FLOW AND DISPERSION IN URBAN AREAS

Annu. Rev. Fluid Mech. 2003. 35:469 96 doi: 10.1146/annurev.fluid.35.101101.161147 Copyright c 2003 by Annual Reviews. All rights reserved FLOW AND DISPERSION IN URBAN AREAS R. E. Britter Department of Engineering, University of Cambridge, Cambridge CB2 1PZ, United Kingdom: email: rb11@eng.cam.ac.uk S. R. Hanna Hanna Consultants, Kennebunkport, Maine 04046; email: hannaconsult@adelphia.net Key Words urban boundary layer, turbulence, pollutant dispersion Abstract Increasing urbanization and concern about sustainability and quality of life issues have produced considerable interest in flow and dispersion in urban areas. We address this subject at four scales: regional, city, neighborhood, and street. The flow is one over and through a complex array of structures. Most of the local fluid mechanical processes are understood; how these combine and what is the most appropriate framework to study and quantify the result is less clear. Extensive and structured experimental databases have been compiled recently in several laboratories. A number of major field experiments in urban areas have been completed very recently and more are planned. These have aided understanding as well as model development and evaluation. 1. INTRODUCTION Much of the global population currently live and work in urban areas, and this urbanization is expected to increase. This trend has recently inspired many urbancentered studies. Many are of a fluid mechanical nature, either in isolation or in combination with other disciplines such as chemistry, epidemiology, and pedestrian and vehicular mobility. Large-scale weather prediction and mesoscale meteorological models require the parameterization of urban areas to provide boundary conditions. Regional air-pollution models are used to estimate the transport of pollutants to and from cities. Urban climatology (Oke 1987) addresses the mass, momentum, and energy transfers through an urban area and the resulting temperatures, humidities, radiant fluxes, etc. These changes in surface energy balance and temperatures and humidities in urban areas influence, for example, general urban planning, greenspace provision, and energy usage in cities. Studies of urban air quality (Fenger et al. 1999) focus strongly on the wind flow over and through the city and the sources of pollutants within and beyond the city. A major pollutant source within the city is vehicle emissions, which lead to an 0066-4189/03/0115-0469$14.00 469

470 BRITTER HANNA interaction between mobility, air quality, and the possible regulation of vehicles in cities. The wind flow within cities, in particular, the local turbulence levels, directly affect pedestrian mobility and comfort. This same wind flow, but on a larger scale, represents the wind environment within which new buildings are to be placed and is of concern both for wind-loading problems (Cook 1990) and for the provision of clean air to the buildings and the removal of exhaust air from the urban canopy. This wind environment and the building construction affect some of the exchange processes between the building interior and exterior and, consequently, building air quality and energy use. Hazardous materials in large quantities are normally prohibited from heavily populated areas, but where this is not the case there is a need for emergency authorities and civil defense personnel to have operational tools available to determine what action to take in case of an accident. Very recently there has been increased concern about the nonaccidental release of hazardous materials in urban areas. Understanding the flow of the wind through and above the urban area and/or the dispersion of material in that flow (Hanna & Britter 2002) is necessary. We address these issues by considering the problem at different scales. At each of the scales there are observations from the field and the laboratory that are interpreted in terms of various physical (and possibly chemical) processes. These processes, once recognized, are often combined and reformed into mathematical models that can form a hierarchy of complexity or sophistication: Each model has its own regime of applicability and accuracy. A detailed interpretation at one scale is commonly parameterized to assist interpretation at the next larger scale. We use spatial scales to describe the major urban flow features, although the spatial scales can be related to time scales through the x = ut relation. Roughly speaking the timescales are the spatial scales divided by an appropriate advection (wind) velocity. The discussion is broken down into four ranges of length scales: regional (up to 100 or 200 km), city scale (up to 10 or 20 km), neighborhood scale (up to 1 or 2 km), and street scale (less than 100 to 200 m). The regional scale is affected by the urban area. For example, the urban heat island circulations, any enhanced precipitation, and the urban pollutant plume can extend to these distances. At this scale the mean synoptic meteorological patterns are given and the urban area represents a perturbation, causing deceleration and deflection of the flow, as well as changes to the surface-energy budget and the thermal structure. The city scale represents the diameter of the average urban area. At these scales the variations in flow and dispersion around individual buildings or groups of similar buildings have been mostly averaged out. Wind flow models developed for this range pay little attention to the details of the flow within the urban canopy layer. Most of the mass of any pollutant cloud traveling over this distance will be above the height of the buildings. On the neighborhood scale buildings may still be treated in a statistical way; however, the approach may be different to that on the city scale. At the neighborhood scale we want to know more about the flow within the urban canopy.

FLOW IN URBAN AREAS 471 The wind flow, particularly within the canopy, may also be changing as it moves from one neighborhood to the next. Much of the mass of a pollutant cloud traveling over this distance may remain within the urban canopy. The street (canyon) scale addresses the flow and dispersion within and near one or two individual streets, buildings, or intersections. This would be of interest when considering turbulence affecting pedestrian comfort and the direct exposure of pedestrians and near-road residences to vehicular emissions. It can be of particular interest when regulatory pollutant monitoring stations are placed within street canyons. Figure 1 is an illustration of the results from a model prediction of annual mean NO 2 concentrations for greater London for the year 2005. The street scale is obvious; the neighborhood scale is evident in Heathrow airport, which is to the west of the center, whereas the city scale is an area of approximately 10 km in radius. The relevant regulatory limit in the United Kingdom is 21 ppb (parts per billion). 2. THE CITY SCALE AND THE REGIONAL SCALE The city scale is, essentially, the urban area: an area distinguished from its surroundings by its relatively large obstacles (buildings and other structures) and, hence, by a large drag force; by the infusion of heat and, perhaps, moisture from man s activities; and by the large heat-storage capacity of concrete, other building materials, and parking lots. The city scale can include variations in urban building types and spacings and primarily concerns the boundary layer above the average building height, H r. The regional scale is the larger surrounding area that is influenced by or influences phenomena at the city scale. Bornstein (1987) has observed that the flow is deflected over and around the urban area. The vertical flow displacement sometimes is visible when it is marked by formation of a cap cloud over the urban area, just as seen over a mountainous island. The horizontal flow displacement can be seen in the turning of sea-breeze or cold-front flow directions and in the bending of frontal surfaces. The deflections are in part kinematic owing to the volume of the buildings and in part dynamic owing to the drag forces on the buildings. The surface-energy balance for the urban area and its surroundings is significantly affected by differences in heat and moisture inputs due to human activities. The additional heat fluxes and the heat storage of the urban surface lead to the urban heat island phenomenon that has been widely observed and discussed (Oke 1987). This phenomenon causes a convergence of horizontal wind into the city and vertical motions over the city. The mixing depth over a city is increased as a result of the vertical motions as well as the enhanced heating. The transport and dispersion of pollutants over the urban area is altered as a result of increased mechanical turbulence caused by the relatively large obstacles over which the pollutants must travel. Furthermore, the urban heat island causes the

472 BRITTER HANNA boundary layer over an urban area to become more unstable as thermal turbulence increases. Both of these effects enhance dispersion. The wind velocities are altered in magnitude and direction. Wind speeds over the urban area are slower owing to the increased roughness of these areas, and wind directions could change as a result of the heat island circulations or the bending of the flow around and over the urban area. Most urban areas have large urban plumes that can be easily observed 100 to 200 km downwind of the urban area. Mathematical modeling of flows on the regional scale requires the parameterization of the effects of the urban surface on the flow. For example, weather forecast models are routinely run with a horizontal grid scale of 20 40 km. Mesoscale meteorological models with horizontal grid scales of 2 8 km are run for special purposes around urban areas and can better resolve the types of land use (Brown 2000). The mesoscale models use a typical vertical grid spacing of approximately 20 m near the surface and approximately 200 m in the middle and top of the boundary layer. Such models often parameterize the urban area by using a simple prescription of land use. Thus an urban land-use grid may be assigned a single surface roughness length, soil moisture, albedo, and Bowen ratio. At the city scale a similar need for the parameterization of the urban surface exists. Section 2.1 below addresses the effects of the momentum, the energy, and the pollutant changes provided by the urban areas. 2.1. Momentum: Surface Shear-Stress and Wind Profiles Urban obstacles exert a relatively large drag force on the atmosphere. This can be treated by standard atmospheric boundary-layer formulas (Stull 1997), as long as the mean building height, H r, is small compared to the surface boundary-layer depth, which is usually approximately 100 or 200 m, and the surface has some statistical homogeneity. Figure 2, adapted from Grimmond & Oke (2002), depicts a cross section of a typical urban area, showing the three major sublayers: the urban canopy sublayer, the roughness sublayer, and the inertial sublayer. The inertial sublayer is the area where the boundary layer has adapted to the integrated effect of the underlying urban surface, and it is treated by standard atmospheric boundary-layer formulas. In the urban canopy sublayer the flow at a specific point is directly affected by the local obstacles, and in the roughness sublayer the flow is still adjusting to the effects of many obstacles. The surface shear stress (averaged over the urban surface) defines a friction velocity, u, that can be used to derive wind and turbulence profiles. It is assumed that, regardless of the underlying surface, the wind speed at the top of the boundary layer (i.e., at a height of 500 to 1000 m) is approximately equal to the equilibrium wind speed defined by the geostrophic wind speed, which is based on the synoptic pressure gradient. For larger surface roughness the drag force (or u ) is larger, and the wind speed at any given level in the boundary layer is smaller.

