The world’s population is increasingly concentrated in large urban areas. Many observational and modeling studies have explored how these large, population-dense cities modify local and mesoscale atmospheric phenomena. These modeling studies often use an urban canopy model to parameterize urban surfaces. However, it is unclear whether this approach is appropriate for more suburban cities, such as those found in the Great Plains. Thus, the Weather Research and Forecasting Model was run for a week over Oklahoma City, Oklahoma, and results were compared with observations. Overall, four configurations were examined. Two simulations used the Noah LSM, one with all urban areas removed (CTRL), and the other with urban areas parameterized by a modified Noah land surface model with three urban categories (LSMMOD). Additional simulations utilized a single-layer urban canopy model (SLUCM) either with default urban fraction values (SLUCM1) or with urban fractions taken from the National Land Cover Database (SLUCM2). Results from the three urban runs compared favorably to high-density temperature observations of the urban heat island. The SLUCM1 run was the most realistic, although the urban fractions applied were the least representative of Oklahoma City. All urban runs also produced a drier and deeper planetary boundary layer over the city. The prediction of near-surface winds was most problematic, with the two SLUCM runs unable to correctly reproduce reduced wind speeds over the city. The modified Noah LSM provided best overall agreement with observations and represents a reasonable option for simulating the urban effects of more-suburban cities.
Meteorological conditions in and around urban areas vary from their rural surroundings because of the special properties of urban areas (Oke 1976, 1981, 1982; Arnfield 2003; Barlow 2014). To better understand urban–rural differences, researchers have undertaken efforts in the past few decades to observe and quantify the range of effects that urban areas can have on the environment. These efforts found that urban areas, when compared with nearby rural and naturally vegetated areas, have generally warmer air temperatures, especially at night (e.g., Gedzelman et al. 2003; Yow and Carbone 2006; Alonso et al. 2007; Yang et al. 2013; Hu et al. 2016), a feature frequently referred to as the urban heat island (UHI; Oke 1982). Although not as well documented, in situ measurements also suggest slower wind speeds, most prominently during the day (Dou et al. 2015; Hu et al. 2016), and decreased humidity (Dou et al. 2015) in urban areas.
While the exact mechanisms through which urban areas modify their environment are not fully understood, the meteorological conditions observed in urban areas are likely caused by 1) changes in surface properties, such as albedo, emissivity, and thermal conductivity, due to the replacement of natural vegetation with man-made materials; 2) higher anthropogenic heat emission; 3) increased surface roughness caused by the geometry of buildings and street canyons; 4) decreased surface evapotranspiration potential due to less exposed soil and decreased vegetative cover; and 5) urban aerosol emissions (Oke 1982; Oke et al. 1991). These land–atmosphere interactions can also be modulated by larger-scale meteorological conditions; moderate wind speeds (e.g., Alonso et al. 2007; Hu et al. 2013b, 2016), cloud cover (e.g., Alonso et al. 2007), precipitation (e.g., Oke et al. 1991), low rural soil moisture (e.g., Winguth and Kelp 2013; Husain et al. 2014), and snow cover (e.g., Malevich and Klink 2011; Smoliak et al. 2015) can decrease, eliminate, or even reverse the UHI effect (i.e., create an urban cool island).
Microscale atmospheric modifications by urban areas can also affect mesoscale weather phenomena. For example, researchers have observed urban areas across the globe to cause earlier arrival of spring (Alonso et al. 2007), fewer freezing rain events (Changnon 2003), thunderstorm and precipitation pattern changes (e.g., Changnon et al. 1991; Shepherd et al. 2002; Dou et al. 2015), altered lightning frequency (e.g., Orville et al. 2001; Rose et al. 2008; Tan et al. 2016), and more severe floods (e.g., Smith et al. 2002). Thus, understanding and accurately predicting the effects that urban areas have on their environment will continue to be important as urbanization increases worldwide.
Toward this goal, recent studies have used the Noah land surface model (LSM; Chen and Dudhia 2001; Ek et al. 2003) coupled with a single-layer urban canopy model (SLUCM; Kusaka et al. 2001; Kusaka and Kimura 2004) in the Advanced Research version of the Weather Research and Forecasting (WRF-ARW; Skamarock and Klemp 2008) Model to investigate land–atmosphere interactions in urban areas. Researchers have performed modeling studies in large, dense cities across the globe, such as New York City, New York (Gutiérrez et al. 2015b); Taipei, Taiwan (Lin et al. 2008); Beijing, China (Miao et al. 2009); Baltimore, Maryland–Washington, D.C. (Zhang et al. 2011; Li et al. 2013); Mexico City, Mexico (Cui and de Foy 2012); Hangzhou, China (Chen et al. 2014); Tokyo, Japan (Adachi et al. 2014; Takane et al. 2015); and Nanjing, China (Chen et al. 2016). Although less numerous, some studies have used the LSM–SLUCM framework in the WRF to model urban effects of smaller or less dense cities such as Las Vegas, Nevada (Kamal et al. 2015); Dallas–Fort Worth, Texas (Hu et al. 2016); Oklahoma City, Oklahoma (Hu et al. 2013a); Houston, Texas (Chen et al. 2011); and Phoenix, Arizona (Grossman-Clarke et al. 2010; Shaffer et al. 2015). These investigations showed that using the SLUCM to parameterize urban areas improves model representation of UHI intensity (UHII), or the difference between urban and rural near-surface temperatures, and urban boundary layer development.
In the WRF LSM–SLUCM modeling framework, the Noah LSM is used to compute surface fluxes at all nonurban grid points based on vegetative and soil properties such as surface roughness, emissivity, and albedo, as well as soil moisture and temperature. In urban areas, the SLUCM is used to compute fluxes in the urban canopy, taking into account differences between building materials, building height, and more complex surface geometry. The LSM is also used at these locations to compute fluxes; however, the LSM considers the grid point to be “natural” (i.e., nonurban). In this study, grassland was used to parameterize the natural, nonurban surface. Each urban grid cell is then partitioned based on urban fraction (i.e., how much of a grid cell is urbanized), and surface fluxes from the LSM and SLUCM are aggregated.
