COAMPS [43], MM5 [44], MESONH [45], and
UK Met. Office unified models [46], use eddy
viscosity or even simpler parameterizations (e.g.,
assuming a known form of the velocity profile).
The experience from field studies in the USA is
that for urban areas located on sloping terrain
where buoyancy forces are significant (i.e.,
F ≲ 1), the models with the more complex turbulence models are more accurate in these conditions. But the greatest improvements in accuracy,
especially in predicting the mean wind speeds and
wind direction in changing meteorological conditions, and in complex terrain, arise from using
these models with finer resolution of the order of
1 km or less. However even with finer resolution
and complex modelling, such models still cannot
predict some significant features of surface layer
turbulence, such as evening transition of upslope
flows and formation of gravity currents, or calculate relevant statistics of the turbulence for dispersion models, unless they are used as large eddy
simulations [47], which, because of computational requirements, is currently only possible in
research mode.
As explained in Table 4, these models take
many hours to compute a single meteorological
situation. Nevertheless, they are sufficiently reliable indicators to be used operationally. Fast
approximate models, FAMs, for air flow and
meteorology over mesoscale distances are being
developed, based on recent research. These
models are useful as qualitative guides to the
complex flow that might occur on the mesoscale,
for quantitative predictions as input to dispersion
computations.
As shown in Table 4, when local buoyancy
effects are weak (i.e., U G /U B ! 1), the air flow
over an urban area is generally a perturbation of
the oncoming flow. Then, from semi-analytic
models using perturbation methods, faster computational schemes have been devised, and are
being developed in a general way to allow for
orography, roughness change, and some effects
of surface heating [8, 48]. Although these are
perturbation methods, the changes in wind speed
and direction predicted (and verified) by these
models can be quite large (~50% or more).
The perturbation modelling approach being
adopted in this range of meteorological flows is
similar to that used for sub-mesoscale/
neighborhood scale orographic flows (for example, in the FLOWSTAR model [49] and
RIMPUFF [50]), the main difference being that
over the larger scale the internal layer l reaches the
top of the boundary layer h, so that the inversion
height is affected and Coriolis effects have to be
included (which, for example, significantly influences flows along coasts and up and down large
river valleys, [51]). Typically these models are run
176
Urban Air Quality: Meteorological Processes
On the mesoscale range, the inputs to the models
are the representation/parameterization of the
topography and the natural features of the surface
necessary for computations down to the smallest
scale L M /R M (which typically is of order
1/3–1 km). This is therefore the scale over which
the surface conditions are averaged, including
ground surface elevation z s (x, y), roughness length
z 0 (x, y), surface heat flux F θ (x, y, t), and surface
temperature θ s (x, y, t) (which is usually derived in
operational mesoscale models by coupling the
atmospheric model to a thermal model for ground
temperature, allowing for radiation from/to the
earth’s surface). In fully computational models,
FCMs, the boundary conditions at the edges of
the domain (and, in some models, above the
domain) have to be specified and are usually
derived from larger scale regional or global
numerical models. These are usually updated at
regular intervals, for example, every 3–12 h, with
observational data taken at all levels in the atmosphere. This process of “data assimilation” is also
beginning to be applied directly to mesoscale
models, for example, in urban areas with many
measurement sites.
A critical feature of any FCM, especially in the
boundary layer and when the flow is influenced by
mountains, buildings, surface heating etc., is the
representation of the effects on residual
(or computed) scales greater than L M or R M of
the turbulence at the smaller space and timescales.
Some models, such as HOTMAC [41], contain
quite complex sub-models of the turbulence statistics (with extra equations for the turbulent
kinetic energy and turbulent dissipation rate
(k-ε)), while others, for example, WRF [42],
COAMPS [43], MM5 [44], MESONH [45], and
UK Met. Office unified models [46], use eddy
viscosity or even simpler parameterizations (e.g.,
assuming a known form of the velocity profile).
The experience from field studies in the USA is
that for urban areas located on sloping terrain
where buoyancy forces are significant (i.e.,
F ≲ 1), the models with the more complex turbulence models are more accurate in these conditions. But the greatest improvements in accuracy,
especially in predicting the mean wind speeds and
wind direction in changing meteorological conditions, and in complex terrain, arise from using
these models with finer resolution of the order of
1 km or less. However even with finer resolution
and complex modelling, such models still cannot
predict some significant features of surface layer
turbulence, such as evening transition of upslope
flows and formation of gravity currents, or calculate relevant statistics of the turbulence for dispersion models, unless they are used as large eddy
simulations [47], which, because of computational requirements, is currently only possible in
research mode.
As explained in Table 4, these models take
many hours to compute a single meteorological
situation. Nevertheless, they are sufficiently reliable indicators to be used operationally. Fast
approximate models, FAMs, for air flow and
meteorology over mesoscale distances are being
developed, based on recent research. These
models are useful as qualitative guides to the
complex flow that might occur on the mesoscale,
for quantitative predictions as input to dispersion
computations.
As shown in Table 4, when local buoyancy
effects are weak (i.e., U G /U B ! 1), the air flow
over an urban area is generally a perturbation of
the oncoming flow. Then, from semi-analytic
models using perturbation methods, faster computational schemes have been devised, and are
being developed in a general way to allow for
orography, roughness change, and some effects
of surface heating [8, 48]. Although these are
perturbation methods, the changes in wind speed
and direction predicted (and verified) by these
models can be quite large (~50% or more).
The perturbation modelling approach being
adopted in this range of meteorological flows is
similar to that used for sub-mesoscale/
neighborhood scale orographic flows (for example, in the FLOWSTAR model [49] and
RIMPUFF [50]), the main difference being that
over the larger scale the internal layer l reaches the
top of the boundary layer h, so that the inversion
height is affected and Coriolis effects have to be
included (which, for example, significantly influences flows along coasts and up and down large
river valleys, [51]). Typically these models are run
176
Urban Air Quality: Meteorological Processes