FLOW IN URBAN AREAS 473 Figure 2 Schematic of the flow through and over an urban area. Adapted from Grimmond & Oke (2002). The friction velocity, u, is the key scaling velocity in Monin-Obukhov similarity theory (Stull 1997). The other key scaling parameter is the Monin-Obukhov length, L, which accounts for the effects of atmospheric stability and is proportional to u 3 divided by the surface turbulent (or sensible) heat flux from the ground surface: H s :L = (u 3 /κ)/(gh s/c p ρt ), where g is the acceleration due to gravity, C p is the specific heat of air at constant pressure, ρ and T are the air density and temperature, respectively, and κ is von Karman s constant taken to be 0.40. H s is positive in the day and is negative at night (discussed in more detail in Section 2.3). The wind-speed profile conforms to Monin-Obukhov similarity theory, with the addition of the two scaling lengths, z 0 (the surface roughness length) and d (the surface displacement length): u = (u /κ)(ln [(z d)/z 0 ] + ψ[z/l]), (1) where ψ(z/l) is a universal dimensionless function (Stull 1997) that equals zero in neutral or adiabatic conditions (i.e., when L = or z/l = 0). For the purposes of the following discussion of the urban boundary layer, neutral conditions are assumed for which Equation 1 reduces to u = (u /κ)ln[(z d)/z 0 ]. (2) Thus, given a wind-speed observation at a height greater than approximately 2H r (see Figure 3) and given an estimate of z 0 and d, then u can be estimated from Equation 2. Note that u and u at some reference height will define a drag coefficient

474 BRITTER HANNA Figure 3 The spatially averaged mean velocity profile near an urban area. Adapted from Bottema (1997). (for meteorologists) or skin-friction coefficient (for engineers). Consequently, these coefficients relate directly to the parameters z 0 and d. Parameterizations for z 0 and d are possible using several approaches. Land-use methods are used in most applied dispersion models. These methods are based on the tables of z 0 versus descriptive land-use types (e.g., Stull 1997). The problem for those interested in urban sites is that the land-use categories are usually very broadly defined, as they have to cover all land uses, ranging from ice flats and deserts to water surfaces and crops and forests. For example, the categories suggested by Stull (1997) do not account for commercial or industrial sites, even though there are four separate categories for towns and cities, with z 0 ranging from 0.3 m for outskirts of towns to 2 m for centers of cities with very tall buildings. The methods by Auer (1978), based on land-use categories within a radius of 3 km from the source and on approximate descriptions of building geometries, are used in the U.S. Environmental Protection Agency s Guidelines and form the basis for the decision as to whether the rural or urban dispersion correlations should be used but not to assign a z 0. Davenport et al. (2000) have proposed a more detailed description of the relation of z 0 to urban/industrial land use. A portion of their table is given in Table 1, where surface roughness lengths, z 0, are listed for five categories of buildings and industrial obstacles. The z 0 values range between approximately 0.1 and 2 m, in agreement with the z 0 suggestions for towns and cities by Stull (1997). A set of 12 urban land-use types has been proposed by Grimmond & Oke (1999), based on an extensive review of wind-profile observations in many urban areas. Their data show that z 0 /H r ranges from approximately 0.06 to 0.20, and d/h r ranges from approximately 0.35 to 0.85. A further complementary set is provided by Theurer (1999).

FLOW IN URBAN AREAS 475 TABLE 1 Updated surface roughness lengths, z 0, for five urban and industrial categories Surface roughness Category length, z 0 Urban/industrial site description Roughly open 0.1 m Moderately open country with occasional obstacles (e.g., isolated low buildings) at relative separations of at least 20 obstacle heights. Rough 0.25 m Scattered buildings and/or industrial obstacles at relative separations of 8 to 12 obstacle heights. Analysis may need displacement length, d. Very rough 0.5 m Area moderately covered by low buildings and/or industrial tanks at relative separations of 3 to 7 obstacle heights. Analysis requires displacement length, d. Skimming 1.0 m Densely built-up area without much obstacle height variation. Analysis requires displacement length, d. Chaotic 2.0 m City centers with mixture of low-rise and high-rise buildings. Analysis by wind tunnel advised. This table is an abridged version of a table in Davenport et al. (2000). More precise estimates of z 0 and d can be made using information about building sizes and spacing. The total building plan area, A p, and the total building frontal area, A f, in a building lot area, A T, can be used to define the lambda parameters that are used in many empirical urban boundary-layer formulas: λ p = A p /A T ; λ f = A f /A T. The dimensionless frontal area, λ f, is more important to drag because it represents the surface facing the wind flow. Typical values of λ f are 0.1 for areas with a moderate density of buildings and 0.3 for downtown areas. λ f would be larger in urban downtown areas were it not for the presence of large parking lots and parks. Note that if the spacing between square buildings of dimension H r equals 1 / 2 and 1.0 times H r, then λ f = 0.44 and 0.25, respectively. Grimmond & Oke (1999) review and evaluate many competing techniques for determining z 0 and d from λ p and λ f. They also acknowledge three types of urban flow. At small λ f the buildings act in isolation, at larger λ f the building wakes interfere with each other, and at even larger λ f a skimming flow over the buildings has limited direct penetration into the spaces between the buildings. Hanna & Britter (2002) have considered several field and laboratory data sets as well as theoretical and empirical formulas in the literature and recommend the following formulas: z 0 /H r = λ f for λ f < 0.15, (3a) z 0 /H r = 0.15 for λ f > 0.15, (3b) d/h r = 3λ f for λ f < 0.05, (4a)

476 BRITTER HANNA and d/h r = 0.15 + 5.5(λ f 0.05) for 0.05 <λ f <0.15, (4b) d/h r = 0.7 + 0.35(λ f 0.15) for 0.15 <λ f <1.0. (4c) Of course, λ p and λ f are not necessarily adequate to fully describe urban areas. One aspect omitted from most techniques used for describing urban areas is the great variability in building heights. Ratti et al. (2002) show that the ratio of the standard deviation of building heights to the mean building height, H r, ranges up to 1.0 for some urban areas. As a consequence the skimming-flow regime may not be well represented by laboratory studies that use obstacles of constant height. Because of the difficulties involved in processing geometrical data from multiple individual buildings that are needed to apply some of the morphological methods, these methods have not been widely used in the past for estimating the roughness length for air-quality modeling applications. However, very recently there has been demand for more detailed modeling, and this has led to the increased use of digital elevation models of cities (Ratti et al. 2002) for morphological studies. A more extensive discussion on urban parameterization schemes for mesoscale models may be found in the report by Brown (2000). These schemes are directly applicable to city-scale modeling. The parameterization is often similar to the use of wall functions in engineering-based computational fluid dynamics. It is important to recognize that, for all techniques described, an equilibrium boundary layer develops only after the air has flowed over many individual obstacles or rows of obstacles. This roughness change problem has been well studied (see Smits & Wood 1985 for a review), and correlations for the growth of the internal boundary layer are readily available; the internal boundary layer is the region of the flow that adjusts to the change of roughness. Within the internal boundary layer there is an equilibrium layer that can be broken into the new roughness sublayer and a new inertial sublayer. In a laboratory study Cheng & Castro (2001) found that it took 160z 0 for evidence of approximate similarity and 300z 0 for the equilibrium layer to reach the upper limits of the roughness sublayer. These distances correspond to approximately four to seven rows of obstacles. This finding led them to question whether urban areas ever allowed sufficient fetch for an inertial sublayer to develop and whether z 0 was an appropriate scaling parameter for urban areas. However, urban areas might be better represented as regions (or neighborhoods) of gradually varying roughness. The flow then is continuously readjusting to these changing surface conditions. The effective roughness length over terrain that consists of well-defined repeating patches of two different roughness surfaces was studied by Goode & Belcher (1999), who used a linearized perturbation model. They define the blending height as the top of the highest extent of the internal boundary layers from individual obstacles, above which the flow is fully adjusted to the combined roughnesses. In one of their numerical modeling tests, where the roughness patches alternated every 50 m with values of z 0 = 0.004 m (typical of mowed grass) and z 0 = 0.4 m (typical of an area of industrial or residential

