Driving physical mechanisms of flow and dispersion in urban canopies

Transcription

1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 27: (2007) Published online 11 September 2007 in Wiley InterScience ( Driving physical mechanisms of flow and dispersion in urban canopies P. Klein, a * B. Leitl b and M. Schatzmann b a School of Meteorology, University of Oklahoma, 120 David L. Boren Blvd., Norman OK, 73072, USA b Meteorological Institute, Center for Marine and Atmospheric Science (ZMAW), Environmental Wind Tunnel Laboratory, University of Hamburg, Hamburg, Germany Abstract: The paper summarizes results from recent full-scale and wind-tunnel studies and discusses the complexity of urban canopy layer (UCL) flow and related challenges for urban modeling. Wind-tunnel data for idealized street-canyon intersections demonstrate that street-level flow and dispersion patterns are significantly altered for configurations with non-uniform building heights. For most wind directions, concentration maxima were upto a factor of 2 higher for cases with taller buildings near the intersection. Enhanced vertical downward mixing in the wakes of the taller buildings apparently causes higher street-level winds close by, but simultaneously, other regions become sheltered from active mixing and poor ventilation. The analysis of Joint Urban 2003 (JU2003) full-scale data also pointed out that high-momentum fluid can be effectively mixed down in the wakes of high-rise buildings, and that UCL flow is predominantly dynamically driven by roof-level winds. Building-height variability is thus, a key factor for UCL dynamics and mixing. Additionally, a linear model, which assumes that the along- and across-canyon velocity components are directly proportional to their above roof-level counterparts, has then been tested. The complexity of UCL flow patterns cannot be captured by such a simple model, but it may be useful for estimating street-level winds. For the along-canyon component, more than 70% of the calculated values were within a factor of 2 of the measured values, while for the across-canyon component less than 50% of the predictions agreed within a factor of 2. The practical applicability of the model was further assessed using JU2003 wind-tunnel data and a comparison between JU2003 wind-tunnel and full-scale flow profiles is also presented. It becomes obvious that UCL flow patterns are significantly altered by changes in upwind wind direction of less than 10. While the field and laboratory results are qualitatively in good agreement, noticeable differences exist at certain sites. Copyright 2007 Royal Meteorological Society KEY WORDS urban meteorology; street-canyon flow; atmospheric dispersion; wind-tunnel modeling Received 18 October 2006; Revised 22 May 2007; Accepted 25 May 2007 INTRODUCTION Owing to an ongoing trend of urbanization, the majority of the world s population lives in cities (Brockerhoff, 2000; United Nations, 2004), and urban areas face a number of serious environmental challenges such as flooding, air quality management, heat-waves or emergency response to accidents with release of toxic materials (Dabberdt et al., 2000). More recently, the possibility that cities become targets for terrorist attacks with releases of chemical, biological or radioactive agents into an urban area has also become a major concern (Britter and Hanna, 2003). The physical and chemical processes in the urban atmosphere are thus of great concern, but their complex nature still poses a challenge to the research community (Dabberdt et al., 2000). In particular, emergencyresponse forecasting is extremely challenging, because it requires models that capture essential features of urban * Correspondence to: P. Klein, School of Meteorology, University of Oklahoma, 120 David L. Boren Blvd., Norman, OK 73072, USA. pkklein@ou.edu flow and dispersion processes, and at the same time provide fast-exposure predictions (Brown, 2004 and NRC, 2003). Recognizing the critical role of accurate information about the atmospheric conditions for predictions of transport, diffusion and removal of atmospheric pollutants, the 11th Prospectus Development Team of the U.S. Weather Research Program (Dabberdt et al., 2004) stated Improving atmospheric forecasts to provide improved air quality forecasts suitable for decision makers and the public is a major challenge. As the key technical problem for urban areas, they identified the specification of vertical wind, turbulence, and temperature profiles, both below and above the urban canopy. Britter and Hanna (2003) provide a good overview of the different spatial scales relevant for urban flow and dispersion patterns as well as the different sub-layers of the urban boundary layer (UBL). The urban canopy layer (UCL) corresponds to the lowest part of the UBL where the flow at a specific point is directly affected by obstacles in its near surroundings. The UCL is also the portion of the UBL where effects of urbanization on atmospheric Copyright 2007 Royal Meteorological Society

