Influence of urban morphometric modification on regional boundary-layer dynamics


Corresponding author: J. C. H. Fung, Division of Environment, The Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong. (


[1] Fidelity in simulating urban boundary-layer (UBL) physics is recognized to prescribe the prognostic skill of subsequent regional air pollutant transport modeling. Conventional mesoscale meteorological models (MMM) deployed over the South China coast among urban locales have often yielded positive bias in surface wind speed. This bias has been hypothetically attributed to model parameterizations that yield inaccurate meteorological predictions due to underrepresentation of urban aerodynamic roughness. Chemical transport model (CTM) simulations that are forced by the overestimated UBL wind field may undergo excessive advection which results in negative bias in predicted pollutant concentration. This study aimed to corroborate the proposed causality between parameterized urban morphometry and UBL meteorology. Focus was placed on the urban meteorological adjustments induced by urban morphometry modifications rather than prediction improvements attributable to urban canopy parameterization (UCP). Case studies were devised to assess the sensitivity of an urban-meteorology model to a pervasive, region-wide urban morphometry modification. Performance of a UCP scheme was evaluated for the Pearl River Delta (PRD) region, a meso- β-scale subtropical coastal megalopolis. To benchmark the limits of UBL adjustments that were predominantly attributable to urban morphometric transformation, numerical experiments were conducted against two urban fabrics of vastly dissimilar morphometric compositions, each occupying identical topographic tracts. Differences in the diurnal evolution of UBL structure and in the mean and turbulent flow characteristics were analyzed. This UCP sensitivity study suggests that improved urban morphological realism is able to reduce positive wind speed bias observed in conventional mesoscale meteorological models when applied to the PRD region.

1 Introduction

[2] The Pearl River Delta (PRD), situated on the southern coast of Guangdong Province, China, is a densely and aggressively built megalopolis. Moreno et al. [2010] estimated its 2010 population to be 120 million and growing. In the immensely commercialized city of Shenzhen, population density approaches 17,150 km  − 2, in contrast to 13,400 km  − 2 in Shanghai, and 10,430 km  − 2 in New York City. Figure 1 illustrates the rate of PRD land use conversion from rural to urban between 1991 and 2003. Urban land use increased from less than 1% of the region to nearly 13%. The Hong Kong Trade Development Council reported ( that the PRD experiences an annual GDP growth of 12.2% and produces 9.4% of China's national GDP, 10.3% of its gross industrial output, and 27.4% of its total export. After the liberalization of economic policies in the 1970s, the regional GDP composition in 1980 versus that in 2010 had shifted from 25.8% to 2.1% in the raw materials sector, from 45.3% to 48.6% in the manufacturing sector, and from 28.9% to 49.2% in the services sector. This extraordinary industrial and commercial buildup has brought with it an escalation in power consumption and direct industrial emissions into the atmosphere. The high population and building densities of the region exacerbate concerns over air pollution-induced health risks and consequences of accidental atmospheric release of harmful substances. For over a decade, the air-quality research community has striven to delineate the pollutant transport and dispersion mechanisms that affect the PRD environs in pursuit of an effective regional air pollution mitigation strategy.

Figure 1.

Pearl River Delta land use conversion between 1991 and 2003. Red denotes urban land use category. Urban land use rose from < 1% to nearly 13% in the intervening period.

[3] Considering the expansive perimeter of inhabited areas in the PRD, the high proportion of urban coverage, and the geographic complexity besetting the emission sources and the affected population centers, numerical air-quality studies [Fung et al., 2005; Huang et al., 2006] ordinarily favored the use of MMM coupled to multi-scale CTMs. In contrast, employing computational fluid dynamics (CFD) solutions to provide meteorological forcing to the CTM, while technically possible, can be computationally cost-prohibitive in a research environment. Nonetheless, the use of MMM can be inherently problematic in air-quality applications [Lo et al., 2006, 2007] where the accuracy of land surface representation and land-atmosphere exchange parameterizations centrally affects model skills. Deng et al. [2004] observed in an inter-regional transport study for the U.S. Appalachian region that the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5) overpredicted surface wind speed by 2.37 ms − 1, yielding a relatively poor index of agreement of 0.57. Further investigation showed that a combination of reducing the horizontal and vertical grid spacing, selecting a suitable set of physics options, and applying four-dimensional data assimilation (FDDA) helped to improve prognostic skills. Tuccella et al. [2012] found in a 2007 study of the European airshed that the WRF-Chem model overestimated wind speeds by 76%, or ≈ 1 m s − 1, despite a well-reproduced diurnal variation of wind speed. Wu et al. [2008] cited an MM5/Community Multiscale Air Quality model (CMAQ) study [Byun and Schere, 2006] covering the North Carolina vicinity in which surface wind speed was overpredicted by 0.9 m s − 1 while the benchmark bias was ≤ ± 0.5 ms − 1. Hanna and Yang [2001] found in an MM5 study deployed over Central California that the mean bias of surface wind speed was + 1.5 ms − 1 while the mean observed value was 3.1 ms − 1. Further, Buccolieri et al. 2011 found in a combined wind tunnel and CFD study that air quality was significantly altered by avenue-like tree planting, street canyon aspect ratio, and angle of the approaching wind with respect to the street axis. These studies reveal a pattern in which conventional MMMs often overestimate surface wind speed. These studies also highlight the importance of realistically representing the urban surface features and their physical forcing upon the atmosphere. While this study principally focuses on the regional wind field, it is important to be aware of other forms of meteorological variability induced by aggressive urban land use development [Cheng and Chan, 2012].

