The Influence of Refined Urban Morphological Parameters on Dynamical and Thermal Fields in a Single-Layer Urban Canopy Model

Abstract

In this study, localised and non-uniform urban morphology (UM) and urban fraction (UF) parameters are implemented in a single-layer urban canopy scheme in the Weather Research and Forecasting (WRF) mesoscale meteorological model. The purpose of this research is to evaluate the effect of the refined parameterisation scheme on the simulation of dynamic and thermal fields in the urban canopy of the Guangzhou metropolitan area. The results showed that, compared with the default urban canopy parameters of the WRF model, using the localised UM parameters resulted in the most significant improvement in the 10 m wind speed simulation. In urban districts, the mean bias between the observed and simulated 10 m wind speed was reduced significantly by 59% from 2.63 m/s to 1.09 m/s during the daytime. For the thermal environment simulation during the daytime, higher UF and UM values resulted in lower surface albedos and generated narrower street canyons compared with the default modelling setting, which caused more heat to be trapped in the urban canopy and ultimately led to an increase in the surface skin temperature (TSK) and a largely increased ground heat flux (GRD). As a result, at night, more heat was transferred from the ground to the surface, producing a higher TSK. The effect of the localised UF on the sensible heat flux (HFX) was closely related to the near-surface temperature gradient. The UM caused the HFX to increase during the daytime, which was related to the near-surface heat exchange coefficient in the lower model layers. As the high-resolution UM significantly altered the urban geometry, the dynamic environment simulation resulted in a large increase in friction velocity and a decrease in wind speed.

1. Introduction

The development of the global economy and rapid urbanisation have led to changes in urban land use and land cover. With the natural land cover being replaced by engineered materials, urban areas are increasingly being covered by impervious surfaces [1,2]. At the same time, with the continually changing urban geometry, the urban canopy layer is becoming increasingly complex [3,4,5]. These changes directly affect the exchange of matter, energy, momentum, and moisture between the near-surface layer and the atmosphere, and alter the physical processes in the atmospheric boundary layer, thereby impacting the local atmospheric circulation [6,7,8].

Early numerical atmospheric models treated urban areas as homogenous, flat surfaces, neglecting the heterogenous character of urban spaces, which greatly limited the ability of such models to simulate the urban atmospheric environment. Since the 1990s, much research has been conducted on the development of urban canopy parameterisation schemes; for example, Masson [9] used street canyons and different urban surfaces to represent the topography of the urban canopy, upon which he established the Town Energy Budget (TEB) scheme. Later, Kusaka, et al. [10] combined the street canyon model with the Noah Land Surface Model (Noah LSM) in the Weather Research and Forecasting (WRF) model to establish the two-dimensional Single-Layer Urban Canopy Model (SLUCM). As the system of coupling an urban canopy model with the WRF model is a relatively mature technique, it has become an effective tool for analysing the impact of urbanisation on the regional climate and atmospheric environment. The SLUCM has been applied primarily to the study of the urban heat island effect [11,12,13,14] and the impact of urbanisation on regional climates and urban pollution [15,16,17,18].

For the urban canopy parameters in the urban canopy model, Loridan, et al. [19] used a systematic and objective model response analysis to assess the sensitivity of the single-layer urban canopy parameterisation scheme implemented in the WRF model to the input parameters and found that the primary factors affecting the simulation of energy fluxes were albedo, roof thermal properties and materials, and canyon geometry. Wang, et al. [20] used an advanced Monte Carlo method to quantify the impact of surface input parameters on the WRF model output and demonstrated that heat fluxes and surface temperature were highly sensitive to the uncertainties in urban geometry, indicating that urban morphology had a significant effect on the surface energy balance. In addition to urban morphology, the urban fraction also affected the energy balance of the surface layer of the urban canopy model [21]. The spatially homogenous parameters used in common numerical urban canopy models cannot adequately express the heterogeneous urban texture; to overcome these limitations, Adachi, et al. [22] and Lin, et al. [23] used the WRF model coupled with the SLUCM scheme and introduced high-resolution land surface data (the urban fraction and spatial distribution of anthropogenic heat) to study the urban heat island effect in Tokyo and Taiwan, respectively, and found that the modified model produced better simulation results. Touchaei and Wang [24] used aerial images to refine the urban morphology of four neighbourhoods in Montreal, Canada, in the WRF/SLUCM model and found that the modified model provided more accurate predictions of 2 m temperature and the urban heat island intensity. Darmanto, et al. [25] proposed a method for deriving 1 km-resolution urban parameters from globally available satellite images and used the calculated parameters as inputs for the modified WRF/SLUCM model to simulate the meteorological conditions in Jakarta, Indonesia. The results of their study showed that the refined urban morphology parameters derived from global datasets improved the performance of the SLUCM. He, et al. [26] established a dataset of three-dimensional urban canopy parameters (UCPs) for Beijing and applied it to the WRF model to investigate the effect of high-resolution (1 km) UCPs on the simulated meteorological variables under clear-sky conditions. Their results indicated that the high-resolution UCP dataset was able to improve the simulation of diurnal variations and the spatial distribution of the 2 m surface air temperature and 10 m wind speed in the urban canopy. It is evident that the accuracy of UCPs, such as the urban fraction and urban morphology, directly affect the performance of numerical models, so deriving more refined and localised urban fraction and urban geometry parameters is essential for improving the quality of urban canopy schemes and producing more accurate urban canopy simulations [27,28,29,30,31].