2 1888 P. KLEIN ET AL. processes are most significant and, in turn, city dwellers are most directly affected by urban weather phenomena and environmental problems. As an example, UCL flow and its interactions with the UBL strongly affect air quality in cities, where most air pollutants result from traffic and industrial processes, and are emitted within the UCL. Previous UCL studies have often focused on flow, turbulence and dispersion phenomena in typical urban building configurations (Kastner-Klein et al., 1997 and Theurer, 1999), with a number of studies concentrating on arrays of cubical buildings (e.g. Cheng and Castro, 2002a,b and Macdonald et al., 2002). Of particular interest have also been street canyons that are characterized by long buildings flanking narrow streets, for which the flow and mixing inside the street is driven by a recirculationtype, quasi-2d vortex (see e.g. Vardoulakis et al., 2003). Such vortices, with a rotation axis parallel to the street, have been observed for narrow canyons under perpendicular approach flow, particularly in wind-tunnel studies conducted with buildings of simple and regular geometries (Brown et al., 2000 and Kastner-Klein et al., 2001). The dispersion patterns inside street canyons are then typically characterized by exponentially decreasing vertical concentration profiles and significantly larger values on the leeward than on the windward building walls (Dabberdt and Hoydysh, 1991; Meroney et al., 1996 and Kastner-Klein and Plate, 1999a). However, more recently Kastner-Klein et al. (2004) have shown that vortices developing at the lateral building edges can extend over significant portions of street canyons, and accordingly, 2D flow regimes are not necessarily typical for shorter street canyons. These studies have also shown that small-scale features of building architecture, such as different roof shapes, can significantly alter the UCL flow dynamics. For example, the flow within the canyon can become rather stagnant for so-called step-down configurations with a taller building on the upwind than on the downwind canyon side. In this case, the recirculation developing on the roof of the upwind building does not fully penetrate into the canyon but rather shelters the exchange between the canyon and above roof-level winds which often causes increased street-level pollution and a shift of the concentration maxima to the windward canyon side (Dabberdt and Hoydysh, 1991 and Kastner- Klein and Plate, 1999a). The variability of building heights in general, appears to be a key factor regarding the development of typical skimming flow types with limited UCL UBL interactions. Particularly, in central business districts (CBDs) of US cities, wakes developing downwind of high-rise buildings can cause rapid vertical mixing and dominate the UCL UBL interactions (Heist et al., 2004). The formation of typical street canyon vortices becomes, thus, questionable in complex urban settings with highly variable building geometries and fluctuating meteorological conditions. New insights on the nature of UCL flow and dispersion patterns can be expected from the data collected during Joint Urban 2003 (JU2003); a major urban tracer experiment, which took place in July 2003 in Oklahoma city, USA (Allwine, 2004). Two important, unprecedented components of JU2003 were (1) a street canyon sub-experiment for which more than 40 3D-sonics were continuously operated inside a relatively narrow downtown street Park Avenue (PA) and (2) extensive, concurrent wind-tunnel measurements at the Environmental Wind Tunnel Laboratory (EWTL) at Hamburg University, Germany. Both the field and wind-tunnel data from JU2003 will be analysed in detail in the current paper. First, an overview of all relevant experimental setups will be given in the next Chapter, followed by a discussion of flow and dispersion characteristics near idealized street-canyon intersections. Such intersections can be considered as a typical UCL building configuration, and the available wind-tunnel data from studies at the University of Karlsruhe, Germany, highlight the complexity of flow and dispersion characteristics within such simple building configurations. As a next step, data from high-resolution laser doppler anemometry (LDA) flow measurements from the JU2003 wind-tunnel studies are then employed together with JU2003 full-scale data to study the flow fields within PA, a complex urban street canyon, in more detail. Finally, conclusions and open questions are presented. EXPERIMENTAL SETUP AND DATA ANALYSIS Street-canyon intersection wind-tunnel studies At the University of Karlsruhe, Germany, a series of wind-tunnel experiments were performed to study the flow and dispersion characteristics in the vicinity of typical urban street-canyon intersections. The experimental setup and the wind-tunnel boundary-layer flow parameters during the intersection studies were shortly summarized in Kastner-Klein et al. (1997), but so far most results were only presented in a not widely available project report (Kastner-Klein and Plate, 1999b). The basic intersection configuration consisted of two perpendicular street canyons separating four building blocks of the same geometry. The buildings in each block were 0.6 m long, 0.12 m high and wide, and the street width was also 0.12 m. The model scale M was 1 : 150, and the wind profile was represented by the power law ( ) u(z) z α = (1) u r z r where u r = u 100 = 7.7 m/s and α = 0.23 with z r = 100 m (in nature). The studied configurations and building dimensions can also be seen in Figure 1(a). In the basic configuration, all buildings had the same height and flat roofs, and the wind direction was altered in 15 increments. Additionally, variations of roof shapes and building heights were studied for which the wind direction was altered in 15 to 45 increments. The number

3 PHYSICAL MECHANISMS OF FLOW AND DISPERSION IN URBAN CANOPIES 1889 Figure 1. Overview of the experimental set-up and investigated building configurations (a) during the wind-tunnel street-canyon intersection studies and photos (b), (c) of the LDV probe mounted in the wind tunnel. The metal test tubes seen in the close-up view (c) correspond to the sampling points of the tracer gas dispersion studies. This figure is available in colour online at of wind directions investigated for each case is listed in Table Ia. Traffic emissions were simulated by two ground-level line sources which used a design similar to the one established by Meroney et al. (1996) to assure that a sulphur hexafluoride (SF 6 )-air mixture was continuously and homogeneously released. They were installed in one street canyon in the future referred to as street 1 and each source was studied separately. In the current paper, only results for studies with the source