[4] Circa 1990s, agricultural motivations to numerically reproduce the meteorological mechanisms that advected airborne particles through and above crop fields led to the scientific research and development in vegetation canopy parameterization (VCP) [Finnigan, 2000]. Semblance in morphological composition between vegetation arrays and urban structures facilitated the adaptation of VCP to UCP in the decade that followed. Martilli et al. [2002] introduced the “Drag Approach (DA),” which accounted for pressure drag and friction force against built structures. Otte et al. [2004] implemented an analogous DA UCP in the MM5. Dupont et al. [2004] further enhanced the implementation by including the presence of urban vegetation and water and their meteorological effects. This urban parameterized derivative of MM5 has been entitled uMM5 (urbanized MM5) amongst its users. The fundamental distinction between DA and conventional MMM is in the treatment of the surface layer. Conventional MMM often assumes that the significant physical exchanges between land and atmosphere occur entirely within the surface layer and are subject to scaling by the Monin-Obukhov Similarity Theory (MOST). In simulations over predominantly natural land surfaces, MOST is assumed to be valid at the center of the surface layer thickness. MMM physics, then, treats the land cover canopy as a two-dimensional surface that possesses particular physical properties, such as aerodynamic roughness length, moisture availability, and thermal inertia, at each horizontal location. Such a parameterization paradigm precludes investigation of atmospheric processes that originate in urban street canyons and propagate their effects up through the urban canopy. In contrast, DA parameterizations address the appreciable roughness heterogeneity in urban canopies and the unsteady and continually meandering (accelerating) wind flow between obstacles. These environmental factors invalidate MOST assumptions. Instead of using MOST to extrapolate street canyon flow properties, DA treats the urban roughness elements as spatially averaged objects that occupy volume, and allows physical fluxes to interact with parameterized vertical and horizontal roughness surfaces. This, in effect, advances the land surface representation from two to three dimensions and enables the in situ numerical resolution of atmospheric processes that transpire in the street canyons and influence the planetary boundary layer (PBL) flow above the urban canopy. Taha [2008a, 2008b], Ching et al. [2009], Bornstein [2008], and Fernando [2008] demonstrated skillful uMM5 predictions of urban influences upon the PBL.

[5] The present study deployed uMM5 over the PRD to investigate the urban meteorological impacts that result from a pervasive modification to the dimensions of urban structures (“urban morphometry”). Numerical experiments were conducted to evaluate the limits to which urban morphometric modification could influence the diurnal evolution of PBL properties.

2 Synopsis of the Urban Canopy Parameterization

2.1 Urban Meteorology

[6] The uMM5 has capabilities to simulate meteorological exchanges that transpire within and above the urban canopy. Dynamical urban effects such as pressure drag, skin friction, and adjustments to turbulence generation and dissipation are captured by augmenting the momentum and turbulence kinetic energy (TKE) equations. The momentum and TKE equations are coupled in the Gayno-Seaman PBL physics scheme through an E − l turbulence closure model [Wilson et al., 1998], where E and L correspond to TKE and a parameterized length scale, respectively. Highlighted in Appendix A is the basic formulation of the UCP, whilst Dupont et al. [2004] provides details of the theoretical basis and implementation.

[7] Incorporating the DA into MM5 has the implicit effect of retarding surface wind speed due to the greater effective aerodynamic roughness. With a vertical grid structure that reaches into the urban canopy, evolution of exchange processes in street canyons can be simulated. Otte et al. [2004, Figure 6] illustrate results of a DA-augmented MM5 simulating the Philadelphia, Pennsylvania urban airshed. TKE profiles show differences between daytime and nighttime boundary layer structure, with details in the canopy layer. The wind profile shows significant reduction relative to MOST based simulations. The dotted curves illustrate that the significant change in the flow dynamics is not attributed to vertical grid density differences, but to physical parameterization improvements. As shall be observed in section 4, this study produces results similar in character to these illustrations.

2.2 Urban Morphometry

[8] Complementary to meteorological parameterization in the UCP, a compatible parameterized representation of the urban fabric is also necessary. The urban parameterized PBL and land-surface schemes are based on volumetric averaging principles. Grid-cell averaged morphometric values are substituted for physical presence of urban objects. In turn, transport processes are confined to the fraction of a grid cell that discounted the volume occupied by buildings and vegetation. The urban morphology model is constructed from grid-cell averaged building morphometry as described by Burian et al. [2002]. Formulation of uMM5 relevant morphometric parameters are highlighted in Appendix B.

3 Experiment Design and Execution

3.1 Urban Meteorology

[9] The numerical experiments utilized MM5 version 3.6.2 and its urban parameterized derivative, uMM5 (section 2.1), to perform meteorological simulations. The methodology entailed running MM5 followed by uMM5. MM5 was responsible for generating initial and boundary meteorological fields ingested into uMM5. uMM5 was responsible for resolving the high resolution meteorological fields that accounted for the 3-D urban effects. The uMM5 output fields constituted a superset of those produced in MM5 and included UCP values such as land-cover-type-specific exchange fluxes, selected TKE budget terms, turbulent diffusivity, height-dependent morphological densities, and others. Decoupled execution of MM5 and uMM5 provided flexibility in permuting a given synoptic meteorological forecast against multiple urban morphology models.

[10] The MM5 computational domain comprised three nested grids as shown in Figure 2, whereas the uMM5 computational domain was not nested and coincided with the MM5 inner-most domain, “D03.” The outer MM5 domain, “D01,” was configured to cover central and southern China and the adjacent coastal waters. The western edge of the domain was situated well to the east of the Tibetan Plateau, so as to minimize dynamical perturbations caused by katabatic winds. The eastern edge of the domain was set to the east of Taiwan, providing fetch for approaching typhoons to develop. Subsidence of dense air ejected from typhoon cores has been observed to correlate with poor air quality along the South China coast [Huang et al., 2006].

Figure 2.

Model grid structure. Three-nested MM5 domain. The inner domain, D03, has 268 × 286 grid points at 1 km spacing and is coincident with the un-nested uMM5 domain D01.