In the original WRF model, the urban morphology and urban fraction parameters are fixed and spatially homogenous. However, because of the differences in economic development, the urban structure, and green space ratio in different parts of the world, the spatial distribution of different urban parameters, including building height, building density, and the proportion of impervious surfaces, shows great regional variations. The common parameterisation schemes used in urban canopy models cannot accurately represent the actual conditions of the underlying urban surface. In addition, there is a lack of detailed three-dimensional urban morphology and building geometry databases around the world—even more so in China—and very few related studies can be found. The objective of this study was to investigate the effect of non-uniform and localised underlying urban surface parameters, such as urban morphology and urban fraction, on the meteorological conditions of the boundary layer using the Guangzhou metropolitan area as the study area. Guangzhou is the capital of Guangdong Province. It is one of the fastest-growing megacities in the world. Its built-up area measures 1237 km² and is characterised by complex urban morphology. The underlying urban surface data were obtained from the localised urban canopy dataset provided by the Guangzhou Urban Planning and Design Survey Research Institute and implemented into the WRF mesoscale meteorological model to simulate the regional meteorological conditions. The model performance and the effect of the refined urban canopy parameterisation scheme on the simulated thermodynamic environment of the boundary layer were analysed and evaluated.

2. Methodology

2.1. Model Configuration

In this study, the WRF model was set up with four domains (Lambert projection) centred on Guangzhou (23.3° N, 113.5° E) and configured with 30 vertical layers, with the top level at 50 hPa. A sufficient number of vertical layers was allocated within the boundary layer to better capture and reproduce the boundary layer’s structure. The eta levels of the lowest 15 layers were 1.000, 0.997, 0.995, 0.990, 0.985, 0.980, 0.975, 0.970, 0.960, 0.950, 0.940, 0.930, 0.920, 0.900, and 0.850. They had average heights below 1500 m in Guangzhou. The four nested domains were set up with horizontal resolutions of 27 km, 9 km, 3 km, and 1 km. The fourth-layer domain covers the entire Guangzhou metropolitan area, which consists of 11 districts: Yuexiu District (YX), Haizhu District (HZ), Liwan District (LW), Tianhe District (TH), Baiyun District (BY), Huadu District (HD), Huangpu District (HP), Panyu District (PY), Nansha District (NS), Conghua District (CH), and Zengcheng District (ZC). The simulated domains are shown in Figure 1a. As the CH and ZC districts are predominantly mountainous and contain very few areas that can be classified as urban land cover, they were excluded from this study. Therefore, the study area includes nine districts of the Guangzhou metropolitan area (22.55°–23.65° N, 112.95°–113.70° E), as shown in Figure 1b. The National Centers for Environmental Prediction (NCEP) final analysis (FNL) data with a horizontal resolution of 1° × 1° were used for the initial and boundary meteorological conditions in the WRF simulation. The boundary conditions derived from these data were created every 6 h. For the land-use data, Global Land Cover 2009 (GLC 2009) was adopted for the study area because it was more suitable for the Pearl River Delta (PRD) region [32]. The physical parameterisation schemes used in the simulation experiments included the New Goddard scheme for short-wave radiation, the RRTM scheme for long-wave radiation, the Lin scheme for microphysics parameterisation, the YSU scheme for the planetary boundary layer, the MM5 similarity scheme for the surface layer, and the Kain–Fritsch scheme for cumulus parameterisation (used only in the first two domains). The simulations were run with the Noah LSM coupled with SLUCM for the land surface processes parameterisation. For this study, the distributions of the local UCPs of Guangzhou were also set in the model, which will be presented in Section 2.2.

A single, 12-day WRF simulation was run. It ran from 00:00 UTC 30 October 2017 to 00:00 UTC 11 November 2017. There was a model spin-up period of 40 h, from 00:00 UTC 30 October 2017 to 15:00 UTC 31 October 2017, before any of the WRF output was analysed. After this spin-up period, the actual period analysed was from 00:00 1 November (16:00 UTC 31 October) to 23:00 LST 10 November (15:00 UTC 10 November) 2017.

2.2. Updating the Distribution of UCPs

In the coupled WRF-SLUCM, the UCPs, including the urban fraction, building height and width, street width, and standard deviation of building height, were categorised under various urban land-use types. In the default numerical model, the above parameters were spatially homogenous, so they were not able to accurately reflect the actual proportion of impervious urban surfaces and the detailed characteristics of urban morphology.