4 1890 P. KLEIN ET AL. Table I. Summary of statistical parameters describing the wind-tunnel concentration distributions near idealized streetcanyon intersections, whereby c max corresponds to the maximum dimensionless concentration of all sites, cs1 mean to the mean value of all sites in street 1, and cs2 max to the maximum dimensionless concentration measured in street 2. a) Number of wind directions Case Investigated With c max > c max,a With cs1 mean > cs1 mean,a With cs2 max > cs2 max,a A 24 B 18 4 (22.2%) 2 (11.1%) 7 (38.9%) C 8 7 (87.5%) 6 (75.0%) 5 (62.5%) D 8 7 (87.5%) 6 (75.0%) 4 (50.0%) b) Highest value among all investigated WD and corresponding WD for Case c max cs1 mean cs2 max A / /255 B /255 C D c) Highest value among all investigated WD and corresponding WD for Case c max /c max,a cs1 mean /cs1 mean,a cs2 max /cs2 max,a B / / /195 C D located at a distance of 85 mm from the walls of the southern buildings (Figure 1(a)) are presented. Concentration samples were taken at two levels 0.01 and 0.08 m above ground, using small metal tubes, seen in Figure 1(c), that were placed 7 mm away from the building walls and connected to air-tight containers. During one measurement cycle, 19 samples were simultaneously collected over a period of 1 min and then analysed one by one with a MELTRON LH108 leak detector. The measured SF 6 concentrations c are normalized according to c = c u 100 H (2) Q SF6 /L q where H is the building height, u 100 the reference wind speed 100 m above ground (full-scale) in the upstream boundary-layer flow, and Q SF6 /L q the SF 6 emission rate per unit length (L q = 1.44 m : length of source). In addition to the dispersion studies, high-resolution LDA flow measurements were also conducted. For these measurements a two-component LDA probe was mounted inside the wind tunnel (Figure 1(b), (c)). The focal length of the probe was mm, the beam spacing 15 mm and the half angle 3.9. The wavelengths were and 488 nm respectively, and time series with a sampling frequency of 100 Hz and even-time sampling mode were recorded over a period of 1 min. The probe itself was approximately 100 mm long and had a diameter of 25 mm. Owing to its size and the relatively short focal length, flow disturbances can occur particularly when measuring close to the ground within the street canyons since the probe is then placed in a very sensitive flow region the shear layer developing at roof level (Figure 1(c)). To evaluate the probe influence additional measurements were taken with a large LDA probe that had a focal length of 1.20 m and was placed outside of the wind-tunnel with the laser beams passing through a glass window. While in the upper part of the canyon the measurements with the two different probes agreed within ±10%, in the lower quarter of the canyon the differences were significant. The reverse flow observed in the canyon with the small probe tended to be stronger than with the large probe, whereby, depending on the particular site, the differences were upto a factor of 2. JU2003 wind-tunnel studies A wind-tunnel model at the scale of M = 1 : 300 of the CBD of Oklahoma city was constructed and installed in the atmospheric boundary-layer wind tunnel Wotan at the University of Hamburg, Germany (Figure 2(d) (e)). The building footprints and heights were derived from recent aerial photos and LIDAR data. The 25 m-long facility Wotan provides an 18 m long test section equipped with two turntables and an adjustable ceiling. The cross section of the tunnel measures 4 m in width and m in height, depending on the adjustment of the variable ceiling. A 1 : 300-scaled counterpart of the atmospheric BL was established by means of turbulence generators and floor roughness elements (seen in the background in Figure 2(d)). The mean flow profile can be described by a power law with an exponent α = 0.18 and by a logarithmic wind profile with a roughness length z 0 = 0.13 m (full-scale). The turbulence intensities and integral length scales are in good agreement with field data for the categories moderately rough to rough, and the flow spectra match reference spectra of Kaimal et al. (1972) and Simiu and Scanlan (1986). For precise probe positioning, the test section of the tunnel is equipped with a computer controlled traverse system with a positioning accuracy better than 0.1 mm on all three axes. For reference flow velocity measurements a Pitot tube connected to a laboratory grade differential pressure transducer is used. High-resolution flow measurements are carried out non-intrusively by means of a 2D fiber-optic LDA with a focal length of 800 mm. Time series were collected over a period of several minutes until a repeatable mean value was guaranteed. For laser light-sheet visualization experiments, the light of a low-power Argon-Ion-laser is transmitted via

5 PHYSICAL MECHANISMS OF FLOW AND DISPERSION IN URBAN CANOPIES 1891 Figure 2. View from the NE of the CBD of Oklahoma city (a); map (b) of the PA study area; photos (c) of the OU towers installed in PA (left: tower 1; right: tower 2); as well as photos (d), (e) of the OKC wind-tunnel model installed in the WOTAN wind tunnel. Photo (d) shows a view of the wind tunnel CBD from the north, while (e) is a close-up view of the wind tunnel model from the south with the First National Center being the high-rise building on the southeast corner at the intersection of PA with Robinson Ave (Map courtesy of May Yuan and Mang Lung Cheuk, Center for Spatial Analysis, University of Oklahoma). This figure is available in colour online at an optical fiber into the test section of the wind tunnel. At the end of the fiber, several lenses and mirrors are used to transform the circular laser beam into a triangular plane of laser light, which can be focused to a thickness of approximately 2 mm. The entire optical system is attached to the probe positioning system enabling the light sheet to be moved to any location within the model. A similar optical system was also used for visualizations in the intersection studies at the University of Karlsruhe. JU2003 field studies A general overview of the JU2003 field studies is given in Allwine (2004). During July 2003, comprehensive meteorological studies and ten intensive observation periods (IOPs) with tracer releases were conducted in the CBD of Oklahoma city. As mentioned already in the introduction, one downtown street canyon PA was chosen as study Copyright 2007 Royal Meteorological Society area for a street-canyon sub-experiment. Details on the PA measurement sites, instrumentation and data analysis for the wind and turbulence measurements within the PA street-canyon can be found in Brown et al. (2004). It is important to note that the PA building geometry is rather complex with buildings heights on the southern side ranging from m and on the northern side from m. From the GIS data (courtesy of May Yuan and Mang Lung Cheuk, Center for Spatial Analysis, University of Oklahoma), the average building height H (each building is weighted according to its relative footprint) has been calculated as roughly 65 m, the average street width S is roughly 25 m, and the study area between the intersections with Broadway and Robinson Avenue is approximately 156 m long (see also Figure 2(b)). The following analysis focuses mostly on sonic measurements which were performed by the University of Oklahoma (OU) on two 15-m towers installed in the central part of PA (Figure 2(c)). Each of the two Int. J. Climatol. 27: (2007)

