As NWP and climate models continue to evolve toward finer grid spacing, efforts have been undertaken to better represent urban effects. For this study, the single-layer urban canopy model (SLUCM) of the High-Resolution Land Data Assimilation System (HRLDAS) and WRF Model was used to investigate the sensitivity of near-surface air temperatures and energy fluxes to SLUCM parameters in uncoupled (land) and coupled (land–atmosphere) predictions. Output from HRLDAS and WRF was compared with observations from the Oklahoma Mesonet and Joint Urban 2003 experiment. Variations in roof albedo (0.04–0.4) produced 40–135 W m−2 changes in net radiation and sensible heat fluxes. Sensible and ground heat fluxes varied by 40–100 W m−2 with changes in roof thermal conductivity (0.05–1.4). The urban fraction was found to be the only SLUCM parameter to significantly impact latent heat fluxes. Near-surface air temperatures, particularly during the daytime, did not show significant variations with SLUCM parameters (remaining within the 0.5-K range). Differences in urban air temperatures due to the change in boundary layer scheme were greater than the temperature changes due to SLUCM parameter variations. The sensitivity of near-surface air temperatures to SLUCM parameters depended on the method used to calculate the skin temperature of the impervious surface. For all simulations, predicted 2-m urban air temperatures were consistently higher than observations, with deviations approaching 8 K during the day and below 3 K at night. These large errors affected the model’s skill in reproducing the diurnal cycle of UHI intensity.
To date, much of the current understanding of the impacts of urban areas on atmospheric processes results from field programs conducted in a variety of cities worldwide (Barlow 2014; Loridan and Grimmond 2012), with the majority centered in North American cities: St. Louis, Missouri (Changnon et al. 1971, 1976; Lowry 1974); Chicago, Illinois (Changnon and Semonin 1978; Grimmond and Oke 1995); Los Angeles, California (Grimmond and Oke 1995); Vancouver, British Columbia, Canada (Steyn et al. 1997); Montreal, Quebec, Canada (Mailhot et al. 1998); Mexico City, Mexico (Doran et al. 1998); Tucson, Arizona (Grimmond and Oke 1995); Salt Lake City, Utah (Allwine et al. 2002; Doran et al. 2002); Phoenix, Arizona (Grimmond and Oke 1995; Fast et al. 2000); and Oklahoma City, Oklahoma (Allwine and Flaherty 2006). Because of the complex atmospheric processes occurring in urban areas, field experiments of this nature are critical to scientific advancements in urban meteorology.
While observational studies have long shown that cities significantly impact local weather and climate, the prediction of urban effects is a major weakness of current LSMs (Sailor and Fan 2002; Best 2005; Jin et al. 2007). Most LSMs were developed for coarse-resolution atmospheric models in which urban effects were once thought to be negligible (Jin et al. 2007; Oleson et al. 2008). More recent efforts focused on better representation of urbanized areas in NWP and climate models (Masson 2006; Martilli 2007; National Research Council 2012). One common approach is to couple an urban canopy model with an LSM (Masson 2000; Kusaka et al. 2001; Martilli et al. 2002; Wang et al. 2011; Grimmond et al. 2010, 2011). The inclusion of urban meteorological processes and anthropogenic heat release within the LSMs of numerical models is vital for assessing potential impacts of extreme weather events and climate change in heavily populated areas (Betts and Best 2004).
The most documented urban-induced weather phenomenon is the UHI effect whereby surface and air temperatures in the urban canopy are higher than in the rural surroundings (Landsberg 1981; Oke 1987; Arnfield 2003) because of differences in the urban and rural radiation and surface energy balances (SEBs). Urbanized models are often evaluated by comparing predictions of air temperature, surface temperature, wind speed, wind direction, and turbulence statistics with observations (Taha 1999; Martilli et al. 2002; Dupont et al. 2004; Otte et al. 2004; Best 2005; Chin et al. 2005; Fan and Sailor 2005; Grossman-Clarke et al. 2005; Kondo et al. 2005; Holt and Pullen 2007; Lin et al. 2008; Miao et al. 2009; Salamanca et al. 2011; Zhang et al. 2011). However, such evaluations with surface observations do not demonstrate whether the energy exchange processes represented in the urban scheme are physically realistic (Masson et al. 2002; Samuelsson et al. 2003). Thus, several evaluation studies focused on directly comparing the predicted components of the SEB with their observed counterparts (Masson et al. 2002; Lemonsu et al. 2004; Best et al. 2006; Dupont et al. 2006; Grimmond et al. 2010, 2011; Loridan et al. 2010; Loridan and Grimmond 2012; Best and Grimmond 2015). Still, open questions remain about the computation of near-surface (2 m) air temperatures in urbanized weather and climate models (Li and Bou-Zeid 2013, 2014), and evaluation studies should ideally address both predictions of SEB components and atmospheric variables.
The performance of LSMs also depends on whether the model is run in coupled (land–atmosphere) or uncoupled (land) mode because of feedback mechanisms in coupled systems (Margulis and Entekhabi 2001; Samuelsson et al. 2003; Best et al. 2006; Loridan and Grimmond 2012). Several studies used coupled land surface–atmospheric models (Martilli et al. 2002; Dupont et al. 2004; Otte et al. 2004; Best 2005; Chin et al. 2005; Fan and Sailor 2005; Kondo et al. 2005; Liu et al. 2006; Jin et al. 2007; Holt and Pullen 2007; Lin et al. 2008; Lemonsu et al. 2009; Miao et al. 2009; Salamanca et al. 2011; Zhang et al. 2011), which makes it difficult to distinguish the biases of the urban scheme from those of the atmospheric model. Appropriate evaluation of an urban surface scheme must include both uncoupled and coupled model predictions (Best et al. 2006).