To address these limitations, we implemented the non-uniform 1 km-resolution urban fraction (UF) and urban morphology (UM) data in the default WRF model. In the following two subsections, we describe the implementation of the non-uniform UF and UM data in the urban canopy model for the Guangzhou metropolitan area.

2.2.1. Urban Fraction (UF)

In the Noah LSM coupled with the urban canopy scheme, UF represents the fraction of impervious surface in an urban grid identified by the land use data [21]. In the default urban canopy scheme, the UF value in low-density residential areas is fixed at 0.5, as shown in Figure 2a.

In order to refine the UF values in the urban canopy scheme, a local and detailed 2D spatial distribution of UF with 1 km resolution (Figure 2b) was established according to 30 m high-resolution land-use information obtained from Landsat in 2010 [33]. It is evident from the difference between the localised and default UF values shown in Figure 2c that, in most urban-centre districts of Guangzhou, the proportion of impervious surfaces has increased greatly, whereas in the HD, NS, and HP districts, it has decreased. The statistical analysis results of the UF values in different districts of Guangzhou (

Table 1. Statistical details about the urban fraction (UF) in different districts.

District	YX	HZ	TH	LW	HP	BY	HD	PY	NS
Urban land-use type proportion (%)	100	48.94	51.97	53.42	17.94	26.29	6.22	14.70	12.36
Proportion (UF > 0.5) (%)	50.00	60.87	44.30	71.79	31.03	46.03	40.00	35.71	3.64
Proportion (UF < 0.5) (%)	50.00	39.13	54.43	25.64	68.97	53.97	60.00	61.90	96.36
Mean UF	0.48	0.58	0.46	0.58	0.34	0.44	0.39	0.44	0.17
Maximum UF	0.88	0.82	0.85	0.83	0.77	0.94	0.85	0.94	0.88

) showed that, in the YX, HZ, and LW districts, the proportions of areas with UF values greater than 0.5 were 50.00%, 60.87%, and 71.79%, respectively, and, in other districts, the proportion of areas with a UF value greater than 0.5 was less than 50%, and it was especially low in the NS district at 3.64%. Although mean UF values in all districts, except HZ and LW, were lower than the default model value of 0.5, the maximum UF values of all districts were greater than 0.75, indicating that there were large differences in the distribution of UF. These results demonstrate that the high-spatial-resolution and non-uniform UF dataset reflects the distribution of UF more accurately.

2.2.2. Urban Morphology (UM)

UM is another important urban-scale UCP. The single-layer urban canopy scheme includes four urban morphology parameters, namely building width (BW), street width (SW), mean building height (BH), and standard deviation of building height (SBH). Similar to UF, the UM values were fixed in the default model; in low-density residential areas, the default values of BH, BW, and SW were 5 m, 8.3 m, and 8.3 m, respectively, and SBH was set at 0 m.

In order to refine the values of UM parameters, this study used detailed and localised urban morphological data provided by the Guangzhou Urban Planning and Design Survey Research Institute. All urban morphology parameters were available in the WRF Pre-processing System (WPS) as an array called URB_PARAM. The main UM parameters in the URB_PARAM array and their equations are shown in

Table 2. Local refined urban morphology parameters of different urban canopy schemes in the WRF model.

Parameter	Equation	Definition
Plan Area Fraction	${\lambda }_p=\frac{A_p}{A_T}$	The ratio between the plan area occupied by the buildings to the total area of the study region, where $A_p$ is the plan area of buildings and $A_T$ is the total plan area of the study region.
Mean Building Height	$\overline{h}=\frac{{{\displaystyle \sum }}_{i=1}^n{h}_i}{n}$	Where $h_i$ is the height of building I and n is the total number of buildings in the area.
Standard Deviation of Building Height	$S_h=\sqrt{\frac{{{\displaystyle \sum }}_{i=1}^n{(h_i-\overline{h})}^2}{n-1}}$	Where $h_i$ is the height of building I and n is the total number of buildings in the area.
Area-Weighted Mean Building Height	$\overline{h_{AW}}=\frac{{{\displaystyle \sum }}_{i=1}^n{A}_i{h}_i}{{{\displaystyle \sum }}_{i=1}^n{A}_i}$	Where $A_i$ is the plan area of building i; $h_i$ is the height of building i; and n is the total number of buildings in the area.
Building Surface-to-Plan Area Ratio	${\lambda }_B=\frac{A_W+A_R}{A_T}$	The total building surface area divided by the total plan area of the study region, where $A_W$ is the total wall surface area; $A_R$ is the total plan area of the rooftops; and $A_T$ is the total plan area of the study region.
Frontal Area Index	${\lambda }_f\left(\theta \right)=\frac{A_{proj}}{A_T}$	The total area of the building projected on a plane perpendicular to the windward direction divided by the total plan area of the study region.