FLOW IN URBAN AREAS 477 buildings measuring 4 5 m in height), wind-speed perturbations of as much as 20% occur at heights of 3 m. Slower wind speeds were found over the rough surface, as expected. At heights of 10 m or above the wind-speed perturbation was calculated to be less than 1%, implying that the blending height or the top of the internal boundary layer was at a height of 10 m, or 2H r, for this combination of roughness lengths and other conditions. In this study the roughness was imposed as a boundary condition rather than by physical obstacles. However, the study does emphasize that the internal boundary layer can be limited to lie within the roughness sublayer and that a z 0 for the combined surfaces may be appropriate. Thus there is some evidence for the pragmatic approach of determining a z 0 and a d for various areas of a city or for the entire city. Very few operational transport and dispersion models applied to the city scale allow inputs of space-varying z 0. Instead, it may be more robust to simply estimate an average or representative z 0 over an area. For example, an average z 0 can be assigned to 30 wind sectors and to distance increments representing at least 20% of the along-wind distance of interest. Because the wind flow and the cloud dispersion rates are relatively insensitive to minor (say, a factor of 2) variations in z 0, it is sufficient to use a single averaged value of z 0 for situations with minor variations of z 0 that are associated with land use (e.g., forest to industrial site to suburb). However, if z 0 varies by a few orders of magnitude (e.g., dense urban area to bay to industrial park), then a weighting scheme can be used to estimate a representative z 0. Following Goode & Belcher (1999), Hanna & Britter (2002) argue that a representative ln(z 0 ) and ln(d ) can be estimated by summing the distance-weighted values of ln(z 0 ) and ln(d ) over the area of interest. The flow at a particular position depends on the urban surface upwind and, consequently, on the direction of the wind and on the relative influence of different upwind portions of that urban surface. Therefore different flow conditions may need to be specified for the different wind sectors surrounding a point of interest. In most cases there are no observations of wind in the urban area, but there may be some at a nearby airport or other measurement site. A simple method is available for using the wind speed at a height of, say, 10 m, at a nearby site to a height at the urban area. If we assume that the wind speed over both sites is the same at a height of 30 m, for example, then Equation 1 can be applied at both sites to link them. If the atmospheric stabilities are neutral, Equation 2 suffices. Because the roughness length at the airport is a few orders of magnitude less than that at the urban area, the wind speed at a height of 10 m is greater at the airport than at the urban area. However, because of the increased drag over the urban area, the friction velocity and the turbulence components are larger. 2.2. Momentum: Urban Turbulence Because the rate of dispersion of a pollutant cloud is proportional to the turbulent velocity components (σ u, σ v, σ w ), investigators have expressed much interest in observations of these turbulent velocity components over urban, suburban,

478 BRITTER HANNA commercial, and industrial surfaces. The results of an extensive measurement program in the St. Louis, Missouri, area, as reported by Clarke et al. (1978) show that during the night the turbulence over the residential and commercial surfaces is approximately twice that over the rural surface. During the day the difference is less, approximately 20% or 30%. These diurnal differences are expected because at night the roughness obstacles not only generate more turbulence, but also force the atmosphere to be less stable. In addition at night human activities add heat to the atmosphere. Rotach (1995) and Roth (2000) have measured and reviewed, respectively, the turbulence above urban areas. The turbulent velocity components, scaled on a u calculated from the local height-dependent Reynolds stress, were approximately constant. Hanna & Britter (2002) confirm that this is consistent with σ u = 2.4u, σ v = 1.9u, and σ w = 1.3u, where u is based on the surface shear stress, usually assumed for the atmospheric boundary layer at heights above approximately 2H r. At lower heights measured in the roughness sublayer the turbulent velocity components decrease somewhat as the ground surface is approached. 2.3. Urban Surface-Energy Balance Figure 2 can be used to understand the surface-energy balance in an urban area. This surface (surface layer, actually) is warmed by the net radiation flux, Q, which is the sum of incoming short-wave solar radiation and the difference between incoming and outgoing long-wave radiation. The surface layer is cooled by the loss of heat due to the sensible heat flux, H s, that is positive upwards during the day, and a latent heat flux, H e, that is positive upwards when water is evaporating from the surface. These three fluxes may be defined at a height of approximately 2H r (i.e., at the top of the roughness sublayer). Oke (1987) and Grimmond & Oke (2002) define a heat-storage term, Q s, that describes the net change in heat within the surface layer. This change in heat applies to the layer consisting of the buildings and the air around them as well as to the ground surface down to a depth of approximately 1 or 2 m where there is insignificant diurnal change in temperature. A hysteresis in the heat-storage term, Q s, has been found in urban areas. Large positive values are noted in the morning. Negative values begin in early afternoon, and then Q s is equal to the net radiation flux, Q, from late evening through the rest of the night. The latter equality leads to the conclusion that the sensible heat flux, H s, equals zero; therefore, neutral (adiabatic) conditions prevail for most of the night in urban areas. A major result of the urban surface-energy balance is that urban boundary layers do not stabilize and cool down at night, which contributes to formation of the urban heat island. Sometimes the urban area is 10 C warmer than the surrounding area, although a more common magnitude of the urban heat island is 2 or 3 C. This warm urban core may cause a thermal circulation with winds tending to flow inward at the surface, forming a convergence zone. The heat island perturbations can be seen as far as 10 to 20 km outside of the city.

FLOW IN URBAN AREAS 479 As the air flows out of the urban area and over the surrounding rural area, the boundary layer at night will restabilize (cool) at the surface. Consequently, there may be a warm urban plume aloft (at heights of 100 to 200 m) over downwind rural surfaces. This urban plume may contain pollutants and is measurable up to 100 km downwind. 2.4. Urban Moisture Effects In the absence of irrigation, the urban area tends to be drier than its surroundings because of the pavement and buildings. This could lead to smaller latent heat flux, H e, and a larger sensible heat flux, H s. Oke (1987) discussed how cities in some arid environments (especially in the United States) can be moister than their surroundings because of the use of irrigation in urban parks, lawns, and other vegetation. Consequently, the latent heat flux can be larger than the sensible heat flux in irrigated cities. Changnon et al. (1977) summarized the results of the St. Louis Metromex field study, where the excess heat flux from urban areas was shown to combine with the convergence zone associated with an urban heat island to form updrafts. These upward vertical motions can lead to condensation and cloud formation, sometimes with precipitation. Cloud physicists and climatologists claim to have found significant increases in clouds and precipitation over and just downwind of urban areas. In some cases, the urban-enhanced clouds and precipitation do not occur until 10 or 20 km downwind of the urban area. 2.5. Urban Stability Several physical effects have been described above that all conspire to force the stability over urban areas toward neutral (adiabatic) conditions. For example, the fundamental stability parameter in Monin-Obukhov similarity theory is z/l, which is proportional to the sensible heat flux, H s, and inversely proportional to the cube of the friction velocity, u. Because H s is not overly enhanced over urban areas, and u can be much larger (because of the increased drag force due to the roughness obstacles) over urban areas, z/l is forced closer to zero by the dominant u 3 effect in the denominator of z/l. At night, the urban storage term causes the heat flux to be close to zero; hence, nearly neutral stability is assured. This assumption is made in several urban dispersion models. 2.6. Urban Dispersion Characteristics At the regional scale, the urban plume is observed to extend downwind of urban areas. A complex mix of pollutants is present, and there are likely to be chemical reactions and gas-to-particle conversions. The sides of the urban plume generally grow at a rate of approximately 0.5 m/s, and the maximum vertical extent of the plume is the daytime mixing depth, usually at a height of 500 1000 m. In the urban plume at night, the near-surface layer may be relatively pollutant free and stable over downwind rural surfaces, whereas the air aloft (heights of 100 m or