[11] Horizontal grid spacing of the MM5 outer domain was set to 9 km, with a recursive factor-of-three reduction at each enclosed nesting level. This translated to a 1 km horizontal resolution for MM5 D03 and uMM5 D01. The uMM5 domain comprised 268 points in the west-to-east direction and 286 points in the south-to-north direction. The fine resolution and expansive grid coverage were devised to suitably represent key geographic features of the PRD and to accommodate protracted advection and accumulation of meteorological effects.

[12] The MM5 vertical grid was made of 34 layers, ranging from the ground to a model-top pressure of 50 mbar. This model-top specification was raised above the customary 100 mbar set for North American experiments, so as to avert model top reflection of Tibetan mountain waves.

[13] Design of the uMM5 vertical grid embraced the UCP ability to simulate physical exchanges that take place within the canopy layer. There were a total of 48 grid layers extending from ground level to a model-top pressure of 50 mbar. The lowest grid-layer thickness was approximately 4 m, which was then moderately stretched at each ascending layer to produce 12 layers within the first 100 m AGL. The 34 layers of MM5 output fields were vertically interpolated to 48 layers via the post-MM5 interpolation utility, NESTDOWN, to become the initial condition and boundary condition for uMM5.

[14] The key MM5 and uMM5 physics options were set as shown in Tables 1 and 2, respectively. The land-use data set was based on the USGS 24-category classification scheme. The land use map was updated to reflect the extent of PRD urbanization in 2003. Morphometric values were assigned only at locations designated as “Urban” in the land use map. Execution of MM5 employed two-way nesting. There was no fine-to-coarse mesh feedback between MM5 and uMM5 since they operate in decoupled configuration.

Table 1. Key MM5 Physics Options
Parameterization Option
Microphysics Simple Ice
Cumulus Grell, disabled in D03
Atmospheric radiation RRTM
Soil temperature Multi-layer
Table 2. Key uMM5 Physics Options
Parameterization Option
Microphysics Simple Ice
Cumulus Disabled
PBL Gayno-Seaman
Atmospheric radiation RRTM
Soil temperature Pleim-Xiu
Urban soil model SM2-U

[15] An experimentation period of relative meteorological consistency was selected to reduce environmental variability caused by weather events. For the benefit of any subsequent air quality analysis, this would help to explore effects of protracted exposure under a given weather pattern. A period that started on 6 January 2008 18Z and ended 72 h later was chosen. During this period, wind speed was relatively calm in the PRD. Wind speed over the expansive Guangzhou metropolitan area was below 2 ms − 1, with a strengthening that reached 3.5 ms − 1 near the end of the duration. Air pollution level at general monitoring stations in Hong Kong fluctuated in the Air Pollution Index (API) range of 51 to 100, or “High,” as categorized by the Hong Kong Environmental Protection Department. The Central District Roadside API reached as high as 121 on 8 January, with dominant contributing pollutant being NO2 (Figure 3). Of interest was evidence of an afternoon ozone burst detected at north-facing locations along the perimeter of the Hong Kong. This would usually be an indication of significant daytime regional transport from PRD industrial centers. On balance, this experiment period offered mild synoptic conditions and sufficient diurnal variations that would be conducive to air quality investigations.

Figure 3.

Time series of NO2 concentration at the Central District, Hong Kong air quality monitoring station, observed during 6–8 January 2008. NO2 was the contributing species to the high pollution index experienced throughout this period.

3.2 Urban Morphology

[16] Constrained by scarcity of urban meteorological observations in the PRD, a precise calibration of morphometric sensitivity in the UCP remained impracticable. Instead, this study attempted to arrive at a first-order approximation by seeking the limits of UBL adjustments that were attributable predominantly to urban morphometric transformation. Urban fabrics representative of two morphometric extremes set the backdrop to this comparative study. One urban morphometric extreme corresponded to the PRD urban morphometry and is henceforth referred to as “Urban,” and so are modeling results that pertain to it. The other morphometric extreme comprised an urban fabric that occupied an identical topographic distribution as Urban, although, with all built obstacles reduced to 1 m in height. This truncated morphometric variant is henceforth referred to as “Stub,” and so are modeling results that pertain to it. Stub is scientifically pertinent for a number of reasons. The land cover materials were identical to those in Urban, the built objects occupied the same footprints, and the urban green fraction and moisture drainage ratio were unchanged. And importantly, the perimeter of roughness transition from rural to urban remained constant. However, the horizontal heterogeneity in roughness were drastically dissimilar. The net effect was a region-wide morphometric transformation to reduce street-canyon height-to-width ratio, frontal area density, and volume of wall material, which is proportional to urban heat storage. This morphometric rearrangement had the potential to modify pressure drag and skin friction, street-canyon heat trapping, and canopy layer and roughness layer turbulence intensity.

4 Analysis

4.1 Organization of Results

[17] The uMM5 results were captured at every 30 model-minutes and were evenly grouped into eight time periods within each diurnal cycle—periods “p1” through “p8,” each spanning 3 h. “p1” designates 00:00–03:00 local time, “p2” designates 03:00–06:00, and so forth. Model output of each period was averaged over the three consecutive simulation days.

[18] In the following sections, vertical profiles of flow properties are displayed for two vertical extents. Figure 4 is an example of wind speed vertical profiles. The vertical axis of Figure 4a, ranging 0–3000 m AGL, is intended to cover maximum PBL height, while the vertical axis of Figure 4b, ranging 0–800 m AGL in most cases, is intended to approximate the combined roughness sublayer and canopy sublayer depths. Each graph contains a solid black profile and a solid blue profile. Black corresponds to Urban, while blue corresponds to Stub. Each profile represents the spatial average of selected urban grid cells. Urban cells surrounded by five adjacent urban cells are chosen, and they amount to 522 qualified grid locations. This topological configuration is intended to orient the analysis toward flow regimes that typically fall between isolated roughness flows and skimming flows [Oke, 1988 ,Figure 8.2]. Each graph also contains light color profiles which denote member profiles of the spatial average. With respect to the two morphometric extremes under study, Urban is considered an analogue to realism, hence the control case. Stub was considered the variant, whose region-wide characterization illustrated the effects of morphometric attenuation and embodied the essence of discovery. Therefore, Stub member profiles are shown in light color, whereas Urban members are hidden to improve legibility. The ensemble of light-colored profiles enhances intuition regarding the manner by which member profiles can depart from the mean. A high degree of variability amongst member profiles alludes to potential environmental influences that are not considered in the site selection logic.

