Evaluation Strategies on the Thermal Environmental Effectiveness of Street Canyon Clusters: A Case Study of Harbin, China

Abstract

Urban overheating significantly affects people’s physical and mental health. The addition of street trees is an essential, economical, and effective means by which to mitigate urban heat and optimize the overall thermal environment. Focusing on typical street canyon clusters in Harbin, China, landscape morphology was quantified by streetscape interface measurements (sky view factor, tree view factor, and building view factor). Through ENVI-met simulations, the correlation mechanism between streetscape interface measurements and thermal environment was evaluated, and optimization methods for assessing the thermal environment of urban streets were proposed. The results revealed: (1) The thermal environment optimization efficiency of general street canyon types was greatest when street tree spacing was 12 m. At present, the smaller spacing has not been simulated and may yield better thermal environment results. The average decrease in temperature (Ta), relative humidity (RH) and mean radiant temperature (MRT) was 0.78%, 2.23%, and 30.20%, respectively. (2) Specific street canyon types should adopt precise control strategies of streetscape interface according to their types to achieve the optimal balance between thermal environment optimization and cost. (3) Streetscape interface measurements and thermal environment indexes show quadratic correlation characteristics, and are critical points for further investigation. The conclusions are more specific than previous research findings, which are of great significance for decreasing the urban heat island effect at the block scale, improving residents’ physical and mental health, and improving the urban environment quality.

1. Introduction

Under the dual effects of global warming and China’s rapid urbanization, cities in the severe cold regions of China, which used to be cooler in summer, have frequently experienced “overheating in the street environment” in their central areas for the last 50 years [1]. The temperature in the city center of Harbin, China is significantly higher than in the suburbs, and the urban heat island accumulation effect is significant. Since 2003, the annual average temperature and seasonal temperatures in northeast China have shown a warming trend (0.043–0.063 °C/year), with the most significant warming in autumn and winter [1]. Existing studies have shown that the overheating of the street environment can significantly reduce the time of citizens’ outdoor activities [2], and further correlate with a significant increase in heat-related mortality of urban residents [2,3,4,5].

Thermal environment improvement of the streetscape is a critical location for improving the overall thermal environment of the city. In urban streets, buildings, underlying surfaces, and trees are the three main factors that affect the thermal environment [6]. The urban center area of Harbin is the historical foundation of the city. Due to the regulatory requirements to preserve the historical character of the neighborhood, it is difficult to fundamentally change the shape and material of buildings and the underlying surfaces of the streets. However, effective management actions for trees such as renewal, replanting, and pruning of existing trees can efficiently block long-wave radiation by changing the street landscape morphology, increasing the humidity, reducing the air temperature and mean radiant temperature (MRT) of the street environment, and significantly improve the microclimate environment of urban streets [3,4,7,8]. This is a practical, economic, and feasible technical approach for Harbin and other cities in northeast China.

2. Literature Review

2.1. Urban Vegetation and Thermal Environment

There is an established understanding of the impacts urban vegetation has on the thermal environment and their surroundings [3,4,9,10,11,12,13]. Methods used to quantify these impacts range from field measurements to complex computational simulations. The specific focus of research areas has expanded from residential land uses to streets, and quantitative indicators of urban vegetation have changed from relatively simple, single measures to complex, multiple indicator evaluations. Early studies focused on the cooling effect of the green urban regions [14] and urban forests [15], the cooling effect of single or group planting [16], and surface–plant–air interactions inside urban structures [17].

There have been several studies on the thermal environment of cities in middle and high latitudes, the same latitude as Harbin, such as in the Netherlands [9], Beijing [10], Tianjin [12], Xi’an [13] in China, and Montreal [11] in Canada. While the climatic conditions of these cities are not identical, the common latitude of these cities with Harbin creates an opportunity to draw from this research. One such study examined courtyards in Dutch urban blocks, and found that adding pools and vegetation could cool the microclimate better than increasing the albedo of the exterior walls [9]. Another found that the high sky view factor (SVF) area of outdoor space in Beijing’s central business district (CBD) could be enhanced with the addition of shade-providing trees to extend the comfort zone of the outdoor space throughout the year [10]. Urban planning researchers Wang and Akbari compared the effects of tree size and spacing on outdoor comfort and found that at midnight in summer, the correlation (R2) between SVF and the urban air temperature was about 0.64, which means urban tree cover reduces the urban heat island (UHI) effect at night. During the day, trees over a height of 20 m can cool an area by 4 °C, and, at the height of 60 m, can cool an area by 2 °C in Montreal [11]. One study tested the influence of urban elements on the thermal environment in a new area of Tianjin. The researchers found that green plants and water bodies increased relative humidity and had an apparent cooling effect in summer. In winter, trees improve thermal comfort by reducing wind speed [12]. Another study examined the relationship between the shape of street trees and the thermal comfort of street canyons in summer in Xi’an, China. They found that the best place for street trees was in the center median of the street on north-south streets, and, on east-west streets, thermal comfort improvement was highest when the trees were located close to the buildings [13]. Li et al. studied the relationship between landscape morphology and thermal environment of the street canyon in Harbin, China. They found that SVF, temperature, relative humidity, and average radiant temperature showed a significant quadratic correlation. A planting distance of 9 m between the trees was the most economical under similar thermal environment conditions [18].

In recent years, thermal environment research has been focused on urban communities, parks, squares, educational campuses, and other urban spaces [19,20,21,22,23,24,25]. One study on the thermal comfort of a park in Beijing found that both artificial and tree shading devices had the best cooling, though artificial shading devices had the best energy-saving effect [19]. Another study examined the tree layout of a community in Phoenix in the United States. They identified that the impact of an equidistant arrangement of trees improved thermal comfort the most, as opposed to a more clustered arrangement of trees on site [20]. In Cesena, Italy, researchers studied the microclimate of Bufalini square, and found that a green-colored surface treatment significantly improved outdoor comfort conditions compared with the untreated pavement [21]. Xu et al. studied the thermal comfort of Xi’an city parks in winter. They found that global radiation in winter had the most significant impact on human comfort, followed by air temperature and relative humidity, and wind speed had the most negligible effect [22]. Deciduous tree shading provided the most efficient design measure to reduce thermal stress [23]. Apreda et al. studied the thermal environment of a residential community in the Mediterranean. The results showed that density, geometry, and air temperature distribution were correlated, and that perimeter-type housing block layouts were the optimal configuration in a Euro-Mediterranean context [24]. A study in the high latitude city of Trondheim, Norway, examined the microclimate of university campuses and found that solar radiation was a crucial parameter for improving thermal comfort, and that shading did not play a significant role in small-scale space. Under the conditions of high wind speed, the effect of wind shielding was better than increasing solar radiation [25].

From this assessment of recent research, we find that current quantitative indicators of landscape vegetation can be divided into two categories: the morphological index of vegetation and urban space, including SVF, tree spacing, and green coverage; and the attribute index of vegetation, including tree species, trunk height, leaf area index (LAI), leaf area density (LAD), crown shape, and crown aspect ratio. The identification and extraction of these quantitative indicators, relying on street view maps, big data, and image recognition technology in recent years, has begun to be applied to quantitative research on urban street space form and landscape vegetation. The quantitative distribution map of indicators can be assessed to determine indicators like the amount of street greening, tree shading, vegetation form, etc., and can be accurately evaluated in a given urban location.