. The equations for BW and SW are as follows:

$$\mathrm{BW}=2\overline{h_{AW}}{\lambda }_p({\lambda }_B-{\lambda }_p)$$

(1)

$$\mathrm{SW}=2\overline{h_{AW}}{\lambda }_p(\frac{UF}{{\lambda }_p}-1)/({\lambda }_B-{\lambda }_p)$$

(2)

where $\overline{h_{AW}}$ is the area-weighted mean building height; ${\lambda }_p$ is the plan area fraction; ${\lambda }_B$ is the building surface-to-plan area ratio; and UF is the urban fraction.

Figure 3 shows the differences in the distribution of the UCPs of the localised SLUCM scheme and the default model parameters. Compared with the default values, the localised BH and SBH values were higher, whereas BW and SW were lower in some areas.

Table 3. Statistics regarding the urban morphology (UM) in different districts.

District	YX	HZ	TH	LW	HP	BY	HD	PY	NS
Proportion (BW < 8.3 m) a	97%	96%	84%	100%	57%	83%	83%	89%	44%
Mean BW (m)	7.35	6.47	7.03	6.09	9.25	7.12	7.49	5.88	12.58
Proportion (SW < 8.3 m) a	77%	67%	53%	59%	32%	50%	38%	70%	15%
Mean SW (m)	6.12	8.49	9.27	8.32	12.85	13.45	12.24	7.84	18.86
Proportion (BH > 5 m) b	100%	100%	95%	97%	70%	83%	75%	83%	44%
Mean BH (m)	11.24	11.22	12.04	7.86	6.28	8.16	7.04	7.51	4.07
Maximum BH (m)	22.11	28.49	38.80	14.07	14.46	18.86	16.01	20.98	11.78
Proportion (SBH > 0 m) c	100%	100%	99%	100%	95%	93%	92%	98%	62%
Mean SBH (m)	12.61	11.82	11.82	7.99	4.50	6.04	5.62	5.05	3.34
Maximum SBH (m)	25.22	44.88	43.85	18.61	12.58	23.42	21.62	20.07	18.58

^a 8.3 m is the default value of building width (BW) and street width (SW) in SLUCM for low- and medium-density residential areas. ^b 5 m is the default value of mean building height (BH) in SLUCM for low-density residential areas. ^c 0 m is the default value of the standard deviation of building height (SBH) in SLUCM.

shows the statistical analysis results of the UM parameters in different districts of Guangzhou. It shows that, in all districts, except for HP and NS, the BW was lower than the default value (8.3 m) in more than 80% of the corresponding areas; the mean BW was less than 7.5 m; SW was lower than the default value (8.3 m) in more than 50% of the YX, HZ, LW, TH, BY, and PY districts; BH exceeded the default value (5 m) in 95% of YX, HZ, TH, and LW; and except for NS, in all other districts, SBH exceeded the default value (0 m) in more than 90% of the respective areas. To summarise, with the localised UCP scheme, the building heights in the YX, HZ, LW, TH, BY, and PY districts increased, and the differences in building heights became more obvious. In more than half of the study districts, the building density increased obviously because of the reduced SW and BW.

2.3. Experimental Design

In this study, we used sensitivity experiments to analyse the effect of the localised UF and UM parameters on the meteorological environment in the Guangzhou metropolitan area. The default model parameters were used in the Base scenario, and the localised, non-uniform UCPs were used in three separate sensitivity scenarios named Case 1, Case 2, and Case 3. In Case 1, only UF was considered; in Case 2, only UM was considered; and in Case 3, the effects of UF and UM were considered simultaneously. The effects of UF, UM, and the combination of UF and UM on the urban environment were analysed through a series of Case–Base sensitivity experiments.

Sensitivity experiments were designed to study the impacts of the high-resolution UCPs (UF and UM) on the dynamic and thermal fields. Four sensitivity scenarios were tested. The UF and UM involved in the Base experiment followed the homogenous settings in the model. In the Case 1 and Case 2 experiments, the homogenous UF and UM were updated separately. In the Case 3 experiment, the UF and UM were both updated using high-resolution data. Accordingly, the effects of UF (Case 1—Base), UM (Case 2—Base), and UF and UM combined (Case 3—Base) could be analysed.

3. Results

3.1. Model Evaluation

In this section, we compare the simulation results of different meteorological factors for the four modelling experiments with the observation data. We evaluate the model’s performance for the following meteorological factors: 2 m air temperature (T2), 10 m wind speed (WS), and 2 m relative humidity (RH). The observed data consist of hourly observations collected by a total of 41 automatic weather stations from 0000 1 November to 2300 10 November 2017 obtained from the Guangzhou Meteorological Bureau. Observations of T2 and WS were collected by 41 stations, and observations of RH were collected by 33 stations. These automatic surface weather stations were set up and monitored according to the Specifications for Automatic Weather Stations (GB/T 33703-2017) issued by the Standardization Administration of the People’s Republic of China. The locations of these sites were strictly selected and had good representativeness of the local areas.