The Noah LSM (Ek et al. 2003) has been widely used for UHI studies (e.g., Li and Bou-Zeid 2013, 2014; Hu et al. 2013a) after it was coupled with a single-layer urban canopy model (SLUCM; Chen et al. 2011). It can be run in uncoupled (land) mode using the High-Resolution Land Data Assimilation System (HRLDAS; Chen et al. 2007) for prescribed atmospheric conditions. For coupled predictions that simulate feedback processes between the land surface and the atmosphere that are ignored in HRLDAS, the Noah LSM and SLUCM can be run as part of the WRF Model (Skamarock et al. 2008). Since the HRLDAS and WRF runs use identical parent and nested grids, the SLUCM can be evaluated in both uncoupled and coupled mode.
The performance of the SLUCM depends, in part, on the accuracy of approximately 30 input parameters (Loridan et al. 2010). Some previous studies either used the default values or did not specify whether SLUCM parameter values were modified (Holt and Pullen 2007; Lin et al. 2008; Miao et al. 2009; Tewari et al. 2010; Zhang et al. 2011; Hu et al. 2013a), while some authors adapted the parameter values based on the city of interest (Grossman-Clarke et al. 2010; Flagg and Taylor 2011; Carter et al. 2012; Loridan and Grimmond 2012; Kim et al. 2013). Because of the high level of uncertainty in the specification of SLUCM parameter values, some studies have investigated the sensitivity of a small number of urban parameters using the fully coupled WRF–Noah–SLUCM. For example, Lin et al. (2008) examined the impact of variations of anthropogenic heat flux (QF) on the PBL depth and surface air temperature. They also examined the impacts of heat capacity, thermal conductivity, albedo, and roughness length of the roof, building wall, and ground surfaces. Similarly, Miao et al. (2009) conducted four tests by increasing and decreasing building heights and QF.
Two studies have implemented complex methods to thoroughly determine the sensitivity of the offline SLUCM to urban parameter values. Loridan et al. (2010) used the Multiobjective Shuffled Complex Evolution Metropolis (MOSCEM) algorithm (Vrugt et al. 2003) to examine the sensitivity of the components of the SEB in the offline SLUCM to variations in the SLUCM and Noah LSM parameters. Wang et al. (2011) used an advanced Monte Carlo simulation tool, subset simulation (Au and Beck 2001), to conduct an analysis of the sensitivity of heat fluxes and surface temperatures to changes in individual urban canopy parameters in the offline SLUCM. These two studies revealed a strong sensitivity of SLUCM prediction to roof parameters and weak sensitivity to road parameters (Loridan et al. 2010; Wang et al. 2011). The two studies disagreed on the importance of roof, building, and road emissivities. Wang et al. (2011) found that accounting for emissivity of roof, building, and road surfaces had minimal impacts, while Loridan et al. (2010) found that with the roof emissivity, sensible heat flux (QH) increased. Urban fraction (furb) was the only urban parameter to impact latent heat flux (QE), as it is handled entirely by the Noah LSM (Loridan et al. 2010).
In summary, much progress has been made over the last decade to develop and evaluate urbanized land surface schemes such as the SLUCM (Chen et al. 2011). However, open questions remain concerning the sensitivity of model predictions to changes in urban parameters and the computation of near-surface variables in atmospheric models like WRF (Loridan and Grimmond 2012; Li and Bou-Zeid 2014). The choice of boundary layer scheme in WRF was also shown to affect the prediction of UHI intensity (Hu et al. 2013a,b).
b. Objectives of the study
The current study is unique in that it tests two different PBL schemes (section 4a), examines the sensitivity to SLUCM parameter changes in both coupled (WRF–Noah–SLUCM; section 4b) and uncoupled (HRLDAS–Noah–SLUCM; section 4c) modes, and investigates the sensitivity of the WRF Model toward different approaches in computing skin temperature (sections 3e and 4d). For this study, the findings of Loridan et al. (2010) and Wang et al. (2011) were used to narrow down the list of the 30 SLUCM parameters to 4: roof albedo (αroof), roof thermal conductivity (kroof), roof width (Wroof), and furb. For Oklahoma City (OKC), coupled and uncoupled predictions with the Noah–SLUCM were obtained using default model parameter values, values more representative of the morphology of OKC (control), and also extreme parameter values (section 3d). The predictions were compared with comprehensive observations of near-surface temperature and components of the SEB that were collected during the Joint Urban experiment in July 2003 (section 2).
a. Joint Urban 2003
During June and July 2003, one of the largest urban field experiments, Joint Urban 2003 (JU2003), was conducted in OKC (Clawson et al. 2005; Allwine and Flaherty 2006). Among other factors, OKC was selected as the study area because of the extensive weather-observing infrastructure in central Oklahoma that includes the Oklahoma Mesonet (Brock et al. 1995; Shafer et al. 2000; McPherson et al. 2007), the NOAA/NSSL research polarimetric radar, the NOAA/NWS upper-air station, and four NEXRAD Doppler radars. JU2003 produced perhaps the largest dataset ever compiled to quantify the impact of urban areas on atmospheric processes within the PBL. Between 28 June and 31 July 2003, a vast array of instrument systems, installed specifically for JU2003, collected high-resolution observations of meteorological variables in and around OKC, which were centrally archived (Dugway Proving Ground 2003). The quality assurance (QA) of JU2003 data was performed by the organization that collected the data and by an independent QA contractor (Halvorson et al. 2006).