2.2. Image Recognition and Urban Vegetation

With deep learning and image recognition technology, street view images are widely used in urban morphology research. A study in Hong Kong proposed a method for extracting street view image features with deep learning to accurately calculate the distribution of SVF, tree view factor (TVF), and building view factor (BVF) of street canyons in a high-density urban environment [26]. A study in the United States examined the spatial distribution and shade provision of street trees [27], and another used Google Street View (GSV) imagery to obtain big data, calculate SVF in the urban area, evaluate the local climate zone map of Singapore classified by Google Earth Engine, and calculated hourly solar hour maps for New York and San Francisco [28]. The urban form and composition of Philadelphia County and three Philadelphia neighborhoods (suburban, central, and low-income) were also analyzed using Street View imagery [29].

Studies using Baidu Street View (BSV) imagery and View Factors (VFs) to calculate the solar duration offer a low-cost approach to assess a wide range of spaces, and have found that green vegetation and buildings blocked 41% and 19% of the typical sunshine duration, respectively [30]. Nagata et al. proposed a new streetscape walkability evaluation method based on semantic segmentation and statistical modeling, using the combination of streetscape elements such as buildings and street trees to establish a regression model and score [31]. Furthermore, Li et al. used a convolutional neural network (SegNet) to segment Baidu Street View images and calculated the shading efficiency of trees in Harbin historical blocks using SVF as an indicator. They found that the average shading effect of trees was 56.3%, and that tree shade has significant spatial aggregation characteristics in Harbin historical blocks [32]. A study of the City of Greater Bendigo in Australia developed a deep learning model to classify street types and proportions at the pedestrian view level, automatically using Google Street View imagery to assess the shade accessibility and propose an improved shading strategy [33]. Finally, Xia et al. developed a method for street SVF measurement based on semantic segmentation processing, extracting sky area data from street view images (SVIs) and estimating the sky view factor of fisheye-images (SVFf) revealing high reliability and effectiveness for SVF estimation in combination with the generation of an SVF map of the street [34].

The research of the influence of urban landscape vegetation on the thermal environment is progressing, but is predominately focused on middle and low latitude cities. The types of urban spaces evaluated range from single streets to larger urban residential areas. The primary research method is numerical simulation, supplemented by deep learning and image recognition technology, to achieve the rapid extraction and evaluation of quantitative index of street trees.

However, due to the cool weather of early summer, there are few studies on the street thermal environment in Harbin, China, a high latitude city in a cold climate zone. With predictions of increasing global warming, the “street heating” conditions become more severe in summer, and there is an immediate need to identify opportunities to manage the thermal environment of streets in Harbin. As a medium-scale, low-density city, Harbin contains buildings of similar height as the trees along the roads, and therefore, the thermal environment of the street canyon is more affected by street trees compared to high-density cities. Especially for the street canyon clusters randomly oriented in the city, a remaining question is whether there is a difference in the thermal environment between the street intersections and the street interior, and how to renew the street trees accurately.

Based on the existing studies of single street canyon, this research further explores more universal optimization strategies of the thermal environment of the street canyons with multiple orientations in Harbin, proposes the concept and index of streetscape interface measurement, and aims to answer the following questions: (1) What is the internal influence mechanism of streetscape interface measurement on the thermal environment of the street canyons with random orientations? (2) What is the precise regeneration strategy of trees in street canyon clusters guided by thermal environment optimization in Harbin?

3. Methods

3.1. Study Area

Harbin (125°42′–130°10′ E; 44°04′–46°40′ N) is located in northeast China, as shown in Figure 1a. According to Koppen-Geiger classifications, Harbin belongs to the Dwa climate type, with cool summers and dry and cold winter temperatures. The city was founded in 1898, as part of Russia’s construction of the Chinese Eastern Railway. The 1927 urban master plan for the city established a foundation of urban form for the Beixiu District (the study area) seen today with multiple street orientations, as shown in Figure 1b [35].

The research area includes five streets and four main orientations, shown in Figure 1c,d: Northwest-Southeast Street (NW-SE), Northeast-Southwest Street (NE-SW), Nearly North-South Street (N-N-S), and Nearly East-West Streets (N-E-W). In the study area, the underlying surface of the street canyons is of the same material, with asphalt roadway and red brick pavement for pedestrians. Adjacent residential land uses dominate the street edges on both sides, with average heights ranging from 15–21 m and a street aspect ratio between 0.65 and 0.9 (

Table 1. Sample block street profiles.

Street Name	Street Orientation	Building Stories	Width of the Street (m)	Street Aspect Ratio	Street Tree Species	Underlying Surface Material
Engineer street	NW-SE	Layer 7	23.5	0.89	Ulmus Pumila	Asphalt, red brick
Shangjiashu street	NE-SW-1	Layer 6–7	26.1	0.8	Salix Babylonica	Asphalt, red brick
Xiajiashu street	NE-SW-2	Layer 6–7	22.6	0.8	Ulmus Pumila	Asphalt, red brick
Songhuajiang street	N-N-S	Layer 7	32.5	0.74	Ulmus Pumila	Asphalt, red brick
Manchurian street	N-E-W	Layer 7	31.2	0.67	Ulmus Pumila, Salix Babylonica	Asphalt, red brick

).

3.2. Research Framework

The research framework of this paper is shown in Figure 2, and includes: (1) Street view interface measurement data and microclimate data were collected. The streetscape interface measurements indicators constitute the landscape interface in the street canyon. The streetscape interface measurements refer, in this context, to the collection of landscape interface quantitative indicators obtained by quantifying the existing landscape elements in the street canyon. Specific indicators include: SVF, TVF, and BVF. The streetscape interface measurements (SVF, TVF, and BVF) were assessed using street view image recognition for sampling points, and microclimate data were measured at these points. (2) The validity of the model was verified. Modeling based on the current site range, view factors were adjusted and the microclimate data is output, and subsequently compared with the measured microclimate data for verification. (3) The thermal environment characteristics of the general street canyons were simulated and design suggestions were put forward. Therefore, three simulated research scenarios, output microclimate data and streetscape interface measurement data, a study of the street thermal environment characteristics of each street canyon cluster with correlations assessed between the streetscape interface measurements and thermal environment measures, and design and planning recommendations for street tree spacing in general street canyon types, were provided. (4) Optimization strategies of specific street canyon types were discussed by establishing spatial optimization strategy models for three specific street canyon types (the inner space of street canyon, street intersection space, park and green open space), and comparison of the results of strategy simulation. The reason for choosing these three types of spaces is that they have the potential to further enhance the thermal environment based on the results of simulation studies. As a final result, proposals for the highest optimization plan coupled with landscape design guidelines are provided.