In this study, the administrative districts of Guangzhou were categorised into urban-centre and suburban districts based on the proportion of the urban land surface type in each district (Table 1). Urban-centre districts were those composed of more than 45% urban land surface, and they included HZ, YX, LW, and TH; the suburban districts were HP, PY, HD, BY, and NS.

Table 4. Statistical comparison of the simulated and observed 2 m temperature (T2), relative humidity (RH), and 10 m wind speed (WS) in the urban centres and suburbs during the daytime (08:00–16:00 LST) and night-time (17:00–07:00 LST).

Statistical Period	Meteorological Factor		Central Districts				Suburbs
			Base	Case 1	Case 2	Case 3	Base	Case 1	Case 2	Case 3
Daytime	T2 (°C)	MB	1.34 **	1.20 **	1.16 **	1.39 **	1.34 **	1.25 **	1.10 **	1.30 **
		MAE	2.06	1.91	1.77	1.97	1.87	1.80	1.64	1.81
		RMSE	2.59	2.36	2.16	2.46	2.34	2.26	2.07	2.28
		R	0.76 **	0.81 **	0.85 **	0.80 **	0.81 **	0.82 **	0.84 **	0.82 **
	RH (%)	MB	−0.96	−0.35	−0.60	−1.42	−0.01	0.37	0.86	0.09
		MAE	9.98	9.40	8.60	9.62	7.05	6.82	6.17	6.78
		RMSE	12.47	11.85	10.68	12.24	8.83	8.54	7.81	8.56
		R	0.80 **	0.81 **	0.84 **	0.80 **	0.64 **	0.64 **	0.67 **	0.64 **
	WS (m/s)	MB	2.63 **	2.52 **	1.09 **	1.17 **	2.29 **	2.26 **	1.85 **	1.87 **
		MAE	2.68	2.59	1.52	1.57	2.41	2.40	2.05	2.07
		RMSE	3.03	2.98	1.80	1.84	2.82	2.82	2.45	2.45
		R	0.41 **	0.44 **	0.43 **	0.41 **	0.53 **	0.54 **	0.55 **	0.53 **
Night-time	T2 (°C)	MB	0.13	0.36	0.40	0.41	0.43	0.64 *	0.51	0.58
		MAE	1.13	1.26	1.33	1.33	1.27	1.42	1.37	1.36
		RMSE	1.39	1.65	1.70	1.69	1.55	1.81	1.70	1.70
		R	0.88 **	0.86 **	0.87 **	0.86 **	0.88 **	0.85 **	0.87 **	0.87 **
	RH (%)	MB	−1.51	−2.19	−1.78	−2.23	−0.59	−1.05	−0.15	−0.61
		MAE	9.42	9.20	8.49	9.35	7.79	7.44	7.11	7.40
		RMSE	11.79	11.48	10.72	11.68	10.01	9.69	9.34	9.65
		R	0.74 **	0.76 **	0.79 **	0.76 **	0.58 **	0.60 **	0.61 **	0.59 **
	WS (m/s)	MB	3.61 **	3.58 **	1.88 **	1.90 **	2.77 **	2.76 **	2.28 **	2.27 **
		MAE	3.62	3.60	2.03	2.08	2.84	2.83	2.40	2.41
		RMSE	3.94	3.94	2.27	2.33	3.24	3.22	2.79	2.79
		R	0.34 **	0.33 **	0.42 **	0.37 **	0.43 **	0.44 **	0.46 **	0.43 **

** The result is above the 99% confidence level; * The result is above the 95% confidence level.

shows the statistical analysis of the simulated and observed meteorological factors in the urban-centre and suburban districts during the daytime and night-time under various sensitivity scenarios. Four statistical parameters were used to evaluate the simulated value against the observation: Mean Bias (MB), Mean Absolute Error (MAE), Root-Mean-Square Error (RMSE), and Correlation Coefficient (R). Their equations are presented as follows.

$$\mathrm{MB} =\frac{1}{\mathrm{N}} {{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{N}}\left({\mathrm{S}}_{\mathrm{i}}-{\mathrm{O}}_{\mathrm{i}}\right)$$

(3)

$$\mathrm{MAE} =\frac{1}{\mathrm{N}} {{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{N}}\left|{\mathrm{S}}_{\mathrm{i}}-{\mathrm{O}}_{\mathrm{i}}\right|$$

(4)

$$\mathrm{RMSE} =\sqrt{\frac{1}{\mathrm{N}} {{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{N}}{\left({\mathrm{S}}_{\mathrm{i}}-{\mathrm{O}}_{\mathrm{i}}\right)}^2}$$

(5)

$$\mathrm{R} =\frac{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{N}}\left({\mathrm{S}}_{\mathrm{i}}- {\mathrm{S}}^{\prime }\right)\left({\mathrm{O}}_{\mathrm{i}}- {\mathrm{O}}^{\prime }\right)}{\sqrt{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{N}}{\left({\mathrm{S}}_{\mathrm{i}}- {\mathrm{S}}^{\prime }\right)}^2}\sqrt{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{N}}{\left({\mathrm{O}}_{\mathrm{i}}- {\mathrm{O}}^{\prime }\right)}^2}}$$