6 1892 P. KLEIN ET AL. Figure 3. Mean horizontal wind speeds for three different wind directions near a wind-tunnel street-canyon intersection with identical buildings of height H = 120 mm in all four blocks of the intersection (case A). The measurement level was 30 mm above the street, i.e. at a height equivalent to H/4, and the free-stream velocity u 4H, indicated by the vectors showing the wind direction, was 10 m/s. The length of all vectors plotted is proportional to the wind speed, and the gray-scale contour maps further illustrate the wind speed patterns in the intersection. towers was equipped with 5 RM Young sonic anemometers with the lowest measurement level being at 1.5 m and the highest at 15 m above street level. The velocity components were defined according to the meteorological convention with positive u-components corresponding to winds from the west, positive v-components corresponding to winds from the south, and positive w- components corresponding to upward motions. Since PA has an east west orientation, the u component describes the along-canyon flow, and v is the across-canyon flow component. Further details on the experimental setup for these measurements and the quality assurance processing of the datasets are provided in Klein and Clark (2007). As reference data that characterize the non-urban atmospheric conditions, wind speed and direction from the PNNL sodar, which was operated by the Pacific Northwest National Laboratories (PNNL) and located approximately 2 km south-southwest of downtown OKC in the Wheeler Park near Shartel & SW 12th (Allwine, 2004; De Wekker et al., 2004) have been used. Klein and Clark (2007) have found that the UCL flow is mostly driven by boundary-layer dynamics at average roof level, and shown that the wind speed from the 80-m sodar level is an appropriate choice for the reference wind speed U r used to normalize the street-canyon flow data. The PNNL sodar was chosen as reference site because of its close proximity to the downtown area, and the fact that for most wind directions the sodar measurements were not influenced by local building effects such as vortices developing on the roofs of buildings, which was not guaranteed for the roof-top anemometers deployed in the downtown area during JU2003. The 80-m level was selected as reference height because it matches the average building height pretty well and for lower measurement heights the quality of the sodar data becomes questionable. Another advantage was that at the Tyler Media (TM) tower, which was operated by the University of Indiana and located 5.5 km south of OKC downtown (Grimmond et al., 2004), sonic data were collected at 79.6 m above ground which were used

7 PHYSICAL MECHANISMS OF FLOW AND DISPERSION IN URBAN CANOPIES 1893 Figure 4. Same as Figure 3 but 90 mm above the street, i.e. at a level equivalent to 3H/4. for comparisons with the 80-m sodar data (Klein and Clark, 2007). However, for northerly wind directions, the flow characteristics at the sodar site may be influenced by urban effects and for these wind directions the wind speed U r may thus not correctly reflect non-urban reference conditions. To study the influence of the structure of the upstream BL on the UCL flow, the stability parameter z/l in the non-urban atmospheric BL was determined based on heat and momentum flux measurements with sonic anemometers at two levels (37.3 and 79.6 m) installed at the TM tower. Details on how exactly the stability parameter z/l was calculated are also given in Klein and Clark (2007). Based on the non-urban z/l-values, the street-canyon data were then grouped into three categories (1) unstable for z/l < 0.1 (2) quasi-neutral for z/l < 0.1 and (3) stable for z/l > 0.1. It must be noted that the local stability parameters within the CBD differ from the non-urban stability parameters, and one may argue that the non-urban conditions do not apply to PA. However, as our study focused on how the structure of the approach flow affects flow and turbulence inside the UCL, it was decided to group the data based on the non-urban stability rather than local stability since the latter is already a result of the enhanced mixing inside the UCL. WIND-TUNNEL FLOW AND DISPERSION PATTERNS NEAR IDEALIZED INTERSECTIONS Near an intersection of two street canyons, it can be expected that the flow is dominated by interactions of channelling-type and recirculation-type flows. To get a better understanding of these interactions and the resulting flow characteristics near intersections was the motivation for the wind-tunnel studies at the University of Karlsruhe, Germany. Flow measurements with a LDA system were conducted near idealized intersections for three different wind directions and datasets with a high spatial resolution have become available (Figure 3 4).

8 1894 P. KLEIN ET AL. Figure 5. Wind-tunnel flow visualizations illustrating the building height influence on flow and dispersion patterns near street-canyon intersections. Snap shots of instantaneous flow patterns (a), (c) for the reference configuration with identical buildings in all four blocks of the intersection are shown in comparison to two situations (b), (d) for which the dark-shaded building was 1.5 times taller than the other buildings. The black arrows indicate the above roof-level wind direction. This figure is available in colour online at For wind direction 180, for which the mean approach flow is perpendicular to street 1 and parallel to the second street, it can be seen that channelling remains the dominant flow type in the intersection. Within street 1, lateral vortices develop near the intersection and even at a distance of 2.5 H away from the intersection small along-canyon velocity components can be noted, i.e. a classic 2D street-canyon vortex does not fully develop. For wind directions with an oblique angle with respect to both streets, channelling flow becomes dominant in both canyons but is superposed by recirculation-type flow as indicated by the reversed across-canyon components in the lower part of the canyons. These helical-type flow patterns then meet and interact at the intersection, where, depending on the specific wind direction channelling along one street canyon may still have a dominant influence (WD 210 ) or the flow at the intersection may become best described as bifurcation (WD 225 ). Compared to wind direction 180, the average wind speed at z = H/4 within street 1 increased by a factor of 2.5 for WD 210 and 3.5 for WD 225, and also in the intersection region higher wind velocities were measured for the oblique wind directions but the differences were less significant (30 50%). It can be expected that the flow patterns near intersections become even more complex for more realistic configurations with complicated building geometries near the intersection. Unfortunately, LDA measurements are very timeintensive and not free of technical constraints (see also Chapter 2.1). It was thus, not possible to conduct further flow measurements, and flow variations due to changes in building geometry could only be indirectly investigated by means of visualizations and dispersion studies. To illustrate the influence of building-height variations near the intersections, flow visualizations with a laser-light sheet and a smoke release with a nozzle in the intersection centre are shown in Figure 5. The flow field characteristics near the intersection were strongly influenced by an increase in building height of one of the buildings the dark shaded buildings are 50% taller (0.18 m) compared to the buildings in the three other blocks. A significant lateral flow component in the intersection region was only observed for the asymmetric configurations with a higher building at one of the four blocks (Figure 5(b), (d)). The effect was stronger for the situation with a higher building downwind of the intersection (Figure 5(d)). In this case, the parallel flow component along street 2 almost totally disappears and the lateral flow component in street 1 is directed away from the higher building. In the case of a higher building placed upwind of the intersection (Figure 5(b)) the north south channeling flow was not totally blocked and the lateral component in street 1 is directed to the leeward region