The Pacific Northwest National Laboratory (PNNL) deployed 32 Onset Computer Corp. “HOBO Pro Temp/External Temp” temperature dataloggers (Whiteman et al. 2000) throughout OKC to measure air temperature at 2 m AGL (T2m). HOBOs were deployed along a north–south transect and an east–west transect centered on the central business district (CBD; Fig. 1).
Seven SEB sites were placed across the metropolitan area of OKC (Table 1). The SEB sites were maintained during JU2003 by three organizations: Arizona State University (ASU), NOAA/ARL’s Atmospheric Turbulence and Diffusion Division (ATDD), and Indiana University (IU). The sites were located over various land-use types across the city, including suburban lawns, gravel, and concrete. The site locations for the SEB measurements provided data upwind, within, and downwind of downtown OKC (Fig. 2).
b. Oklahoma Mesonet
The Oklahoma Mesonet is an automated network of over 100 remote hydrometeorological stations across Oklahoma (Brock et al. 1995; Shafer et al. 2000; McPherson et al. 2007). Mesonet observations for 10 core variables (including T2m) are collected as 5-min averages, with the exception of soil temperature (15-min averages) and soil moisture (30-min averages).
In 1999, as part of the Oklahoma Atmospheric Surface-Layer Instrumentation System (OASIS) Project, 10 mesonet sites (OASIS supersites) were instrumented to measure the components of the SEB with enhanced accuracy (Brotzge et al. 1999; Brotzge 2000; Basara and Crawford 2002) using, among others, eddy-covariance techniques. Because of their proximity to OKC and availability of data, the OASIS supersite of interest in this study was the Norman, Oklahoma, site (NRMN; Fig. 2).
c. Study period and data processing
To minimize the impacts of varying cloud cover between observation sites and focus on the local urban–atmosphere interactions, days during JU2003 characterized by strong radiative forcing (solar insolation near theoretical maximum values) and weak wind shear were chosen for this study.
Data from the 32 PNNL HOBO temperature dataloggers and 6 Oklahoma Mesonet sites surrounding the OKC metropolitan area (Fig. 2) were block averaged to hourly values. The mesonet hourly values were then averaged across all six sites to create a mean diurnal cycle of rural T2m (Basara et al. 2008). Similarly, the HOBO hourly values were averaged across all 32 HOBO sites to obtain a mean diurnal cycle of urban T2m. The mean diurnal cycle of the UHI intensity ΔTu-r was also calculated.
Data from the OASIS supersites (5-min averages) and ATDD sites (30-min averages) were block averaged to hourly values to reduce the variability of the flux values and ensure consistency across the JU2003 dataset. Mean diurnal cycles were compiled for the following variables: net radiation (Q*), ground heat flux (QG), QH, and QE. Components of the SEB not directly measured (QE at NRMN; QG at ATDD gravel, IU TMA, IU TMB, and IU WH) were estimated as the residuals of the SEB equations, when possible. It should be noted that the residual approach allowed any measurement errors in the other terms of the SEB to accumulate in the estimated term. Thus, the residual terms were interpreted as upper limits to the variables estimated.
The time interval for the case investigated in the current study was 0000 UTC 14 July 2003 to 0000 UTC 16 July 2003. This interval was chosen because it includes two consecutive days where few, if any, clouds were present and the low-level jet was relatively weak (Fast et al. 2005; Hu et al. 2013a).
a. Uncoupled model
NCAR developed HRLDAS (Chen et al. 2007) to initialize the land surface state in WRF coupled model studies. The foundation of the HRLDAS is the Noah LSM, a soil–vegetation–atmosphere transfer scheme initially developed at Oregon State University (Pan and Mahrt 1987). Since then, it has been continuously modified by the NCEP and collaborators for use in the NCEP’s regional and global prediction models and data assimilation systems (Chen et al. 1996, 1997; Betts et al. 1997; Koren et al. 1999; Ek et al. 2003). In this study, the HRLDAS is run in an uncoupled mode on the same parent and nested grids as the WRF Model and with the same land-use data, soil texture data, terrain height, vegetation properties, and Noah LSM parameters (Chen et al. 2007). This allows the HRLDAS soil variables (soil moisture and soil temperature) to be directly ingested into the coupled WRF Model without spatial interpolation.
This project used output from the NCEP NARR (Mesinger et al. 2006) to set initial and lateral boundary conditions for the WRF Model and used NARR near-surface atmospheric output for the HRLDAS. The NCEP NARR is a long-term (1979–present) atmospheric and land surface dataset for North America (Mesinger et al. 2006). Because of the project emphasis on the SEB, downward shortwave radiation data from Phase 2 of the North American Land Data Assimilation System (NLDAS-2; Xia et al. 2012) were used for the HRLDAS. The NLDAS-2 uses a ratio-based algorithm to correct the downward shortwave radiation from the NARR to a 5-yr (1996–2000) GOES-retrieved solar radiation (Pinker et al. 2003; Cosgrove and Alonge 2008).