To analyze the performance of the LSM–SLUCM modeling system, many investigations have made modifications to land-use categories, urban fraction, vegetation percentage, and various urban canopy and LSM parameters. For example, Miao et al. (2009) used the WRF Noah LSM–SLUCM system with modified land-use categories, building height, and anthropogenic heat release within the urban area. They concluded that increasing urban development intensity resulted in a warmer and drier urban environment extending to 1.8 km AGL. In addition, their results suggested that increased building height slows nocturnal wind speeds in the lowest 1 km AGL because of increased friction but enhances vertical mixing during the day, resulting in stronger wind speeds. Cui and de Foy (2012) found that increasing vegetative cover in an urban area moderates the UHI effect. Others, such as Li et al. (2013) and Chen et al. (2014, 2016), simulated urban expansion by changing the urban land-use representation to that from various years. As urban extent and intensity increased, so did sensible heat, ground heat storage, and air temperature throughout the boundary layer. Additionally, Adachi et al. (2014) found that a compact city with the same population as a disperse one (simulated by modifying urban fraction) had a weaker average UHI but had higher temperatures in the urban core. These findings suggest that accurate urban land-cover representation is essential for accurately simulating meteorological conditions in urban areas.
While most studies have found the WRF to properly reproduce many aspects of the boundary layer, some weaknesses still exist, particularly the simulation of boundary layer winds. Zhang and Zheng (2004) and Ngan et al. (2013) found that phase and amplitude errors exist across all WRF PBL schemes throughout the depth of the PBL, both during the day and at night. In addition, Lee et al. (2011), Li et al. (2013), and Hu et al. (2016) found errors in boundary layer winds in urban areas, particularly near the surface. However, there is a paucity of observational studies that investigate the modification of winds in urban areas alongside UHI effects against which to compare these modeling results (Klein 2012; Klein and Galvez 2015; Hu et al. 2016). In addition, as many modeling investigations of the urban canopy focus on intense heat wave events with weak synoptic wind speeds (e.g., Chen et al. 2014; Takane et al. 2015), few closely examine urban modification of the wind field. Additionally, studies that examine time periods with moderate wind speeds frequently focus on land–sea breeze processes (e.g., Chen et al. 2011; Gutiérrez et al. 2015b; Sharma et al. 2016), a phenomenon separate from the daytime low-level wind speed increase that results from radiation-induced turbulence. Chen et al. (2016), using a robust analysis of WRF-simulated 10-m wind speed modifications in a continental urban area, found that while increasing urban extent improved the accuracy of simulated winds, urban wind speeds were still too high.
Therefore, this study will investigate the sensitivity of boundary layer temperature, moisture, winds, and structure over Oklahoma City to urban land surface parameterizations within the WRF Model under conditions typical of U.S. continental springtime. We will focus on changes to the simulated representation of this Great Plains urban area that result from the choice of various urban parameterizations. In the next section, the study area, model configuration, and code modifications will be discussed, followed by sections covering results and conclusions.
2. Data and methods
a. Study area and synoptic background
Oklahoma City (OKC) (35°28′56″N, 97°32′06″W) is located in central Oklahoma in the Sandstone Hills region of Oklahoma, which is characterized by 75–120-m-tall rolling hills and sits on the North Canadian River. The metropolitan area, however, is quite flat (Fig. 1). OKC has a humid subtropical climate, with daily mean temperatures ranging from 4°C in January to 29°C in July and an average of 912 mm of annual precipitation. The average prevailing wind in May is from the south-southeast at 3.5 m s−1. OKC ranks as the 7th largest city in the United States by land area (~1610 km2), and, as of 2010, 27th largest by population (~580 000; U.S. Census Bureau 2010). While it has a small central business district (~27 km2; Burian et al. 2005) with a number of buildings with more than 20 stories (>60 m), the majority of OKC is a mixture of suburban homes with 1–3-story (3–10 m) commercial buildings along the major arterial roadways. As most examinations using the SLUCM are preformed over dense cities (e.g., Holt and Pullen 2007; Lin et al. 2008; Miao et al. 2009), and only a few have modeled less dense cities (e.g., Grossman-Clarke et al. 2010; Shaffer et al. 2015; Kamal et al. 2015; Hu et al. 2016), it remains unclear if an urban canopy model can handle appropriately this type of city structure common in the Great Plains.
The time period chosen for this study, from 1 to 7 May 2010, was selected because it is characterized by conditions typical of midspring in central Oklahoma, it occurs just prior to the 10 May 2010 severe weather outbreak in central Oklahoma, and it resides within the time period during which Hu et al. (2016) analyzed observations from the OKC Micronet (OKCNET; Basara et al. 2011). One goal of this research is to select an urban parameterization that is appropriate for use in a severe weather simulation; hence, the proximity of the study’s time period to the 10 May 2010 outbreak is ideal. This week at the beginning of May was characterized by a 500-hPa trough over the mountain west (Fig. 2), as well as minimal cloud cover, very little rain, high temperatures between 20° and 35°C, and moderate wind speeds in the OKC area (Fig. 3). A UHI was present during each day of this 7-day period (Fig. 3a), and rural wind speeds were consistently stronger than urban wind speeds at similar times (Fig. 3b).
b. Model configuration
The simulations for this study were performed using the WRF-ARW Model, version 3.6.1. Three one-way nested domains with grid spacing of 4.5, 1.5, and 0.5 km on grids of size 300 × 300, 400 × 400, and 399 × 399, respectively, were used (Fig. 4a). Each domain had 119 terrain-following vertical levels from the surface to 100 hPa (~16 km), with vertical spacing of ~50 m from the surface to 800 hPa (~2 km) and ~150 m above that. Approximately 20 of these levels were below 1 km, providing fine vertical resolution in the PBL.
