How 2D and 3D Built Environment Impact Urban Vitality: Evidence from Overhead-Level to Eye-Level Urban Form Metrics

Abstract

The built environment is the key to creating vibrant urban spaces that contribute to the health and sustainability of cities. Studies have demonstrated that a reasonable built environment helps to stimulate urban vitality. Nevertheless, there are limitations to the understanding that three-dimensional (3D) built environment indicators from the ‘human perspective’ can substantially affect urban vitality. This study provides an empirical analysis of Xi’an, a city with both traditional historical blocks and a modern city landscape. By applying the ordinary least square model and the geographically weighted regression model, this study explores the impacts of the two-dimensional (2D) and 3D built environments on urban vitality. Results show: (1) the urban vitality exhibits significant spatial and temporal difference characteristics; (2) the 3D built environment exerts a greater influence on urban vitality than 2D; (3) taking weekdays for instance, the indicators of green space and road space (e.g., normalized difference vegetation index (−0.092), green view index (−0.104), road density (−0.021), and enclosure (−0.089)) are negatively correlated with urban vitality, while the indicators of building space and mixed function (e.g., building density, floor area ratio, points of interest (POI) mixing degree, and 3D mixing degree) present a positive effect. To improve urban vitality, the study provides suggestions from the perspective of 3D and human perception, which will contribute to the meticulous practice of urban design.

1. Introduction

1.1. Background

The economic development and urbanisation process in China have shifted to a high-quality development stage. In this context, urban vitality emerges as a crucial point in evaluating the quality of urban development [1]. High urban vitality is indicated by a robust innovation ecosystem, well-developed infrastructure, and abundant cultural resources. Increases in vibrant urban spaces help attract capital and talent [2] and create healthy and sustainable urban environment [3]. Urban vitality is used as a basic goal of urban planning and development [4]. Therefore, exploring the spatial distribution and understanding the impact mechanisms of urban vitality are important in the process of urban planning and policy making.

Urban vitality can mainly be measured in two ways. The more traditional way is to use questionnaires and field observations. However, these field research methods are labour-intensive and time-consuming, which may not suitable for assessing urban vitality at a large scale [5]. Recently, big data have been adopted to measure urban vitality. Big data can collect precise sequences of human mobility data, thus describing the spatial distribution of human activities. Furthermore, many mapping platforms supply free urban spatial data, which provide the data base for quantifying the built environment [6].

More attention is paid to the relationship between the built environment and urban vitality, but the conclusions are not completely consistent. For example, Lu et al. found that floor area ratio (FAR) is positively related to vibrancy, whereas building density is negatively correlated in the study of Beijing. However, in Chengdu, the two indicators have different effects [6]. Inconsistencies in research conclusions may be explained by the differences in the measurement of built environment and urban vitality or the differences in the development scale and stage of each city. In the existing literature, the index system of the built environment is diverse, such as density [7], diversity [3], and design [8]. It is worth noting that most of these indicators are at the overhead level (the view overlooking from a high altitude, e.g., normalized difference vegetation index) and few are based on the eye-level (the view from humans at eye level, e.g., green view index). In sum, these previous works only reflect the built environment in the horizontal dimension, and less attention is paid to the vertical dimension [5,9].

Existing studies mainly applied the Ordinary Least Square (OLS) model. However, this model can only reflect the global relationship, overlooking the spatial autocorrelation and spatial non-stationarity, potentially leading to biased estimates. Therefore, the Geographically Weighted Regression (GWR) model was employed to eliminate these limitations. Moreover, since the distribution of population varies greatly in different time periods, we considered the temporally heterogeneous impacts of the built environment on urban vitality.