9 PHYSICAL MECHANISMS OF FLOW AND DISPERSION IN URBAN CANOPIES 1895 Figure 6. Dimensionless concentrations c measured along the building walls at a height of 10 mm (= 0.083H) above street level for the four different street-canyon intersections described in Figure 1. The plotted circles indicate the location of the sampling points; the size of the circles is proportional to the measured concentrations. The wind direction was 180, i.e. perpendicular to the line source located at y = 25 mm (black horizontal line). of the higher building. By comparing the two snapshots for case A with equal building heights in all four blocks (Figure 5(a), (c)), one can get an impression of the instationary nature of UCL flow. Even though both images were taken for the same approach-flow conditions and building geometry, the location of the plume varies. In one case, it drifts more to the eastern side of street 2, while just a few seconds later it shifts west. It can thus be expected that an air sampler near an intersection, depending on its specific site, will record significant concentration spikes followed by longer periods with much lower or even negligible concentrations. To further investigate the influence of building height variability on the dispersion patterns near intersections, distributions of mean, near-ground tracer gas concentrations (z = 0.01 m) for four different building configurations and four wind directions are shown in Figure 6 9. As described in chapter 2.1, the tracer gas was released from a ground-level line source inside street 1, and samples were taken along the building walls. The concentrations were normalized according to Equation (2), and to illustrate the spatial distributions, circles are plotted that are scaled proportional to the dimensionless concentrations at each site. For southerly and northerly wind directions, for which the roof-level winds are perpendicular to the line source, the dispersion patterns for case A in street 1 are similar to the ones in 2D-type street canyons (Figure 6(a) and 7(a)). The concentrations are significantly larger on the leeward than windward building walls, which is in

10 1896 P. KLEIN ET AL. Figure 7. Same as Figure 6 but for wind direction 360. agreement with the vortex-type flow patterns discussed above. However, the lateral recirculation zones developing near the intersection cause along-canyon concentration variations whereby the concentration maxima are found at a distance of 2 H away from the intersection. These effects are less pronounced for case B with pitched roofs (Figure 6(b) and 7(b)). For wind direction 45, the leeward-windward concentration differences are still significant and the plume remains mostly within street 1 for cases A and B (Figure 8(a), (b)). As it could already be expected from the visualization studies (Figure 5), roof height variations near the intersection strongly affect the dispersion patterns for all studied wind directions (Figure 6(c), (d) 9(c), (d)). The spatial concentration distributions are altered and often significantly higher maximum concentrations are observed in the wake regions of the taller buildings. As noted above, the pollutants can also be transported away from the taller building if it is downwind from the source (Figure 7(c), (d)). Without flow-field data it becomes difficult to get a clear picture of what causes this phenomenon, but it appears that strong downward motions develop near the wind-exposed wall of the taller building triggering flow divergence near the surface in the region of the tall building and strong lateral flow motions that shelter other portions of the street from being well ventilated. To further summarize the wind-tunnel results, the following statistical parameters have been computed for each measured near-ground concentration distribution (1) maximum c of all sites, c max (2) mean c of all sites in street 1, cs1 mean and (3) maximum c of all sites in

11 PHYSICAL MECHANISMS OF FLOW AND DISPERSION IN URBAN CANOPIES 1897 Figure 8. Same as Figure 6 but for wind direction 45. street 2, cs2 max. The latter two parameters have been chosen to evaluate the ventilation of street 1 and the transport of pollutants into the emission-free street 2. The highest values of these three parameters for all measured wind directions, and the corresponding wind direction for which these maxima were observed are shown in Table Ib. While the influence of building configuration on the overall ventilation of street 1 and transport into street 2 is not very pronounced the maxima for cs1 mean and cs1 mean differ only by 10 20% the overall concentration maximum can increase by a factor of 1.5 if the roof shapes and heights are altered. While these overall statistics give an idea of the variations that can be expected regarding the longer-time air quality in certain building configurations, they mask the dramatic differences in UCL ventilation that exist for certain wind directions. These differences are captured by the maxima of the ratios c max /c max,a, cs1 mean /cs1 mean,a, cs2 max /cs2 max,a (Table Ic), which relate the parameters found for cases B D to the conditions found in the reference case A. It becomes obvious that for situations with taller buildings the near-ground concentration maxima can increase by almost a factor of 2, pollutant transport into the emissionfree street 2 by more than a factor of 3 and the overall ventilation of street 1 can be reduced by about 30%. Also, for these configurations (cases C and D), all three statistical parameters tend to be higher for the majority of the studied wind directions (Table Ia). These findings are important as current numerical models typically are not capable of resolving the full complexity of urban settings and use rather simplified building configurations instead. It can be summarized that building-height variations are important factors for UCL ventilation. While taller buildings increase the downward mixing of above