3.3. ENVI-met Model Descriptions

This study was modeled using ENVI-met 4.4.5. The ENVI-met model is a geometric method used to generate a discrete space through grids to maintain the approximate form of plants [12,36]. The basic parameters such as building height, vegetation layout, underlying surface material, and road width of the study area were obtained through field investigation. Meanwhile, the latitude and longitude geographical coordinates and time zone were recorded. The model size was 660 m × 750 m × 188.8 m, the number of grids was 220 × 250 × 25, the model grid was 3 m × 3 m × 3 m. The setting of model size did not only cover the research block and its surrounding buildings, but also ensured sufficient distance from the model boundary. Most studies have adopted a grid resolution of 1.5 m–5 m [14,15,17,18,21]. The 3 m grid provided the best accuracy that could run on the computer host configuration. Furthermore, a 3 m grid was sufficient to obtain sampling points with a spacing of 30 m. The model was set for 11% magnification over 21 m vertical height and a compass orientation of 49.93°, as shown in Figure 3. The physical parameters of the material in the model are shown in

Table 2. Parameters of building and pavement material.

	Roughness Length (z0)	Reflectivity	Emissivity
Hollow concrete (building material)	0.01	0.30	0.90
Dark concrete (paving material)	0.01	0.20	0.90
Asphalt surface (paving material)	0.01	0.20	0.90
Redbrick surface (paving material)	0.01	0.30	0.90
Loamy soil	0.15	0.00	0.98

.

A plant canopy image analyzer (FS-2000) photographed the vegetation in the model, and the leaf area index (LAI) was calculated, as shown in Figure 4. FS-2000 adopted the principle that canopy porosity is related to the canopy and its structure. According to Beer’s law of light fading through the medium, it analyzed the image of the canopy hemisphere under the canopy, measured canopy porosity, and calculated canopy structure parameters. The leaf area density (LAD) value of the vegetation was estimated by MATLAB (see Supplementary Materials File S1 for MATLAB code) [18]. The conversion equations of the two are shown in (1) and (2), and were derived from the empirical relationship describing the vertical distribution of LAD [18,37]. The elm and willow vegetation models were built using the Albero plant library in ENVI-met (see Supplementary Materials File S4).

To verify the effectiveness of the simulation, we conducted an ENVI-MET simulation by obtaining background meteorological data and testing it with measured data. The background meteorological configuration file used in the model is shown in Figure 5, which is derived from the meteorological data website weatherspark.com, and on-site with a Harbin 50953 Weather Station at the airport. We obtained the following data including temperature, relative humidity, and wind speed. The wind speed calculation was obtained from Equation (3) [38]. The calculated value of 10 m high wind speed was obtained by inputting a wind speed of 2 m high, which is the formula after data adjustment according to the recommended wind profile power law [39,40]. Other initial conditions were established as shown in Table 3.

$$L\left(z\right)=L_m{(\frac{h-z_m}{h-z})}^n\exp [n(1-\frac{h-z_m}{h-z})]$$

(1)

in,

$$n=\{\begin{array}{c} \\ \\ \end{array}\begin{array}{cc}6& 0\le z<z_m\\ \frac{1}{2}& z_m\le z\le h\end{array}$$

$$\begin{array}{cc}\mathrm{LAI}& ={{\displaystyle \int }}_0^h L\left(z\right)dz\\ & ={{\displaystyle \int }}_0^h L_m{(\frac{h-z_m}{h-z})}^n\exp [n(1-\frac{h-z_m}{h-z})]dz\end{array}$$

(2)

where,

• L: The leaf area density (LAD) of vegetation;
• ${\mathrm{L}}_{\mathrm{m}}$: The maximum value of LAD;
• h: The height of the plant;
• ${\mathrm{z}}_{\mathrm{m}}$: Plant height corresponding to the maximum LAD;
• $\mathrm{z}$: Height of plant layering;
• $\mathrm{n}$: The number of layers to be layered.

$$W{S}_{10}=W{S}_h{(10/h)}^{0.35}$$

(3)

where,

• h: Data sampling height
• WS_h: Wind speed at the height of h meters (m/s)

Figure 5. Background weather data in the model.

Figure 5. Background weather data in the model.

Table 3. The setting of initial conditions for simulation.

	Parameter	Initial Value
Basic parameters	Start date of the simulation	13 September 2020
	Start time of the simulation	5:00
	Total simulation time	13 h
	Meteorological settings	Simple forcing
	Wind speed at 10 m height	1.76 m/s
	Wind direction	75°
	Roughness length	0.01
Cloud cover	Cover of high clouds	0.00
	Cover of medium clouds	0.00
	Cover of low clouds	0.00
Plant parameters	Foliage shortwave transmittance	0.30
	Use the tree calendar	Yes
	Background concentration of CO2	400 ppm
Soil conditions	Upper layer (0–20 cm)	Soil humidity 60%, initial temperature 22.85 °C
	Middle layer (20–50 cm)	Soil humidity 65%, initial temperature 22.85 °C
	Deep layer (50–200 cm)	Soil humidity 75%, initial temperature 21.85 °C
	Bedrock layer (below 200 cm)	Soil humidity 75%, initial temperature 21.85 °C
Radiation	Adjustment of radiation	1.00

Table 3. The setting of initial conditions for simulation.

	Parameter	Initial Value
Basic parameters	Start date of the simulation	13 September 2020
	Start time of the simulation	5:00
	Total simulation time	13 h
	Meteorological settings	Simple forcing
	Wind speed at 10 m height	1.76 m/s
	Wind direction	75°
	Roughness length	0.01
Cloud cover	Cover of high clouds	0.00
	Cover of medium clouds	0.00
	Cover of low clouds	0.00
Plant parameters	Foliage shortwave transmittance	0.30
	Use the tree calendar	Yes
	Background concentration of CO2	400 ppm
Soil conditions	Upper layer (0–20 cm)	Soil humidity 60%, initial temperature 22.85 °C
	Middle layer (20–50 cm)	Soil humidity 65%, initial temperature 22.85 °C
	Deep layer (50–200 cm)	Soil humidity 75%, initial temperature 21.85 °C
	Bedrock layer (below 200 cm)	Soil humidity 75%, initial temperature 21.85 °C
Radiation	Adjustment of radiation	1.00

3.4. Thermal Environment Model Verification

3.4.1. Data Acquisition

The on-site thermal environmental measurement time was from 8:30 to 17:30 on 13 September 2020 (9 h in total); the weather was clear and breezy. It should be noted that the results of this study are valid only in approximate climate conditions. The parameters and accuracy of the thermal environment measuring instruments (testo-435, FS-2000) are shown in

Table 4. Instrument parameters and accuracy.