Figure 4.

Sample of vertical profile of wind speed. Figure 4a shows the profile from ground level to 3 km AGL. Figure 4b shows the same data set from ground level to 800 m AGL. Black solid line with dot markers represents the spatially averaged Urban profile. Blue solid line with circle markers represents the spatially averaged Stub profile. Light colored lines represent Stub profiles that make up the spatial average.

4.2 Evolution of the Turbulence Kinetic Energy Profile (TKE)

[19] Figure 5 illustrates the diurnal evolution of the TKE vertical profile. During the hours 00:00 through 03:00 LST, the Urban profile indicates a significant TKE content in the first few hundred meters of the PBL, with a peak located near the urban canopy top. This can be explained by the relative strong shear TKE production immediately above the urban canopy top. Owing to a weak nocturnal urban heat island effect (not shown), the Stub profile exhibits minimal levels of TKE throughout the vertical profile. Dissipation of residual heat from the urban canopy continued to limit TKE production during 03:00–06:00, as evident in a general reduction of magnitude in both the Urban and Stub profiles. Due to longer nights in the winter, only part of the 06:00–09:00 period received significant solar radiation, hence a slight increase in TKE is observed in both Urban and Stub. Noteworthy is that from 00:00–09:00, the maximum TKE value maintained a greater than 4:1 ratio between the Urban and Stub profiles. From a turbulent mixing perspective, Urban was able to support intense mixing levels in spite of the weak background winds before sunrise. Continuing onto the morning hours of 09:00–12:00, insolation rapidly heated the land surface. The Urban profile maximum doubled in value, while that of Stub quadrupled. The Stub profile's remarkable switch in shape, compared to the previous period and relative to that of Urban, revealed an appreciably smaller thermal storage capacity in the shallow Stub canopy. Under solar radiation, a small thermal storage capacity resulted in larger release of sensible heat fluxes, enabling the Stub profile to surpass that of Urban at elevations above 120 m AGL. At elevation levels above urban canopy top, both Urban and Stub TKE profiles diminished with height. The profiles converged at about 950 m AGL (not shown), suggesting an elevation above which the atmosphere tends to become insensitive to significant urban morphometric variations. In the early afternoon period of 12:00–15:00, peak TKE of Urban increased by 25%, that of Stub increased by 67%, and the two were nearly equal in magnitude. Buoyancy combined with upward momentum flux resulting from sea-breeze (section 4.6) interacting with urban morphology to produce a convective urban boundary layer. This is evident in the nearly uniform wind speed profile in the well-mixed layer (Figure 10). In the vertical distance from Urban canopy top to the elevation where the profiles converge, Stub could become twice as energetic as Urban. Comparison of potential temperature profiles (not shown) indicates that Stub is consistently warmer than Urban by 0.4 °K from near the surface to approximately 700 m AGL. This condition in conjunction with evidence of stronger buoyancy TKE production during this period (Section 4.4) is a possible explanation for the more energetic Stub TKE profile above 100 m AGL. In the late afternoon period of 15:00–18:00, again, due to short winter days, only part of the period received sunlight. The Stub TKE profile weakened rapidly, while the greater Urban heat storage sustained a stable TKE level. During the 18:00–21:00 period, Urban TKE saw a slight increase, while the Stub profile reverted to a form similar to that observed overnight. Despite the absence of a persistent nocturnal low-level jet (LLJ) in Urban (Section 4.7), this Urban TKE increment could be the result of a transient onset of nocturnal LLJ, whose formation is delayed by the residual turbulent momentum flux energized by nocturnal urban heating. Beyond the immediate vicinity above the Urban canopy top, TKE levels receded notably.

Figure 5.

Diurnal evolution of TKE vertical profiles. Each panel represents 3 h time-averaging of corresponding periods over 3 days. The Urban maximum is greater than that of Stub for the entire diurnal cycle. Urban TKE is much greater than Stub TKE in the nighttime.

[20] The diurnal TKE profiles indicate that peak Urban TKE is greater than Stub TKE throughout the day. The Urban TKE profile shape changes at a lower rate than does Stub, likely a result of Urban's greater heat storage capacity and aerodynamic roughness. During the period 09:00–15:00, the combination of stronger turbulent lifting and higher wind speeds in Stub poses the possibility of Stub emissions advecting further downstream than do Urban emissions.

4.3 Shear Production of TKE

[21] The relative contribution to total TKE by shear production versus that by buoyancy production facilitates a more thorough understanding of of urban turbulent mixing. Figure 6 illustrates the diurnal evolution of TKE shear production. The shape of the Urban profiles generally resemble those [Christen et al., 2009 ,Figure 4a] presented for near-neutral stability conditions in urban setting. They show shear production increases with height and reaches a peak at an elevation of 1.0–1.5 times canopy depth, then gradually decreases with height, with an inflection at about 2.0 times canopy depth. During periods when a shear production peak is discernible, it is located at an elevation of approximately 83 m AGL, or 1.56 times canopy depth. The normalized height of maximum shear production agrees reasonable well with Christen et al. [2009]. On the other hand, the weak morphometric presence in Stub causes its shear production profile to resemble that of daytime rural land surface [Stull, 1988 ,Figure 5.4].

Figure 6.

Diurnal evolution of TKE shear production. Each panel represents 3 h time-averaging of corresponding periods over 3 days. A peak in shear production is observed at ≈ 1.56 canopy depth, similar to [Christen et al., 2009]. An increase in TKE is seen very near the ground, indicative of flow interaction with the reduced urban porosity.