(6)

where i = 1, 2, …, N. N is the total number of samples used to calculate the statistical parameters. S_i and O_i are the simulated (three modelling experiments) and observed hourly values of meteorological factors, respectively. S’ and O’ represent the average of the simulated and observed hourly values, respectively. The equations for calculation are presented as follows.

$${\mathrm{S}}^{\prime }=\frac{1}{\mathrm{N}} {{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{N}}{\mathrm{S}}_{\mathrm{i}}$$

(7)

$${\mathrm{O}}^{\prime }=\frac{1}{\mathrm{N}} {{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{N}}{\mathrm{O}}_{\mathrm{i}}$$

(8)

For the daytime, the simulated and observed hourly values from 08:00 LST to 16:00 LST for 10 days (from 1 to 10 November) were used to derive the statistical metrics (N = 90 h). Meanwhile, for the night-time, those from 17:00 LST to 07:00 LST were adopted (N = 150 h). We conducted significance tests for the mean biases (MB) and correlation coefficients (R) between the simulated and observed meteorological factors and the differences in the simulated values between different modelling experiments. We typically used a confidence level of 99%. The results showed that the correlation coefficients passed the significance test (p < 0.01), suggesting that the variations in the simulated meteorological factors were well captured. The MB values of the simulated RH in the different experiments were not statistically significant. The low mean biases of RH in the different experiments (i.e., generally under 1%) indicate that they performed well in simulating this parameter. The T2 values simulated by different schemes were significantly (p < 0.01) different to the observed values. The WS MB value varied greatly for different simulation schemes during the day and night, and the MB passed the significance test (p < 0.01).

The comparison of the R values between the simulated and observed meteorological factors under the different simulation schemes shows that the refined localised UF and UM parameters produced better R values for each meteorological factor; all R values of T2 were greater than 0.76 and showed a very consistent trend. The trend consistency of RH and WS during the day was better than that at night. In the urban-centre districts, the MB between the simulated and observed T2 values during the day changed from 1.34 °C (Base) to 1.16 °C (Case 2), and the MB between the simulated and observed T2 values during the night changed from 0.13 °C (Base) to 0.40 °C (Case 2), indicating that only the Case 2 scenario (UM only) produced a better simulation of the daytime T2 and had little effect on RH. The WS MB value varied greatly between the different simulation schemes during the day and night. In the urban-centre districts, the MB between the simulated and observed WS values during the day decreased from 2.63 m/s (Base) to 2.52 m/s (Case 1), 1.09 m/s (Case 2), and 1.17 m/s (Case 3), corresponding to percentages of −4.11%, −58.59%, and −55.33%, respectively, relative to the Base scheme. In the suburban districts, the MB of the simulated and observed WS values during the day decreased from 2.29 m/s (Base) to 2.26 m/s (Case 1), 1.85 m/s (Case 2), and 1.87 m/s (Case 3), corresponding to percentages of −1.28%, −19.45%, and −18.35%, respectively, relative to the Base scheme. For the night-time, we also observed reductions in the MB with slightly smaller percentages compared with those during the daytime. These results show that the default parameterisation scheme overestimated the WS values in Guangzhou. Compared with the Base scenario, the Case 2 and Case 3 schemes had a much lower MB value in the urban-centre districts, which means that these schemes produced better WS simulations.

3.2. Impacts of the Updated UCPs on Thermal Fields

In this section, we analyse the effect of using high-resolution UF (Case 1—Base) and UM (Case 2—Base) on the urban thermal environment factors, including the surface skin temperature (TSK), sensible heat flux (HFX), and ground heat flux (GRD).

Figure 4 shows the effect of the high-resolution UF and UM parameters on the simulation of the daytime (08:00–16:00 LST) and night-time (17:00–07:00 LST) mean TSK in the underlying urban surface of Guangzhou. An increase (decrease) in UF led to an increase (decrease) in the daytime and night-time TSK values, which was related primarily to the changes in surface albedo; when UF increased, the surface albedo decreased greatly (Figure 5), resulting in increased absorption of solar short-wave radiation by the ground and ultimately leading to a large increase in TSK. The results in Figure 4 also indicate that using the high-resolution UM caused a large increase in the daytime and night-time TSK values, especially in the urban-centre districts (daytime: 0.453 °C and night-time: 0.498 °C). According to the data in Table 3 and Figure 3, the refined UM parameters decreased the SW values in 64% of the urban-centre district areas (compared with the default value of 8.3 m), with the mean SW being about 8 m. During the daytime, narrower street canyons trap more heat in the urban canopy and decrease the wind speed, preventing heat dissipation and leading to higher TSK values. In addition, higher UM values largely reduce the surface albedo, which is conducive to an increase in the TSK. The increase in the daytime TSK values can be attributed to the decreased surface albedo and greater building widths, and the change in the night-time TSK values can be related to the change in GRD.