480 BRITTER HANNA more) may be well mixed and polluted. Observations by satellites, aircraft, and surface monitors have shown that the urban plume can sometimes be detected several hundred kilometers downwind of the urban area and may have a width of 100 or 200 km (White et al. 1983). At the city scale, where it is assumed that a pollutant plume extends vertically over a layer of depth at least 2H r, there is no need to account for specific effects around individual buildings. Consequently, dispersion can be calculated with standard approaches that apply for general boundary layers. Larger surface roughness produces larger z 0, u and turbulence levels, and these lead to greater dilution of a plume and reduced concentrations downwind (Hanna et al. 1982, Roberts et al. 1994). There is a hierarchy in complexity for mathematical models for treating dispersion in urban areas on the city scale. The simplest (and most commonly used in operational modeling) are of the Gaussian plume/puff type with empirical (urban category) correlations for the growth with distance of the plume dimensions. Less empiricism is apparent when the plume growth rates are determined from univeral functions that are based on the turbulence levels (and these can be related to u discussed earlier in this section) and estimates of the Lagrangian integral timescales (Hanna et al. 1982). Lagrangian stochastic dispersion models have been developed for urban situations. Rotach & de Haan (1997) reported on improved model performance when a more detailed description of the flow in the roughness sublayer is incorporated. Computational fluid dynamics models (i.e., those including turbulence closure models) abound. Many mesoscale flow models include a dispersion extension that may be attained through an Eulerian modeling approach or a Lagrangian stochastic approach, and these models or the modeling approach have been extended to the city scale (Schatzmann et al. 1997). However, there has been a lack of comprehensive experimental data for dispersion in urban areas, particularly from low-level sources (apart from routine air quality monitoring), and this has restricted the evaluation and development of dispersion modeling. Very recently several urban dispersion experiments have been undertaken. An experiment in Birmingham, United Kingdom, is discussed in Section 3; results from more comprehensive experiments in San Diego, California; Los Angeles, California; and Salt Lake City, Utah are awaited. A follow-up to the Salt Lake City experiment is planned for Oklahoma City in 2003. The results from these experiments will likely lead to a reassessment of dispersion modeling in urban areas. 3. THE NEIGHBORHOOD SCALE The neighborhood scale is a spatial scale of 1 2 km; a spatial scale at which a gross parameterization of the flow can be attempted and also a scale at which detailed computational study is feasible although, currently, at some expense. It is also a scale over which some statistical homogeneity may be anticipated; the city then is

FLOW IN URBAN AREAS 481 seen as being composed of these neighborhoods. Dispersion studies on this scale will likely require a more refined knowledge of the flow within and immediately above the urban canopy. This is of particular relevance when considering the consequences and risks associated with the accidental or deliberate release of hazardous materials within cities. There are two often-studied aspects of the flow within and near the urban canopy when viewed on the neighborhood scale. First, the flow is assumed to have a long fetch over a statistically homogeneous surface, and some quasi-equilibrium flow has been established. Second, the flow is assumed to have developed as a result of a change from one to another region (Smits & Wood 1985). Of course, real urban morphology will not provide such convenient distinctions, and some pragmatism is required when addressing real problems. 3.1. Flow Characteristics Observations from field and laboratory experiments have been used to develop an understanding of flow characteristics. In wind tunnels and water flumes, laboratory experiments allow for detailed flow measurement, the use of idealized urban areas consisting of simple geometric shapes in ordered arrays, as well as the modeling of real urban areas. Field measurements are far less common, difficult and expensive to perform, and provide limited data. They are, however, essential in providing data to ensure that the laboratory modeling has a sound basis. The flow over and through urban areas is characterized in Section 2; isolated, wake interference, and skimming are used as descriptive terms. Real urban areas often have large variations in building heights. Hall et al. (1996) noted the importance of height variability in inhibiting the skimming-flow regime. This suggests a fundamental difference between flows in an urban canopy and those in a typical vegetative canopy (Finnigan 2000). Raupach et al. (1980) and Rotach (1995) have highlighted the existence of a roughness sublayer within the atmospheric boundary layer below the inertial sublayer. The roughness sublayer is thought of as a region in which the underlying buildings lead to a spatial horizontal inhomogeneity of the flow. Raupach et al. (1980) provide a criterion based on obstacle height and spacing, whereas Rotach (1995), concerned particularly with cities, uses building height alone. For typical urban areas these approaches are similar. These, and other similar comments, have commonly been interpreted as a restriction that the roughness sublayer extends to approximately twice the average building height. The mean velocity profile in Figure 3 is, by implication, a horizontally spatially averaged profile. The profile well above the urban canopy is of a conventional logarithmic form. However, closer to the buildings, individual velocity profiles deviate as they respond directly to the real surface that produces the spatial inhomogeneity (see the wind tunnel simulation of a section of Nantes, France, by Kastner-Klein et al. 2000). The twice the average building height criterion may be too demanding when considering the variation of the horizontally spatially averaged profile from the

482 BRITTER HANNA logarithmic velocity profile, noted above, extrapolated downward. Hanna & Britter (2002) suggest that this profile will approximate the logarithmic form down to 1.5 times the average building height or even lower. Here approximate means that the error in estimating an advection wind speed for a dispersion model would be of no practical consequence. At the same time velocity measurements at these heights should not be used for profile fitting to obtain gross surface parameters such as u, z 0 and d (Snyder 1981). 3.2. Shear-Stress Profiles At positions away from a roughness change and for constant height obstacles the maximum shear stress occurs at approximately the height of the obstacles (Macdonald et al. 2000, Cheng & Castro 2002). Below this height the shear stress carried by the fluid decreases to zero as the buildings take up part of the stress through the drag forces on them. This is consistent with studies of vegetative canopies (Kaimal & Finnigan 1994). The shear stress approaches zero at the underlying surface for small λ p or λ f, although it approaches zero at elevation for large λ p or λ f. Above the obstacle height the shear stress decreases with height with some limited evidence of a constant shear-stress region. Finite fetch experiments make conclusions concerning a constant shear-stress region difficult. A change of roughness (from small to large roughness, say) viewed at the neighborhood scale appears as a high velocity flow impinging on an array of obstacles. This produces a large drag force on the most upstream obstacles (producing a large surface shear stress) and a divergence of the flow as it is turned vertically up and laterally out of the array. Further downstream the velocity within the array decreases to a level in some quasi-equilibrium with that above the urban canopy. This vertical deflection near the upper edge of the roughness change produces an elevated maximum shear stress quite distinct from that discussed above. A change of roughness from large to small produces a decline in the elevation of the maximum shear stress. Field measurements by Louka (1998) with an array of four canyon-type buildings placed the maximum shear stress close to the height of the buildings. In a wind tunnel study of an array of street canyons Brown et al. (2000) found similar results but noted that the maximum shear stress was larger and more confined in the first few rows; this shear stress then became smaller and more diffuse further downstream, a behavior consistent with such a roughness change. Rafailidis (1997) had observed much the same behavior in a similar study but with various roof geometries. Field (Rotach 1993a, 1993b, 1995; Oikawa & Meng 1995; Feigenwinter et al. 1999) and laboratory measurements with a model of a 400 m diameter section of central Nantes, France, (Kastner-Klein & Rotach 2001) in which the building heights vary produce somewhat different results. A maximum is evident in the shear stress although this occurs well above the average building height. For varied-height obstacles the maximum shear stress should occur at approximately the height of the highest obstacle, decreasing to zero shear stress in much the

FLOW IN URBAN AREAS 483 same manner as for constant-height obstacles. This statement relies on the dispersive stresses, those due to the inhomogeneity of the mean flow, being negligible, as demonstrated by Cheng & Castro (2002). The maximum shear stress should equate to the surface shear stress and determine the surface friction velocity. As a consequence the maximum shear stress occurs (well) above the mean height of the buildings. 3.3. Mean-Velocity Profiles Some disagreement exists in the literature over the form of the (spatially averaged) mean-velocity profile over an urban canopy or very rough surface. There is agreement that above the surface roughness layer the velocity profile has a conventional logarithmic form based on u, z 0 and d (for the neutrally stratified boundary layer). Below this level Raupach et al. (1980) showed clearly that the velocity is increased above that expected by extrapolation toward the surface of the logarithmic profile. However, other velocity profiles (Macdonald 2000) show the mean velocity to be decreased (below that given by the extrapolated logarithmic form) in the surface roughness layer at positions above the mean building height and increased at positions below the mean building height. This apparent inconsistency has arisen because of the difficulty of specifying u, z 0, and d from experimentally determined velocity profiles. All three cannot be determined with any accuracy from curve-fitting. The determination of u from direct measurement of the Reynolds stress is often made. In the work by Macdonald (2000) a further constraint was added. The fitted logarithmic velocity profile had to contain the same volume flux as the measured profile, an approach that requires there be regions of velocity excess and deficit (compared to the extrapolated logarithmic form). This problem observed with laboratory data will be even more apparent with field measurements. There is general agreement that the wind speeds are more uniform with height below rather than above the average building height, except for positions very close to the underlying surface where the velocity must decrease to zero. Cionco (1965) developed a model for the velocity profile within a vegetative canopy of constant height; this model has been extended to the urban canopy. Using a simple mixing-length turbulence model he analyzed the flow through a regular array of obstacles of constant cross-section and constant height. The wind velocity u(z), scaled on the velocity at the (constant) building height, had an exponential profile: u/u Hr = exp [ a(1 z/h r )]. (5) The profile requires specification of an empirical constant a. Macdonald (2000) observed exponential profiles in laboratory experiments with λ f between 0.05 and 0.20. He was also able to link the constant a to λ f directly with a = 9.6λ f. This approach is simple and describes the laboratory experiments well; however, it does rely on knowledge of the wind speed at the building height (the height at which the velocity profile is changing most rapidly), which may not be easily defined. The approach might easily be extended to a real urban canopy with a statistical