Identical physics schemes were used across all domains, including the double-moment, six-class, graupel particle density–predicting NSSL microphysics scheme (Mansell et al. 2010), the Goddard short- and longwave radiation schemes (Chou and Suarez 1999; Chou et al. 2001; Matsui and Tao 2007), the Yonsei University (YSU) PBL scheme (Noh et al. 2003), the MM5 Monin–Obukhov surface layer scheme (Monin and Obukhov 1954; Paulson 1970; Dyer and Hicks 1970; Webb 1970), and the Noah LSM. The YSU PBL scheme was chosen because it reproduces PBL moisture and temperature profiles accurately in warm, moist severe weather environments (Coniglio et al. 2013; Clark et al. 2015) in addition to its ability to simulate surface energy balance terms in rural and urban areas well (Shaffer et al. 2015). In two of the four simulations, noted SLUCM1 and SLUCM2 (Table 1), the single-layer urban canopy model was employed to simulate the effects of urban land use. The SLUCM default values for heat capacity, conductivity, albedo, emissivity, and roughness lengths for heat and momentum of roof, road, and wall surfaces (Loridan and Grimmond 2012) were employed. Anthropogenic heating in OKC, estimated at <5 W m−2 in the summer by Sailor et al. (2015), was set to zero for all SLUCM simulations. Additionally, observations indicate that soil moisture was at normal levels in early May 2010, so although irrigation in urban areas is not explicitly parameterized in the present model configuration, it would likely not play a large role, if any, in modifying observations.
Each WRF simulation was initialized at 0000 UTC 1 May 2010 (1800 CST 30 April 2010) and was integrated for 168 h until 0000 UTC 8 May 2010 (1800 CST 7 May 2010). Initial and lateral atmospheric boundary conditions were taken from the 13-km Rapid Update Cycle (RUC; Benjamin et al. 2004) at 1-h intervals. Soil state variables were initialized with offline North American Land Data Assimilation System-2 (NLDAS-2) 0.125° Noah model output (Xia et al. 2012) provided by the Goddard Earth Sciences Data and Information Services Center.
c. Update of urban land-use and urban fraction data
The default land-use and land-cover (LULC) arrays used by the Noah LSM are computed from (at best) 30-s Global Land Cover Characterization (GLCC) data [available from the U.S. Geological Survey (USGS)], which is derived primarily from 1992–93 Advanced Very High Resolution Radiometer (AVHRR) 10-day normalized difference vegetation index (NDVI) composites (Loveland et al. 2000). These data are categorized according to the USGS 24-category LULC system. Using these datasets results in coarse, sometimes inaccurate, representations of finer-scale land surface features that change frequently over time, such as urban areas (Figs. 4c,e). In addition, the USGS LULC data have only one category of urban land use. Consequently, complicated urban morphology is represented by just one urban class, such that the majority of the OKC metropolitan area is defined simply as urban.
As an accurate representation of the OKC urban area is critical to this study, LULC and urban fraction data were modified (for some simulations; Table 1) using data from the 2011 National Land Cover Database (NLCD; Homer et al. 2015). The NLCD provides a 10-m, 20-category, four-urban-type continental U.S. land-cover classification (LCC; Fig. 4b) derived from Landsat data. Unfortunately, the NLCD LCC data use a different classification scheme than the USGS data, and LULC data classified using the USGS categories are required by the Noah LSM. Thus, the NLCD data were reclassified to match USGS’s 24-category classification system (Table 2), with the developed open space (DOS) and developed low intensity (DLI) categories merged to allow for three urban categories.
When the SLUCM is used, its default strategy is to assign gridcell urban fraction percentage based on urban category. If the single-urban-category USGS LULC dataset is used, all urban grid cells have 90% urban fractional coverage. However, when paired with NLCD LULC data with remapped three urban categories, urban fraction values of 50%, 90%, and 95% are assigned to the low-intensity residential (LIR), high-intensity residential (HIR), and commercial (COM) urban LULC types, respectively. In addition, NLCD’s impervious surface area (ISA; Xian et al. 2011) data are used as a direct substitute for urban fraction, such that urban fraction is no longer assigned a default value according the land-use type. To arrive at the data used in the WRF, raw ISA data were median aggregated to 510-m cell sizing then resampled to the same 500-m grid sizes as the LULC data using the nearest-neighbor (NN) method. The WRF ingests these data and creates the urban fraction array on the model grid using NN interpolation. Note that combining the DOS and DLI categories into the USGS’s LIR category tends to overestimate the default urban fraction () of the grid points that were originally categorized as NLCD DOS, which is typically assigned to areas with 0% ≤ ISA ≤ 20% (Homer et al. 2004; Shaffer et al. 2016).
d. Noah LSM modifications
In its original form, the Noah LSM uses a single land-use category to represent all urban areas (e.g., a bulk-urban scheme; Liu et al. 2006). To take advantage of the NLCD three-category urban land-use information, the Noah LSM code and parameter table were modified. The added entries from the Noah LSM’s parameter table are shown in Table 3. Two variables from this table, and α, both empirical vegetation-dependent parameters, are used to parameterize latent heating within the model through their control on canopy resistance. In the Noah LSM (Chen and Dudhia 2001), canopy resistance is computed as
where is the minimum possible canopy resistance; LAI is leaf area index; and the , , , and factors correspond to stomatal response to insolation, vapor pressure deficit, ambient air temperature, and soil moisture, respectively. The factor is defined as
where is the maximum possible canopy resistance (typically set to 5000 s m−1), is insolation reaching the surface, and is an empirical scaling parameter assigned for each vegetation type, typically varying from 30 W m−2 for forests to 100 W m−2 for cropland (Jacquemin and Noilhan 1990). Holding all else constant, larger values of typically lead to larger values of canopy resistance and hence lower evapotranspiration. In addition, the factor, which parameterizes stomatal response to vapor pressure deficit, is dependent on α as
where is the saturation water vapor mixing ratio of the near-surface air, is the water vapor mixing ratio, and α is an empirical vegetation-dependent parameter, varying from 40 for crops to 150 for forests and 300 for shrubland. As with , larger values of α produce greater canopy resistance and less evapotranspiration.