Measuring Instrument	Parameter	Range of Measurement	Instrument Precision	Instrument Image
Multi-function measuring instrument (testo-435)	Air temperature	−50... + 150 °C	±0.3 °C
	Relative humidity	0... + 100%RH	±2% RH (+2–+98% measurements)
	Wind velocity	0.6... + 40 m/s	±(0.03 m/s + 4% measurements)
Fish-eye lens (FS-2000)	Sky view factor	-	Focal length: 1.42 mm Electronic shutter: 1/3.2″ Pixel: 5 MP

. A bidirectional movement measurement method was utilized with two people conducting measurements on the sidewalks simultaneously on both sides, holding the instruments at a height of 1.5 m (Figure 6), and walking at a more or less constant speed with an average pace. Here, the constant speed refers to the perceived constant speed, rather than the absolute constant speed in the physical sense. The instrument automatically recorded the air temperature (Ta), relative humidity (RH), and mean wind speed (Va) every 30 s at the current position. The sample street location is shown in Figure 1c. The measurement points were set at 30 m intervals, with 79 measurement points, including the interior space and intersections of the street canyons. The measuring points of street canyons are marked in yellow, and the measuring points of intersections are marked in red, as shown in Figure 1c. The points were selected according to the equal spacing of the map, and the geographic coordinates were obtained by ArcGIS. The measured verification value was the average value of the sampling points in each street canyon.

The coordinates of the 79 measurement points were used to obtain the panoramic image ID from the Baidu Street View Image Open Platform:

http://api.map.baidu.com/panorama/v2?ak=AK&width=1024&height=512&location=LAT,LON&fov=FOV&pitch=PITCH&heading=HEADING (accessed on 20 October 2020).

The pre-stored point coordinates were read through Python and connected to the Application Programming Interface (API) to achieve the batch download of street view images at the corresponding location (see Python code in Supplementary Materials File S2). The neural network algorithm DeepLabV3+ was used to extract street features from Baidu Street View (BSV) images and perform semantic segmentation [41]. At the same time, the Cityscapes Dataset [42] was selected for training [43].

The panoramic image was converted into a fisheye image through image processing and SVF, TVF, and BVF were calculated. To produce a fisheye image, a hemispherical environment (cylindrical projection) was projected onto a circular plane (azimuthal projection). As shown in Equations (4) and (5) [27], Cx and Cy are the coordinates of the pixels in the center of the fisheye image, and n is the number of concentric rings. The projection is achieved by constructing a relationship between pixels ($x_f$, $y_f$) on the fisheye image and pixels ($x_p$, $y_p$) on the panoramic image. According to Equation (6) by Johnson and Watson [44], the landscape factor was calculated. The fisheye image was divided into several concentric hollow rings of equal width. Then SVF was calculated by summating all the annular sections representing the sky. We used Python to calculate the sky, vegetation [27,32], and building pixel ratios in the fisheye image as the SVF, TVF, and BVF (see Python code in Supplementary Materials File S2). The generated recognition results and their conversion into fisheye images, are shown in

Table 5. Some examples of the generated recognition results.

	High SVF: N-N-S Street		Low SVF: NE-SW Street
Baidu panoramic picture (2019)
Feature extraction
Fisheye image
[SVF, TVF, BVF]	[0.73,0.14,0.13]		[0.15,0.68,0.17]

.

$$x_p=\{\begin{array}{c} \\ \\ \end{array}\begin{array}{cc}\left(\pi /2+{\tan }^{-1}\left[\left(y_f-C_y\right)/\left(x_f-C_x\right)\right]\right)\times W_p/2\pi ,& x_f<C_x\\ \left(3\pi /2+{\tan }^{-1}\left[\left(y_f-C_y\right)/\left(x_f-C_x\right)\right]\right)\times W_p/2\pi ,& x_f>C_x\end{array}$$

(4)

$$y_p=(\sqrt{{\left(x_f-C_x\right)}^2+{\left(y_f-C_y\right)}^2}/r_0)\times H_p$$

(5)

$${\mathsf{\Psi }}_x=\frac{1}{2\pi }\sin \frac{\pi }{2n}{\displaystyle \sum }_{i=1}^n \sin [\frac{\pi \left(2i-1\right)}{2n}]{\alpha }_{i,x}$$

(6)

3.4.2. Model Verification

The measured SVF and microclimate data were compared with the simulated values of the corresponding points to verify the accuracy of the model. As shown in Figure 7, the correlations of SVF, Ta, and RH are higher than 0.7, indicating that the model accurately describes the actual block’s architectural form and landscape vegetation distribution. While ENVI-met can accurately simulate the thermal environment of the street canyon, there is still a difference between the measured and simulated values of the thermal environment index, which is due to the influence of urban traffic flow, outdoor street vendors, and other artificial heat, as well as the error range of the instruments. The limited correlation between the measured and the simulated wind speed, indicates a large deviation between them, which is due to the large instantaneous change in wind speeds of the measurement method.

3.5. Simulation Scheme

In the model, the streetscape measurement indicators can be adjusted by changing the spacing and number of street trees. Then, through thermal environment simulation, the overall tree renewal strategy of street canyon cluster guided by thermal environment optimization is proposed. According to the current situation, scene CG was established as the control group model, and the experimental group EG-12, EG-15, and EG-18 were simulated by selecting row spacing of 12 m, 15 m, and 18 m in the original vegetation location and replanting the same tree species (Figure 8,

Table 6. Control and experimental group of the street view interface indicators.

The Scene Name	Spacing of Vegetation	Average SVF	Average TVF	Average BVF	Number of Trees
CG	Current spacing (12–18 m range)	0.39	0.30	0.31	310
EG-12	12 m	0.21	0.53	0.26	668
EG-15	15 m	0.22	0.52	0.26	534
EG-18	18 m	0.26	0.48	0.26	445

). The boundary conditions of the simulation in the experimental group are the same as those in the simulation verification stage, as shown in Table 3. The measuring points are set in the street canyon and the intersection of the street separately, and 3–4 sampling points are placed inside each street canyon.

4. Simulation Results

4.1. Spatial Distribution and Types of Streetscape Interfaces

The identified SVF, TVF, and BVF values for each measurement point were imported into ArcGIS and an inverse distance weighted interpolation was used to visualize the spatial distribution map of landscape interface measures (Figure 9), presenting three spatial distribution features as a whole. According to the spatial distribution characteristics and the numerical values of SVF, TVF, and BVF, the streets are divided into three types by typology classification method [45].

The first type is the TVF-dominated streetscape space, i.e., street tree-dominated (Figure 10a). The BVF value of such streets is relatively low, and SVF value is mainly affected by TVF values, such as NW-SE street, N-N-S street, and NE-SW-1 street. The TVF values of local areas in NW-SE streets were significantly higher. One side of N-N-S street has the highest SVF due to the building being setback from the street and lack of tree cover. The overall BVF value of NE-SW-1 street is lower, TVF value is higher, vegetation shade is more, and SVF value is lower.

The second type is the streetscape space jointly dominated by BVF and TVF; that is, buildings and trees are jointly dominated (Figure 10b). The BVF value of such streets is significant, while the TVF and SVF values are low so that the TVF value can be increased by increasing landscape vegetation, such as NE-SW-2 Street and N-E-W Street. The BVF value of NE-SW-2 street is larger than that of NE-SW-1 street with the same orientation. The BVF value of N-E-W street is considerable, while SVF and TVF values are small due to the shade provided by the building.

The third type is the SVF-dominated streetscape space, namely street open space (Figure 10c). This kind of street space has a significant SVF value, and both BVF and TVF values are small. The increase of landscape vegetation has almost no effect on the reduction of SVF value, such as street intersections.