484 BRITTER HANNA description of the buildings, thereby reducing the velocity gradients in the vicinity of the average building. An alternative, even simpler, approach is to define a spatially and temporally averaged characteristic velocity within the urban canopy. Bentham & Britter (2002) showed, with some reasonable assumptions, that this can be related to u, λ f, and the average drag coefficient for the buildings C DB with U c /u = (C DB /2) 1/2 λ 1/2 f. (6) Comparison of this result with laboratory data produced good agreement with an assumed drag coefficient of unity. This approach is simple and direct, as is the exponential profile approach above, but for both it is unlikely that λ f is the only parameter required to describe the urban canopy. Laboratory data show that flat plates produce a larger surface stress than cubes with the same λ f, that cubes in a staggered array produce a larger surface stress than cubes in an aligned array with the same λ f and λ p, and that variable-height elements produce a larger surface stress than cubes with the same λ f and λ p (Macdonald et al. 2000, Cheng & Castro 2002). Macdonald (2000), using the exponential profile, and Hanna & Britter (2002), using the characteristic velocity approach, connected the in-canopy profile with the velocity profile above the surface roughness layer by direct extrapolation of the velocity profile above the surface roughness layer down through that layer. Both regard the deviations of the velocity profile within the surface roughness layer from the downward extrapolated profile as not being essential for simple modeling purposes. For the flow above the roughness elements the change of surface roughness is interpreted as the growth of an internal boundary layer. At lower levels near the roughness elements there are large alterations to the flow field at the upwind edge of the roughness change that then relax back to a nearly unchanging, fully developed velocity field downwind. Jerram et al. (1994), using laboratory and field data with cubes, showed that a distance corresponding to approximately five obstacle heights was necessary to approach a fully developed flow. Macdonald et al. (2000), also using cubes and with λ f = 0.16, found that the velocity field near to the roughness elements showed little further development beyond the first or second row of obstacles, a distance into the array of only two to four obstacle heights. The flow within the obstacle arrays adjusted more rapidly than the flow well above the array. On the basis of these limited experiments it is concluded that an approximately fully developed velocity field within the array will have been attained within a distance of approximately five obstacle heights into the array. 3.4. Turbulence Intensities Rotach (1995), Roth (2000), and Macdonald et al. (2000) showed that above the average building height the local turbulence intensities scale with the square root

FLOW IN URBAN AREAS 485 of the local kinematic Reynolds stress. The local Reynolds stress is not the same as the surface stress from which u is determined; however, they are comparable in the region near the top of the roughness elements or building heights. Below the average building height this scaling is less appropriate; the Reynolds stress can go to zero above the underlying surface, whereas the turbulence intensities do not. Scaling on the surface shear stress and surface friction velocity may be more appropriate. Only limited data is available: W.H. Snyder (unpublished), Macdonald et al. (2000), and Kastner-Klein et al. (2000) all provide results from laboratory experiments. The turbulence intensities vary (decrease) slowly with height inside the canopy, and spatially averaged (up to H r ) turbulence intensities may be sensibly defined. Snyder s results at a λ f of 0.027 and Macdonald s results at a λ f of 0.0625 and 0.16 are similar and average out to approximately σ u /u = 1.6, σ v /u = 1.4, and σ w /u = 1.1, respectively. Kastner-Klein et al. (2000) used a model of a central part of Nantes, France, with a large λ p of 0.6 0.7. Her results in Figure 4 show roughly uniform turbulence intensities within the canopy, and they can be approximated as σ u /u = 1.0, σ v /u = 1.0, σ w /u = 0.8. Broadly speaking, these results suggest that the turbulence levels may be assumed to be approximately uniform throughout the canopy, to scale on u, and to be less than that above the canopy. The turbulence levels, when scaled on u, appear to decrease with increasing λ f. If this slight decrease was ignored, then σ u, σ v, σ w /(characteristic canopy velocity U c ) must vary as (λ f /2) +1/2. This ratio of turbulence levels to advective velocity is an important variable in the dispersion of pollutants. The turbulence levels within the canopy may be estimated using a production-dissipation balance argument. The conclusions depend on the choice of length scale and averaging procedures; however, it can be argued that the above ratios should depend on (λ f /2) +1/2 in the isolated obstacle regime, (λ f /2) +1/3 in Figure 4 Measured vertical profiles of the turbulent velocity components (σ u, σ v, σ w ) at various positions within a wind-tunnel study in Nantes, France. Printed with permission from P. Kastner-Klein.

486 BRITTER HANNA the wake-interference regime, and (λ f /2) 0, i.e., a constant for the skimming-flow regime. This simple analysis is consistent with the observed decrease in σ u, σ v, σ w /u with increasing λ f. There will be an additional contribution to turbulent dispersion arising from the spatial variation of the mean velocity in the horizontal plane as a result of the buildings. Thus effective levels of turbulence within the canopy are comparable to those above the canopy. The discussion above has assumed that it is a flow through the canopy that generates the turbulence. This view is less appropriate for cities in which λ f or λ p are large. The turbulence may be more attributable to the interaction of the highspeed flow near the top of the urban canopy with the building tops and subsequent advection of this turbulence into the canopy. Similarly, turbulence may be generated by the shear layer atop a street canyon for a skimming flow and advected into the canyon. Another approach to studying the flow in urban areas is to consider exchange processes. The surface shear stress is a measure of the momentum exchange from the flow to the surface. The surface stress nondimensionalized with the fluid density and a reference mean wind speed (such as the geostrophic wind or the wind speed at a convenient reference height) becomes a surface-drag coefficient for meteorologists or a skin-friction coefficient for engineers. The two disciplines have slightly different definitions; engineers conventionally include a factor of one half in the denominator that is often omitted by the meteorologists. Note that the surfacedrag coefficient (and the reference height for which the nondimensionalizing wind speed is measured) can be directly related to z 0 and d. The aerodynamic conductance reported by Grimmond & Oke (1999) is an exchange velocity that transfers momentum at the reference level to the surface where the velocity is zero. Using the meteorologists definition of C D researchers note that the surface conductance is equal to C D u (z = z ref ) and the ratio of the surface conductance to the wind speed at the reference level is just C D. Grimmond & Oke (1999) estimate C D as 0.008 for residential and warehouse sites, 0.016 for a central site in Mexico city, and between 0.03 and 0.05 for a downtown site in Vancouver. This approach can be usefully extended to consider an exchange velocity between the in-canopy and the above-canopy flow, that is, an exchange velocity relevant for momentum transfer into and out of the canopy. Here we view the momentum exchange as being in two stages: an exchange between the reference level and the in-canopy flow and an exchange between the in-canopy flow and the surface (including the building surfaces). On this basis and choosing a reference level of 2.5H r the exchange velocity can be calculated exactly, and it is typically between 0.2 and 0.3u for a wide range of scenarios. This exchange velocity, based on momentum transfer, is also the exchange velocity for ventilating the canopy and for the convective exchange of heat, moisture, and pollutant between the canopy and the flow above. The drag coefficient C D is not the same as the equivalent coefficients for heat or moisture because the transfer processes from the flow to the surfaces are quite different for momentum and heat or moisture. For the same reason z 0 calculated for temperature or moisture profiles

FLOW IN URBAN AREAS 487 can be several orders of magnitude different from the z 0 for the velocity profile (Voogt & Grimmond 2000). 3.5. Dispersion Characteristics It was shown in Section 2 that an increase in surface roughness produces a significant increase in turbulence levels. These cause a reduction in ground-level concentrations that arise from ground-level releases for situations where the pollutant cloud depth is much larger than the height of the surface obstacles. It is less clear whether the same conclusion can be drawn when the pollutant cloud is of a depth comparable to or smaller than the height of the surface obstacles. Laboratory experiments (Davidson et al. 1995; Macdonald et al. 1997, 1998) showed that a conventional Gaussian plume model provides an appropriate structure for the problem. The increased turbulence levels within the urban canopy (over those occurring in the absence of obstacles) produce larger dispersion coefficients that tend to reduce concentrations. However, the accompanying reduction in the advection velocity within the canopy tends to increase concentrations. The relative magnitudes of these opposing effects determine whether the obstacles lead to increased or decreased concentrations as the roughness is increased. The Kit Fox field experiments (Hanna & Chang 2001) showed clear evidence of substantial reductions in ground-level concentrations from a ground-level source for conditions in which the plume centroid was comparable to or smaller than the obstacle heights. The reductions were a factor of approximately 3 as z 0 changed from 0.002 m to 0.02 m and were a further factor of approximately 3 as z 0 changed from 0.02 m to 0.2 m. These experiments used a dense gas, carbon dioxide, as the pollutant; however, the preceding statement is based only on experiments at the highest wind velocities when the dense gas effects would be minimal. Other small-scale field experiments (Davidson et al. 1995, Macdonald et al. 1997) and related laboratory experiments (Davidson et al. 1996, Hall et al. 1996) found that the obstacle arrays produced little effect on the maximum ground-level concentrations. The results noted above for the advection velocity and the turbulence levels allow for an estimate of the effect of increasing surface roughness on the groundlevel concentrations that arise from a ground-level source within the urban canopy. If we use the turbulence levels as surrogates for the plume growth rates and the characteristic in-canopy velocity as the advection velocity, it is straightforward to show that the ground-level concentration should be proportional to (λ f /2) 1/2 in the isolated-building regime, to (λ f /2) 1/6 in the wake-interfering regime, and to (λ f /2) +1/2 in the skimming-flow regime. Thus, increasing surface roughness initially tends to decrease the ground-level concentration that will go through a minimum and then increase, possibly returning to become equal or larger than that for an unobstructed case. Dispersion-field experiments from low-level sources in real cities are rare. The St. Louis experiments (McElroy & Pooler 1968) were conducted with low-level

