The Mechanism of Street Spatial Form on Thermal Comfort from Urban Morphology and Human-Centered Perspectives: A Study Based on Multi-Source Data

Abstract

The influence of street spatial form on thermal comfort from urban morphology and human-centered perspectives has been underexplored. This study, utilizing multi-source data and focusing on urban central districts, establishes a refined index system for street spatial form and a thermal comfort prediction model based on extreme gradient boosting (XGBoost) and Shapley additive explanations (SHAP). The results reveal the following: (1) Thermal comfort levels display spatial heterogeneity, with areas of thermal discomfort concentrated in commercial zones and plaza spaces. (2) Compared to the human-centered perspective, urban morphology indicators correlate strongly with thermal comfort. (3) The key factors influencing thermal comfort, in descending order of importance, are distance from green and blue infrastructure (GBI), tree visibility factor (TVF), street aspect ratio (H/W), orientation, functional diversity indices, and sky view factor. All but the TVF negatively correlates with thermal comfort. (4) In local analyses, the primary factors affecting thermal comfort vary across streets with different heat-risk levels. In high heat-risk streets, thermal comfort is mainly influenced by distance from GBI, H/W, and orientation, whereas in low heat-risk streets, vegetation-related factors dominate. These findings provide a new methodological approach for optimizing urban thermal environments from both urban and human perspectives, offering theoretical insights for creating more comfortable cities.

1. Introduction

Global warming has emerged as one of humanity’s most significant challenges in the coming decades [1,2,3]. The Intergovernmental Panel on Climate Change (IPCC) report titled “Climate Change 2023” [4] projects that global temperatures will exceed 1.5 °C by 2030. As urbanization intensifies, compounded by both global climate change and urbanization effects, aggravated heat island phenomena will result. The frequency of extreme heat waves has increased, posing significant risks to urban residents’ physical and mental health [5,6,7]. Thermal comfort, which refers to individuals’ subjective perception of satisfaction with the thermal environment, is influenced by both objective and subjective factors [8]. Streets, as key public spaces in cities, significantly influence residents’ thermal comfort [8,9,10]. Therefore, understanding how street spatial form affects thermal environments is essential for improving residents’ quality of life and overall health.

Existing research on street thermal comfort primarily focuses on quantifying the components of street spatial form, such as road surfaces, building facades, the sky view created by these elements, and street landscapes [11,12,13]. Common indicators often include building form, street aspect ratio (H/W), orientations, landscape configurations, and sky visibility factors (SVF) [14,15,16,17,18,19]. Modifying these urban morphology indicators can reduce surface temperatures and enhance urban thermal environments and human thermal comfort effectively by adjusting physical elements such as shadow areas and wind conditions [20,21,22]. However, most studies mainly reflect urban spatial form and thermal environments with field measurements and numerical simulations [23], which are limited in scope and cost, making it difficult to assess the spatial environment on a larger scale of neighborhoods in a fine-grained manner.

Although the impact of street spatial form on thermal comfort has been extensively examined from an urban morphology perspective, the relationship between street form and thermal comfort from a human-centered perspective is often neglected. Subjective perception is a crucial factor influencing human thermal comfort [24,25,26]. Environmental elements perceived by individuals (such as vegetation, buildings, sky, and materials) can affect emotions, thereby influencing thermal comfort levels both physiologically and psychologically [27,28,29,30]. Due to limitations in data collection and processing, few studies have integrated subjective data into spatial indicator systems. Most existing research relies on objective metrics, such as physiological equivalent temperature (PET) and mean radiant temperature (T_mrt) for thermal comfort evaluation [19,31], while subjective assessments depend on questionnaires and interviews, making it difficult to apply in larger study areas. Consequently, it is essential to analyze and integrate spatial indicators and subjective thermal comfort evaluations of streets from a human perspective to support the development of thermal environment optimization strategies that prioritize the experiences of street users.