4.2. Influence of Streetscape Interface Measurement on Air Temperature

Each value point in this section, and in Section 4.3 and Section 4.4, is an average value over the simulation period for each location in the street canyon/intersection. To allow readers to better judge the subtle changes in street microclimate, the difference value (the value of CG minus the value of EG) is used to express the degree of change.

4.2.1. TVF−Dominated Streetscape Space Type

For the streetscape space dominated by TVF, the air temperature difference between each EG and CG scenes was calculated as Δta (Figure 11). NE−SW−1 street had the best cooling effect (Δta = 0.13 °C, 0.12 °C, 0.11 °C), with a maximum cooling of 0.64%, followed by N−N−S street (Δta = 0.11 °C, 0.10 °C, 0.08 °C). For NW−SE street, air temperature did not drop, but increased (Δta = −0.01 °C, −0.02 °C, −0.02 °C). Compared with the control scene, the optimized scene increases the row number and density of trees in the NW−SE street canyon, which affects the ventilation of the street canyon and increases the air temperature (see Supplementary Materials File S6). However, the building setback distance of N−N−S street was more significant than that of NE−SW−1 street, and the scene of the experimental group led to limited shading increases and relatively further minor cooling in the street canyon.

EG−12 scenes were determined to be the most efficient at reducing temperatures in the street canyon, and NW−SE street canyon had a concentrated data point distribution throughout the day, indicating a minor variation range of air temperature throughout the day. N−N−S street and NE−SW−1 street had similar dispersion degrees of data point distribution; the temperature drop was more significant than in NW−SE street canyon.

4.2.2. BVF+TVF Co−Dominant Streetscape Space Type

For the street scene space jointly dominated by BVF and TVF, the air temperature difference between each EG and CG scenes was calculated as ΔTa (Figure 12). N−E−W street canyon had the most significant temperature drop (ΔTa = 0.21 °C, 0.19 °C, 0.15 °C), with a maximum temperature drop of 1.04%. Centralized data distribution proved that the optimized scene can provide a cooling effect all day long. The air temperature of NE−SW−2 decreased less (ΔTa = 0.08 °C, 0.0 7 °C, 0.07 °C). The cooling effect of the three optimized scenarios was not significantly different, and the large range in temperature differences indicates that there is a large variation during the day in the change in air temperature when trees were added in the NE−SW−2 street.

The reason for this result is that N−E−W street is wider and the building setback distance is greater than in NE−SW−2 street. The street orientation is near east−west so that the sunshine duration is relatively short, which means the addition of landscape vegetation can provide shade and does not affect the ventilation of the street. Hence, the vegetation cooling effect is noticeable.

4.2.3. SVF−Dominated Streetscape Space Type

For SVF-dominated streetscape space, the air temperature difference between each EG and CG scenes was calculated as ΔTa (Figure 13). Each optimization scenario revealed a decrease air temperature, in which NE-SW-2 street intersection had the largest value (ΔTa = 0.13 °C, 0.13 °C, 0.11 °C), with a maximum cooling of 0.65%, followed by N−SW−1 street intersection (ΔTa = 0.08 °C, 0.08 °C, 0.07 °C). Next was the intersection of NW-SE street (ΔTa = 0.05 °C, 0.03 °C, 0.05 °C). The change of air temperature at the intersection of N−E−W street was relatively insignificant (ΔTa = 0.03 °C, 0.04 °C, 0.01 °C), and the air temperature at the intersection of N-N-S street decreased the least (ΔTa = 0.01 °C, 0.03 °C, 0.00 °C).

The results show that, at street intersections, the smaller the planting spacing is not necessarily better for reducing temperature, and the differences are in general small and most of them are not noticeable (<0.1 °C). According to the urban road design code, a safe visual distance must be considered when planting trees at street intersections; that is, enough space should be reserved at road intersections to ensure driving sight. Therefore, as a street intersection space, the number of trees is less than that of general streets, which makes the thermal environment change limited. For NW−SE street intersections, the EG−18 scene revealed the best results and the data distribution was concentrated, indicating that the whole-day cooling effect was positive. For the intersection of N−N−S, NE−SW−2, and N−E−W streets, the EG−15 scene had the best cooling effect. The data distribution at the intersection of NE−SW−2 street had the largest range, indicating that the air temperature varied greatly at different times throughout the day. For the intersection of NE−SW−1 street, EG−12 had the best cooling effect, and data distribution was relatively concentrated.

According to Figure 14, ΔSVF, ΔTVF, and air temperature show a piecewise linear correlation, and the fitting “inflection point” appears with the change of air temperature. When ΔSVF ≤ 0.14, ΔTa increases by 0.07 °C for every 0.1 increase of ΔSVF. When ΔSVF > 0.14, as ΔSVF increases by 0.1, ΔTa only increases by 0.03 °C. When ΔTVF ≤ 0.16, ΔTa increases by 0.09 °C for every 0.1 increase of ΔTVF. When ΔTVF > 0.16, as ΔTVF increases by 0.1, ΔTa increases by 0.03 °C. Therefore, SVF and TVF values should be improved to more than 0.14 and 0.16, respectively, based on the original optimization strategy scenario to achieve higher cooling efficiency. There was no correlation found for BVF, which may be because street trees are generally planted in front of buildings, and vegetation changes building shade, making it difficult to determine the ΔBVF.

4.3. Influence of Streetscape Interface Measurement on Relative Humidity

RH is strongly correlated with temperature and depends on the prevailing temperature.

4.3.1. TVF−Dominated Streetscape Space Type

For the TVF−dominated streetscape space, the relative humidity difference between each EG and CG scenes was calculated as ΔRH (Figure 15). Each optimized scene had the best humidification effect on NE−SW−1 street (ΔRH = 1.3%, 1.1%, 0.8%), with a maximum increase of 1.85% in EG−12. N−N−S street was the second placed (ΔRH = 0.9%, 0.8%, 0.5%), and has the least significant humidification effect on NW−SE street (ΔRH = 0.6%, 0.5%, 0.4%).

EG−12 had the largest effect on increasing relative humidity for all street canyons, showing that increasing landscape vegetation can significantly affect the relative humidity inside street canyons. The NE−SW−1 street had the largest spread and RH changed during the day due to large fluctuations in RH. This was because the increase in landscape vegetation significantly increased the TVF value of the street canyon, and the evapotranspiration of vegetation can significantly increase localized relative humidity. The change in relative humidity distribution of NW−SE street was moderately concentrated. The humidity distribution of N−N−S street was the most concentrated because the setback distance of street canyon buildings is large, which promotes the evaporation of humidity and reduces the increase in humidity.