488 BRITTER HANNA Figure 5 The concentration-time history at a receptor 1 km downwind of a shortduration tracer release in central Birmingham, United Kingdom. Two tracers, PMCH and PMCP, were used. sources. A study in Birmingham, United Kingdom, (Britter et al. 2002) was only able to run three experiments owing to resource constraints. The recent and planned dispersion-field studies referred to in Section 2 are intended to provide data on the neighborhood scale in particular. Figure 5 is a concentration-time history for a receptor in central Birmingham 1 km away from the source. The release was a 20-min finite-duration release. The concentration history shows a time delay consistent with an estimated advection velocity and a rapid rise in concentration. There is also some evidence for a plateau as the release behaves like a continuous plume and then a slow decay. A time constant of 4 min can be determined from the early part of the decay; this is an obvious result of pollutant trapped in recirculating regions among the buildings. It should be possible to relate the time constant to the morphological characteristics of the surface even though this has not been done. In addition there appears to be another, much longer, time constant as the concentration approaches the background level. This may be due to pollutant that is taken into buildings and released on a building ventilation timescale. Model development at the neighborhood scale has proceeded along several diverse lines. Jerram et al. (1994) developed a linearized perturbation model with a force distribution representing the drag on the buildings. Theurer (1995) based a Gaussian plume model on extensive wind-tunnel data; this work allowed for the plume direction to be influenced by the urban layout. Kaplan & Dinar (1996)

FLOW IN URBAN AREAS 489 derived a mass-consistent wind model accounting for flow recirculation behind buildings and combined this with a Lagrangian particle model for dispersion. The extensive laboratory experiments by Hall et al. (1996) have been used to develop a simple urban dispersion model (Hall et al. 1997) based principally on data correlations. Similarly the arguments in Hanna & Britter (2002) have been recast into another simple urban dispersion model. Computational techniques of much greater complexity including large eddy simulation [called LES only recently (Boris 2002)] have been used. It is only now that data that allow formal scientific evaluation of such models are becoming available. At present there has been only limited effort (e.g., Soulhac et al. 2002) to use advanced computational techniques to develop a general understanding of the flow and to give guidance as to how best to view and parameterize the flow as distinct from a direct application to specific scenarios. 4. THE STREET SCALE The street (canyon) scale is particularly studied in the context of urban air quality where the dominant source of pollution, vehicle emissions, are in close proximity to the pollutant receptors of concern, people (and the regulatory pollutant-monitoring stations). In cities both the source and the receptor are within a very confined geometry, the street canyon, and this confined geometry exhibits a sheltering effect from the diluting influence of the wind. Flow and dispersion near street intersections are also of interest as it is the acceleration of vehicles away from traffic lights and/or pedestrian crossings that gives rise to large pollutant emission rates and consequently poor urban air quality. The location of building ventilation intakes is sensitive to the anticipated distribution of pollutants on the spatial scale of the buildings or the street. 4.1. Flow and Dispersion The archetypal street-canyon flow is basically that of a turbulent shear flow above a rectangular cavity with the mean flow direction perpendicular to the axis of the street canyon. For a canyon of near unity aspect ratio, and provided that the rooflevel wind speed exceeds 1.5 2 m/s, a recirculating flow is set up in the canyon driven by momentum transport from the shear layer above. The flow at the bottom of the canyon is then upwind toward the lee wall of the canyon. Observations in the laboratory (see Hosker 1985) and the field (Nakamura & Oke 1988) support this structure. Typically the recirculating velocities relative to the roof-level wind speed are of the order of 1/3 1/2, and the turbulence levels are of the order of 1/10. The recirculating flow is neither steady nor symmetric. The asymmetric flow appears as a stronger, more concentrated downflow on the downwind wall and as a weaker, more extensive upflow on the lee wall. There are also variations on this archetypal flow as the geometry and flow directions vary. For example, regular rectangular street canyons with a small height

490 BRITTER HANNA to width ratio allow for possible reattachment of the separating shear layer off the lee wall to the floor of the street canyon. For large aspect ratios there can be a counter-rotating vortex below the main recirculation flow. Similarly and even for aspect ratios near unity there can be a counter-rotating vortex in the corner between the lee wall and the floor. However, these latter possibilities are more common in idealized laboratory experiments rather than in field experiments. Irregular street canyons with the lee wall and the downwind wall having different elevations introduce further complications to the flow, but the resulting flows are generally qualitatively consistent with expectations (Pavageau et al. 2001). When the wind flow direction is not perpendicular to the street axis, a recirculating flow is still present as is an alongstreet flow. The flow in the street canyon can also depend on the characteristics of the flow above, e.g., whether it is a fully developed rough wall boundary layer, what turbulence scales are present, or whether the street canyon is one of several canyons in parallel. These considerations might reflect an overly detailed view of a complex, essentially inhomogeneous problem and might be based on a view of the problem that is too idealized to ever be relevant to real situations. Recent work has been done on how thermal effects will influence the flow. Can street canyon walls heated by the sun influence the mechanically driven recirculating motion? Although laboratory flows and computational studies do show an effect, this seems unlikely to be operationally important in most scenarios. Any effect is far less evident in field measurements (Louka et al. 2002) probably because, in the field, the physical width of the free convective boundary layer on the heated wall is small compared with the scale of the mechanically driven motion, which is on the street scale. Of course when the external wind is very weak these thermal effects are of greater consequence. Thus the basic flow in a street canyon is a recirculating flow filling the canyon unless the canyon has a large aspect ratio, in which case the recirculating flow may not reach to the floor. Another basic aspect of the flow is an exchange of air between the street canyon and the flow above. This might be estimated assuming that a free shear layer separates the street canyon from the flow above (Soulhac et al. 2002; F. Caton, personal communication). Surprisingly few experiments, in the laboratory or in the field, direct attention to the exchange mechanism, to exchange fluxes, or to determining an exchange velocity (the velocity that is ventilating the street canyon) (but see Louka et al. 2000, Robins et al. 2001). These two aspects are the principal phenomena that require inclusion in operational models of street-scale pollution from vehicles (Yamartino & Wiegand 1986, Berkowicz 2000). Laboratory experiments can be used to estimate the magnitude of adjustable parameters; however, tuning and testing of the models is more commonly done with field experiments by using roadside monitoring stations and extensive traffic data (Lohmeyer et al. 2002). Tracer dispersion measurements in wind tunnels (e.g., Hoydysh & Dabberdt 1988, Dabberdt & Hoydysh 1991) are qualitatively consistent with expectations. A low-level pollutant source produces a vertical concentration profile that is