Cosgrove et al. (2003) determined that rough equilibrium with the Noah LSM could be reached within 1 to 2 yr of model run time. This estimate is consistent with the 1.33-yr spinup required to reach rough equilibrium for sandy clay loam soils noted by Chen et al. (2007). Given the multiple years of the NARR forcing data available (1979–2003) prior to the period of interest and results of spinup tests for the study domain (Nemunaitis 2014), the current study opted for the more conservative approach and used a 5-yr spinup time for the HRLDAS. All domains (Fig. 3) were initialized with the NARR output on 1 June 1998. The Noah LSM was run within the HRLDAS using the NARR output and NLDAS-2 solar radiation data through 31 July 2003. A restart file was saved for 5 February 2003 and used to allow each variation in urban parameter to spin up for four months. The HRLDAS output for 0000 UTC for 14 July 2003 replaced the NARR surface states in the WRF input files for each domain. This process ensured that any trends seen in the HRLDAS and WRF Model output were not the result of land surface drift toward model equilibrium.
b. Coupled model
The WRF Model is a fully compressible, nonhydrostatic model that utilizes the terrain-following η coordinate and a staggered Arakawa C grid (Skamarock et al. 2008). It allows one-way, two-way, and moving nesting. In addition, users can choose various subgrid-scale parameterization schemes in seven categories: microphysics, cumulus convection, atmospheric radiation, land surface, surface layer, PBL, and turbulent mixing. The following settings were used in the current study: WRF single-moment 6-class (Dudhia 1989; Hong et al. 1998, 2004; Dudhia et al. 2008) mixed-phase scheme, as mixed-phased schemes should be used for grid sizes less than 10 km (Skamarock et al. 2008); Kain–Fritsch scheme (Kain 2004) for domains with grid spacing larger than 5 km because of its thorough testing within the Eta scheme (Kain 2004); Rapid Radiative Transfer Model (Mlawer et al. 1997) and Dudhia (1989) shortwave radiation scheme, as they have been extensively applied in high-resolution urban WRF studies and allow for a more consistent comparison with prior results for other cities; Yonsei University (YSU; Hong et al. 2006) and MYJ (Mellor and Yamada 1982; Janjić 1990, 1996, 2001) PBL schemes because they are most widely used in conjunction with the SLUCM; MM5 (Paulson 1970; Dyer and Hicks 1970; Webb 1970; Zhang and Anthes 1982; Beljaars 1995) and Eta (Janjić 1996, 2001) surface-layer schemes because they are linked to the MYJ and YSU PBL schemes; and Noah LSM, as it is the only LSM that can be coupled with the SLUCM (Kusaka et al. 2001; Kusaka and Kimura 2004a,b).
Version 3.4.1 of the WRF Model utilized four spatial domains with one-way nesting, 40 vertical levels, and grid spacings of 27, 9, 3, and 1 km (Fig. 3), with the lowest level at approximately 29 m. The HRLDAS used the same four spatial domains. The innermost domain was centered over OKC.
A SLUCM is coupled with the Noah LSM within both the HRLDAS and WRF models using a tile approach (Chen et al. 2011). Each urban grid box is assigned an furb and a fraction of vegetated surface (1 − furb), which uses grassland as the default vegetated land use. The Noah LSM and SLUCM run separately, and the output fluxes are weighted according to the urbanized fraction of the grid cell (Loridan et al. 2010). The SLUCM represents the furb as two-dimensional street canyons of infinite length with equal height on both sides (Kusaka et al. 2001; Kusaka and Kimura 2004a; Chen et al. 2006), specified by mean average values of building height (ZR), street width (Wroad), and roof width (Wroof). The portions of urban surface covered by walls, roads, and roofs are normalized by the total width (Wroad + Wroof) to determine the contribution each surface type has on the SEB fluxes from the SLUCM.
Within the street canyon, the SLUCM accounts for shadows, reflections, and trapping of radiation. Bulk transfer equations are used to model the QH and QE above roof surfaces in the SLUCM. The Monin–Obukhov similarity theory is applied to determine the turbulent exchange coefficients above the roof and canyon areas; these coefficients are identical to those calculated by the surface-layer scheme. The QH from wall and road surfaces are derived from Jurges’s formula (Kusaka et al. 2001). The roof, wall, and road surface temperatures are derived as temperatures that balance the SEBs on the individual canyon surfaces through their effects on QH and QE.
In the WRF–Noah–SLUCM coupling, the SLUCM does not compute the wind profile from the lowest level to the street level, so a logarithmic profile (neutral stability) is assumed from the height of the first model level to the top of the buildings. An exponential profile is used within the urban canyon. A detailed description of the SLUCM implemented in HRLDAS and WRF, including equations and look-up tables, can be found in Loridan et al. (2010).
d. Urban canopy parameters
The urban land-use categories were derived from the National Land Cover Data 2001 (NLCD 2014) and included low-intensity residential (LIR), high-intensity residential (HIR), and industrial/commercial (I/C). Outside of urban areas, the USGS land-use and soil texture data were used.
In version 3.4.1 of the WRF Model, the SLUCM looks up in a table—with three different urban classes (LIR, HIR, and I/C)—approximately 30 input parameters for the thermal properties of the surfaces, dimensions of canyon geometry, and internal building temperatures (Loridan et al. 2010; Wang et al. 2011).