To achieve very little evapotranspiration from urban areas, the Noah LSM sets and α to 999 W m−2 and 999, respectively. These large parameter values result in very small and factors, thereby creating an abnormally high canopy resistance and little to no evapotranspiration. However, for cities such as OKC, where many urban locations have grassy areas and trees, this treatment of evapotranspiration is not appropriate. To allow some latent heat flux within urban areas, and α were modified to values of 150, 200, and 250 W m−2 and 75, 100, and 125, respectively, for LIR, HIR, and COM, respectively. These values are higher than the highest value used for any vegetation type (maximum of 100 W m−2 for cropland and grassland and maximum α of 60.00 for wetlands). Hence, the chosen parameter values should provide a greater resistance to evapotranspiration than an entirely vegetated area but not eliminate evapotranspiration completely, as is achieved by the default values. In addition, these parameters were scaled below these given values in proportion to urban fraction; hence, the quantities shown here indicate maximum values.
The roughness values assigned to NLCD categories developed low, medium, and high intensity (0.7, 1.5, and 2.0 m, respectively) were used as a first guess and then adjusted until LSMMOD results agreed well with OKCNET wind observations. This adjustment was conducted via a trial-and-error approach wherein different values were tested within a physically reasonable range, given the known characteristics of the land cover, until the results agreed with observations. This method is similar to that discussed in Wang et al. (2011) in which many of the parameters needed by the SLUCM are estimated and calibrated for each case. As a result, surface roughness lengths of 0.5, 1.0, and 2.0 m were used for LIR, HIR, and COM grid points, respectively (Table 3). As surface roughness values are typically ~10% of the height of the average obstruction, the surface roughness values used would correspond to 5-, 10-, and 20-m mean building height for the three urban categories.
e. Numerical experiments
Five numerical simulations were performed for this study: CTRL, LSM, LSMMOD, SLUCM1, and SLUCM2 (Table 1). In CTRL, the modified land-cover data (Fig. 4b) were used but with all representations of urban areas in domain d03 replaced with grassland, cropland, or forest, depending on surrounding rural vegetation types (Fig. 4d). This model run will serve as a basis for how the WRF performs without any urban areas. Simulation LSM used the original Noah LSM bulk-urban code with the default USGS 30-s LULC data (Fig. 4c), while LSMMOD used the modified land-use information (Fig. 4b) with the modified Noah LSM, thus having three urban categories but still using a bulk representation of urban areas. The SLUCM1 and SLUCM2 runs employed the SLUCM scheme, but SLUCM1 used the default urban fraction values (Fig. 4e), while SLUCM2 ingested NLCD impervious surface data as a proxy for urban fraction (Fig. 4f). These latter three simulations will be compared against each other and with the CTRL run as well as against observations to evaluate their ability to properly reproduce the effects of urban areas on their surroundings.
f. Observational data
For verification purposes, two sets of surface observations were used. OKCNET data were obtained for comparisons with model results within the OKC urban area. The OKCNET, which was designed to improve atmospheric monitoring throughout the OKC metropolitan area, consisted of 39 stations, spaced approximately 3 km apart, with 36 traffic signal–mounted (9 m AGL) stations and 3 stations sited according to Oklahoma Mesonet (Brock et al. 1995; McPherson et al. 2007) standards (Fig. 5). These observations were available at 1-min-averaged intervals from November 2008 through November 2010. Oklahoma Mesonet data were used as a rural counterpart to OKCNET observations. The Oklahoma Mesonet (hereinafter "Mesonet"), a rural network of 120 meteorological stations with siting chosen so as to minimize influences from urban landscapes, provides a continuous dataset of 5-min averages of 2- and 9-m temperature, 2-m relative humidity, and 10-m wind speed and direction from across Oklahoma (Brock et al. 1995). Spatial comparison of observed data with model-derived quantities was desired, so point measurements at OKCNET and Mesonet sites were interpolated using the ordinary kriging method of Journel and Huijbrgts (2004). Kriging is a technique commonly used in mapping point observations of meteorological, geological, and chemical quantities to continuous two-dimensional fields (Moore and Rojstaczer 2002; Alfieri et al. 2009; Mercer et al. 2011; Smoliak et al. 2015; Gutiérrez et al. 2015b; Hu et al. 2016). While kriging interpolation is a statistical method subject to errors based on user-defined interpolation options, and may change significantly if various data points are removed, it maintains observational integrity at the observed locations.
a. Model verification
We can evaluate general WRF performance by comparing mean biases (MBs) and RMSEs for near-surface conditions of all five simulations at Mesonet (Table 4) and OKCNET (Table 5) locations, as well as time series of diurnal average model differences from observations (Fig. 6). Modeled rural conditions are too warm during the day and even more so at night (Table 4 and Fig. 6a), a bias common with the YSU PBL scheme (Hu et al. 2010; Coniglio et al. 2013; Clark et al. 2015), though it may also be a by-product of too-dry soil moisture initialization (Fig. 7). Simulated rural wind speeds are in good agreement with observations during the day but are overestimated by ~25% at night in all simulations (Table 4 and Fig. 6c). Simulated urban areas (Table 5 and Figs. 6a,b) are also too warm and dry at night, particularly for SLUCM. However, daytime urban temperatures are underestimated by all urban simulations, but simulated daytime urban mixing ratios agree well with observations. Urban wind speeds have the largest biases and RMSEs of the three variables (Table 5), especially for the SLUCM simulations during the day. The CTRL simulation has the coolest temperatures and highest mixing ratios throughout the day, indicating the urban-induced thermodynamic modifications in the simulations with urban areas. Simulation LSMMOD outperforms all other simulations for both urban MB and RMSE for daytime and nighttime wind speed and mixing ratio. Furthermore, LSMMOD has lower absolute mixing ratio and wind speed MB and RMSE than LSM and a decreased nighttime warm temperature bias, although LSM’s lower absolute MB for daytime temperature is a result of its larger negative moisture bias. Hence, because LSMMOD generally outperforms LSM in urban areas, further analyses will not consider simulation LSM.