Advancements in artificial intelligence have opened new avenues for research through machine learning methods. The application of multi-source data has enhanced the extraction and processing capabilities, enriching the types and precision of urban foundational data. For instance, semantic segmentation models can extract view factor (VF) information from street view images [32,33,34], while keywords can be derived from location check-in data [34]. Furthermore, machine learning and deep learning methods, such as random forest (RF), extreme gradient boosting (XGBoost), and recurrent neural networks (RNN), can be utilized to develop predictive models, improving the accuracy of quantifying nonlinear relationships [35,36,37]. The Shapley additive explanations (SHAP) model helps clarify the relationships between independent and dependent variables in predictive models [38], enhancing the interpretability and readability of machine learning outputs. Despite these advancements, current studies often lack the integration of refined street spatial elements, indicating that the evaluation system for street spaces is still further improvement [39]. Additionally, there is a pressing need for the practical application of computer science technologies in data mining and analysis.

In summary, current research on the impact of street space on outdoor thermal comfort primarily relies on field measurements and numerical simulations, which often have limitations in scope and cost. Street spatial indicators are typically selected from an urban morphology perspective, with limited attention to human-centered approaches. Furthermore, various environmental elements within urban spaces have varying impacts on thermal comfort, and existing studies have not integrated street spatial indicators from both urban morphology and human-centered perspectives, leaving the mechanisms of their influence on thermal comfort unclear. The application of multi-source data and machine learning technologies enhances the richness and analytical precision, underscoring the need for further refinement of the street space evaluation system and deeper exploration of its relationship with thermal comfort.

Therefore, this study focused on the central districts of a coastal city, utilizing multi-source data—such as building forms and street view images—to construct a refined street spatial form indicator system. By integrating both objective (T_mrt) and subjective (Weibo hot check-in) thermal comfort evaluations, XGBoost and SHAP models were employed to establish a street thermal comfort prediction model. This model quantified the mechanisms through which street spatial form influences thermal comfort, providing a basis for managing urban thermal risks and optimizing thermal environment design. The research consisted of two main components: (1) Constructing a refined street spatial form indicator system and accurately assessing the overall outdoor thermal comfort levels of streets in the study area. (2) Utilizing XGBoost and SHAP explainable machine learning algorithms to establish a street thermal comfort prediction model, calculating the contributions of various street spatial form factors to thermal comfort.

2. Materials and Methods

The workflow of this study is illustrated in Figure 1. First, the comprehensive thermal comfort of the study area was assessed by utilizing urban geographic data, street view images, and Weibo data. This process generated a system of urban morphology and human-centered perspectives on street spatial form indicators. Following this, a correlation analysis was conducted between street spatial form and thermal comfort, leading to the selection of relevant indicators. Based on these indicators, a thermal comfort prediction model was developed using XGBoost and SHAP models. Finally, optimization design proposals for urban spaces and key streets were developed based on the research findings to improve the thermal environment.

2.1. Study Area

The study site is situated in the central urban area of Dalian, Liaoning Province, China, as illustrated in Figure 2. Dalian has a temperate continental monsoon climate influenced by maritime conditions, resulting in a generally mild and humid atmosphere and distinct seasons. The annual temperature ranges from −16.8 to 37 °C, with an average of 11.19 °C [40]. In particular, the city center is affected by high-density urban spatial morphology and intense solar radiation, contributing to significant heatwave issues and thermal discomfort [41].

Considering the current high-temperature conditions in Dalian, this study focused on typical streets in the city’s central district, characterized by a significant urban heat island effect and relatively well-preserved street patterns. The study area spans approximately 5.9 square kilometers, with an east-to-west topographical gradient. To ensure accurate calculations of street spatial form indicators and reliable thermal environment simulations, streets with incomplete data, lower classification, or predominantly vehicular functions were excluded. In the end, 107 streets were selected for analysis.

2.2. Thermal Comfort

The study conducted a comprehensive evaluation of thermal comfort from both objective and subjective perspectives. Objective thermal comfort was assessed using the Radiant Heat Stress Intensity (RHSI) index, which evaluates dynamic changes in heat stress based on T_mrt, a key meteorological parameter representing the impact of thermal radiation on human thermal comfort [42]. The formula for calculating RHSI is as Equation (1).

$$\mathrm{RHSI}={{\displaystyle \sum }}_{i=1}^n{T}_{mrt}\left(i\right)=\left\{\begin{array}{c}\left(i\right)-55 ℃, T_{mrt}\left(i\right)\ge 55 ℃\\ 0,T_{mrt}\left(i\right)<55 ℃\end{array}\right.$$

(1)

where RHSI represents the cumulative temperature difference when the average radiant temperature exceeds 55 °C during the study period [43,44]; n is the total hours of sunlight during the study period; T_mrt(i) is the mean radiant temperature at hour i.