GRD is generally negative during the daytime and positive at night. During the daytime, heat is transferred from the surface to the ground, and at night, heat is conducted from the ground toward the surface. Figure 6 shows the effect of high-resolution UF and UM parameters on the mean GRD for the daytime (08:00–16:00 LST) and night-time (17:00–07:00 LST) in the underlying urban surface of Guangzhou. During the daytime (Figure 6a,b), negative values (blue shading) indicated increased GRD and positive values (red shading) indicated decreased GRD. It is evident from Figure 6a that, during the daytime, the increased UF value (0.53) in the urban-centre districts increased the heat transfer from the surface into the ground by 2.690 W/m², whereas in the suburban districts, the reduced UF (0.36) decreased the heat transfer from the surface to the ground by 14.244 W/m². Similarly, the high-spatial-resolution UM parameters also notably increased the GRD in the districts (Figure 6b) with narrowed street canyons (Figure 3d). This phenomenon was related to the TSK in those districts: as the UF increased and street widths decreased, the TSK increased, ultimately resulting in higher heat transfer from the surface to the ground. During the night-time (Figure 6c,d), positive values (red shading) indicated increased GRD, and negative values (blue shading) indicated decreased GRD. As the soil heat storage increased during the daytime, in the districts with greater UF values, the amount of heat transferred from the ground to the surface at night also increased (central districts: 1.076 W/m²). With the high-resolution UM parameters, the street canyon widths in some areas of the YX, HZ, TH, and BY districts were decreased, so the heat storage of soil during the daytime and the amount of heat released at night increased in those districts. In contrast, in the districts in which the street canyon widths increased, the heat storage of soil during the daytime was reduced (Figure 6d). Therefore, in the districts where the street canyon widths decreased, the heat flux from the ground to the surface increased, resulting in an increase in the night-time TSK in the corresponding districts; in the districts where the street canyon widths increased, the heat flux from the ground to the surface decreases, resulting in decreased night-time TSK in those districts.

Figure 7 shows the effect of the high-resolution UF and UM parameters on the daytime and night-time sensible heat flux (HFX) in the underlying urban surface of Guangzhou. In general, the HFX is positive during the daytime and negative at night. In Figure 7a,b, positive values (red shading) indicate increased HFX during the daytime, and negative values (blue shading) indicate decreased HFX during the daytime; in Figure 7c,d, negative values (blue shading) indicate increased HFX at night, and positive values (red shading) indicate decreased HFX at night. It is evident from Figure 7 that, as the UF increased (decreased), the daytime HFX increased (decreased); at night-time, as the UF increased (decreased), the HFX decreased (increased). The high-resolution UM parameters caused a notable increase in HFX during the daytime (urban districts: 20.547 W/m² and suburban districts 22.731 W/m²) and a slight decrease in HFX at night (urban districts: 3.272 W/m² and suburban districts: 1.107 W/m²).

As it was difficult to analyse the effect of the high-resolution UF and UM parameters on HFX based on the spatial distribution diagram, we used the calculation formula for HFX used in the Noah LSM [34]:

$$HFX=\rho C_p C_h U \left({\theta }_{sfc}-\theta \right)$$

(9)

where $\rho$ is the air density, C_p is the specific heat capacity, C_h is the surface heat exchange coefficient, ${\theta }_{sfc}$ is the surface potential temperature, U and $\theta$ are the wind speed and potential temperature at the lowest model layer, and C_p and $\rho$ are constants.

The above equation shows that the HFX value depends primarily on the surface heat exchange coefficient, near-surface temperature gradient, and wind speed. Considering the underlying urban surface layer (starting from height Z₀ upward) of the TH district as an example, it is evident from Figure 8 that the high-resolution UF had little effect on the surface heat exchange coefficient during the daytime and night-time. Therefore, the effect of UF on the HFX was related primarily to the near-surface temperature gradient. As UF had a minimal impact on T2, its effects on the HFX (Figure 7a) and TSK (Figure 4a) were basically consistent. The results in Figure 9 confirm that, as the near-surface temperature increased, more heat was conducted from the surface into the atmosphere so that HFX increased. The effect of the high-resolution UM on HFX was somewhat different because the UM had a greater impact on the surface heat exchange coefficient. Therefore, the role of the surface heat exchange coefficient must be considered when analysing the impact of UM parameters on HFX.

Figure 8 shows the vertical distribution of the simulated heat exchange coefficient for the sensitivity experiments. The heat exchange coefficient was calculated online with the simulation and outputted by the WRF model. It showed that the high-resolution UM caused a large increase in the heat exchange coefficient in the lower model layers. During the daytime, the UM caused the TSK to increase in some parts of the urban districts (Figure 4b), while no obvious change in T2 occurred, which caused the near-surface temperature gradient to increase. At the same time, the heat exchange coefficient also increased, which was conducive to the transfer of heat between the surface and the atmosphere, ultimately leading to increased HFX during the daytime. As heat was generally transferred from the atmosphere to the surface at night, although the heat exchange coefficient in the lower model layers showed an increase, the near-surface temperature gradient was relatively low. Figure 9b shows that, with the high-resolution UM in the night-time, the difference between the TSK and T2 was negligible, which was not conducive to the transfer of heat from the atmosphere to the ground surface, ultimately leading to a slight decrease in the HFX at night.