FLOW IN URBAN AREAS 491 approximately exponential. For wind flows perpendicular to the street axis the lee-side concentration is larger than the windward side by a factor of 2. This is true except for a step-down configuration when the windward side may have the larger concentration. Maximum concentrations occur for wind directions perpendicular and parallel to the street axis with a shallow minimum between. Studies similar to those for street canyons have also been undertaken for more complex intersections where pollutant emission rates are likely to be relatively high (Robins et al. 2001). Street-scale problems are amenable to computational fluid mechanics, and many such studies may be found in the literature. They typically produce reasonable qualitative results, but the performance, when compared with laboratory or field experiments, is little better than the simple operational models described above. Of course the solution using computational fluid dynamics allows for the treatment of more complicated geometrical arrangements. 4.2. Traffic-Produced Turbulence Poor urban air quality is associated with low wind-speed conditions; however, operational urban air quality models commonly perform poorly under these conditions. Typically the models are structured so as to have an inverse relationship between the concentration field and an externally imposed wind speed, and this relationship leads to substantial, and critical, concentration overestimation of models at low wind speeds. Other sources of turbulence are thermal production from the environment or from vehicles, as well as mechanical production from the motion of vehicles, also known as traffic produced turbulence (TPT). Laboratory measurements of the turbulence produced by modeled traffic in a street canyon (Kastner-Klein 1999) showed that TPT can be a significant source of turbulence when compared with that arising from the mean wind. There is an asymmetry of the distributions that reflect the vortex-like structure within the street canyon owing to the external wind. This asymmetry pushes the added turbulence toward the lee wall of the canyon. Vachon et al. (2002) described generally the same influences of traffic motions on the turbulence field in an urban street canyon. The measurements were done on a street in Nantes, France. In the lower part of the street canyon increased levels of turbulent kinetic energy, which could be attributed to turbulence created by vehicle motions, have been found. Close to the traffic region turbulence enhancement has been observed on the leeward and windward side of the street canyon. However, on the leeward side the influence has been more pronounced, and the vertical extent of the region with increased turbulence levels has been much larger than on the windward side. This indicates once again that the advection of turbulence created in the traffic layer toward the leeward wall is due to the wind-induced street-canyon vortex. The magnitude of, or at least the appropriate scaling for, the TPT may be estimated from a production-dissipation balance for turbulent kinetic energy.

492 BRITTER HANNA DiSabatino et al. (2002) considered three regimes: isolated vehicles, wake-interfering vehicles, and congested traffic (analogous to the isolated, wake-interference, and skimming-flow regimes for buildings) to produce estimates of TPT. The turbulence levels depend on the traffic density to the 1/2, 1/3, and zero power as the traffic density increases. When the laboratory results by Kastner-Klein (1999) were scaled with the (appropriate) wake-interfering correlation, the leading coefficient was found to be close to unity. In Kastner-Klein et al. (2002) these results were tested in full-scale observations in Gottingerstrasse, Hanover, Germany, where street-side measurements of NO 2 were available over several years. The measurements were shortterm, that is, half-hour, averaged. In contrast to the predictions in Figure 1 of annual mean concentrations other regulations are noted in terms of the statistics of short-term-averaged measurements. Extreme events are likely to occur when the ambient wind speed is low, and models that ignore TPT will significantly overpredict the magnitude of these events. The incorporation of TPT into an operational model led to a marked improvement in the model performance, in particular for the extreme events. 5. PROBLEMS AND PROGRESS The literature addressing flow and dispersion in urban areas is spread over many disciplines (meteorology, engineering, geography, and others), with each having a fundamental and an operational aspect. In preparing this review it became apparent that there was no clear coherent framework within which the study of the urban area was taking place, which reflects the various disciplines participating, their different goals, and the complexity and essential heterogeneity of the problem. Attention to the development of a common, accepted framework would likely be rewarded. The sensitivity of operational modeling procedures to the uncertainty in the input variables or, possibly, the conceptual underpinning of the procedures is often unclear. This may be due to the lack of data, particularly in the field; however, there is currently much progress in this area, and more formalized model evaluation procedures are being developed. These will also use the extensive laboratory databases that have been recently constructed by D.J. Hall, R.W. Macdonald, M. Schatzmann, and others. In a fluid-mechanical context the most pressing problems include the treatment of atmospheric stability in urban areas, the specification of reference variables (e.g., the wind speed well above the urban area, the wind speed just above or at the average building height, or the wind speed within the urban canopy), and the treatment of arbitrary spatial variations in surface roughness. For dispersion studies it is still unclear how best to address the neighborhood scale and its connections with the street and the city scale, particularly when addressing transient problems.

FLOW IN URBAN AREAS 493 ACKNOWLEDGMENTS The authors thank M.K. Neophytou and S. DiSabatino for assistance in the preparation of this manuscript. Figure 1 was kindly provided by CERC Ltd., Cambridge, United Kingdom. The Annual Review of Fluid Mechanics is online at http://fluid.annualreviews.org LITERATURE CITED Auer AH. 1978. Correlation of land use and cover with meteorological anomalies. J. Appl. Meteorol. 17:636 43 Bentham T, Britter RE. 2002. Spatially averaged flow velocity within large groups of obstacles. Atmos. Environ. Submitted Berkowicz R. 2000. OSPM a parameterised street pollution model. J. Environ. Monit. Assess. 65:323 31 Boris J. 2002. The threat of chemical and biological terrorism. Comput. Sci. Eng. Mar/Apr: 22 32 Bornstein R. 1987. Mean diurnal circulation and thermodynamic evolution of urban boundary layers. In Modeling the Urban Boundary Layer. Boston, MA: Am. Meteorol. Soc. Bottema M. 1997. Urban roughness modeling in relation to pollutant dispersal. Atmos. Environ. 31:3059 75 Britter RE, Disabatino S, Caton F, Cooke K, Simmonds P, Nickless G. 2002. Results from three field tracer experiments at the neighborhood scale in the city of Birmingham, UK. Water Air Soil Pollut.-Focus. In press Brown MJ. 2000. Urban parameterizations for mesoscale meteorological models. In Mesoscale Atmospheric Dispersion, ed. Z. Boybeyi, Chapter 5, pp. 193 255. Southhampton, UK: WIT Press Brown MJ, Lawson RE, Descroix DS, Lee RL. 2000. Mean flow and turbulence measurements around a-d array of buildings in a wind tunnel. Conf. Appl. Air Pollut. Meteorol., 11th, Long Beach, Calif., 2000. Boston, MA: Am. Meteorol. Soc. Changnon SA, Huff FA, Schickedanz PT, Vogel JL. 1977. Summary of METROMEX Vol 1: weather anomalies and impacts. ISWS B-62, Illinois State Water Survey. 260 pp. Cheng H, Castro IP. 2001. Near-wall flow development after a step change in surface roughness. Bound.-Layer Meteorol. 6:1 21 Cheng H, Castro IP. 2002. Near-wall flow over urban-like roughness. Bound.-Layer Meteorol. In press Cionco RM. 1965. Mathematical model for air flow in a vegetative canopy. J. Appl. Meteorol. 4:517 22 Clarke JF, Ching JKS, Binkowski FS, Godowitch JM. 1978. Turbulent structure of the urban surface boundary layer. Proc. Int. Tech. Meet. Air Pollut. Model. Its Appl., 9th, NATO/CCMS. New York/London: Plenum Cook N. 1990. The Designers Guide to Wind Loading of Building Structures Part 2 Static Structures. London: Butterworth Publ. Dabberdt WF, Hoydysh WG. 1991. Street canyon dispersion: sensitivity to block shape and entrainment. Atmos. Environ. 25A:1143 53 Davenport AG, Grimmond CSB, Oke TR, Weiranga J. 2000. Estimating the roughness of cities and scattered country. Conf. Appl. Climatol., 12th, Asheville, NC, 2000, pp. 96 99. Boston, MA: Am. Meteorol. Soc. Davidson MJ, Mylne KR, Jones CD, Phillips JC, Perkins RJ, et al. 1995. Plume dispersion through large groups of obstacles a field investigation. Atmos. Environ. 29:3245 56 Davidson MJ, Snyder WH, Lawson RE, Hunt JCR. 1996. Plume dispersion from point sources upwind of groups of obstacles wind tunnel simulations. Atmos. Environ. 30: 3715 25