In this study, T_mrt was simulated using the SOLWEIG model [21,45,46], with simulations running from 10:00 to 17:00 at hourly intervals. Baseline meteorological data were collected using a HOBO portable automatic weather station and a handheld anemometer at a representative meteorological observation site within the study area on 26 July 2021. Finally, the RHSI was classified into five levels using the natural breaks classification method in QGIS.

Subjective thermal comfort is primarily assessed based on Weibo check-in data from the study area. High-temperature check-in data from Weibo can effectively reflect residents’ perceptions and expressions of thermal comfort. Previous studies have demonstrated a significant positive correlation between temperature and heat-related Weibo data, indicating that such data can objectively reflect the public’s subjective experience of the thermal environment [34]. This study employed the BERT language processing model for semantic analysis to identify and filter heat-related information from 2435 Weibo check-ins in Dalian during July and August 2021. After training, the model achieved an accuracy of 95% [47,48]. 237 heat-related check-ins were processed using QGIS to perform spatial analysis, mapping the data into a 50 m × 50 m grid for the study area. Considering the influence of urban population distribution, this study defined subjective thermal comfort as the ratio of high-temperature-related Weibo check-in data points to the total number of Weibo check-in data points, calculated as Equation (2).

$$W_i=d\left(i\right)/D\left(i\right)$$

(2)

where W_i represents the subjective thermal perception in grid i; d(i) is the number of high-temperature-related Weibo check-in data points in grid i; D(i) is the total number of Weibo check-in data points in grid i.

Both objective and subjective thermal comfort indicators were standardized using the range method. Subsequently, a mathematical function model was developed to create a comprehensive outdoor thermal comfort index (CTC) for the study area. This index was visualized in QGIS and classified into five risk levels using the natural breaks classification method, facilitating a precise qualitative and quantitative assessment of the overall thermal comfort level in the study area. The formula for calculating the CTC is Equation (3).

$$\mathrm{CTC}=AF+W\left(1-F\right)$$

(3)

where A is the standardized heat stress intensity evaluation index; W is the standardized Weibo thermal evaluation index; F is the weight of the indicators (subjective and objective thermal comfort indicators are assigned equal influence on outdoor thermal comfort in this study, each with a weight of 0.5).

2.3. Urban Morphology

2.3.1. Urban Morphology Perspective Street Spatial Indicators

Previous studies, predominantly from a macro-level urban morphology perspective, have investigated how morphological characteristics such as street scale and shading influence neighborhood thermal comfort. Building on these findings, this study integrates common street indicators, including the H/W and SVF, while also accounting for additional urban design factors. These factors encompass street networks, interface continuity, functional diversity, and the presence of blue-green spaces. The primary indicators for quantifying street morphology include H/W, street orientation, functional diversity indices, build-to-line rate, distance from green and blue infrastructure (GBI), and street view factors (VF). The calculation methods for these indicators are provided in Table 1.

The urban road network, building form, land use types, and surface cover data were provided by Dalian Territory Spatial Planning and Design Co., Ltd. The POI data were obtained from the Amap API platform via Python. Fish-eye images were sourced from the Baidu API platform, with zenith-view street images scraped using a Python program at 30 m intervals. A total of 2590 effective 360-degree panoramic photos were collected, as illustrated in Figure 3. After undergoing fish-eye projection processing, the photos were recognized and segmented using the pyramid pooling scene parsing network (PSPNet) model, achieving an accuracy of 80.1% [49,50,51,52]. All data were processed, computed, and analyzed spatially in QGIS to accurately reflect the current state of the urban built environment.

Table 1. Street spatial perception indicators from urban morphology perspective.