Spatial variations in the average daily minimum, maximum, and mean first-model-level (approximately 25 m AGL) temperature from SLUCM1 and LSMMOD are compared with kriging-interpolated Mesonet and OKCNET 9-m temperature for the seven days of the simulations in Fig. 8. The SLUCM1 simulation is selected over SLUCM2 because it has urban fraction values assigned per the SLUCM’s default values. The UHI effect is evident in the mean , maximum , and minimum 9-m temperature T observations (Figs. 8a,d,g). Observed temperatures are spatially correlated with urban fraction (Fig. 4f; correlation coefficient for ), particularly over the Oklahoma City central business district (CBD) at night, with an area of warmer temperatures (1°–2°C) concentrated in central OKC. During the day, this warm air extends north of the CBD, likely attributable to advection by predominately southerly winds (Fig. 3b). Other studies have observed similar spatial distributions of temperature in Oklahoma City (Basara et al. 2008, 2010; Hu et al. 2013a, 2016), although there is no evidence of the daytime cool island observed by Basara et al. (2008) during part of their study period.
Observed mean daily 9-m water vapor mixing ratios q (Fig. 9a) are slightly drier in the urban area, particularly near and north of the CBD. Similar patterns were noted for more dense cities (e.g., Dou et al. 2015). Urban effects on humidity are contradictory; while combustion and lawn irrigation release water vapor into the urban surface layer (not explicitly accounted for in the present study), replacement of natural vegetation with impervious artificial materials decreases potential evapotranspiration. In addition, warmer and rougher daytime urban PBLs experience greater turbulent mixing and slower near-surface wind speeds than cooler, smoother, rural PBLs, increasing the upward flux of moisture while also decreasing the surface moisture transfer coefficient. Together, these mechanisms typically result in lower urban humidity values (Grimmond and Oke 1986; Dou et al. 2015); hence, the dry urban area simulated here is appropriate. Also consistent with this theory, and similar to results from Hu et al. (2016), mean 10-m wind speeds (WS; Fig. 9b) are as much as 2 m s−1 (approximately 50%) weaker over the CBD than the surrounding rural areas owing to increased surface roughness. However, unlike in the temperature observations, these decreased wind speeds do not appear to be plumelike but instead appear to be correlated with urban fraction (Fig. 5a; ), which suggests that weaker winds are a direct result of increased surface roughness, in agreement with Hu et al. (2016).
The SLUCM1 and LSMMOD simulations reproduce the general warm, dry pattern over most of the urban area (Figs. 8b,c,e,f,h,i and Figs. 9b,c), though both are too warm and dry overall, especially at night. Both of these biases are perhaps a result of overzealous turbulent mixing seen frequently with the YSU scheme (Hu et al. 2010; Coniglio et al. 2013; Clark et al. 2015) as well as dry soils (Fig. 7). However, these differences can also be expected given the ~15-m height difference between the observations and the first WRF model level. However, unlike observations and results from LSMMOD (Figs. 9d,g), mean first-model-level wind speeds in SLUCM (Fig. 9e) are only slightly negatively correlated with urban fraction () and are generally higher than in rural areas. These results, which appear to be unrealistic given the OKCNET and Mesonet observations, suggest that vertical transport of momentum may be overestimated by the SLUCM/YSU combination, thereby mixing excess momentum downward from aloft.
b. Model reproduction of diurnal changes in urban–rural differences
The diurnal cycle of observed urban–rural differences in near-surface atmospheric conditions can be analyzed by calculating the averages over all OKCNET sites at each time and subtracting the average of all eight Mesonet sites (Fig. 5) to arrive at an observed urban–rural difference. These sites correspond to those used by Hu et al. (2016) for UHI comparisons; thus, our results are directly comparable to their observations. The observed urban–rural differences are compared with WRF-simulated urban–rural dissimilarities. These are computed by subtracting the mean values at each 5-min output of rural grid points contained in the area covered by Fig. 5 but not within box U from averages of urban grid points within box U in Fig. 5. Both observed and simulated urban–rural differences are averaged over days 2–6 of the simulation. Day 1 was not used because of spinup adjustment time period, and a sharp cold front crossed the study area during day 7, hence the exclusion of both days. The results of these calculations for temperature T, q, and WS are shown in Fig. 10. For each parameter, differences in simulated first-model-level ( m) and diagnostic quantities (2-m T and q and 10-m WS) are compared with observed urban–rural dissimilarities at 9 m. While the diagnostic variables should be closer to the 9-m observations than the model-level computations valid at approximately 25 m AGL, they are parameterized values and are thus more susceptible to errors.
Observed UHII (Fig. 10a) remains constant throughout the night near 1.5°–2°C before decreasing to near 0°C ~3 h after sunrise. The UHII slowly rises again in response to daytime heating, reaching a maximum daytime UHII of 0.75°C around 2200 UTC. Differences between the three urban representations of UHII are evident (Fig. 10d). Given relatively higher urban fractions, SLUCM1 has a greater UHII at night in both the first-model-level T (Fig. 10a) and diagnostic 2-m T (Fig. 10d). UHII is higher at 2 m than at 25 m, likely because model-level computations allow for turbulence and diffusion, while diagnostic variables are based primarily on surface fluxes. However, none of the runs sustains the UHII, either at 2 or 25 m, throughout the night at the level (~2°C) seen in the observations. Chen et al. (2014) simulated a similar slow decrease in UHI throughout the night over Beijing. Although UHII is not maintained as long as observed, all three simulations produce a greater first-model-level UHII throughout most of the night than their maximum daytime UHII, which agrees with the observations.