The current study examines the sensitivity to parameter changes in both coupled (WRF–Noah–SLUCM) and uncoupled (HRLDAS–Noah–SLUCM) modes. The findings of Loridan et al. (2010) and Wang et al. (2011) were used to narrow down the list of SLUCM parameters to investigate. Table 2 lists the parameter values used in the coupled and uncoupled SLUCM predictions for the sensitivity analysis, as well as parameter values that were changed from their default values because of additional data sources specific to OKC. Building height (ZR) and standard deviation of building height (σz) were taken directly from the values derived by Burian et al. (2005) for single-family (single house lots) and multifamily (multiple-unit structures) land-use categories. The I/C values for SLUCM were derived by combining the commercial and services; industrial; transportation, communication, and utility; and the built-up subcategory of the other urban or built-up land-use types (Burian et al. 2005). Roof width (Wroof) values for the control run (CTL) were derived from building plan area of Burian et al. (2005) assuming buildings were square shaped.
For the sensitivity studies, roof width values were increased and decreased by 25% in each category from the CTL values, which represent averages for OKC; this accounts for the variability and uncertainty in the original estimates. Data from Scott (2006) were used to convert the average number of lanes for over 55 000 road segments in the OKC metropolitan area to Wroad using the standard lane width of 3.6 m used by the U.S. Interstate Highway System (Federal Highway Administration 2013). It should be noted that QF was neglected in the current study.
The SLUCM parameters were changed one at a time for both the WRF and HRLDAS predictions. Initially, seven model runs were conducted for each of the two PBL schemes. The CTL set of parameter values and naming conventions for the parameter variations relative to the CTL are presented in Tables 2 and 3. Simulations with the ZR, σz, and furb variations were run with the YSU PBL scheme only. The additional simulations were motivated by the low sensitivity observed in the initial model runs and the differences between the model predictions and observations. We intentionally choose settings that differed significantly from the CTL values to better assess how the model sensitivity compares to the observed errors of the forecasts.
e. Computation of near-surface temperatures
To better explain the response of predicted T2m to variations in urban parameters, the method of calculating the diagnostic T2m within the WRF Model was examined. The 2-m temperature field (T2m) does not affect modeled fields, but is diagnosed from modeled fields and land surface fields (Jiménez et al. 2012; Li and Bou-Zeid 2014),
where Tskin is the land surface or skin temperature, QH is the sensible heat flux at the surface, ρ is the density of air, CH2 is the exchange coefficient at 2 m, and U2 is the wind speed at 2 m.
For urban grid cells when the SLUCM is not used, T2m is calculated according to Eq. (1) using a lookup table for the urban roughness length employed to calculate CH2. When the SLUCM is used, the urban grid cells are divided into impervious and vegetated portions. The vegetated portion of the grid cell is assumed to represent grassland. The assigned fractions of the urban and vegetated portions are based on the furb specified for each urban land-use category in the urban parameter table. The Noah LSM is called to calculate the Tskin for the vegetated fraction. Then the SLUCM is called to calculate the Tskin for the impervious fraction. The Tskin for the entire grid cell is calculated as a weighted mean of the impervious and vegetated surface temperatures:
The surface temperature for the impervious portion of the grid is calculated as a diagnostic variable by the SLUCM following
where TA is the air temperature at the lowest model level, while
is the area-weighted mean of sensible heat fluxes from the roof [QH(roof)], wall [QH(walls)], and road [QH(road)]. The equation for T2m then becomes
Li and Bou-Zeid (2014) presented an alternate method to calculate
where Froof is the fraction of canyon surface covered by roofs, while Troof and Tcanyon are the temperatures of the roof and canyon, respectively, which are prognostic variables calculated by the SLUCM. This quantity is similar to the “complete urban surface temperature” proposed by Voogt and Oke (1997), except that the wall and road temperatures are incorporated through the Tcanyon. A new T2m can be calculated by substituting the Tskin(SLUCM) used to calculate Tskin in Eq. (5) with that presented by Eq. (6).
Validating grid-average model results with instantaneous point-scale observations is a recurring problem. Averaging point-scale observations in time partially remedies this problem by increasing the source area or field of view of the instrument. Because of the heterogeneous nature of urban areas and the relatively small number of SEB sites during JU2003, the observed data were block averaged to hourly values and used primarily to investigate the importance of parameter sensitivity relative to the variability of the observed fields. Because the thermal conductivity of most building materials is higher and the heat capacity is lower than those of rural soils (Landsberg 1981; Oke and Cleugh 1987; Oke 1988), observed values of QG were interpreted as lower limits to the modeled values.
The HRLDAS and WRF Model output were extracted for cells that coincided with three or more PNNL HOBO sites, each of the six mesonet sites surrounding OKC, and each of the ASU, ATDD, and IU sites. Data from the PNNL HOBO sites were averaged to represent the observed mean for that grid cell.