3.3. Impacts of the Updated UCPs on Dynamical Fields

In this section, we analyse the impact of the high-resolution UF (Case 1–Base) and UM (Case 2–Base) parameters on the urban dynamic environmental factors, including the friction velocity (UST) and wind speed (horizontal and vertical distribution).

Figure 10 shows the effects of the high-resolution UF and UM on the horizontal distribution of the friction velocity in the urban underlying surface of Guangzhou. The UF had a negligible effect on friction velocity. On the other hand, the high-resolution UM caused a significant increase (p < 0.01) in the friction velocity of the urban underlying surface, especially in the urban-centre districts, where the friction velocity increased by 0.172 m/s; in the suburban districts, the friction velocity increased by 0.126 m/s. This could be attributed to the fact that the mean SBH values (11.06 m in urban districts and 4.91 m in suburban districts) in those districts exceeded the default SBH value of 0 m; in some districts, this was the case in more than 99% of the district area. As a result of these large changes in urban morphology, the surface roughness increased, leading to a notable increase in the friction speed.

The effect of the high-resolution UM on wind speed was significantly (p < 0.01) greater than that of the UF (Figure 11). The high-resolution UM caused a significant decrease in wind speed in both the urban-centre and suburban districts (urban-centre districts: 1.504 m/s and suburban districts: 0.749 m/s), with changes of 33.50% and 15.77%, respectively. The high-resolution UM generated greater building heights, resulting in increased surface roughness and decreased wind speed. In addition to affecting the spatial distribution of wind speed, the high-resolution UM had a certain impact on the vertical distribution of wind speed. Figure 12a shows how the UM affected the vertical distribution of the horizontal wind speed along the latitude of 23.116° N, passing through the LW, YX, TH, and HP districts. It is shown that the UM caused a notable decrease in the simulated horizontal wind speed vertically in the urban canopy, with its effect on the near-surface wind speed being more obvious. As the height increased, the effect of UM on wind speed decreased, and the effect of UM on wind speed was the most obvious below a height of 400 m. As the building heights in the YX and TH districts (Figure 12b) increased, the wind speed decreased significantly in those districts (around 113.3° E).

4. Conclusions

This study evaluated the effectiveness of a localised non-uniform urban morphology (UM) and urban fraction (UF) parameterisation scheme implemented in the WRF mesoscale numerical model based on a simulation of the single-layer urban canopy of the Guangzhou metropolitan area. We also analysed the impact of a high-resolution parameterisation scheme on an urban-scale simulation of the thermal and dynamic urban environment of the study area.

We found that, in the suburban districts, the high-resolution UM only slightly improved the T2 and RH; the UM had the most obvious effect on the 10 m wind speed, especially in the urban districts, where the MB between the simulated and observed values decreased significantly by 59% (p < 0.01) from 2.63 m/s to 1.09 m/s during the daytime. For the thermal environment simulation, the increased UF and UM decreased the surface albedo during the daytime and generated narrower street canyons compared with the default modelling setting, causing more heat to be trapped in the urban canopy and ultimately leading to a large increase in TSK. The increased TSK caused an increased GRD during the daytime, resulting in increased transfer of heat from the ground to the surface at night-time and ultimately causing an increase in the TSK at night. The impact of the UF on the HFX was closely related to the near-surface temperature gradient; the UM caused a significant increase in the daytime lower layer heat transfer coefficient and increased the transfer of heat from the surface to the atmosphere, ultimately leading to an increased HFX. In the night-time, the near-surface temperature gradient was relatively low, which was not conducive to the transfer of heat from the atmosphere to the surface and led to a slight decrease in HFX at night. For the simulation of the dynamic urban environment, the UF had a negligible effect on various meteorological factors, but the refined UM significantly affected the urban geometry of the urban-centre districts, increasing their surface roughness and friction velocity and decreasing the wind speed by 1.504 m/s significantly (p < 0.01).

The results of this study show that the localised high-resolution parameterisation in the SLUCM had a greater impact on the dynamic urban environment than on the thermal environment. At this time, the anthropogenic thermal parameters in the present parameterisation are defined in the same way as the urban morphology parameters; that is, their fixed values are specified in the UCP list. These fixed values represent the anthropogenic characteristics of the underlying urban surface layer throughout the entire region, so they do not reflect the spatial variations of urban space. In the future, we will establish a localised and non-uniform list of anthropogenic heat emission parameters and carry out research on the impact of a heterogeneous, high-resolution, anthropogenic heat emission parameterisation on the simulation of the urban heat island effect.