[22] During the hours 03:00–12:00, shear production was weak compared to the rest of the day. This was attributed to the generally calm atmosphere prior to buildup of land-sea breeze circulation. From 09:00 to 03:00, shear production increased with time, to peak during early evening (18:00–21:00), then gradually declined to quiescent condition. An inflection in the shear production profile could be observed at deeper levels of the canopy sublayer. The height of this inflection suggests that it relates to the transition from the diminishing roof top shear and the beginning of strong vertical wind shear very near the surface.

[23] The Stub and Urban shear TKE production profiles are dissimilar. The Stub profiles monotonically decrease with height, with maximum values located near the surface where wind shear is the strongest. The Urban profiles exhibit a peak, whose magnitude varies with time of day, at approximately 83 m AGL in the vicinity above the average Urban canopy top. Deep in the Urban canopy, the profiles resemble those of Stub, although with smaller maximum values at the surface due to wind shear attenuation by urban constructs.

4.4 Buoyancy Production of TKE

[24] Figure 7 illustrates the diurnal evolution of buoyancy TKE production. During 00:00–09:00, buoyancy production was nearly zero and slightly negative below 500 m AGL. This implied that much of the nighttime turbulent mixing was energized by canopy top shear production and that the lower portion of the residual layer was thermodynamically neutral to stable. Positive buoyancy production commenced during 09:00–12:00, signifying convective PBL growth. Maximum buoyancy production occurred during 12:00–15:00, in contrast to 18:00–21:00 for maximum shear production. The form of the Stub profile resembled that over rural roughness. The Stub maximum was observed in the lowest grid layer ( ≈ 4 m thick) and was three times greater than the Urban maximum, which was located at ≈ 30 m AGL instead. The local maximum in the Urban profile could be attributed to a combination of heat accumulation in the urban canopy and cooler temperatures near the ground due to shadowing.

Figure 7.

Diurnal evolution of TKE buoyancy production. Each panel represents 3 h time-averaging of corresponding periods over 3 days. Nighttime buoyancy TKE production was negligible for Urban and Stub. Due to less heat storage capacity in Stub, the more intense sensible heat propelled its buoyancy production past that of Urban. The daytime Urban profiles show a reduced production near the ground, which can be a result of shadowing.

[25] Figure 8a exhibits buoyancy TKE production during 00:00–03:00 for standalone urban cells. The Urban profile shows a discernible degree of negative buoyancy production, indicating that thermodynamically driven vertical turbulent motion was being suppressed. Note that the Stub profile exhibits a minuscule positive production at the lowest level, implying a minute amount of sensible heat release. Figure 8b shows buoyancy production in the same period for urban cells surrounded by eight adjacent urban cells, analogous to a location away from rural-urban transition boundaries. Buoyancy production was less negative than that of the standalone case due the relatively warmer urban airshed. Furthermore, the segment of the Urban profile below canopy top, which was ≈ 53 m for urban centers, indicates positive buoyancy production. Although the magnitude is small, it highlights the stronger heat trapping due to urban sprawl and the sustained below-canopy mixing in urban centers.

Figure 8.

Buoyancy production of TKE, 00–03h LST. (a) Standalone urban cell. (b) Urban cell surrounded by eight adjacent urban cells. The UBL over sparse urbanization experienced negative buoyancy production.

[26] Model results indicated that Urban peak shear and buoyant productions were of similar magnitude. Although, Urban shear production was discernible at greater heights, up to 2500 m AGL, versus 500 m AGL for buoyancy production.

4.5 Evolution of Wind Profile

[27] In the UBL, demarcation of a well-formed surface layer is uncertain due to dynamical urban effects. The elevation at which the mean flow can be approximated by a MOST profile, if at all, is determined by the vertical extent of urban effects. Theoretical studies and field and wind tunnel measurements indicate that the near-surface mean wind is retarded by the presence of morphological perturbations, resulting in a momentum deficit. A wind speed reduction and interrelated adjustments to turbulent transport have crucial implications to pollutant transport and dispersion. The vertical structure and magnitude of this deficit are explored below.

[28] Figure 9 summarizes wind profile differences at the surface and at 200 m AGL, where the largest Stub-Urban profile difference appears to occur throughout the day. Figure 9a shows the diurnal Urban surface wind speed time series in dark gray, that of Stub in blue, and the subtraction of Urban from Stub in crimson. Surface wind speed is modeled at approximately 2 m AGL. Figure 9b shows an analogous data set resolved at 200 m AGL. In Urban, diurnal minimum wind speed occurred simultaneously at 200m AGL and at the surface. In Stub, diurnal minimum wind speed at 200 m AGL lagged up to 3 h behind that for ground level. The lag corresponded to the time necessary for Stub to develop sufficient turbulent momentum transport (drag) to reach 200 m AGL. This phenomenon is corroborated by the Stub-Urban TKE profile differences. In all cases maximum wind speed occur during 18:00 to 21:00. The maximum Stub-Urban difference occurred during 21:00–24:00. In the absence of significant synoptic forcing, the after-sunset occurrence of maximum wind suggests a strong residual thermal gradient flow driven by land-sea breeze circulation.

Figure 9.

Urban and Stub wind speed differences at the surface and at 200 m AGL. Urban-Stub wind speed difference was largest after sunset, and the smallest in the afternoon when the PBL was well mixed.