a. Air temperature from the WRF Model predictions
Because of the nonlinear interactions between UHIs and heat waves and resulting heat stress for urban residents (Li and Bou-Zeid 2013), it is critical for numerical models (including WRF) to capture the influence of urban land use on the T2m. Figure 4 illustrates the urban and rural T2m predicted by the WRF Model using the MYJ and YSU PBL schemes. Regardless of the PBL scheme used, the WRF Model significantly overestimated T2m during daytime hours for both urban and rural grid points. Except for the overnight hours on 15 July (0000–1200 UTC), variations in the SLUCM parameters produced an approximately 1-K spread of urban temperatures. The spread of predicted urban temperatures increased during the overnight hours of 15 July 2003, during which a decreased kroof and increased furb produced the largest temperature decreases. The decreased kroof reduced the rate of heat transfer through the top roof layer. As a result, a decrease (increase) in kroof increased (decreased) the amplitude of the diurnal cycle of T2m. After sunset, roof temperatures decreased more quickly and contributed to lower nighttime T2m. The decreased furb shrank the fraction of the grid cell composed of urban surface and increased the vegetated fraction, resulting in decreased nighttime temperatures. During the daytime hours, the decrease (increase) in αroof increased (decreased) the T2m because of the resulting increase (decrease) in energy available to heat the roof surface. However, differences in urban T2m due to the change in PBL scheme were greater than the temperature changes due to urban parameter variations. The changes in the predicted rural temperatures may be attributed to the NRMN grid cell being classified as LIR land use in the NLCD 2001.
The MYJ PBL scheme produced a larger amplitude of the diurnal cycle than the YSU scheme. As a result, nighttime urban T2m predicted using the MYJ scheme compared well to the observed temperatures. However, the daytime predicted temperatures were 2–5 K greater than observed. The temperatures predicted when using the YSU scheme were consistently 1–3 K higher than the observed temperatures for all hours.
After the initial analysis of the sensitivity of the T2m to variations in the αroof, kroof, and Wroof, additional model runs were conducted using only the YSU PBL scheme. The considered variations of SLUCM parameters included decreasing the furb (discussed above), using the default values for σz, using the default values for ZR, and using the default values for both σz and ZR. Varying ZR and σz did not significantly impact the T2m.
Figures 4e and 4f display the predicted difference between urban and rural temperatures, or UHI intensity (ΔTu-r), for each parameter variation using both PBL schemes. The ΔTu-r values predicted with both PBL schemes, regardless of the parameter variations, were consistent with the ΔTu-r observed, but only because both the predicted urban and rural temperatures at 2 m exhibited the same trends. The diurnal cycle of ΔTu-r predicted by the WRF Model peaked 1–2 h earlier than the observed ΔTu-r because of rapid rural cooling at sunset.
b. Components of the surface energy balance from the WRF Model predictions
Figure 5 illustrates the Q*, QE, QH, and QG predicted by the WRF Model using the MYJ and YSU PBL schemes for the grid cell containing the ATDD concrete site. All components of the SEB responded similarly to changes in the urban parameters, regardless of the PBL scheme used. Examination of cloud fraction and downwelling shortwave radiation (not shown) revealed minor differences in predicted cloud cover associated with a stationary front moving southeast across the domain, evident between 1300 and 1600 UTC 14 July 2003 in all components of the SEB. Aside from differences in early morning clouds, the magnitudes of all components of the SEB with the MYJ scheme were typically within 40 W m−2 from respective component magnitudes calculated with the YSU scheme. Because of the consistency in response to variations in urban parameters for each grid cell examined, the discussion forward will primarily focus on the YSU results.
Table 4 summarizes the average largest daytime and nighttime differences in components of the SEB between sensitivity runs and the CTL. The land uses for the grid cells containing the ATDD concrete and ATDD gravel sites were LIR and I/C, respectively (Fig. 6). Varying the furb caused larger variations in the fluxes for the I/C than LIR land use, which agrees with the more drastic change in furb tested for I/C. Consistent across all grid cells examined (see Figs. 7 and 8 and Table 4), αroof was the only SLUCM parameter that significantly impacted Q* (Wang et al. 2011; Loridan et al. 2010).
The predicted daytime values of QE for the grid cell containing the ATDD gravel site (I/C) were 100–150 W m−2 less than the predicted QE values for the LIR grid cells (Figs. 5 and 6). Consistent with Loridan et al. (2010), the only urban parameter to significantly impact QE was furb. The value of QE was only sensitive to furb because the SLUCM predicted values of QE equal to 0 W m−2 for the entire forecast period (Loridan et al. 2010). The furb simply increased the percentage of QE predicted by the Noah LSM that contributed to the calculation. All other parameter variations impacted QE by less than 10 W m−2.
The urban parameter variations that most significantly impacted QH were variations of αroof and kroof. As stated previously, because of the larger associated furb, the QH for the grid cell with I/C land use (ATDD gravel; Fig. 6) varied more with parameter changes than QH for the LIR grid cells (Figs. 5, 7, and 8; Table 4). The αroof impacted QH through altering the amount of energy (Q*) available to heat the roof surface. The decrease (increase) in kroof decreased (increased) the transfer of heat through the roof layer and resulted in higher (lower) roof surface temperatures and QH (Loridan et al. 2010).
The SLUCM parameters that impacted QG the most were kroof and furb. The value of QG is directly proportional to kroof, as it controls the transfer of heat through the roof surface. An increase (decrease) in kroof resulted in an increased (decreased) amplitude of the diurnal cycle of QG. During the day, the increased (decreased) kroof increased (decreased) the transfer of heat to the roof in comparison with the CTL (Figs. 5–8; Table 4). At night, the transfer of heat toward the surface of the roof was increased (decreased). The values of QG predicted by the SLUCM were higher than those predicted by the Noah LSM. The decrease in furb decreased the SLUCM contribution to the gridcell QG.