12 1898 P. KLEIN ET AL. Figure 9. Same as Figure 6 but for wind direction 90. roof-level fluid this does not necessarily improve the local air quality, quite to the contrary, for non-uniform configurations the pollution levels near the intersection typically increased significantly. FLOW FIELD WITHIN PARK AVENUE Within this section, the main findings from flow-profile measurements within the complex urban street canyon, PA, are shortly summarized, a simple model for the evaluation of street-level winds is tested, and a comparison between the JU2003 full-scale and wind-tunnel data is presented. Full-scale data On the basis of the analysis of Klein and Clark (2007) the 80-m level of the PNNL sodar data was chosen as reference level for the normalization of the field data (see also Chapter 2.3). The normalized 30-min mean values and first-order statistics for all three velocity components measured on the northern OU site (tower 1) at a height of 10 m above ground are presented in Figure 10 as a function of reference wind direction WD r. These data were exemplarily chosen, whereas a complete overview of all available data from the OU towers is given in Klein and Clark (2007). The plotted data are grouped into the three stability classes (1) quasi-neutral (2) unstable and (3) stable, which were defined in Chapter 2.3. It becomes obvious that compared to the influence of wind direction the influence of atmospheric stability in the upwind BL is clearly a second-order effect, and given the small stability variations seen in Figure 10, stability effects are ignored in the further analysis. However, as discussed in more detail in Klein and Clark (2007),

13 PHYSICAL MECHANISMS OF FLOW AND DISPERSION IN URBAN CANOPIES 1899 Figure 10. Influence of above roof-level wind direction on mean (a) and turbulent velocity components (b) measured in the field during JU m above ground inside PA on OU tower 1, which was located on the northern side of the street. The data were grouped according to the non-urban boundary-layer stability: quasi-neutral (dark grey asterisks), unstable (light grey dots) and stable (black triangles). The black lines correspond to calculated values for the mean along- and across-canyon velocity components that are based on linear fits with the corresponding above roof-level wind speed components (see text for more details). flow and turbulence data clearly differ for the three stability groups if winds from higher elevations in the atmospheric boundary-layer are used as reference data, and based on such normalization one could infer that atmospheric stability strongly influences UCL flow and particularly turbulence. The fact that upwind stability becomes of minor influence when in-canyon data are normalized with a roof-level wind speed demonstrates that UCL flow is primarily driven by the boundary-layer dynamics at average roof level and that the stability of the upwind atmospheric boundary layer influences in-canyon flow, mostly indirectly, by altering the wind speed at roof level. Such conclusions were also drawn by Bohnenschlegel et al. (2004) based on numerical simulations of urban flows. One could also argue that local rather than upwind stabilities determine urban flow phenomena, and that neutral conditions prevail within the urban-canopy layer because of strong mechanical mixing and nocturnal heat releases from buildings. On the other hand, it was discussed in previous studies (Louka et al., 2002) that differential heating of the canyon walls can affect flow within street canyons and that thermally induced flow motions become important under calm conditions and winds perpendicular to the canyon. A detailed analysis of local stability effects should thus be the focus of future studies before drawing conclusions about their nature and impact on UCL flow and turbulence. For southerly wind directions ( ), for which most full-scale data are available, the turbulent velocity components are rather uniform with the values for the along-canyon component (u sig ) being approximately 50% higher than for the across-canyon (v sig ) and vertical (w sig ) component (Figure 10b). All three mean velocity components are on the other hand very sensitive towards wind direction and the reverse flow (v m < 0) and downwards motions (w m < 0) indicate the existence of recirculation-type flow patterns particularly for south-southeasterly directions (Figure 10a). However, for a wide range of wind directions significant along-canyon flow components (channelling) were also recorded. These findings are consistent with the results for the idealized intersections discussed in Chapter 3 and in general agreement with Dobre et al. (2005) who recently presented

14 1900 P. KLEIN ET AL. Figure 11. Dependence of JU2003 along- (a) and across-canyon (b) mean velocity components on the corresponding non-urban above roof-level components u r = U r sin(wd r ) and v r = U r cos(wd r ).Shownarethedataatz = 10 m from OU tower 1 inside PA and the symbols are the same as in Figure 10. The black lines show linear fits for which the coefficients and correlation coefficients are given in the Table II. an analysis of flow data measured during the DAPPLE campaign (Arnold et al., 2004) near a downtown intersection in London, England. They concluded that flow within the street is the vector sum of a channelling and a recirculation (street-canyon type) vortex. Furthermore, they also found that the along-canyon, channelling velocity component u m, depends linearly on the along-canyon component of the roof-level reference wind u r, whilst the cross-canyon component v m that characterizes the streetcanyon vortex depends linearly on the component v r of the roof-top reference wind perpendicular to the canyon. In the case of the east west oriented canyon PA, these roof-level components are given by the standard equations for the lateral and longitudinal wind components: u r = U r sin(w D r ) (3) v r = U r cos(w D r ) (4) According to a linear model, which Dobre et al. (2005) stated is robust enough to occur in streets of non-ideal geometry, the in-canyon velocity components can then be calculated by: u m = a 1 u r + a 0 (5) v m = b 1 v r + b 0 (6) From the PA studies a much larger dataset, covering a wider range of reference wind directions, is available than from the DAPPLE studies analysed by Dobre et al. (2005). Furthermore, the PA building geometry is far more complex than the one near the intersection in London. In the latter case, the tallest building was only roughly 35 m high, while the high-rise buildings on the western edge of the PA study area were more than 120 m tall. The PA measurements provide thus, an ideal dataset for testing how a linear model will hold in a complex setting over a wide range of approach-flow conditions. One question that must be addressed is the choice of the site for the above roof-level winds. Dobre et al. (2005) used data from an automatic weather station that was located on the roof of a building next to the studied intersection. They discuss that wakes and recirculation zones developing at roof-top building structures may influence the chosen reference winds, but assume that these influences are not significant for the wind directions prevailing during the DAPPLE campaign. However, much taller buildings were located around PA in Oklahoma city than at the DAPPLE site, and while some roof-top anemometers were deployed in the CBD during