Parameters	Representative Features	Definition and Calculation Method
H/W	Street Scale [18,19]	The ratio of the average height of buildings (H) on both sides to the average width (W) between facades
Street orientation		The clockwise angle between the north axis and the long axis of the street, ranging from 0° to 180°.
Functional diversity indices	Street Functional Diversity [15]	$S_d=-{\displaystyle {\displaystyle \sum }_i^n}{P}_{i}ln{P}_i$ n is the number of POI categories; Pi is the proportion of POI category i within the buffer zone.
Build-to-line rate	Street Interface Continuity [53]	$P_i=B_i/L_i\times 100\%$ Bi is the street wall line of road i; Li is the length of street i.
Distance from GBI	Street Ecological Location [15]	$D_i=Min\left(d_{ij}\right)$ dij is the distance between j (park, lake, etc.) and point i on a street
View factors (VF): Sky view factor (SVF) Tree view factor (TVF) Building view factor (BVF)	Street Shading [32,54]	$VF={\displaystyle \frac{\pi }{2n}}{\displaystyle {\displaystyle \sum }_{i=1}^n}sin\left[{\displaystyle \frac{\pi \left(2i-1\right)}{2n}}\right]\left({\displaystyle \frac{a_i}{t_i}}\right)$ n is the number of concentric rings in the fisheye image aliquot; $\left({\displaystyle \frac{a_i}{t_i}}\right)$ is the percentage of each type of streetscape element in the ring i of the segmentation

Table 1. Street spatial perception indicators from urban morphology perspective.

Parameters	Representative Features	Definition and Calculation Method
H/W	Street Scale [18,19]	The ratio of the average height of buildings (H) on both sides to the average width (W) between facades
Street orientation		The clockwise angle between the north axis and the long axis of the street, ranging from 0° to 180°.
Functional diversity indices	Street Functional Diversity [15]	$S_d=-{\displaystyle {\displaystyle \sum }_i^n}{P}_{i}ln{P}_i$ n is the number of POI categories; Pi is the proportion of POI category i within the buffer zone.
Build-to-line rate	Street Interface Continuity [53]	$P_i=B_i/L_i\times 100\%$ Bi is the street wall line of road i; Li is the length of street i.
Distance from GBI	Street Ecological Location [15]	$D_i=Min\left(d_{ij}\right)$ dij is the distance between j (park, lake, etc.) and point i on a street
View factors (VF): Sky view factor (SVF) Tree view factor (TVF) Building view factor (BVF)	Street Shading [32,54]	$VF={\displaystyle \frac{\pi }{2n}}{\displaystyle {\displaystyle \sum }_{i=1}^n}sin\left[{\displaystyle \frac{\pi \left(2i-1\right)}{2n}}\right]\left({\displaystyle \frac{a_i}{t_i}}\right)$ n is the number of concentric rings in the fisheye image aliquot; $\left({\displaystyle \frac{a_i}{t_i}}\right)$ is the percentage of each type of streetscape element in the ring i of the segmentation

2.3.2. Human-Centered Perspective Street Spatial Indicators

As users of street spaces, pedestrians’ perceptions of the street environment can significantly influence their thermal comfort. This study incorporates human-centered spatial factors to explore their relationship with street thermal comfort. The human-centered indicators consist of both subjective and objective components, as detailed in Table 2. Objective indicators include the visibility of various spatial elements (sky, trees, buildings, etc.) from a pedestrian’s view.

Pedestrian-view street images were obtained from the Baidu API platform, and scraped at 30 m intervals, covering the period from May to July between 2016 and 2022. Images were stitched together from four 90° horizontal views at each sampling point, resulting in 2408 360° street scene images. These images were also processed using the PSPNet model to extract and classify the various elements of the street scenes.

Subjective indicators are derived from a combination of manual evaluation models and machine learning assessment models that generate quality scores for street spaces. A group of 20 researchers in urban planning evaluated a sample of 100 images based on cleanliness, permeability, human scale, enclosure, and imagery [55,56]. These evaluations were then used to train an RF model, which was applied to assess street space quality across the study area. The results were visualized using QGIS.

Table 2. Street spatial perception indicators from human-centered perspective.

Parameters	Representative Features	Definition and Calculation Method
Green View Index (GVI)	Level of greening of streets in three dimensions [29,57]	$C_{green }=n_{green}/N$ ngreen is the vegetation element streetscape image pixel value; N is the total pixel value of the streetscape image.
Sky visibility degree	Sky opening from human perspective [58]	$C_{sky}=n_{sky}/N$ nsky is the sky element streetscape image pixel value; N is the total pixel value of the streetscape image.
Interface enclosure degree	The spatial scale of the street [59,60]	$C_{building}=n_{building }/N$ nbuilding is the building element streetscape image pixel value; N is the total pixel value of the streetscape image.
Carriageway visibility degree	Motorization level of the street [9,61]	$C_{car}=n_{car}/N$ ncar is the carriageway element streetscape image pixel value; N is the total pixel value of the streetscape image.
Pavement visibility degree	Pedestrianization level of the street [9,61]	$C_{foot}=n_{foot}/N$ nfoot is the pavement element streetscape image pixel value; N is the total pixel value of the streetscape image.
Street space quality	Human subjective perception of street space [9]	An RF model is built based on the scoring results of a small sample of researchers in the planning field to evaluate the quality of a large sample of streets.