[29] The diurnal evolution of urban wind speed is shown in Figure 10. The Stub and Urban profiles also differ significantly in shape over the course of the day. In the nocturnal boundary layer, not only does the Urban profile exhibit a substantial momentum deficit relative to that of Stub, it also contains a maximum clockwise curvature at approximately 200 m AGL, roughly 3.7 times urban canopy depth. On the other hand, in spite of Stub's finite morphometric variations, its wind profiles exhibit no momentum deficit and resemble those produced in traditional mesoscale models. The Stub profile shape illustrates that the UCP is able to transition to a MOST framework when morphometric heterogeneity is attenuated to an extreme which can be approximated by a simple roughness length. As shown in previous TKE profiles, turbulent mixing was vigorous during daytime. The momentum deficit was efficiently reduced by turbulent exchange with faster moving air above. The result was a nearly constant vertical wind profile typical of a convective boundary layer. Throughout the day, Urban exhibits a momentum deficit relative to Stub, though, at smaller proportions in daytime. The difference between Urban and Stub supports the hypothesis that the regional UBL wind field adjusts to morphometric changes.

Figure 10.

Diurnal evolution of wind profile. Each panel represents 3 h time-averaging of corresponding periods over 3 days. Urban, due to its deep urban canopy, slowed the wind field significantly. A nocturnal low level jet was observed in Stub for the 18:00 to 21:00 period.

4.6 Regional Urban Wind Field

[30] Figure 11 shows region-wide diurnal evolution of Urban wind speed at 10 m AGL. Figures 11a– 11h correspond to a sequence of 3 h periods that begin at 00:00–03:00 LST and ends at 21–24:00 LST. Rural wind was generally stronger than urban wind and suburban wind was stronger than urban center wind. Except for 12:00–18:00, the estuary coastal zone north of Shenzhen experienced particularly weak winds. However, the vast Guangzhou metropolis and the Zhongshan vicinity exhibited the weakest wind throughout the day relative to other densely built urban clusters in the region. During 00:00–03:00, regional wind showed a southeasterly orientation, along the direction of sea-breeze circulation that developed in the previous day. Streamlines were less organized than at the height of sea-breeze circulation in late afternoon. During 03:00–06:00, regional wind reached its weakest point, with urban locations exhibiting wind speeds below 2 ms − 1, which was corroborated by previous wind profiles. By the 06:00 to 09:00 period, a right turning tendency developed in the region's wind pattern. This signified a strengthening land-breeze. The right turning tendencies continue through the 09:00–12:00 period, when a discernible land-breeze pattern appeared in the estuary. As the rural and urban land surface continues to emit sensible heat under insolation during 12:00–15:00, the land-sea thermal gradient reached a necessary threshold, and the regional wind field rapidly returned to a sea-breeze pattern, as evident in the streamlines crossing the estuary (Figure 11e). In the late afternoon, after accumulating more than 6 h of solar irradiation, the land-sea thermal gradient became even stronger, as evident in the estuary wind speed reaching over 5 ms − 1. This strong forcing also rotated the estuary sea-breeze pattern northward. After dusk, the afternoon wind pattern persisted, although with gradually decreasing wind speeds.

Figure 11.

Diurnal evolution of surface wind field [m s − 1]. Each panel represents 3 h time-averaging of corresponding periods over 3 days. In Figure 11a, locations of Guangzhou (triangle), Dongguan (rectangle), Zhongshan (rhombus), and Shenzhen (oval) are denoted. Lower wind speeds were observed over urban clusters. The calmest period was 00:03–00:06, when streamlines were also more chaotic.

[31] Figure 12 shows the vertical evolution of the PRD wind field during 15:00–18:00. By this time a robust sea-breeze pattern had already developed. Wind maps for elevations below 500 m AGL exhibit flow perturbation by urban morphology and by orography. Over the Greater Guangzhou area, wind speed was broadly lower due to morphometric influences. As expected, morphometry induced wind retardation diminished with increasing elevation. At ≈ 800 m above mean sea level (MSL) (pressure = 916 mbar, Figure 12b) the flow approaching Guangzhou from the Pearl River Estuary indicated acceleration. Such a wind field signature signifies the transition from cell-specific vertical urban effects to onset of a skimming flow in which the PRD urban fabric behaved jointly as an orographic protrusion over which wind flow accelerated. Streamlines at this elevation are devoid of divergence about urban obstacles, although, weak bifurcation around orography can still be observed at some locations. At ≈ 3000 m MSL (pressure = 688 mbar, Figure 12c), the wind field exhibited faint signatures that are arguably attributable to urban effects. At ≈ 3800 m MSL (pressure = 618 mbar, Figure 12d), the only discernible land surface effect was orographic. The streamlines have gradually spiraled from a southerly pattern at the surface to a westerly geostrophic pattern at this elevation, signaling an exit from the atmospheric Ekman layer.

Figure 12.

Vertical evolution of regional wind field along σ coordinate surfaces during 15:00–18:00. (a) Locations of Guangzhou (triangle), Dongguan (rectangle), Zhongshan (rhombus), and Shenzhen (oval) are denoted. Flow lines are on the σ surface. Wind speed in [m s − 1] is evaluated at and tangent to the σ surface. (a) Urban cluster signatures were discernible near the surface. (b) Near 800 m AGL, the flow sensed Guangzhou City as a massive orographic protrusion, as if in a skimming flow regime. (c) Note the re-scaling of colors. Urban effects were hardly detected. (d) Only orographic effects were discernible. Streamlines assumed geostrophic direction, signaling an exit from the atmospheric Ekman layer.

[32] To further appreciate the nature of wind field differences between Urban and Stub, Figure 13 shows the subtraction of the 6 m AGL Urban wind field from that of Stub. The streamlines in these diagrams corresponded to Stub, unlike those in Figure 11 which correspond to Urban. Striking differences between Stub and Urban streamlines are apparent throughout the day. Stub streamlines can maintain a steady heading pattern as they cross into urban zones, whereas Urban streamlines exhibit higher degrees of chaos. It is reasonable to deduce that the lack of meandering through urban roughness contributes to stronger surface winds. Moreover, Stub's orderly streamline patterns imply that the Urban wind field is more dispersive than Stub. Therefore, an attenuated morphometric environment had the potential to allow pollution plumes to maintain a narrow spread over greater distances. Wind speed differences in these diagrams are corroborated by those already presented in vertical profiles. Namely, nighttime differences are greater than daytime differences; urban center differences are greater than suburb differences. Depending on time of day, significant differences can also manifest in the Pearl River Estuary where the surface roughness is independent of urban morphometry. This demonstrates the ability of urban effects to reach horizontal distances many times the dimension of urban canopy depth.