4.3.2. BVF+TVF Co−Dominant Streetscape Space Type

For the streetscape space jointly dominated by BVF and TVF, the relative humidity difference between each EG and CG scenes was calculated as ΔRH (Figure 16). The amplitude values of increasing relative humidity in two street canyons were similar, and NE−SW−2 street (ΔRH = 1.8%, 1.7%, 1.4%) was slightly higher than N-E-W street (ΔRH = 1.6%, 1.4%, 1.1%), with the maximum increase of 2.51%. NE−SW−2 had a high degree of dispersion of the relative humidity variation, which showed that the relative humidity variation range of the whole day was more extensive. Due to the lack of trees in the CG scene of NE−SW−2 street, the vegetation addition significantly increased the relative humidity of the street canyon. Due to the significant time of direct sunlight, the relative humidity significantly changed throughout the day. The N−E−W street had shorter periods of direct sun exposure throughout the day, but the street is wider, so the relative humidity in the street canyon had a slight change in range throughout the day.

4.3.3. SVF−Dominated Streetscape Space Type

For the streetscape space dominated by SVF, the relative humidity difference between each EG and CG scenes was calculated as ΔRH (Figure 17). For street intersections, the optimized scenes increased the relative humidity, and the N−SW−2 street intersection had the greatest value (ΔRH = 1.7%, 1.6%, 1.3%), with a maximum increase of 2.34%, followed by the N−E−W street intersection (ΔRH = 1.0%, 0.9%, 0.7%). The NW−SE street intersection had the next greatest value (ΔRH = 1.0%, 0.7%, 0.8%), NE−SW−1 relative humidity change was relatively small (ΔRH = 0.9%, 0.8%, 0.6%), and the relative humidity of N−N−S street intersection increased the least (ΔRH = 0.4%, 0.8%, 0.2%).

The EG−15 scene had the best humidification effect for the N−N−S intersection, and the data distribution was relatively concentrated. At other intersections, the EG−12 scene had the best humidification effect. The data distribution at the intersection of NE−SW−2 street was spread out the most, indicating that the humidification effect varies significantly at each time of the day. The dispersion degree of data at intersections of NW−SE, NE−SW−1, and N−E−W streets was similar.

Figure 18 shows that relative humidity also presents an “inflection point” with ΔSVF and ΔTVF changes. The regularity was similar to that of air humidity, because there was a negative linear correlation between temperature and relative humidity. When ΔSVF ≤ 0.25, ΔRH increased by 0.57% as ΔSVF increased by 0.1. When ΔSVF > 0.25, ΔRH increased by 0.81% with every increase of 0.1. When ΔTVF ≤ 0.3, ΔRH increased by 0.39% as ΔTVF increased by 0.1. When ΔTVF > 0.3, ΔRH increased by 0.68% for every 0.1 increase of ΔTVF. This indicated that planting sufficient trees has a better effect on increasing relative humidity. To increase relative humidity, the ΔSVF and ΔTVF values should be adjusted to above 0.25 and 0.3, respectively. There was no correlation found for BVF, which may be because street trees are generally planted in front of buildings, and vegetation changes building shade, making it difficult to determine the ΔBVF.

4.4. Influence of Streetscape Interface Measurement on MRT

4.4.1. TVF-Dominated Streetscape Space Type

For the TVF-dominated streetscape space, the MRT difference between each EG and CG scenes was calculated as ΔMRT (Figure 19). The average MRT reduced by each optimized scene for the three street canyons were similar. NE−SW−1 street had the largest decrease of mean radiation temperature (ΔMRT = 8.83 °C, 9.91 °C, 6.92 °C), with a maximum reduction of 18.59%, followed by NW−SE Street (ΔMRT = 8.32 °C, 7.13 °C, 6.12 °C), and a minor decrease of N-N-S (ΔMRT = 6.50 °C, 5.33 °C, 4.06 °C). This is because the added street trees of N−N−S street are placed further from buildings because of the width of the street, so they do not provide as much shade as the street trees of the narrower streets, NE−SW−1 and NW−SE, which results in a less effective reduction of the MRT.

The EG−15 scene of NE-SW-1 street had the best effect on reducing the MRT, and the data distribution was concentrated, which proves that there was little difference in the variation of the MRT throughout the day. For the NW-SE and N−N−S streets, the EG−12 scenario reduced the MRT the most.

4.4.2. BVF+TVF Co−Dominant Streetscape Space Type

For the streetscape dominated by BVF and TVF, the MRT difference between each EG and CG scenes was calculated as ΔMRT (Figure 20). The change value of the reduced MRT of NE−SW−2 street (ΔMRT = 18.53 °C, 17.72 °C, 16.03 °C) was significantly higher than N−E−W street (ΔMRT=7.18 °C, 7.80 °C, 6.72 °C), with a maximum decrease of 37.79%. The wide spread data distribution indicated that the variation in the MRT throughout the day was relatively large. We believe the reason for this is that SVF is larger and TVF is lower in the CG scene of NE−SW−2 street. The EG scene with additional planting covers significantly increased the shaded area common to vegetation and buildings, resulting in a significant decrease in MRT. However, N-E-W street has a minor variation in the MRT due to sufficient shading of some buildings and less direct sunlight.

4.4.3. SVF−Dominated Streetscape Space Type

For the SVF−dominated streetscape space, the MRT difference between each EG and CG scenes was calculated as ΔMRT (Figure 21). The optimized scenes reduced the MRT, and the values at the intersection of NE-SW-2 street were the largest (ΔMRT = 15.87 °C, 15.26 °C, 13.15 °C), with a maximum decrease of 34.24%, followed by the intersection of N−N−S street (ΔMRT = 9.37 °C, 11.53 °C, 9.89 °C). The intersection of NW−SE street was second (ΔMRT = 6.87 °C, 5.49 °C, 6.75 °C), followed by a change in average MRT of NE−SW−1 (ΔMRT = 4.32 °C, 5.15 °C, 2.51 °C), and the decrease of MRT of N−E−W was the least (ΔMRT = 3.38 °C, 3.21 °C, 1.86 °C).

The EG−15 scene at the N−N−S and NE−SW−1 streets intersections reduced the maximum MRT. The data were distributed centrally, indicating a slight ΔMRT variation throughout the day. The ΔMRT corresponding to the EG−12 scene at the intersection of NW−SE, NE−SW−2, and N−E−W streets was the largest, and the data distribution of NE−SW−2 street had the largest range, indicating that the variation of ΔMRT throughout the day was relatively large. This was probably due to the fact that the intersections of NW−SE, NE−SW−2 and N-E-W streets provided more shade than the other intersections under the EG−12 scenario.

As seen from Figure 22, the MRT also presents an “inflection point” with the change of ΔSVF and ΔTVF. When ΔSVF ≤ 0.15, ΔMRT increased by 9.0 °C as ΔSVF increased by 0.1 °C. When ΔSVF > 0.15, ΔMRT increased only 3.0 °C for every 0.1 growth of ΔSVF. When ΔTVF ≤ 0.21, ΔTVF increased by 0.1, ΔMRT increased by 7.1 °C. When ΔTVF > 0.3, ΔMRT increased only 2.9 °C for every 0.1 increase of ΔTVF. To make the MRT cost-effective, the ΔSVF and ΔTVF values should be increased to at least 0.15 and 0.21, respectively. There was no correlation found for BVF, which may be because street trees are generally planted in front of buildings, and vegetation changes building shade, making it difficult to determine the ΔBVF.