Table 2. Street spatial perception indicators from human-centered perspective.

Parameters	Representative Features	Definition and Calculation Method
Green View Index (GVI)	Level of greening of streets in three dimensions [29,57]	$C_{green }=n_{green}/N$ ngreen is the vegetation element streetscape image pixel value; N is the total pixel value of the streetscape image.
Sky visibility degree	Sky opening from human perspective [58]	$C_{sky}=n_{sky}/N$ nsky is the sky element streetscape image pixel value; N is the total pixel value of the streetscape image.
Interface enclosure degree	The spatial scale of the street [59,60]	$C_{building}=n_{building }/N$ nbuilding is the building element streetscape image pixel value; N is the total pixel value of the streetscape image.
Carriageway visibility degree	Motorization level of the street [9,61]	$C_{car}=n_{car}/N$ ncar is the carriageway element streetscape image pixel value; N is the total pixel value of the streetscape image.
Pavement visibility degree	Pedestrianization level of the street [9,61]	$C_{foot}=n_{foot}/N$ nfoot is the pavement element streetscape image pixel value; N is the total pixel value of the streetscape image.
Street space quality	Human subjective perception of street space [9]	An RF model is built based on the scoring results of a small sample of researchers in the planning field to evaluate the quality of a large sample of streets.

2.4. Interpretable Machine Learning Algorithms

This study constructs a street thermal comfort prediction model based on XGBoost, utilizing detailed street spatial form. The analysis employs SHAP to assess the significance of various spatial morphology factors on overall thermal comfort in urban areas. XGBoost, developed by Chen et al. [62], has achieved widespread application and success across various domains [63,64]. The study uses a dataset of street spatial form indicators as the explanatory variables and comprehensive thermal comfort indicators as the response variable. Data samples from the study area are input into the XGBoost for training within a Python environment.

SHAP, developed by Lee and Lundberg [38], is an interpretable machine learning model that quantifies the contribution of each feature through a feature attribution method, cumulatively yielding the final prediction result. Its robust data visualization capabilities make it well-suited for interpreting complex algorithmic models [65]. The mathematical formulation of the SHAP model is as Equation (4):

$$y_i=y_0+f\left(x_i,x_1\right)+f\left(x_i,x_2\right)+f\left(x_i,x_3\right)+\cdots +f\left(x_i,x_k\right)$$

(4)

where y_i represents the predicted value for the sample being explained; y₀ is the model’s average prediction across all samples (the baseline of the model); $f\left(x_i,x_k\right)$ represents the contribution of the feature k value of sample i to the model’s predicted value.

3. Results

3.1. Thermal Comfort Conditions and Street Spatial Form

The evaluation results of the thermal environment in the study area are presented in Figure 4. The objective thermal comfort levels across the study blocks are generally medium, with high thermal stress streets primarily located near urban public spaces, while low thermal stress streets are mainly narrow segments with high levels of shading. Subjective assessments reveal a clear spatial clustering of thermal environments, with areas of high environmental sensitivity.

Based on the CTC results, the comprehensive thermal comfort in the study area ranges from a maximum of 0.9 to a minimum of 0, with a standard deviation of 0.2. There is significant spatial variation in thermal comfort levels, with most areas having a thermal risk level below medium. High-risk areas are primarily concentrated in the northwest part of the study area (around the commercial center) and the east-central region (large urban public open spaces). These areas experience high pedestrian traffic, sparse street trees, and weak shading facilities, resulting in a comprehensive thermal comfort index that is moderate to high, leading to low human thermal comfort. In contrast, blocks in the southwest and northeast of the study area have comprehensive thermal comfort indices below the medium level, indicating a favorable urban thermal environment. Specifically, the central north-south oriented blocks in the study area show a distinct “high-low-high” pattern in the thermal comfort level, likely influenced by factors such as the arrangement of surrounding buildings, open spaces, and varying levels of greenery along different segments of the block.