To fill the above research gaps, the GWR model was applied to explore the impact of the built environment on urban vitality in Xi’an and its spatial heterogeneity for addressing the abovementioned gaps. First, Baidu heat map data were collected to portray the spatial–temporal patterns of urban vitality at a grid of 250 m. Second, 2D (two-dimensional) and 3D (three-dimensional) indicators were selected based on the overhead-level and the eye-level, respectively. Particularly, 3D indicators were introduced, including the building indicators (e.g., average building height) and the human-scale streetscape indicators (e.g., green view index) to extend the built environment assessment system. The streetscape indicators characterize the human-scale visual environment, better reflecting how people actually perceive and experience street space [10,11]. Last, both the OLS model and the GWR model were applied to investigate the relationship between urban vitality and the built environment. GWR can also reflect the heterogeneity impact of the built environment on urban vitality in space, providing a more refined reference for improving urban vitality.

The rest of the paper is organized as follows. Section 1 presents a comprehensive literature review of urban vitality. Section 2 introduces the study area, data sources, and methods. Section 3 describes the analysis results of the OLS and GWR models. Section 4 discusses the different effects of the 2D and 3D indicators. Section 5 summarizes the research and presents the limitations.

1.2. Literature Review

Numerous studies have focused on how the built environment influences urban vitality, aiming to provide scientific suggestions for improving urban vitality. However, existing studies of the built environment were mostly carried out from three or five dimensions. Among them, indicators based on 3D and eye-level are less considered. Most of the previous studies adopted linear regression models, which cannot reveal the non-stationarity of space. Accordingly, this section will review the literature through these two aspects: concepts and measurements of urban vitality and the associations between urban vitality and built environment.

1.2.1. Concepts and Measurements of Urban Vitality

Jacobs first proposed the concept of urban vitality. She believed that urban vitality promotes human activities and social interactions, thus fundamentally enhancing urban social life [12]. Lynch believed that vitality was both the condition and ability of a city to produce sustainable development [13]. As the primary indicator in urban form evaluation frameworks, vitality reflects the residents’ activities in the city. As urban vitality is considered by more and more scholars, its research content has also expanded to multiple dimensions such as economy, society, and culture, and it has become a comprehensive indicator to evaluate a city. A deep study of urban vitality distribution patterns and comprehensive analyses of built environment determinants will enable urban planners to optimize facility allocation and enhance environmental quality.

Urban vitality is fundamentally manifested through diverse human activities, and people are the central component [3]. Therefore, the degree of human aggregation serves as a key metric for assessing urban vitality [12]. At present, the evaluation of urban vitality is mainly divided into two ways. Traditional research uses questionnaires and field observations. These methods have disadvantages such as difficult data collection and limited research scope, which are not conducive to further research. Fortunately, big data, such as residents’ daily behavior activities, have become easy to obtain and provide new data sources and research directions for the study of urban vitality. Tencent location data [3], mobile phone data [7,14,15], social-media check-in data [6,16], and Baidu heat maps [17] have been widely adopted for measuring urban vitality. Compared with the traditional field statistics of residents’ activities, these data have the advantages of easy access and existing in large amounts [18] and can provide higher spatio-temporal resolution [19], which can effectively characterize the spatial and temporal distribution [5].

1.2.2. The Associations Between Urban Vitality and Built Environment

The issue of how the built environment affects urban vitality has been widely discussed. Initial studies adopted three dimensions, namely 3D (density, diversity, and design) [20]. Later, distance to transit and destination accessibility were introduced into the study, forming the 5D evaluation system. It has been proved in plentiful studies that it has an important effect on urban vitality [21]. The specific built environment includes building density, distance from metro stations, road intersection density, etc. For example, Lu et al. showed that points of interest (POI) diversity and public transport accessibility indicators were strongly positively associated with vitality. Moreover, FAR and building density indicators have completely different effects on urban vitality in Beijing and Chengdu [6].

The vertical built environment has received more attention in recent studies. Researchers found that urban vitality is also related to some characteristics of architecture. Ye et al. found that the block typology in the urban form has a more significant impact than the block density [9]. Lin et al. selected building-related 3D built environment indicators and found that building coverage, high-rise building density, and plot ratio are important drivers of urban vitality [5]. Xiao and Liu argued that building form notably influences urban vitality, with their case study demonstrating that the ground space index affects urban vitality more significantly than the average height of buildings [22].