Overall, observations collected above mean building height (TMA, TMB) or away from buildings compared more favorably to modeled values than did observations collected within the urban canopy layer. When compared with the observations from the SEB sites, predicted values of Q* either were 50–100 W m−2 greater than the observed values (ATDD concrete, ATDD gravel, and IU GR) or compared favorably to observations (IU TMA, IU TMB, IU WH, and NRMN). Values of QE observed at the ATDD concrete site were significantly smaller than those predicted by the WRF Model because this site was the most urbanized site, while the land-cover category for the grid cell containing that site was LIR. Except for the ATDD gravel site, the predicted LIR QE values were significantly lower than observed values, likely because of site proximity to suburban neighborhoods with lawn irrigation. Even at the ATDD sites, observed QH values were 100–150 W m−2 smaller than the predicted values. However, observed QH values remained positive overnight at the two ATDD sites, while the WRF Model failed to maintain positive QH during nighttime hours. Positive QH at night is considered a significant contributor to the development of the UHI (Yap and Oke 1974; Kalanda et al. 1980; Oke 1988; Grimmond and Oke 1995; Grimmond et al. 2004; Offerle et al. 2006). Finally, the diurnal cycle of QG predicted by the WRF Model consistently peaked earlier than the observed diurnal cycle.
c. Components of the surface energy balance from HRLDAS predictions
The current study also looked at the impact of running the Noah–SLUCM within the HRLDAS in an uncoupled mode (Table 4) to test if the fluxes were more responsive to changes in the SLUCM parameters without the WRF Model feedback mechanisms (Best et al. 2006). While the sensitivity to SLUCM variations was similar between the WRF and HRLDAS predictions, the flux magnitudes differed significantly.
It was not possible to calculate Q* values from the HRLDAS input and output data as the calculated emissivity was not included in the HRLDAS output files. Values of QE predicted by the HRLDAS for all urban land-use types were 50–75 W m−2 smaller than values of QE predicted by the WRF Model for all parameter variations (Figs. 9–12). Values of QH predicted by the HRLDAS for LIR grid cells (all except ATDD gravel) were approximately 25–75 W m−2 greater than values of QH predicted by all WRF Model runs (Figs. 9–12). Values of QH predicted by the HRLDAS for the I/C grid cell (ATDD gravel) were slightly less than QH predicted by the WRF Model for all parameter variations (Figs. 9–12). Values of QG predicted by the HRLDAS were approximately 25 W m−2 greater than those predicted by the WRF Model (Figs. 9–12). The responses of the fluxes to changes in the urban canopy parameters were nearly the same when SLUCM was run in an uncoupled mode with the HRLDAS as compared with coupled mode with the WRF Model. Notable exceptions were the sensitivities of QE and QH to furb (Table 4). Values of QE predicted by the HRLDAS were less sensitive to urban fraction than predicted by the WRF Model. For I/C land use (ATDD gravel), the decrease in furb from 0.95 to 0.6 resulted in peak QE values of 50 W m−2 predicted by the HRLDAS and 100 W m−2 predicted by the WRF Model. Changes in the Wroof, ZR, and σz did not significantly impact the components of the SEB in both uncoupled and coupled land–atmospheric systems. However, the lack of response of T2m to changes in urban parameters was found to be not associated with a lack of response in QH. This led to the investigation of the sensitivity of Tskin to changes in SLUCM parameters.
d. Skin and near-surface temperatures from the WRF Model predictions
Figure 13a shows urban Tskin predicted by the WRF Model using the YSU PBL scheme. In contrast to T2m (shown in Fig. 4), changes in SLUCM parameters produced noticeable changes in predicted Tskin during both daytime and nighttime hours. The value of Tskin was most sensitive during the daytime to increases (decreases) in αroof. Predicted decreases (increases) in Tskin were in the range from 1 to 2 K as αroof altered the amount of energy available for heating the roof surface. The decrease in kroof increased Tskin by 1K during the daytime because of the reduced transfer of heat through the roof layer. The quantity kroof was the most impactful factor during the night, resulting in a 2-K decrease in Tskin. After sunset, the decreased kroof resulted in more rapid cooling of the roof surface and reduced transfer of heat toward the surface of the roof. In contrast, when the kroof was increased, nighttime values of Tskin increased because of the increased upward transfer of heat through the roof surface.
While the sensitivity of Tskin was greater than the sensitivity of 2-m temperature to changes in SLUCM parameters, it was surprising that large changes in QH (up to 120 W m−2) resulted in modest changes in Tskin (1–2 K). Recalling how the WRF Model calculates Tskin(SLUCM), Eq. (3) is only correct if the CH contains roughness lengths for the impervious surfaces (Li and Bou-Zeid 2014). In version 3.4.1 of the WRF Model, the CH used for the urban grid cell was calculated based on the roughness lengths of the vegetated grassland surface as opposed to roughness lengths appropriate for urban land use. This inconsistency in roughness lengths does not capture the bulk influence of the impervious fraction and causes the reduced sensitivity of T2m to changes in SLUCM parameters. The default calculation of T2m (using CH2 for grassland at 2 m) appears to be adequate because T2m is less sensitive to CH2 than the Tskin and is not truly representative of the air temperature at 2 m because of the complexity of the urban surface (Li and Bou-Zeid 2014).