The street spatial form of the study area is illustrated in Figure 5. From an urban morphology perspective, the streets are generally narrow, with an H/W commonly exceeding 0.67, and most streets follow a north-south orientation. The neighborhood exhibits functional richness, with buildings loosely arranged along the streets. The area also benefits from ample public green spaces, a favorable ecological environment, high sky visibility, and limited shading from both buildings and vegetation.

From a human-centered perspective, pedestrian visual access to greenery in the study area is generally low. At the same time, sky visibility remains high, consistent with the urban morphology findings. The overall enclosure of the blocks is weak, with significant local spatial variations. The level of motorization on the streets is high, while walkability is low, resulting in poor street spatial quality and inadequate basic service facilities.

3.2. Pearson Correlation Analysis

Pearson correlation analysis of fine street spatial pattern indicators and CTC was conducted using SPSSAU, and the results are shown in

Table 3. Pearson correlation coefficient between fine street spatial pattern indicators and CTC.

Parameters			r	p
Urban morphology perspective street spatial perception indicators	H/W		−0.164	<0.001
	Street orientation		−0.086	<0.001
	Functional diversity indices		−0.274	<0.001
	Build-to-line rate		−0.160	<0.001
	Distance from GBI		−0.319	<0.001
	VF	SVF	0.222	<0.001
		TVF	−0.108	<0.001
		BVF	−0.130	<0.001
Human-centered perspective street spatial perception indicators	GVI		−0.025	<0.001
	Sky visibility degree		0.245	<0.001
	Interface enclosure degree		−0.180	<0.001
	Carriageway visibility degree		0.071	<0.001
	Pavement visibility degree		0.020	>0.05
	Street space quality		0.097	<0.001

. Of the 14 indicators, 13 showed significant correlations with the CTC. Among these, 9 indicators showed negative correlations (including H/W, street orientation, functional diversity indices, build-to-line rate, distance from GBI, TVF, BVF, GVI, and interface enclosure degree). Meanwhile, 4 indicators demonstrated positive correlations (including SVF, sky visibility degree, carriageway visibility degree, and street space quality). Notably, the correlation between indicators from an urban morphology perspective and CTC was stronger than that of indicators from the human perspective.

Pearson correlation analysis was also conducted on the street spatial form indicators for initial screening. Previous research suggests that a correlation coefficient exceeding 0.7 indicates significant multicollinearity between two variables [66]. The results of this analysis are shown in Figure 6. Based on the correlation between each indicator and thermal comfort, as well as the correlation coefficients among the indicators, 10 key indicators were identified for further evaluation of their impact on thermal comfort in the study area. These indicators include H/W, street orientation, functional diversity indices, build-to-line rate, distance from GBI, sky visibility degree, TVF, interface enclosure degree, carriageway visibility degree, and street space quality.

3.3. Analysis of Overall Characteristics

To further mitigate the effects of multicollinearity and verify the spatial correlation between variables, the variance inflation factor (VIF) and Koenker (BP) statistic were calculated for the 10 selected variables. Variables with a VIF greater than 10 and insignificant BP values were excluded [67]. As a result, six indicators were retained: H/W, street orientation, functional diversity indices, distance from GBI, sky visibility degree, and TVF. After applying XGBoost, the SHAP values of each street spatial indicator in the study area were visualized to assess their contribution to CTC, as shown in Figure 7.

Overall, the relative importance of these indicators on thermal comfort, in descending order, is as follows: distance from GBI, TVF, H/W, street orientation, functional diversity indices, and sky visibility degree. Low values of distance from GBI, TVF, H/W, and street orientation were primarily concentrated in areas where the SHAP values were greater than 0, indicating that areas with higher values for these indicators tend to have lower CTC levels, meaning the environment is more comfortable. Except for TVF, the majority of the spatial points for the other five indicators were located in regions where SHAP values were less than 0, demonstrating a negative correlation with CTC levels. High values of the TVF (red points) were primarily concentrated on the far left, suggesting that both low and high values of TVF enhance thermal comfort, with the most significant improvements occurring when it is high.