It is worth noting that urban vitality is reflected in people’s behavioral patterns, and their behavior is inherently shaped by environmental perception [18]. Accordingly, it is necessary to incorporate human perception into studies. However, only a small number of scholars have studied from the perspective of humanization. C. Wu et al. used street view images to quantify human visual perception of the built environment [1]. Based on the selected indicators (e.g., green visibility), they found that there was a correlation between human perception factors and urban vitality. Yang et al. innovatively introduced the environmental elements of the block from the perspective of people, pointing out that pleasant green areas and appropriate sky openness can promote human activities and thus improve the vitality in a study of Wuhan [18].

The advancement of computer technology has led to the emergence of numerous novel methods and data, which provides the possibility for quantitative analysis of urban issues. For instance, data sources such as Baidu street view and Google street view provide abundant spatial information on urban streets. Compared to traditional methods (e.g., field research methods and remote sensing data), street view images demonstrate significant advantages in analyzing built environment characteristics. Street view images are characterized by their high resolution, enabling them to capture the facade morphology of streets and compensate for the inability of remote sensing data to obtain the vertical surface information of buildings. Additionally, street view images can simulate the pedestrian visual experience, reflecting their actual perception of the built environment and thus providing an indispensable foundation for research on subjective dimensions such as street space perception. Due to these attributes, street view images have become a critical resource for analyzing urban environments, which is widely applied in urban spatial research. Street view images have been utilized in various research areas, including urban greenery assessment [23], building function identification [24], and urban environment quality evaluation [25]. Zhang et al. employed segmentation techniques to extract multiple object categories from street view images, thereby reflecting human visual perception of building space [26]. Gao and Fang employed street view images to identify urban environmental features and evaluate the relationship between the urban environment and cycling amount [27]. According to the existing studies, street view images are mostly used for the measurement of micro-built environments (e.g., community), while being less utilized at the meso scale (e.g., city). Moreover, few studies compare the traditional 2D built environment with the 3D environment represented by street view images, which is a worthy direction for further exploration.

As for the mechanism of urban vitality, scholars have explored a variety of model methods. With the deepening of the study of urban vitality, the model methods used for simulation are constantly improving. Most of the previous studies adopted linear regression models, the simplest of which is OLS. However, this method ignores the interaction between variables and cannot involve spatial correlation. It is usually used as a benchmark model for comparison. Compared with OLS, the spatial lag model (SLM) and the spatial error model (SEM) can consider spatial correlation [28,29]. Other than the above global models, local models can better reveal the non-stationarity of space. The typical representative of this type of model is the GWR model, which is widely used in urban development and other research.

Building on the conclusions from the literature review, this study makes three key contributions: (1) 3D indicators are introduced to enrich traditional built environment framework; (2) indicators at the eye-level are taken into account; and (3) both spatial heterogeneity and temporal dynamics in the impact of the built environment on urban vitality are considered.

2. Materials and Methods

2.1. Study Area

Xi’an, the provincial capital of Shaanxi, is located in Northwest China, at 107°40′−109°49′ E and 33°42′~34°45′ N. It is a famous historical and cultural city in China, and an important scientific research and industrial base. By the end of 2022, it had a total area of 10,108 km² and nearly 13 million people. Since the urban residents are mainly concentrated around the city center, we select the area within the 3rd Ring Road in Xi’an as the study area (Figure 1).

2.2. Data Sources

2.2.1. Baidu Heat Map

Launched in 2011 by Baidu company, the Baidu heat map represents a big data visualization platform that aggregates and displays spatial distribution patterns of Baidu App mobile phone users through location-based service. As smartphone users access Baidu products, the location information generated by them is recorded. According to the location clustering, the direction and location of urban population flow are calculated, and the spatial aggregation degree of population is presented. The data have a strong timeliness, updating every 15 min. In the study, a higher heat value indicates a greater population density, which means the region is more dynamic. Baidu heat map has been widely used in park use, land use, urban spatial structure, and other fields [30,31].