494 BRITTER HANNA DiSabatino S, Kastner-Klein P, Berkowicz R, Britter R, Fedorovich E. 2002. The modeling of turbulence from traffic in urban dispersion models Part I: theoretical considerations. J. Environ. Fluid Mech. In press Feigenwinter C, Vogt R, Parlow E. 1999. Vertical structure of selected turbulence characteristics above an urban canopy. Theor. Appl. Climatol. 62:51 63 Fenger J, Hertel O, Palmgren F. 1999. Urban Air Pollution European Aspects. London: Elsevier Finnigan J. 2000. Turbulence in plant canopies. Annu. Rev. Fluid Mech. 32:519 71 Goode K, Belcher SE. 1999. On the parameterisation of the effective roughness length for momentum transfer over heterogeneous terrain. Bound.-Layer Meteorol. 93:133 54 Grimmond CSB, Oke TR. 1999. Aerodynamic properties of urban areas derived from analysis of surface form. J. Appl. Meteorol. 38:1261 92 Grimmond CSB, Oke TR. 2002. Turbulent heat fluxes in urban areas: observations and a local-scale urban meteorological parameterisation scheme. J. Appl. Meteorol. In press Hall DJ, Macdonald R, Walker S, Spanton AM. 1996. Measurement of dispersion within simulated urban arrays a small scale wind tunnel study. BRE Client Rep. 178/96, Build. Res. Establ., Garston, Watford, UK Hall DJ, Spanton AM, Macdonald R, Walker S. 1997. A simple model for estimating dispersion in urban areas. BRE Client Rep. 169/97, Build. Res. Establ., Garston, Watford, UK. 91 pp. Hanna SR, Briggs GA, Hosker RP. 1982. Handbook on Atmospheric Diffusion. DOE/TIC- 11223 (DE82 002045). Springfield, VA: NTIS/USDOC Hanna SR, Britter RE. 2002. Wind Flow and Vapor Cloud Dispersion at Industrial Sites. New York: Am. Inst. Chem. Eng. Hanna SR, Chang JC. 1992. Boundary layer parameterizations for applied dispersion modeling over urban areas. Bound.-Layer Meteorol. 58:229 59 Hanna SR, Chang JC. 2001. Use of the Kit Fox field data to analyze dense gas dispersion modeling issues. Atmos. Environ. 35:2231 41 Hanna SR, Ramsdell J, Cramer H. 1987. Urban Gaussian diffusion parameters. Modeling the Urban Boundary Layer. Boston, MA: Am. Meteorol. Soc. Hosker RP. 1985. Flow around isolated structures and building clusters: a review. ASHRAE Trans. 91:1671 92 Hoydysh WG, Dabberdt WF. 1988. Kinematics and dispersion characteristics of flows in asymmetric street canyons. Atmos. Environ. 22:2677 89 Jerram N, Perkins RJ, Fung JCH, Davidson MJ, Belcher SE, Hunt JCR. 1994. Atmospheric flow through groups of buildings and dispersion from localized sources. Proc. NATO Adv. Study Inst., Karlsruhe, ed. EJ Plate, JE Cermak. Dordrecht: Kluwer Kaimal JC, Finnigan JJ. 1994. Atmospheric Boundary Layer Flows: Their Structure and Measurement. Oxford, UK: Oxford Univ. Press Kaplan H, Dinar N. 1996. A Lagrangian dispersion model for calculating concentration distribution within a built-up domain. Atmos. Environ. 24:4197 207 Kastner-Klein P. 1999. Experimentelle Untersuchung der strömungsmechanischen Transportvorgänge in Strassenschluchten. PhD thesis. Inst. Hydromech. Univ., Karlsruhe, Ger. Kastner-Klein P, Ketzel M, Berkowicz R, Fedorovich E, Britter R. 2002. The modeling of turbulence from traffic in urban dispersion models Part II: evaluation based on laboratory and full-scale concentration measurements in street canyons. J. Environ. Fluid Mech. In press Kastner-Klein P, Rotach M. 2001. Parameterization of wind and turbulent shear stress profiles in the urban roughness sublayer. Proc. Int. Conf. Urban Air Qual., 3rd, Loutraki, Greece, 2001. London: Inst. Phys. Kastner-Klein P, Rotach M, Fedorovich E. 2000. Experimental study on mean flow and turbulence characteristics in an urban

FLOW IN URBAN AREAS 495 roughness sublayer. 14th Symp. Bound. Layer Turbul., Aspen, CO, 2000. Boston, MA: Am. Meteorol. Soc. Lohmeyer A, Mueller WJ, Baechlin W. 2002. A comparison of street canyon predictions by different modelers: final results now available from the Podbi-exercise. Atmos. Environ. 36:157 58 Louka P. 1998. Measurements of airflow in an urban environment. PhD thesis. Univ. Reading, UK Louka P, Belcher SE, Harrison RG. 2000. Coupling between airflow in streets and the welldeveloped boundary layer aloft. Atmos. Environ. 34:2613 21 Louka P, Vachon G, Sini JF, Mestayer PG, Rosant JM. 2002. Thermal effects on the airflow in a street canyon Nantes 99 experimental results and model simulation. Water Air Soil Pollut.-Focus. In press Macdonald RW. 2000. Modelling the mean velocity profile in the urban canopy layer. Bound.-Layer Meteorol. 97:25 45 Macdonald RW, Carter S, Slawson PR. 2000. Measurements of mean velocity and turbulence statistics in simple obstacle arrays at 1:200 scale. Therm. Fluids Rep. 2000-1, Therm. Dep. Mech. Eng., Univ. Waterloo, Can. Macdonald RW, Griffiths RF, Cheah SC. 1997. Field experiments of dispersion through regular arrays of cubic structures. Atmos. Environ. 31:783 95 Macdonald R, Griffiths RF, Hall DJ. 1998. A comparison of results from scaled field and wind tunnel modelling of dispersion in arrays of obstacles. Atmos. Environ. 32:3845 62 McElroy JL, Pooler F. 1968. St. Louis Dispersion Study Volume II Analysis. Arlington, VA: US Dep. HEW Nakamura Y, Oke TR. 1988. Wind, temperature and stability conditions in an E-W oriented canyon. Atmos. Environ. 22:2691 700 Oikawa S, Meng Y. 1995. Turbulence characteristics and organized motion in a suburban roughness layer. Bound.-Layer Meteorol. 74:289 312 Oke TR. 1987. Boundary Layer Climates. London, UK: Routledge Pavageau M, Rafailidis S, Schatzmann M. 2001. A comprehensive experimental databank for the verification of urban car emission dispersion models. Int. J. Environ. Pollut. 15:417 25 Rafailidis S. 1997. Influence of building, areal density and roof shape on the wind characteristics above a town. Bound.-Layer Meteorol. 85:255 71 Ratti CS, Disabatino S, Britter R, Brown M, Caton F, Burian S. 2002. Analysis of urban databases with respect to pollutant dispersion for a number of European and American cities. Water Air Soil Pollut.-Focus. In press Raupach MR, Thom AS, Edwards I. 1980. A wind-tunnel study of turbulent flow close to regularly arrayed rough surfaces. Bound.- Layer Meteorol. 18:373 97 Roberts PT, Fryer-Taylor RE, Hall DJ. 1994. Wind-tunnel studies of roughness effects on gas dispersion. Atmos. Environ. 28:1861 70 Robins AG, Scaperdas A, Grigoriadis D. 2001. Spatial variability and source-receptor relations at a street intersection. 2001. Water Air Soil Pollut.-Focus. In press Rotach MW. 1993a. Turbulence close to a rough urban surface. Part I: Reynolds stress. Bound.-Layer Meteorol. 65:1 28 Rotach MW. 1993b. Turbulence close to a rough urban surface. Part II: variances and gradients. Bound.-Layer Meteorol. 66:75 92 Rotach M. 1995. Profiles of turbulence statistics in and above an urban street canyon. Atmos. Environ. 29:1473 86 Rotach M, de Haan P. 1997. On the urban aspect of the Copenhagen dataset. Int. J. Environ. Pollut. 8, 3 6:279 86 Roth M. 2000. Review of atmospheric turbulence over cities. Q. J. R. Meteorol. Soc. 126:941 90 Schatzmann M, Rafailidias S, Britter R, Arend M. 1997. Database, Monitoring and Modelling of Urban Air Pollution: Inventory of Models and Datasets. Luxembourg: Eur. Comm. 109 pp.

496 BRITTER HANNA Smits AJ, Wood DH. 1985. The response of turbulent boundary layers to sudden perturbation. Annu. Rev. Fluid Mech. 17:321 58 Snyder WH. 1981. Guidelines for fluid modeling of atmospheric diffusion. Rep. EPA- 600/8-81-0009, EPA, Res., Triangle Park, NC Soulhac L, Mejean P, Perkins RJ. 2002. Modeling transport and dispersion in a streetcanyon. Int. J. Environ. Pollut. In press Stull RB. 1997. An Introduction to Boundary Layer Meteorology. Dordrecht: Kluwer Atmos. Publ. 670 pp. Theurer W. 1995. Point sources in urban areas: modeling of neutral gas clouds with semiempirical models. Wind Climate in Cities, pp. 485 502. Dordrecht: Kluwer Theurer W. 1999. Typical building arrangements for urban air pollution modelling. Atmos. Environ. 33:4057 66 Vachon G, Louka P, Rosant J-M, Mestayer P, Sini J-F. 2002. Measurements of trafficinduced turbulence within a street canyon during the Nantes 1999 experiment. Water Air Soil Pollut.-Focus. In press Voogt JA, Grimmond CSB. 2000. Modeling surface sensible heat flux using surface radiative temperatures in a simple urban area. J. Appl. Meteorol. 39:1679 99 White WH, Patterson DR, Wilson WE. 1983. Urban export to the nonurban troposphere: results from Project MISTT. J. Geophys. Res. 88:10745 52 Yamartino RJ, Wiegand G. 1986. Development of simple models for the flow, turbulence and pollutant concentration fields within an urban street canyon. Atmos. Environ. 20:2137 56

Figure 1 Model predictions of annual mean NO 2 concentrations for 2005 in Greater London.