3.4. Local Characteristic Analysis

Based on the CTC levels within the study area, three street spaces—A, B, and C—were selected for specific analyses. Street spaces A and B demonstrated high thermal comfort levels, indicating greater thermal risk, while street space C exhibited a lower thermal comfort level, reflecting reduced thermal risk. The localized SHAP characteristic analysis results are shown in Figure 8. The spatial form indicators positively influenced CTC in street spaces A and B. For street space A, the four most significant contributors to the thermal comfort level were street orientation, distance from GBI, TVF, and sky visibility degree. In street space B, the primary factors are distance from GBI, H/W, functional diversity indices, and tree visibility degree.

In contrast, the spatial form indicators in the more thermally comfortable street space C negatively impacted its CTC. Street space C’s top four contributing factors were TVF, distance from GBI, sky visibility degree, and functional diversity indices. This finding suggests that dense street trees are crucial in mitigating urban thermal stress and enhancing thermal comfort.

4. Discussion

4.1. Street Spatial Form Indicators and Thermal Comfort

The spatial distribution of subjective and objective thermal comfort in the study area is generally consistent, with some localized differences. In the northwest section of the study area, the objective thermal comfort levels are relatively low, while the number of heat-related check-ins on Weibo is high. This area is a commercial center, and although T_mrt remains within a comfortable range, the large extent of impervious surfaces and insufficient shading lead to a stark contrast between indoor and outdoor thermal environments. The high density of foot traffic exacerbates the difference in thermal sensations between indoor and outdoor spaces, significantly increasing negative responses to hot weather on social media. The overall thermal comfort levels exhibit spatial heterogeneity, with poorer thermal comfort concentrated in areas with high pedestrian flow and limited tree cover, such as commercial districts and public squares. Human perception of comfort involves numerous variables. From a climatic perspective, these urban spaces’ elevated solar radiation and temperature contribute to discomfort. Psychologically, the stark temperature differences between indoor and outdoor spaces, combined with overcrowding, can increase negative emotions, making individuals more prone to feelings of thermal discomfort [68]. Furthermore, factors such as age, physical condition, and behavior influence people’s perception of thermal comfort, which warrants further exploration in future studies.

Based on the correlation analysis, street spatial perception indicators from an urban morphology perspective demonstrate a stronger relationship with thermal comfort than human-centered perspective indicators. Although the human-centered perspective focuses on individual experiences, existing research indicates that overall urban design impacts more on thermal comfort [69]. Factors directly related to solar radiation, such as the distance from GBI, SVF, and sky visibility degree, exhibit the strongest correlation with thermal comfort, consistent with previous research findings [70]. However, the correlation between the pedestrian view of greenery and thermal comfort is relatively weak and lower than that between TVF and thermal comfort. This may be because tree canopies extend beyond the pedestrian’s field of view, with vegetation primarily enhancing thermal comfort through shading [71]. Thus, street-level thermal comfort in the study area is largely influenced by tree shading.

The results from the fine street spatial form thermal comfort prediction model reveal that landscape-related indicators, particularly distance from GBI and TVF, have the greatest influence on the comprehensive thermal comfort level in the study area. Shaded street spaces are critical for mitigating urban heat, significantly improving thermal comfort during the summer [72]. Previous studies have shown that street trees can reduce solar radiation intensity by up to 91.39% [73]. Both high and low values of TVF contribute to improved thermal comfort, with higher values offering more pronounced benefits. This may be because lower TVF values provide less canopy coverage, which provides less wind obstruction, allowing higher wind speeds and enhancing comfort. Meanwhile, higher TVF not only provides greater shading but also induces a psychological cooling effect [71,74]. Other important indicators include H/W, street orientation, functional diversity indices, and sky visibility degree, all of which are negatively correlated with thermal comfort. Street H/W and orientation primarily affect thermal comfort by influencing solar radiation and wind flow, with the H/W having a greater impact, consistent with previous studies [75]. Research by Ibrahim et al. [76] found that the effect of the H/W on thermal comfort varies for streets with different orientations, while optimal orientation also depends on local climate and wind conditions [77]. Further research is needed to determine the optimal H/W and orientation. Additionally, functional diversity is negatively correlated with the CTC, indicating that higher functional diversity reduces thermal risk. A greater mix of POI enhances service availability, improving the comfort of the urban environment [78].