Baidu heat map data for a consecutive week from April 10 to 16, 2023 were selected in this study, and the data for these 7 days did not include holidays and festivals. The interval of data collection is 1 h. The Urban Residential Area Planning and Design Standards specifies an optimal road spacing of 150–250 m for a living block. Therefore, we adopted a 250 m × 250 m grid system for spatial analysis.

2.2.2. Variables

Baidu heat map data are the dependent variable, and the built environment variables are the independent variables. Fourteen indicators were selected to measure the built environment, which were based on the 5D framework (diversity, design, density, distance to transit, and destination accessibility). Moreover, to distinguish from the traditional research ideas, this study divided the indicators into two dimensions: 2D and 3D. Compared with the 2D indicators, 3D indicators can reflect the human scale and perspective, which provide a more solid reference for urban local planning and renewal. Moreover, street view images are adopted in the study, which can measure the 3D built environment more comprehensively. Figure 2 shows the process of selection and classification. Due to it is difficult in quantifying “distance to transit” and “destination accessibility” at the three-dimensional level, 3D indicators were not selected for the above two principles. Therefore, in this research, the built environment is conceptualized as the integrated system comprising both the man-made and natural environment, which is the general term for various environments in the city.

Table 1. Descriptive statistics of independent variables.

Variables (Unit)	Abbreviations	Std	Mean	Min	Max
2D variables
POI mixing degree	POI_MD	0.761	1.553	0	2.622
Normalized difference vegetation index	NDVI	0.063	0.346	0.040	0.638
Building density	BD	0.142	0.183	0	0.853
Road density (km/km2)	RD	5.408	7.345	0	44.710
Road intersection density (1/62,500 m2)	RID	4.596	2.759	0	62
Metro station density (1/62,500 m2)	MSD	0.171	0.026	0	2
Bus stop density (1/62,500 m2)	BSD	5.110	2.472	0	43
Metro accessibility (m)	SA	1058.118	1066.778	7.498	6530.163
3D variables
Average building height (m)	ABH	11.939	12.497	0	120
Floor area ratio	FAR	1.335	1.070	0	8.748
Green view index	GVI	0.132	0.169	0	0.744
Enclosure	EN	0.153	0.195	0	0.867
Sky view index	SVI	0.102	0.126	0	0.470
3D mixing degree	3D_MD	0.639	1.338	0	2.094

Note: Std = Standard deviation; Min = Minimum; Max = Maximum.

shows the specific index composition and statistical description.

For all the variables in Table 1, Identity and Summarize tools are introduced in ArcGIS 10.6 to calculate the indicators within each 250 m × 250 m grid. Semantic segmentation technology was used to calculate the street view data captured in the Baidu map, the Identity tool helped to obtain the street view location points, and then the Summarize tool helped to calculate GVI, SVI, EN, and 3D_MD. Figure 3 shows the 3D spatial distribution of urban forms in Xi’an. Furthermore, the detailed calculation process of some indicators is as follows:

(1)

POI mixing degree

The POI data used in the study include 23 types of consumer services, such as shopping services, science and education cultural services, public facilities, etc. When the proportion of each type of POI is equal, the index reaches the maximum value.

$$POImixingdegree=-\sum _{i=1}^n\left(p_i\times \ln p_i\right)$$

(1)

where $p_i$ represents the proportion of the $i$th POI category and $n$ represents the total number of POI categories.

(2)

Street view images

Street view images can record the street from a perspective that is similar to human vision, thus reflecting the 3D characteristics of streets [32,33]. Based on the road network, 33,217 sampling points at 100 m intervals were generated, and 4 images for each sampling points were collected. Overall, 132,868 images were obtained. Each image was divided into four images on the basis of 0–90°, 90–180°, 180–270°, and 270–360°. Then, segmentation was employed to quantify proportions at the pixel level, followed by calculating the average of different features at each point [34] (Figure 4).