Figure 13.

Diurnal evolution of wind speed differences, WindStub − WindUrban [m s − 1]. In Figure 13a, locations of Guangzhou (triangle), Dongguan (rectangle), Zhongshan (rhombus), and Shenzhen (oval) are denoted. Color temperature denotes the amount that Urban wind was slower than Stub wind. Regional differences were the largest during 18:00–24:00. Differences over urban clusters were greater than those over rural land.

4.7 Nocturnal Low Level Jet

[33] As solar irradiation recedes during dusk, a nocturnal inversion develops over land. Surface air density increases and turbulent mixing subsides. In effect this curtails the turbulent momentum transfer (drag) between surface roughness elements and the subgeostrophic mixed layer wind that has built up over the course of the day. At this point horizontal pressure gradient accelerates the winds back toward geostrophic. Concurrently, Coriolis forcing causes an inertial oscillation in the wind which becomes supergeostrophic in the ensuing night hours [Stull, 1988].

[34] Figure 10g shows the mean wind profile for the early evening hours 18:00–21:00. A nocturnal LLJ developed in Stub, evident in the well-correlated member profiles that culminated to show an accelerated layer that reached from ground level to 600 m AGL. Maximum wind speed in the Stub LLJ occurred at 200 m AGL and was 7 ms − 1, 2.5 ms − 1 stronger than Urban wind, and 1.5 ms − 1 stronger than mixed-layer wind at large. Meanwhile, the Urban profile exhibited no sign of LLJ development. Continuing through 21:00 to 24:00, Urban continued to exhibit no clearly distinguishable nocturnal LLJ (Figure 10h). As the night progressed, the Stub LLJ height increases, and the maximum speed decreased. The onset of a weak Urban nocturnal LLJ was discernible during 00:00–03:00, indicating a 6 h lag behind Stub (Figure 10a). The strongest Urban LLJ profile occurs during 21:00–24:00 (Figure 10h). However, vertical wind shear in the Urban nocturnal LLJ is much weaker than that in Stub. Meanwhile the strength and elevation of the Stub nocturnal LLJ continue to rise. By 06:00 (Figure 10c), neither Urban nor Stub registers a nocturnal LLJ signature in the wind profile.

[35] Relative to Stub, Urban morphology provided

  1. [36] stronger aerodynamic resistance, hence shear turbulence production, and

  2. [37] greater heat storage, hence buoyancy turbulence production.

[38] These Urban conditions enabled sustained turbulent mixing between the stable nocturnal boundary layer and the residual layer. It is possible for the turbulent drag to delay the initiation of layer slippage which signified development of the nocturnal LLJ. By 00:00–03:00, urban heat storage was so depleted that it became insufficient to maintain turbulent momentum exchange with the residual layer and an LLJ developed. However, by this time the momentum imbalance between the residual layer and the nocturnal boundary layer was weaker than it was when the Stub nocturnal LLJ developed. This led to a weaker Urban LLJ as evident in the wind profile. A weak or nonexistent nocturnal LLJ above built areas can deprive the urban airshed of nocturnal ventilation of pollutants.

4.8 Summary of Urban Meteorological Modeling

[39] Through urban parameterized mesoscale meteorological modeling, this study has demonstrated the physical impacts that morphometric adjustments impose on the PRD regional boundary layer. Meteorological predictions with respect to two extreme morphometric environments have been reviewed and compared in detail. A nocturnal Low Level Jet formed over the deep Urban canopy, albeit a delayed commencement and at reduced strength. During nighttime, due to the greater amount of stored heat and aerodynamic drag inherent to the Urban morphology, it was able to sustain greater levels of TKE production. TKE shear production profiles for Urban exhibit a peak at an elevation 1.6 times canopy depth, which is in reasonable agreement with published field observations. The height of the shear production peak indicates the existence of a shear layer just above the urban canopy top. Nocturnal TKE buoyancy production was minuscule for both Urban and Stub, though, in the daytime buoyancy production was much stronger over shallow canopies. Mean wind was stronger over Stub while the difference was larger during nighttime. Land-sea breeze circulation was observed in both Urban and Stub. Deep canopies in Urban caused approaching streamlines to divert and meander around urban obstacles more so than did Stub.

5 Summary

[40] This study set out to explore the influences that urban morphometric modifications exert on the dynamical properties of the urban boundary layer. Using an urban parameterized mesoscale meteorological model, uMM5, experiments were conducted to evaluate the limits of boundary layer adjustments in response to urban morphometry attenuation. uMM5 predictions for a deep urban canopy (Urban case) and a shallow canopy (Stub case) were compared. The morphometric dissimilarities were found to cause spatiotemporal adjustments to the diurnal evolution of the PBL structure. A majority of adjustments arose from differences in aerodynamic drag and friction, and the storage and release of heat by the urban fabric. This UCP sensitivity study suggests that improved urban morphological realism is able to reduce positive wind speed bias observed in conventional mesoscale meteorological models when applied to the PRD region. Urban meteorological measurements will enable model calibration and improve realism. Future studies which allow for coupling of uMM5 meteorological output fields to chemical transport models can facilitate the assessment of regional urban morphometry impacts on air quality.