15 PHYSICAL MECHANISMS OF FLOW AND DISPERSION IN URBAN CANOPIES 1901 Figure 12. Horizontal wind fields in PA at z = 10 m measured in the Hamburg wind-tunnel studies for wind direction 180 (a), 170 (b) and 190 (c). The circles indicate the field site of OU towers 1 and 2, the stars the position of additional wind-tunnel profiles. The length of all vectors is proportional to the wind speed. This figure is available in colour online at JU2003, none of these anemometers was placed on a mast high enough to assure that the measured data were not affected by local building effects. Owing to the large variability of the building heights near PA it is also not obvious which building would be the best site for measuring the roof-level winds, and it was thus decided to again use the 80-m level from the PNNL site as reference. The variation of in-canyon velocity components at the 10-m level on OU tower 1 with their above rooflevel counterparts is shown in Figure 11 as an example.

16 1902 P. KLEIN ET AL. Figure 13. Wind fields in PA measured with a LDA system in the wind-tunnel studies at three vertical planes located at y 15 m (a), y 45 m (b), and y 90 m. The OU towers 1 and 2 were located at y 79m(tower1)andy 83 m (tower 2), i.e. closer to plane (c). This figure is available in colour online at Table II. Parameter values for the linear fits (Equations (5), (6)) of in-canyon velocity components as functions of their non-urban above roof-level counterparts measured at the 80-m level of the PNNL sodar site. Tower 1 Along-canyon component u m Across-canyon component v m Level Slope a 1 Intercept a 0 Corr. coeff. R u FAC2 in % Slope b 1 Intercept b 0 Corr. coeff. R v FAC2 in % Tower 2 Along-canyon component u m Across-canyon component v m Level Slope a 1 Intercept a 0 Corr. coeff. R u FAC2 in % Slope b 1 Intercept b 0 Corr. coeff. R v FAC2 in % The black lines show the linear best fits for data from all three stability classes. For the data from all five levels on both OU towers, the best-fit values for the slope, intercept, correlation coefficients, and percentage of data points for which the predicted wind speeds are within a factor of 2, FAC2 (0.5 u pr /u obs 2or 0.5 v pr /v obs 2), are summarized in Table II. Simply based on the correlation coefficients, one could conclude that the linear model works also for PA for most cases. Furthermore, the parameters for both towers are very consistent and it appears reasonable that for the alongcanyon component the slope is positive and increases with height while it is negative and its absolute value decreases with height for the across canyon component. However, Figure 10 and 11 illustrate that a linear model is obviously not able to capture the full complexity of the PA flow patterns. Differences between the calculated and measured velocities are most prominent for northeasterly

17 PHYSICAL MECHANISMS OF FLOW AND DISPERSION IN URBAN CANOPIES 1903 Figure 14. Comparison of normalized along-canyon u m (a), (b) and across-canyon v m (c), (d) wind velocity components as measured in the field (black symbols) and wind-tunnel studies (grey lines) for wind direction 180 on OU towers 1 (a), (c) and 2 (b), (d). For the field studies the average values of all available profiles with reference wind directions within the range have been calculated and the error bars illustrate the variability (±σ ) of these datasets. The different lines from the wind-tunnel studies correspond to repeated measurements with the same settings and at the same locations. wind directions (Figure 10). One explanation for the higher observed values could be that, in particular for northeasterly wind directions, the tall towers near the intersection of PA and Robinson Ave cause strong downward mixing of high-momentum fluid which then results in flow divergence near the surface, and enhanced channelling and recirculation near the ground, similar to effects discussed in the previous studies for the idealized street-canyon intersections with variable roof heights. It can also be noted, that the performance of the linear model in the case of the across-canyon component decreases with height, which may indicate that the upper levels are close to a vortex centre in which the acrosscanyon component becomes undefined. The percentages of data points with model predictions within a factor of 2 from the observations FAC2, allow further evaluation of the overall performance of the linear model in a statistical sense. For the along-canyon component, u, the FAC2 values are around 70% or higher, while for the across-canyon component, v, particularly on tower 2, FAC2-values of less than 50% are observed. The model performance is possibly also affected by the choice of the roof-level winds and it should be tested in future studies if a better agreement between predictions and observations can be obtained by using a roof-top sensor closer to the study area as reference site. Given its simplicity, the model may still be useful in getting a rough estimate of the street level winds, but for practical applications it is also questionable how universal the fit parameters are even within one street, or in other words, how uniform the flow becomes within the street. While the flow within the street canyons at some distance away from the intersection became rather uniform for the idealized intersections, more significant spatial variations can be expected in complex urban settings with significant building geometry variations such as in PA.