We selected the following 3D street view indicators. GVI quantifies the proportion of green pixels in the image, which indicates the amount of green landscape in the space. EN indicates the sum of the proportions of buildings, walls, fences, poles, and other similar pixels in the image. SVI refers to the proportion of sky pixels in the image, which is used to measure the sky openness of street space. 3$\mathrm{D}\mathrm{m}\mathrm{i}\mathrm{x}\mathrm{i}\mathrm{n}\mathrm{g}\mathrm{d}\mathrm{e}\mathrm{g}\mathrm{r}\mathrm{e}\mathrm{e}$ refers to the variety of 3D street view landscapes. The calculation formulas are as follows:

$$Greenviewindex={\displaystyle \frac{\sum _{i=1}^4{GreeneryPixels}_i}{\sum _{i=1}^4{TotalPixels}_i}}$$

(2)

$$Enclosure={\displaystyle \frac{\sum _{i=1}^4{BuildingPixels}_i+\sum _{i=1}^4{WallPixels}_i+\sum _{i=1}^4{FencePixels}_i+\sum _{i=1}^4{PolePixels}_i}{\sum _{i=1}^4{TotalPixels}_i}}$$

(3)

$$Skyviewindex={\displaystyle \frac{\sum _{i=1}^4{SkyPixels}_i}{\sum _{i=1}^4{TotalPixels}_i}}$$

(4)

$$3Dmixingdegree=-\sum _{i=1}^{\mathrm{n}}\left(p_i\times \ln p_i\right)$$

(5)

2.3. Methodology

OLS and GWR were employed to analyze the relationship between the built environment and urban vitality in this study. However, while OLS provides a global estimate of the association, it fails to account for spatial non-stationarity between variables. To address this limitation and better capture spatial heterogeneity in the relationship, this study adopts GWR as the research model, which considers spatial autocorrelation. GWR has become increasingly prominent in spatial analysis, particularly in geography, urban planning, and other fields to reflect spatial non-stationary [35,36,37,38]. Figure 5 shows the research framework.

The OLS model expression is as follows:

$$Y=C+{\beta }_k{x}_k+\epsilon$$

(6)

where $Y$ represents the value of vitality; $x_k$ represents the value for built environment index $k$; $C$ represents the constant; ${\beta }_k$ represents the regression coefficient for built environment index $k$; and $\epsilon$ is the residual value.

The GWR model expression is as follows:

$$y_i={\beta }_0\left(u_i,v_i\right)+\sum _k{\beta }_k\left(u_i,v_i\right)x_{ik}+{\epsilon }_i$$

(7)

where $\mathrm{y}$ represents the value of vitality; $\left(u_i,v_i\right)$ is the coordinate; ${\beta }_0$ is the regression coefficient of the $i$-point; ${\beta }_k\left(u_i,v_i\right)$ is the local coefficient for built environment index $k$; and ${\epsilon }_i$ is the error term.

3. Results

This section is structured as follows. First, two temporal classifications are defined: weekdays/weekends and morning/afternoon/night. Then analyze the spatial-temporal variations in urban vitality. Second, before constructing the GWR model, the spatial autocorrelation of urban vitality needs to be tested. Third, compare the results between OLS and GWR. Describe the spatio-temporal heterogeneity in built environment effects on urban vitality.

3.1. Spatial–Temporal Variations in Urban Vitality

Urban vitality exhibits distinct spatial clustering characteristics with polycentric distribution patterns. The high-value areas are distributed around the Gulou in the middle and the Xiaozhai business district in the south of Xi’an. The low-value areas are distributed around the Han Chang’an City Site in the northwest and the Chanba ecological zone in the northeast. In addition, the southern areas exhibit significantly higher urban vitality compared to those in the north. The reason for this is that the development and construction steps in Xi’an are ‘from south to north’. As a result, a variety of businesses, colleges, and cultural facilities are located in the south, which attracts a greater number of people, leading to a high population density.