5. Discussion

5.1. Precise Optimization Strategy for Specific Streetscape Types

Based on the simulation results of general street canyon clusters in the 12-m planting interval scenario (EG-12), the precise optimization strategies of three special street canyons were further examined: interior space of street canyons (type A), street intersection space (type B), and park and green space (type C). Space A and space B represent space with high SVF and low SVF, respectively. Although both optimize the thermal environment by appropriately reducing the spacing, the mechanism is different. The former improves ventilation by reducing tree occlusion, while the latter increases tree shading and solar radiation. The thermal environment optimization strategies were formulated and simulated for each type of street canyon cluster combined with the Harbin city road planting code (

Table 7. Optimization strategies of various spaces in street canyon clusters.

Optimized Simulation Scenarios	Plan	Street Section
A-1 Building setback space
A-2 Narrow street canyon space
B-1 Multi-directional intersection
B-2 Crossroad
C-1 Street pocket park
C-2 Fragmented green space

). Simulation results were obtained, and landscape design guidelines were developed (

Table 8. Optimization strategy of the thermal environment of street canyon clusters.

C-2 Fragmented green space				ΔTa = 0.31 °C, ΔRH = 1.52%, ΔMRT = 19.59 °C	Tree arrays should be planted 15 m apart.	For the whole site, fragmentary green space should be excavated and utilized as much as possible, and landscape vegetation should be increased to improve the thermal environment.
C-1 Street pocket park				ΔTa = 0.13 °C, ΔRH = 1.26%, ΔMRT = 10.57 °C	Trees should be added, spaced 15 m apart, and street-facing activity space should be provided.	The optimization strategy of street pocket parks should strike a balance between air cooling efficiency and the openness of walking space.
B-2 Crossroad				ΔTa = 0.17 °C, ΔRH = 1.22%, ΔMRT = 5.7 °C	Add 20% green patches on both sides and use 15-m planting intervals.	For all street intersections, space potential should be explored, the number of green patches should be increased, and street trees should be planted appropriately.
B-1 Multi-directional intersection				ΔTa = 0.21 °C, ΔRH = 2.07%, ΔMRT = 16.23 °C	Within 30 m at street intersections, planting spacing of 15 m should be adopted.	For multi-directional intersections, vegetation spacing should be expanded within a specific range from the intersection while paying attention to safety visual distance.
A-2 Narrow street canyon space				ΔTa = 0.09 °C, ΔRH = 0.92%, ΔMRT = 1.48 °C	Single rows should be planted, spaced up to 18 m apart.	Single-row planting should be considered appropriate for narrow street canyon space to prevent street canyon overcrowding from affecting the overall thermal environment.
A-1 Building setback space				ΔTa = 0.20 °C, ΔRH = 1.34%, ΔMRT = 16.3 °C	Two rows of street trees should be added, and the planting spacing increased to 18 m.	When increasing the number of rows and columns of street trees, it is necessary to ensure the optimization effect of the thermal environment and the economy.
Spatial types	Current situation	The original scene	Optimization strategy scene	Best optimized effect	Optimal planting strategy	Guidelines for landscape design

; see Supplementary Materials File S5).

For the building setback space A-1 (represented by N-N-S street), the simulation showed that adding two rows of street trees and increasing the planting spacing to 18 m can optimize the thermal environment and, as an added benefit, be more cost-effective (ΔTa = 0.20 °C, ΔRH = 1.34%, ΔMRT = 16.3 °C). For the narrow street canyon space A-2 (represented by NW-SE street), instead of planting two rows of street trees, it was found that single row planting and increasing the spacing to 18 m had the best thermal environment optimization efficiency (ΔTa = 0.09 °C, ΔRH = 0.92%, ΔMRT = 1.48 °C).

For the multi-directional intersection B-1, simulation results showed that the optimal thermal environment was best when a planting interval of 15 m is adopted within 30 m range of street intersection (ΔTa = 0.21 °C, ΔRH = 2.07%, ΔMRT = 16.23 °C). For the crossroad space B-2, while increasing the green patches (20%) on both sides, the planting spacing of 15 m was used to improve the thermal environment (ΔTa = 0.17 °C, ΔRH = 1.22%, ΔMRT = 5.7 °C).

For the street pocket park C-1, and considering the efficiency of air temperature reduction and the open balance of walking space, the planting method with the spacing of 15 m was selected as the best balance between them (ΔTa = 0.13 °C, ΔRH = 1.26%, ΔMRT = 10.57 °C). For fragmented green space C-2, it was found that the optimal thermal environment of tree array with planting spacing of 15 m was the best (ΔTa = 0.31 °C, ΔRH = 1.52%, ΔMRT = 19.59 °C).

5.2. Correlation Analysis between Streetscape Interface Measurement and Thermal Environment

The critical threshold of Ta/RH/MRT changes for two street types with correlation are discussed. Each value point in this section is an average value over the simulation period for each street canyon/intersection location. To allow readers to better judge the subtle changes in street microclimate, the difference value (the value of CG minus the value of EG) is used to express the degree of change.

5.2.1. TVF-Led Streetscape Interface Measurement

The N-N-S street is taken as an example because adding street trees in the N-N-S street can significantly increase TVF and significantly decrease SVF compared to the other TVF-led streetscape streets. As can be seen from Figure 23, both ΔTVF and ΔSVF show significant quadratic correlation with ΔTa, ΔRH, and ΔMRT.

The quadratic correlation curves of ΔTVF, ΔSVF and ΔTa open upward, with ΔTVF = −0.18/ΔTa = 0.27 °C; ΔSVF = 0.13/ΔTa = 0.28 °C as the critical threshold, at which ΔTa is the minimum. The changing trend around the crucial point ΔTa shows a prominent quadratic correlation characteristic of “first fall and then rise”. For the precise optimization strategy of landscape vegetation in N-N-S street canyon, the streetscape interface measurement with ΔTVF < −0.18 and ΔSVF < 0.13 should be adopted to achieve the best air temperature reduction efficiency.

The quadratic correlation curves of ΔTVF, ΔSVF, and ΔRH open downward, with ΔTVF =−0.07/ΔRH = −0.26 %; ΔSVF = 0.04/ΔRH = −0.25% as the critical threshold, when ΔRH is maximum. The shifting trend around the crucial point presents a prominent quadratic correlation of “first rise and then fall”. For the N-N-S street canyon, the landscape interface measurement index ΔTVF < −0.07 and ΔSVF > 0.04 should be adopted to increase relative humidity greatly.

The correlation curves of ΔTVF, ΔSVF, and ΔMRT open upward, and are close to that of ΔTVF and ΔTa, with ΔTVF = −0.11/ΔMRT = 0.82 °C; ΔSVF = 0.07/ΔMRT = 1.93 °C as the critical threshold, at which time ΔMRT is the minimum. In optimizing N-N-S street canyon landscape vegetation, ΔTVF < −0.11 and ΔSVF < 0.07 have the best effect on reducing the MRT.