The SHAP analysis of typical street spaces and thermal comfort revealed that the factors influencing thermal comfort differ across various street spaces. In high-risk streets, the key factors were distance from GBI, H/W, and street orientation, while in low-risk streets, landscape-related factors were the main contributors. Moreover, the effect of each indicator on thermal comfort varied depending on the street. For instance, distance from GBI positively affects thermal comfort in streets A and B but negatively on street C. This highlights that the spatial features contributing to thermal environmental issues vary between street spaces, suggesting that future optimization of urban heat environments should be tailored to the specific spatial characteristics of each neighborhood.

4.2. Optimization Strategies

Extreme heat waves are closely related to urban planning, as both the existing built environment of urban neighborhoods and future planning decisions by city managers significantly affect thermal comfort levels. This study proposes optimization strategies for the overall study area and specific streets. The optimization strategies for the overall study area are as follows:

• Spatial Layout: Implement thermal environment control measures in spatial layout planning by developing a spatial structure plan based on heat risk.
• Green Space Systems: Add cluster green spaces and street landscape green belts, improve the permeability of green areas, and enrich the vertical landscape structures within green spaces to form a node-axis green space planning system.
• Ventilation Corridors: Develop a hierarchical spatial planning structure for ventilation corridors to promote optimal airflow within street spaces and ensure effective ventilation.
• Disaster Shelters: Establish a hierarchical disaster shelter system by selecting high-temperature shelter locations based on the “city-community-street” structure, integrating both indoor and outdoor public spaces.

Street space A is identified as a priority area for thermal environment optimization. It is a high-density commercial district with a high CTC level and significant thermal risk. Based on the local prediction model and field surveys, optimization efforts should focus on four key areas: public space, landscape and greening, service facilities, and surface material. The proposed optimization plan is illustrated in Figure 9.

4.3. Limitations

This study has several limitations that warrant further exploration. First, the data sources used, particularly meteorological and vegetation elevation data, have limitations in terms of accuracy and comprehensiveness. Second, there is room for improvement in the comprehensive thermal comfort evaluation methods. Regarding subjective thermal comfort, the use of Weibo data may present a representational bias, as it cannot fully capture the subjective thermal comfort of all residents in the study area. Additionally, the current analysis lacks consideration of human behavioral characteristics and the social traits of street users. Future research could incorporate factors such as age, behavioral patterns, and specific population groups within the study area, enabling more targeted optimization studies based on different thermal environments and population demographics. Furthermore, the method for determining the relative weights of subjective and objective thermal comfort also requires further refinement. Lastly, the proposed optimization strategies lack empirical validation of their effectiveness. Future work could employ simulation tools, such as Grasshopper, to model and predict the thermal conditions of optimized street spaces, thereby improving the reliability of the proposed strategies.

5. Conclusions

This study focuses on the central urban streets of a coastal city, utilizing multi-source urban data to construct a refined street spatial form indicator system from both urban morphology and human-centered perspectives. XGBoost and SHAP were developed to quantify the relationship between street spatial form and thermal comfort. The main conclusions are as follows:

• The spatial distribution of subjective and objective thermal comfort is generally consistent, with some local variations. The overall thermal comfort levels exhibit spatial heterogeneity.
• In the univariate analysis, H/W, street orientation, functional diversity indices, build-to-line rate, distance from GBI, TVF, BVF, GVI, and interface enclosure degree were negatively correlated with thermal comfort. In contrast, SVF, sky visibility degree, carriageway visibility degree, and street space quality were positively correlated with thermal comfort. Overall, urban morphology indicators had a stronger correlation with thermal comfort than human-centered perspectives of street spatial perception indicators.
• Of the overall street characteristics, landscape-related indicators had the highest impact on thermal comfort levels in the study area, followed by street form. In localized analyses, the factors influencing thermal comfort varied between streets, and the impact of the same indicator on thermal comfort also differed across different neighborhoods.

Based on the findings from the street thermal comfort prediction model, this study proposes both global and targeted optimization strategies, offering a novel approach to climate-adaptive urban planning and enhancing the human living environment from both urban and human perspectives. Future research should refine data granularity, further explore street space indicators from human-centered perspectives, incorporate social factors into the street thermal comfort evaluation system, and validate the proposed optimization strategies to strengthen the theoretical foundation for improving urban environments.