Measured urban mixing ratios are consistently drier than those in rural areas (Fig. 10b), especially when rural downward turbulent transport of drier air decreases overnight, when the mean urban mixing ratio is as much as 1.5 g kg−1 drier. These observations agree with those of Hu et al. (2013a), which suggested that in the absence of strong low-level wind speeds, turbulence decreases dramatically in rural areas after sunset, while rougher and warmer surface characteristics of urban areas result in enhanced vertical mixing, which then inhibits the formation of a cool, moist, surface-based stable layer. None of the WRF runs correctly reproduces this diurnal pattern. While daytime urban–rural mixing ratio differences are comparable to observations (~0.5 g kg−1), particularly for 2-m q (Fig. 10e), nighttime urban and rural mixing ratio differences are near zero. This difference is perhaps because of the strong wind speed and dry biases in rural areas (Table 4), which may inhibit the formation of the rural stable layer. Both simulated and observed urban–rural mixing ratio differences decrease dramatically near sunset. However, modeled urban–rural differences reduce shortly afterward, while observed differences remain constant for several hours before slowly equalizing.
The most noticeable difference between the three urban runs is in their simulation of near-surface wind speeds (Figs. 10c,f). The LSMMOD run, similar to observations, produces lower urban wind speeds throughout the diurnal cycle, with the greatest differences (~1 m s−1, ~15%) occurring during the daytime. However, nighttime urban and rural wind speeds in SLUCM1 and SLUCM2 are very similar at 25 m, and urban 10-m wind speeds are upward of 1 m s−1 (~23%) greater than over rural areas, which is in disagreement with observations. These results are in accordance with the diurnal mean wind speeds seen in Fig. 9d and those simulated by Hu et al. (2016).
c. Urban modification of surface and near-surface properties
To ascertain exactly how the various representations of the Oklahoma City urban area modify their environment, the analysis from here will consider differences between CTRL and each of the three urban runs, LSMMOD, SLUCM1, and SLUCM2, averaged over days 2–6 of the simulation. Figures 11a–f shows that urban–CTRL differences in near-surface T, q, and WS are similar to the urban–rural differences seen in Fig. 10. First-model-level and 2-m temperatures are warmer in the urban runs, especially at night, with SLUCM1 producing the warmest nocturnal near-surface temperatures. Additionally, all urban runs produce generally drier urban areas, especially in the early evening, while SLUCM1 and SLUCM2 (LSMMOD) have higher (lower) wind speeds over the city during the day.
To further investigate how these average urban–CTRL differences are produced, we examine spatial distributions of T (Fig. 12) throughout the diurnal cycle. During the day (1500–2300 UTC), all three urban runs have urban temperatures ~0.0°–0.5°C greater than the CTRL run (Figs. 12a–c), with somewhat warmer temperatures in LSMMOD (Fig. 12a) and SLUCM1 (Fig. 12b) compared with SLUCM2 (Fig. 12c). Evening (0000–0400 UTC; Figs. 12d–f) urban near-surface temperatures are even warmer compared with CTRL than during the day, with SLUCM1 the warmest during the evening (Fig. 12e; >2°C over northwestern OKC). SLUCM2’s (Fig. 12f) urban area is also warmer during the evening than that of LSMMOD (Fig. 12d) but cooler than that of SLUCM1, and SLUCM2’s warmest area is more concentrated over and downwind of the CBD than the other two runs. The spatial patterns of nocturnal (0400–1100 UTC) near-surface temperatures in LSMMOD (Fig. 12g), SLUCM1 (Fig. 12h), and SLUCM2 (Fig. 12i) are similar to those in the evening, though the urban warming is ~0.5°–1.0°C less intense.
The diurnal cycle of spatial distributions of first-model-level wind speeds are also examined in Fig. 13. During the day, the wind speeds in the LSMMOD run (Fig. 13a) are 1–3 m s−1 (15%–20%) slower over urban areas than in the CTRL run, with the slowest wind speeds located in central OKC. However, neither SLUCM1 (Fig. 13b) nor SLUCM2 (Fig. 13c) shows similar patterns of wind speed differences, with few concentrated areas of winds faster or slower than in CTRL anywhere near the city. Near sunset, urban wind speeds remain slower in LSMMOD (Fig. 13d) but to a lesser degree than during the day. Evening urban wind speeds in SLUCM1 (Fig. 13e) are, in many locations, faster than those in CTRL, particularly on the southwest side of OKC. However, the pattern of SLUCM2 − CTRL wind speeds (Fig. 13f) now more closely resembles those of LSMMOD − CTRL, with mostly slower wind speeds over OKC. At night (Figs. 13g–i), wind speeds in LSMMOD and SLUCM2 are 0.5–2.0 m s−1 slower over all of OKC and 0.0–1.5 m s−1 slower in SLUCM1.
Analyzing all observations within an urban area as a single category, when each is affected by the different underlying surface structure and LULC, may be inadequate to properly analyze urban effects (Stewart 2011; Stewart and Oke 2012). The spatial plots of T, q, and WS show that simulated temperature and wind speed are dependent on urban category or urban fraction, as the warmest first-model-level temperatures and slowest first-model-level wind speeds are often concentrated near the CBD (e.g., Fig. 12f and Fig. 13a). Analysis of OKCNET observations by urban category by Hu et al. (2016) suggested that this should be the case for OKC. Analysis of the times series of the 5-day mean and standard deviation of urban − CTRL temperature (Figs. 14a–c) and wind speed (Figs. 14d–f) differences for each urban category indicates that this is only sometimes the case. While the distribution means of daytime wind speeds associated with the three urban categories in LSMMOD − CTRL are separated noticeably (Fig. 14d), little distinction is present between wind speed differences over the same points in SLUCM − CTRL (Fig. 14e) and SLUCM2 − CTRL (Fig. 14f). In contrast, first-model-level nighttime urban − CTRL temperatures associated with the three urban categories are all very similar in LSMMOD − CTRL (Fig. 14a) and SLUCM2 − CTRL (Fig. 14c), while in SLUCM1 − CTRL, nighttime temperatures in low-intensity residential areas are cooler than those of the more urbanized categories (Fig. 14b). Both of these results corroborate the relatively uniform distribution of SLUCM1 − CTRL and SLUCM2 − CTRL wind speeds (Fig. 13) and LSMMOD temperatures (Fig. 12) throughout the day.