18 1904 P. KLEIN ET AL. Figure 15. Comparison of normalized along-canyon u m (a), (c), (e) and across-canyon v m (b), (d), (f) wind velocity components from the JU2003 field (diamonds: tower 1; stars: tower 2) and wind-tunnel studies (x = 0m,y = 60 m: solid lines, x = 0m,y = 90 m: dashed lines) for wind direction 170 (a), (b), 180 (c), (d) and 190 (e), (f). From the full-scale data, the average of all available profiles with reference wind directions within the range ±2 from the specified value has been calculated and the error bars illustrate the variability (±σ ) of these datasets. Comparison of wind-tunnel and full-scale data To further study spatial flow variations within PA, windtunnel datasets are ideal since they can provide the necessary high spatial resolution of the 3D flow fields, whilst the PA field data provide only limited information on the 3D nature of the flow despite the relatively large number of sensors deployed. As an example, horizontal velocity fields measured in the Hamburg wind tunnel at a height corresponding to 10 m at full-scale are shown in Figure 12 for three wind directions, 170, 180 and 190. It can be clearly seen that lateral vortices developing near both intersections penetrate deep into the PA canyon, which Brown et al. (2004) also observed when analyzing all available streetlevel sonic data for PA. For the first two cases (170 and 180 ), the flow dynamics are dominated by multiple recirculation patterns (see also Figure 13) which results in rather inhomogeneous and complex flow patterns and limits the applicability of parameters determined from the OU towers to predict wind speeds in other regions of PA. For wind direction 190, the flow follows a more channelling pattern, as it was also

19 PHYSICAL MECHANISMS OF FLOW AND DISPERSION IN URBAN CANOPIES 1905 observed in the field, and is consequently more uniform, and the linear model would likely apply to a larger region. By comparing the wind-tunnel results for the three wind directions it becomes again obvious how sensitive the UCL flow-patterns are to changes in reference wind directions. Particularly near the two OU towers, a transition between lateral corner vortices and vertical recirculation patterns takes place, which is also supported by the velocity fields in three vertical cross-sections at different positions within the canyon (Figure 13). Such transitions, and the strong sensitivity towards wind direction variations must also be taken into account when directly comparing wind-tunnel and full-scale data as the flow is rather non-stationary, and any small deviations between the modelled and real building geometry or sampling position can result in large velocity or concentration differences. Such comparisons between wind-tunnel and full-scale flow profiles are shown in Figure 14 and 15. In the first case (Figure 14), the position of two wind-tunnel profiles for wind direction 180 was chosen to match the position of OU towers 1 and 2 as closely as possible. The wind-tunnel measurements were repeated several times and the mean velocities from the different measurement cycles shown by the grey lines in 0 agree generally very well. From the field data, all available full-scale profiles for wind directions were selected for the inter-comparison, and the average profiles were calculated as well as standard deviations, which are shown as error bars in Figure 14 to illustrate the natural flow variability. The wind-tunnel and full-scale data were both normalized by a reference wind speed taken at z = 80 m in the non-urban, upstream BL flow. As can be seen in Figure 14, both the channelling (u m ) and the recirculation (v m ) components are stronger in the full-scale measurements than in the wind tunnel, whereby in the wind tunnel no recirculation (v m > 0) is predicted. This is somewhat surprising as one may expect to see stronger recirculation patterns in a wind-tunnel simulation with a neutrally stable and more controlled approach flow than in an urban environment. However, taking into account the above discussed transition zones near the OU towers such differences can be expected. Figure 12a illustrates that just slightly to the west of the OU towers the flow in the wind tunnel model was also less stagnant with both, a higher along-canyon wind speed, and a tendency for reverse flow (v m < 0). Any slight deviation in the measurement position could thus cause the observed differences, but of course other factors, such as differences in the mean and turbulent properties of the upstream BL may play a role too. Additionally, also the non-stationary nature of the UCL flow may contribute to the observed differences. For the field data, longer averaging periods than 30 min may be required to obtain a statistically representative mean value which is, however, difficult to realize since the non-urban conditions are often not steady for long time periods. Furthermore, the performance of the RM Young sonic anemometers used in the field studies should be further evaluated, as their design is not optimal for deployment inside urban canyons where significant vertical flow components may occur that may be affected by the relatively large shaft of the instruments. In Figure 15, flow profiles for the wind directions, 170 and 190, were also employed in the inter-comparison. It must, however, be taken into account, that for these two wind directions the position of the OU towers could not be matched in the wind tunnel due to technical constraints. From the wind-tunnel measurements two centre line (x = 0 m) profiles at y 60 and 90 m (see Figure 12 for definition of coordinates and position of profiles) are thus consistently plotted in Figure 15 for all three wind directions. Their locations with respect to the OU towers is illustrated in Figure 12a. Since the measurements in the field and in the wind tunnel were taken at different sites the results can only be compared qualitatively. For wind direction 190, the along-canyon components, u m, measured in the wind tunnel and field agree well, despite the differences in the site locations, which supports again, the hypothesis that for this wind direction channelling becomes dominant and the streetlevel winds thus, rather homogeneous. For wind direction 170, on the other hand, the wind-tunnel and full-scale along-canyon components, u m, differ in both magnitude and direction, which can be explained by the recirculation zones developing near the tall buildings that result in very complex, spatially variable flow patterns at street level for this wind direction. Differences between the wind-tunnel and full-scale data can also be noted for the across-canyon component, v m, particularly for wind direction 190, for which reverse flow in the lower part of the canyon was observed in the field but not measured in the wind tunnel. Concluding from the differences between the wind-tunnel profiles for v m, the development of reverse flow generally appears to be very site-specific. In summary, the presented high-resolution wind-tunnel flow fields and qualitative comparison of wind-tunnel and full-scale profiles demonstrates the complexity and spatial variability of UCL winds, and allows evaluating how universally a linear model that is based on measurements at one site would apply to a whole street or neighbourhood. For a complex site such as PA, the applicability of the linear model is clearly limited. CONCLUSIONS AND OPEN QUESTIONS The analysis of flow and dispersion data for idealized intersections and a complex downtown street have highlighted the complexity of UCL flow and dispersion mechanisms. Quasi-2D street-canyon vortices, which have often been the focus of previous studies, are not very typical for realistic urban settings and a wide range of wind directions. The UCL flow becomes dominated by lateral vortices forming near intersections and wakes developing behind buildings with above-average roof heights. Whilst such tall buildings can cause rapid vertical mixing, the