Appendix A: Urban Canopy Parametrization: Meteorology

[41] The Reynolds averaged momentum equation is written as

display math(A1)

ui⟩ denotes the Reynolds averaged velocity, ui is the horizontal wind speed component, and j is the land cover type, such as building, vegetation over artificial surface, or vegetation over natural land. inline image represents the canonical forcing terms in the Reynolds averaged momentum equation before urban parameterization. inline image represents frictional forces induced by building roofs. In expanded form, it is written as

display math(A2)

ρ(k) is the air density at grid level k. Va(k) is the volume air density, with unit (m 3m − 2), at the level k. fb is the horizontal surface density of buildings, with unit (m 2m − 2), and is defined as the ratio between the top area of buildings projected vertically at the ground level and the area of the horizontal section of the grid cell. ξb(k) is the top area density at level k above the ground and is defined as the ratio between the top area of buildings at the level k and the entire top area of buildings projected vertically at ground level. u * b(k) is the friction velocity induced by roofs. The form of the expression shows that frictional drag on roofs is dependent on height at level k, and is averaged by the free space in the cell, Va(k). The term inline image is computed as in [Guilloteau, 1998], which is based on Monin-Obukhov Similarity Theory. In other words, it is assumed that a well-formed surface layer existed over individual roofs in a cell. While this method offers computational efficiency, the suitability of its underlying assumption has yet to be verified by experiment. inline image represents pressure and viscous drag forces induced by vertical surfaces. This term is summed over land cover types, implying that the flow can exchange momentum fluxes with vegetation, in addition to built objects. In expanded form, it is written as

display math(A3)

The form of the equation resembles the simple aerodynamic drag formula

display math(A4)

where FD is the form drag, ρ is air density, A is the reference area, CD is the drag coefficient, and V is the wind speed. Term inline image in equation (A3) is the effective drag coefficient. The effective drag coefficient theoretically should account for sheltering between canopy objects, though for simplicity, the sheltering effect is not considered here. The land-cover-type index, j, in inline image implies that the term varied by land cover type. In this study, inline image for buildings, inline image for vegetation over natural ground and artificial surfaces, respectively.

[42] The urban canopy parameterized TKE equation is written as

display math(A5)

where E is the TKE, θv is the virtual potential temperature, g is the gravitational acceleration, and Km is the eddy diffusivity for momentum. Term  I represents advective transport by the mean flow. Term  II represents dynamic shear production. Sa is defined as the ratio between the air area at level k and the area of the horizontal section of the computational cell. inline image represents shear production by roofs. Term  III represents buoyancy production. HE correspond to production from urban sensible heat flux and anthropogenic heating (Equations (A8) and (A9)).Term  IV represents turbulent transport. Term  V represents turbulent dissipation. Term  VI represents wake production. Term  VII represents accelerated turbulence cascade prompted by eddy impingement against urban surfaces.

[43] Urban surface fluxes, such as latent and sensible heat, moisture, and evapotranspiration, are accounted by combining the SM2-U(3D) soil model [Dupont et al., 2002] with the Pleim-Xiu land surface scheme. The general form of the heat equation is

display math(A6)

where θL is the liquid-water potential temperature, which is written as

display math(A7)

where θ is the potential temperature, L is the latent heat of evaporation, π is the Exner pressure function, qL is the cloud-water mixing ratio, and Cp is the specific heat of air. Rθ represents the general forcing terms. Dθ represents sensible heat sources from roofs and vegetation, and is written as

display math(A8)

HS(k) is the mean sensible heat flux emitted at level k by buildings and vegetation per unit of ground area. Elaboration of this term is documented in equation (23) of Dupont et al. [2004] and its Appendices A and B. The term Aθ represents anthropogenic heat sources, such as vehicle emission and residential heating and cooking. Aθ is written as

display math(A9)

where Qυ is the anthropogenic heat flux. Formulation of Aθ is detailed in equations (B9)–(B11) in Dupont et al. [2004].

[44] In addition to moisture flux exchange with ground soil, moisture in the canopy layer arose from evapotranspiration from urban vegetation and from evaporation of water stored in urban water bodies and intercepted by structures. The atmospheric water content is expressed as

display math(A10)

where qw is the total water content, Rq is the general moisture forcing, and Dq is the moisture forcing from structures and vegetation. Dq is defined as

display math(A11)

where ζ(k) is the mean moisture flux per unit ground area from urban surfaces located at level k in the canopy. Formulation of this moisture flux variable is detailed in Appendix B of Dupont et al. [2004] as Emean(k).

Appendix B: Urban Canopy Parameterization: Morphology

[45] The street canyon height-to-width ratio, λS, is written as

display math(B1)

where H1 and H2 are heights of buildings 1 and 2. S12 is the horizontal distance between buildings 1 and 2, where building 1 is upwind of building 2. The building wall-to-plan-area ratio is written as

display math(B2)

where AR is the plan area of rooftops. AW is the total area of non-horizontal roughness element surfaces. For constant cross-section structures, this is equivalent to the total area of walls. AT is the total plan area of the grid cell. The area-averaged building height, inline image, is written as

display math(B3)

where Ai is the plan area of building i at ground level. hi is the height of building i. N is the number of buildings in the cell. The top area density as function of height, Ar(z), is written as

display math(B4)

Δz is the height increment, set to 1 m in this study. AT is the total area within which the buildings are contained. Ar(z) is the roof area at height z. ap(z) is the plan area of buildings at height z. The frontal area density, af(z,α), is written as

display math(B5)

where inline image is the area of building surface projected into the plane normal to the approaching wind direction for a specified height increment, Δz. α is the wind direction. The height dependent plan area density, ap(z), is written as

display math(B6)

where Ap(z ′ ) is the plan area of buildings at height z ′ . Δz is the height increment. Calculations for the frontal area, plan area, and top area densities are repeated for urban trees. Equations (B1) and (B5) are anisotropic. The absence of morphological orientation sensitivity in the UCP was mitigated by averaging these variables over multiple directions.


[46] This research effort was made possible by grants N_HKUST630/04, N_HKUST631/05SB106/07.SC06, RGC612807, and RGC615406, collectively administered under the auspicious of The Research Grants Council of the Government of the Hong Kong Special Administrative Region of the People's Republic of China. The deepest appreciation is extended to each and every JGR reviewer who helped to make this presentation coherent and to improve its scientific quality.