Figures 13b–d illustrate the Tskin, T2m, and ΔTu-r for urban grid cells resulting from implementation of Eq. (6) in the SLUCM. The method for calculating Tskin(SLUCM) led to a 1–3-K increase in Tskin during the day for all parameter variations relative to the original method. Except for decreased Tskin at night when the kroof was decreased, the nighttime spread of predicted Tskin with variations in urban parameters did not change with the use of Eq. (6). Variations in Wroof and furb still resulted in very small (<1 K) changes of Tskin relative to the CTL.
Daytime values of T2m were higher because of larger values of Tskin calculated using Eq. (6). Nighttime values of T2m were consistent with trends in Tskin and the originally calculated T2m. The quantity kroof was the only urban parameter to which nighttime T2m were sensitive. Despite none of the components of the SEB being sensitive to the Wroof when combined with the new formulation for Tskin(SLUCM), the values of T2m were most sensitive to increased Wroof and decreased αroof during the daytime and decreased kroof at all hours of the day. The higher sensitivity of T2m to Wroof was due to Wroof being used directly in the calculation of Tskin(SLUCM) combined with the lack of response of QH to Wroof. Because T2m is a function of the difference between Tskin and QH, a slight sensitivity of Tskin to Wroof and a lack of sensitivity of QH to Wroof produced a T2m sensitive to Wroof. The higher sensitivity of T2m to furb was similar to that of Wroof.
Implementing Eq. (6) in the SLUCM increased the sensitivity of T2m to urban canopy parameters, especially during the daytime. However, its implementation also increased the daytime UHI intensity, increased the disparity between observed and predicted daytime temperatures, and failed to improve the temporal difference. While the observations show a daytime cool island developing with urban temperatures 1°C cooler than surrounding rural temperatures, the WRF Model predicts a daytime heat island ranging from 1° to 2.5°C. Depending on the choice of urban parameters, the predicted daytime UHI intensity is higher than the nighttime UHI intensity, which was never observed during JU2003 and is generally rarely observed. Because of the sensitivity of T2m to the method of calculating Tskin, additional cases and approaches to calculate Tskin, the sensitivity of temperatures at higher levels, and temporal differences should be investigated.
5. Summary and conclusions
This study investigated changes in the surface energy and radiation balances that contribute the development of the UHI. The data collected from the field measurements during JU2003, combined with data from existing atmospheric observing systems in central Oklahoma, were used in conjunction with numerical modeling.
Numerical model runs were conducted using the uncoupled (HRLDAS–Noah–SLUCM) and coupled (WRF–Noah–SLUCM) models to investigate the sensitivity of T2m and components of the SEB to variations of urban canopy parameters of the SLUCM. The T2m during the day were significantly overestimated and not sensitive to changes in SLUCM parameters, regardless of the PBL scheme used in the WRF Model. Predicted nighttime temperatures decreased in response to a decrease in kroof because of the reduced rate of heat transfer through the roof layer. The T2m were more sensitive to the PBL scheme used in the WRF Model than to changes in SLUCM parameters.
The components of the SEB were more sensitive to SLUCM parameters than T2m. Changes in the Q* due to changes in αroof were accompanied by compensating changes in QH due to increased roof surface heating, consistent with Loridan et al. (2010). Similarly, changes in QH due to changes in kroof were accompanied by compensating changes in QG. The furb was the only urban parameter to significantly impact QE (Loridan et al. 2010) because QE predicted by the SLUCM was erroneously 0 W m−2 for the entire period. The furb simply determined what percentage of QE predicted by the Noah LSM contributed to the gridcell value. The fluxes predicted for I/C grid cells showed larger responses to changes to αroof, kroof, and furb than those predicted for LIR grid cells. The responses of the HRLDAS fluxes to changes in the SLUCM parameters were nearly the same as the responses in the coupled mode. The exception was QE, where the response of QE to furb in uncoupled mode was less than the response in coupled mode. While the sensitivity to urban parameter variations was similar between the WRF and HRLDAS predictions, the flux magnitudes differed significantly. Thus, it may not be optimal to tune an LSM in uncoupled mode to reproduce observations of the SEB because of potential feedback processes in coupled mode.
Upon examination of Tskin predicted by the WRF Model, it was determined that the diagnostic Tskin calculated by the SLUCM used the roughness length for grassland instead of a value appropriate for urban land cover. A different, recently proposed method (Li and Bou-Zeid 2014) was used to calculate the Tskin within the SLUCM. The results revealed higher values of Tskin, especially during the day, and increased sensitivity to the SLUCM parameters. When the new values of Tskin were used to calculate the diagnostic T2m, Wroof emerged as an additional urban parameter to which T2m was sensitive. Implementing Eq. (6) in the SLUCM increased the sensitivity of T2m to urban canopy parameters, especially during the daytime. However, its implementation worsened the disparity between observed and predicted daytime temperatures, failed to correct the temporal difference, and consequently caused large errors in and unrealistic diurnal cycles of the predicted UHI intensity. These large discrepancies highlight that, while progress in urbanizing weather and climate models has been made, further improvements of these models are needed, additional cases and approaches to calculate Tskin should be investigated, and the sensitivity of air temperatures at the surface (2-m level) and also at higher levels should be further analyzed.
This study was supported, in part, by funding from National Aeronautics and Space Administration (Award NA17RJ1277); U.S. Department of Defense; and NOAA/Office of Oceanic and Atmospheric Research under NOAA–University of Oklahoma Cooperative Agreement NA11OAR4320072, U.S. Department of Commerce. Some of the computing for this project was performed at the University of Oklahoma Supercomputing Center for Education and Research (OSCER).