5.2.2. Streetscape Interface Measurement Jointly Led by TVF and BVF

Taking the NE-SW-2 street canyon as an example, the BVF value of the streetscape interface measure is relatively high, and the height of trees and buildings is similar, which provides shade and thus affects the thermal environment of the street canyon. The influence of TVF and SVF changes on the overall thermal environment is limited. Figure 24 shows the effect of ΔTVF and ΔSVF on thermal environment indexes.

The quadratic correlation curve of ΔTVF, ΔSVF, and ΔTa opens downward, with Δ TVF = −0.17/ΔTa = 0.35 °C; ΔSVF = 0.15/ΔTa = 0.37 °C as the critical threshold, and ΔTa is the largest at this time. The changing trend of ΔTa around the critical point shows a quadratic correlation characteristic of “first rise and then fall”. For the precise optimization strategy of landscape vegetation in NE-SW-2 street canyon, the measurement indexes of ΔTVF = −0.17 and ΔSVF = 0.15 should be adopted to achieve the best cooling effect.

The correlation curve between ΔTVF and ΔRH opens upward, with Δ TVF = −0.30/ΔRH = −1.85%; ΔSVF = 0.26/ΔRH = −1.81% as the critical threshold, and ΔRH is the smallest at this time. The changing trend around the critical threshold ΔRH presents a relatively prominent quadratic correlation characteristic of “first fall and then rise”. For the NE-SW-2 street canyon, if the relative humidity is significantly improved, the landscape interface measurement indicators of ΔTVF = −0.30 and ΔSVF = 0.26 should be adopted.

The quadratic correlation curves of ΔTVF, ΔSVF, and ΔMRT open downward, TVF = −0.30/ΔMRT = 18.74 °C; ΔSVF = 0.26/ΔMRT = 18.08 °C is the critical threshold, but the correlation R² is only 0.33 and 0.35; these correlations are too weak to make strong conclusions. This is because the NE-SW-2 street is narrow and shaded by vegetation and buildings, affecting changes in mean radiant temperature. Due to building shading, the shift in landscape vegetation has a limited impact on the average radiant temperature.

5.2.3. Comparison with Previous Literature Conclusions

A study in the high-latitude city of Montreal found that at midday, SVF and Ta were linearly correlated, with a R2 of 0.45 between SVF and Ta, and 0.44 between SVF and MRT in the shaded area. At midnight, the R2 between SVF and Ta was 0.64, and that between SVF and MRT it was 0.69 [15]. A previous study in Harbin found that SVF of street canyons had a secondary correlation with mean temperature, mean relative humidity, and MRT. The results showed that SVF was significantly correlated with these factors. According to current fitting results, when the SVF is about 0.2, the three main factors related to the thermal environment (air temperature, relative humidity, and MRT) show an opposite trend [18]. Montreal and Harbin are both high latitude regions, but the relationship between thermal environment indicators and SVF is different. This conclusion may be due to the unique urban morphological characteristics of Harbin. In this study, thermal environment indicators have a quadratic correlation with SVF and TVF, and there are different inflection points toward different streets. This may imply that there may also be a secondary correlation between other forms of streets and thermal environment in Harbin, and the inflection point of the secondary relationship will be the focus of future research.

5.3. Limitations and Future Studies

This research examined the relationship between the measurement of streetscape interface and thermal environment and the optimization strategy of typical blocks in Harbin. However, there is still work to be done. The planting spacing of all the street canyons in the study area is suggested to be 12 m. For special street canyons with potential for optimization, it is suggested to increase planting spacing appropriately, and design guidelines are proposed. First, the overall building height of the street canyon cluster is similar to the height of street trees, and buildings and vegetation jointly play a role in shading streets in this research. However, for other blocks in Harbin, where the height of buildings and vegetation are different, the leading role of streetscape interface measurements may be different. Furthermore, adding trees can theoretically lead to reductions in temperature, but the changes are of such an order of magnitude that they cannot be measured. This paper uses the typology method to classify streets, which is a common qualitative classification method in the preliminary study of urban morphology. For the future study of street canyons in Harbin city area, cluster analysis can be used, which is suitable for a large number of samples at macro urban scale. Therefore, targeted thermal environment optimization strategies should be further studied to find the best optimization value.

Second, the thermal environment optimization strategy of the typical block from street canyon clusters in Harbin is not simply analogous to the cold regions in other cities in China. Different latitudes lead to differences in streetscape form, sunshine angle, and duration of exposure, as well as other contextually specific conditions in different cities. This study provides a method framework, and specific optimization strategies should be further simulated and formulated according to the actual situation.

Finally, the simulation of ENVI-met has certain limitations. The ENVI-met model simplifies the complex vegetation morphology into cubic units with different leaf areas, and does not accurately depict the complexity of trees. Similarly, the ENVI-met architectural model also ignores the anthropogenic heat factor. In future research, the accuracy of this methodology could be improved by increasing the simulation accuracy and subdividing the layers and forms of vegetation and buildings.

6. Conclusions

In this research, the effect mechanism of street view interface measurements on the thermal environment of different types of street canyons in Harbin was studied by field measurements combined with ENVI-met numerical simulation. The overall optimization strategy of the thermal environment of street canyons in Harbin was discussed. In general, street canyons and intersections are affected by landscape interface measurements by different mechanisms. All street canyons should be planted with 12 m spacing as a whole, and locally with appropriate amplification spacing. The main conclusions are as follows:

(1) For the general street canyon type (0.27 < mean SVF < 0.51), the optimal thermal environment optimization efficiency is achieved when the street trees are located 12 m apart. The average air temperature decrease was 0.78%, and the reduction of street temperature dominated by BVF+TVF was the most significant (1.04%). The average relative humidity increased by 2.23%, and the increase was highest for streets dominated by SVF and BVF+TVF (2.34%, 2.51%). MRT decreased by 30.20% on average, with the most significant decrease (34.24%, 37.79%) in streets dominated by SVF and BVF+TVF.

(2) For the distinct street canyon types (SVF ≤ 0.27 or SVF ≥ 0.51), a precise streetscape interface regulation strategy should be adopted according to the streetscape space type to achieve the optimal balance between thermal environment optimization and cost. ① Building setback space (A-1) should adopt three rows of 18 m spacing planting strategy. ② Narrow street canyon space (A-2) should be a single row with 18 m spacing planting strategy. ③ Multi-directional intersection (B-1) should adopt the planting strategy of 15 m spacing within 30 m. ④ For the crossroad space and park and green space (B-2, C-1, C-2), the optimal strategy of increasing green patches with planting spacing of 15 m should be adopted.

(3) The streetscape interface measure and thermal environment index show quadratic correlation characteristics and different critical thresholds. ① For streets similar to N-N-S and dominated by TVF, the correlation between ΔTVF, ΔSVF, Ta (R² = 0.92/0.91), RH (R² = 0.95/0.95), and MRT (R² = 0.95/0.94) is significant. ② In street canyon dominated by BVF+TVF, ΔTVF, and ΔSVF have a significant quadratic correlation with air temperature (R² = 0.93/0.94), a relatively substantial correlation with relative humidity (R² = 0.63/0.64), but is not strongly correlated with MRT (R² = 0.33/0.35).