d. Urban modifications of the planetary boundary layer
Urban effects are also seen throughout the PBL. Vertical cross sections of q and potential temperature θ near peak heating (2100 UTC), along the north–south line in Fig. 8 for LSMMOD − CTRL, SLUCM1 − CTRL, and SLUCM2 − CTRL are shown in Fig. 15. This location was chosen for the cross section because it is aligned along the mean boundary layer wind over the study period and because it transects the majority of the OKC metro, including downtown OKC. The lowest ~1 km of the PBL is drier in each urban run compared with the CTRL simulation, with the drying extending farther north and south, and of greater intensity in LSMMOD and SLUCM1. The drying is also most concentrated over the urban area, as indicated by the trace of run-specific urban fraction below each plot. In addition, SLUCM1 is 0.25°C warmer just above the surface, while warmer air extends over a deeper portion of the PBL in LSMMOD, with a few locations of +0.5°C air near the ground. The warmer temperatures in the lower portions of the PBL of SLUCM1, and especially in LSMMOD, result in deeper PBLs, denoted by the black (blue) line for the CTRL (urban) run. A higher PBL top leads to moistening and cooling at the PBL top over the city (given that the PBL is typically topped by an inversion), especially in LSMMOD, where the atmosphere is up to 0.5 g kg−1 moister and 0.5°C cooler. Cross sections across other parts of OKC (not shown) show similar urban–CTRL run differences.
To quantify how each urban parameterization modified the PBL throughout the day, CTRL (Figs. 16a,e) and urban − CTRL θ, q, WS, and vertical wind speed w, averaged over urban grid points in box U (Fig. 5a), are plotted in the lowest 2000 m above urban grid points in box U of Fig. 5 at 1200 (Figs. 16b,f), 2100 (Figs. 16c,g), and 0100 UTC (Figs. 16d,h). At 1200 UTC, just after sunrise, all urban runs are 0.5°–1°C warmer just above the surface, but owing to minimal nighttime turbulent mixing, this warming does not extend above the shallow PBL (Fig. 16b). In addition, because of surface roughness, slower wind speeds are evident in the PBL in all runs, especially nearest the surface (Fig. 16f), accompanied by weaker near-surface sinking motion than in CTRL (Fig. 16e) because of warmer surface temperatures. Above the PBL, while temperature, moisture, and horizontal wind speed remain relatively unchanged from CTRL, stronger sinking motion is evident just below average daytime PBL height (Figs. 16c,g), particularly in LSMMOD and SLUCM2.
Near peak heating at 2100 UTC, the lower urban PBL has dried 0.5–1 g kg−1 (Fig. 16c) in all runs compared with CTRL (Fig. 16a), with somewhat drier air through most of the PBL. The majority of the PBL is also slightly warmer though not as much near the surface. As was evident in analysis of first-model-level conditions (Fig. 11), the LSMMOD run produces ~1.5 m s−1 weaker daytime wind speeds near the surface of urban areas, while low-level wind speeds in both SLUCM runs differ little from the CTRL simulation (Fig. 16g). The decreased wind speeds in LSMMOD extend through the depth of the PBL, though they are lowest near the surface. Just above the most slowed wind speeds, at ~150 m AGL, LSMMOD has weaker sinking motion, nearly neutralizing the downward vertical motions at this height in CTRL (Fig. 16e). All runs feature weaker sinking motion near the CTRL PBL top (Fig. 16g), particularly in SLUCM1, indicating the increase in PBL height seen in Fig. 15.
At sunset (0100 UTC), all runs are 1°–2°C warmer and 1–1.5 g kg−1 drier than CTRL just above the surface (Fig. 16d), in agreement with observations in Fig. 11. However, as was the case at sunrise (Fig. 16b), these modifications are limited to the air nearest the ground because of lower turbulent mixing than during the day. Wind field deviations near sunset (Fig. 16h), however, extend over a larger depth. LSMMOD and SLUCM2 have lower horizontal wind speeds just above the surface but reach their maximum difference in speed 50–100 m higher. SLUCM1 also has lower wind speeds ~200 m AGL; however, SLUCM1’s wind speeds nearest the surface are higher than CTRL’s, a result echoed in the time series (Fig. 11d) of first-model-level SLUCM1 wind speeds. In addition, the daytime weaker sinking motion near the surface evident in LSMMOD (Fig. 16g) is now apparent in both SLUCM1 and SLUM2 and extends through the depth of the collapsing PBL. In the SLUCM runs, this weaker sinking motion also continues above the evening PBL, a residual feature from the daytime convective boundary layer (CBL).
4. Discussion and conclusions
Earth’s population is increasingly concentrated in urban areas, with nearly two-thirds of the world’s population expected to live in urban areas by 2050 (United Nations 2015). As the number of people within cities grows, it is becoming more important to understand, and to be able to correctly predict, the interactions between urban environments and the atmosphere. Using the WRF-ARW, run at 500-m horizontal grid spacing with 25-m vertical grid spacing near the ground, model-simulated urban–atmosphere interactions in Oklahoma City were investigated. The CTRL simulation with no urban area served as a comparison point for the urban simulations. These urban runs included the LSM and LSMMOD, using the original and a new, modified version, respectively, of the Noah land surface model to parameterize the urban surface. Also analyzed were two SLUCM simulations using the more complex single-layer urban canopy model parameterization, differing only in their urban fraction values. Simulation SLUCM1 had the default urban fraction values assigned by the urban canopy model, while SLUCM2, through a novel approach, used NLCD 2011 impervious surface data as a proxy for urban fraction, resulting in generally less dense urban areas. The results presented in this study suggested that it is more appropriate to use the modified Noah LSM alone (LSMMOD), instead of the SLUCM–Noah LSM combination, for lower-density cities, which are common in the Great Plains.
The SLUCM1 and LSMMOD runs simulated near-surface temperatures in the OKC area reasonably well. First-model-level daily maximum, minimum, and mean temperatures were, on average, ~1°C warmer over the urban area than nearby rural areas. Both simulations also reproduced the daytime dry urban area seen in observations, although this was not maintained through the night. However, maximum and mean daily temperatures were slightly warmer and mean mixing ratios were drier than observations, biases seen frequently in models using the YSU PBL scheme. Furthermore, given the dry bias of the initialization soil moisture content, future work should investigate how using the High-Resolution Land Data Assimilation System (HRLDAS; Chen et al. 2007) to spin up the soil state before WRF simulation, as done by Sharma et al. (2016) and Nemunaitis-Berry et al. (2017), might improve these results. The SLUCM1 run also was unable to produce weaker winds over urban areas as seen in observations, while LSMMOD’s urban wind speeds more closely agreed with observations.
All simulations (LSMMOD, SLUCM1, and SLUCM2) reproduced the nocturnal UHI relatively well, with urban–rural differences ranging from ~1° to 1.5°C overnight. However, SLUCM1 produced a more intense, and thus more realistic, nocturnal UHI than LSMMOD and SLUCM2, although the urban fractions in SLUCM1 are less representative of the city. Evaluations of rural and urban observations separately suggest that the models’ weak nocturnal UHIIs are a result of a warm bias in rural areas as opposed to a cool bias in urban areas. Results show that while the daytime UHI was accurately reproduced as less intense than the nocturnal UHI, all three urban simulations produced only a slightly warmer (~0.25°C) daytime urban area relative to observations of an ~+0.5°–1°C daytime UHI. Owing to lower urban densities in OKC, the SLUCM2 simulation had a weaker nighttime and daytime UHI than in the SLUCM1 run. However, given that the UHI in SLUCM1 was already too weak, the UHI in SLUCM2 had an even greater cool bias.
Additionally, observed wind speeds were lower in the urban area, especially during the day. The biggest difference between the three urban simulations was the failure of either SLUCM run to produce this phenomenon. However, the diurnal cycle of urban–rural differences in wind speed in the LSMMOD simulation agreed well with observations. Wind speeds as a function of urban category were analyzed to investigate the role of surface roughness in producing these differences. While increasingly rough urban surfaces (i.e., higher-intensity urban areas) resulted in progressively slower wind speeds in the LSMMOD run than in the CTRL run, wind speeds over all urban surfaces were roughly the same in both SLUCM runs, regardless of urban category.
Hu et al. (2016) theorized the cause of the higher near-surface wind speeds in the WRF during the evening to be an imbalance of friction-induced speed reduction and turbulent transport of increased momentum from aloft caused by inadequacies of the PBL scheme. Other investigations have had similar difficulties correctly simulating evening and nocturnal wind speeds with the WRF (e.g., Zhang and Zheng 2004; Lee et al. 2011; Ngan et al. 2013). However, as all of the runs examined here use the same PBL scheme, and the differences in wind speeds are noticeable during the daytime, their theory cannot account for the discrepancies shown here between the SLUCM and LSMMOD simulations. Miao et al. (2009) noted increased wind speeds in the city center of Beijing while using the SLUCM, which they attributed to increased turbulence; however, they did not have a simulation using only the LSM with which to compare these findings. It is possible that the core of OKC is not large or aerodynamically rough enough to result in the turbulence-induced acceleration of the winds that Miao et al. (2009) observed in Beijing. Further testing of the SLUCM is needed to ascertain the cause of these erroneous wind speeds in OKC.
It is troublesome that providing the SLUCM with more accurate surface characteristics would result in a less realistic UHI representation. Contrary to these results, Li et al. (2013) found that using urban fraction computed directly from 30-m NLCD 2006 data (Fry et al. 2011) improved the representation of surface energy balance terms. The SLUCM, designed around the “urban canyon” observations of Nunez and Oke (1977), assumes that all urban areas are covered by closely spaced buildings with canyon-like streets and alleys in between. Indeed most studies that have used the SLUCM–Noah LSM modeling system to simulate UHIs are performed over large, densely populated cities such as New York City (Holt and Pullen 2007), Taipei (Lin et al. 2008), and Beijing (Miao et al. 2009). Few studies have used the SLUCM to simulate the UHI of cities that are less dense and have greater suburban sprawl. The results herein suggest that the SLUCM may not always be suited for use in less dense urban areas, particularly if accurate reproduction of urban wind speeds is important. However, a more accurate aerodynamic parameterization for the SLUCM, as suggested by Varquez et al. (2015), could be used to attempt to remedy this problem in future simulations.
Recent developments aimed at eliminating these deficiencies from urban canopy parameterizations have focused on developing multilayer urban canopy models, particularly the Building Energy Parameterization (Martilli 2002) and Building Energy Model (Salamanca et al. 2010) in the WRF. Studies using these parameterizations have indicated their ability to improve PBL temperatures and wind speeds in urban areas (e.g., Gutiérrez et al. 2015a,b; Sharma et al. 2016). However, both of these parameterization options are incompatible with the nonlocal YSU PBL scheme, which was chosen for this study because of its suitability for use in severe weather environments (Coniglio et al. 2013; Clark et al. 2015). However, the local-mixing Mellor–Yamada–Nakanishi–Niino PBL scheme (MYNN; Nakanishi 2000, 2001; Nakanishi and Niino 2004, 2006), which is also appropriate for use in severe weather situations (Coniglio et al. 2013), could be used with the multilayer urban parameterization options; hence, this is an avenue for future investigation.
This research was supported in part by the NASA Interdisciplinary Science program Grant NNX12AM89G. This research is also part of the Blue Waters sustained-petascale computing project, which is supported by the National Science Foundation (Awards OCI-0725070 and ACI-1238993) and the state of Illinois. Blue Waters is a joint effort of the University of Illinois at Urbana–Champaign and its National Center for Supercomputing Applications. Some of the data used in this study were acquired as part of the mission of NASA’s Earth Science Division and archived and distributed by the Goddard Earth Sciences (GES) Data and Information Services Center (DISC). The authors thank David Parsons for his guidance and support during this research and Jess Walker for acquiring and processing NLCD data that were used in these simulations. We also thank the anonymous reviewers for their detailed commentary and helpful insight.