The distribution of urban vitality presents different characteristics in different time periods. Areas with high values in urban vitality in the morning (6:00–12:00) and afternoon (13:00–18:00) are mainly Kangfu Road and Nanshaomen. The high-value areas of urban vitality in the evening (19:00–23:00) are mainly Grand Tang Mall and Ganjiazhai. The Gulou and Xiaozhai areas maintain high vitality all day. This situation is mainly due to the strong attractiveness and influence of the abovementioned districts, generated by their favourable locations and convenient transport links as the well-established business districts. In terms of time periods throughout the day, afternoon shows the highest urban vitality, closely followed by evening.

Differences in the daily change patterns of urban vitality are observed on weekdays and weekends (Figure 6). From the perspective of urban vitality level, except for the morning and evening commuting periods (6:00–10:00 and 18:00), weekdays are higher than weekends. In the comparison of the rest of daily time, weekends exhibit higher vitality levels than weekdays. On weekdays, the relatively high level of urban vitality lasts longer (8:00–22:00). On weekends, it lasts from 10:00 to 22:00, and residents’ daily schedule on weekends shows the characteristics of ‘keeping late hours’. Figure 7 and Figure 8 show the spatial distribution of vitality; high-value areas are concentrated in business and employment centres on weekdays. On weekends, these areas are only concentrated in business centres. The urban vitality on weekends is more evenly distributed in time and space than that on weekdays. This situation is because residents’ activities on weekends are less constrained by time and space, which means their travel arrangements are more flexible and diverse.

3.2. Spatial Autocorrelation Analysis of Urban Vitality

It is necessary to use Global Moran’s I to analyze the spatial autocorrelation of urban vitality before constructing the GWR model.

Table 2. Spatial autocorrelation analysis of urban vitality on weekdays and weekends.

Period	Weekday			Weekend
	Moran’s I	z-Score	p-Value	Moran’s I	z-Score	p-Value
Morning	0.823	124.801	0.001	0.817	124.149	0.001
Afternoon	0.788	119.376	0.001	0.690	105.631	0.001
night	0.822	127.195	0.001	0.756	117.117	0.001

Note: Number of permutations = 999.

shows that the Moran’s I values of each time period are 0.823 (morning), 0.788 (afternoon), and 0.822 (evening) on weekdays. On weekends, the Moran’s I values are 0.817 (morning), 0.690 (afternoon), and 0.756 (evening), all of which pass the significance level test (p < 0.001). Therefore, the analysis reveals a significant spatial correlation in each time period, and the agglomeration characteristics of weekdays are more obvious than those of weekends. Accordingly, the GWR model can be used for following analysis.

3.3. Results of the OLS and GWR Models

3.3.1. OLS Model Results

Table 3. Results of OLS coefficients on weekdays and weekends.

Variable	Weekday			Weekend
	Morning	Afternoon	Night	MORNING	afternoon	Night
POI_MD	0.210 ***	0.181 ***	0.245 ***	0.221 ***	0.156 ***	0.210 ***
	(0.013)	(0.013)	(0.013)	(0.013)	(0.014)	(0.014)
NDVI	−0.092 ***	−0.083 ***	−0.068 ***	−0.097 ***	−0.078 ***	−0.065 ***
	(0.010)	(0.011)	(0.010)	(0.010)	(0.011)	(0.011)
BD	−0.062 ***	−0.061 ***	−0.042 **	−0.058 ***	−0.066 ***	−0.033 *
	(0.014)	(0.015)	(0.015)	(0.014)	(0.016)	(0.016)
RD	0.000	0.021	0.031 *	−0.031 *	−0.006	0.031 *
	(0.013)	(0.014)	(0.014)	(0.013)	(0.015)	(0.015)
RID	0.104 ***	0.104 ***	0.076 ***	0.122 ***	0.127 ***	0.090 ***
	(0.013)	(0.013)	(0.013)	(0.013)	(0.014)	(0.014)
MSD	0.102 ***	0.115 ***	0.074 ***	0.093 ***	0.138 ***	0.098 ***
	(0.010)	(0.010)	(0.010)	(0.010)	(0.011)	(0.011)
BSD	0.123 ***	0.125 ***	0.090 ***	0.117 ***	0.122 ***	0.089 ***
	(0.010)	(0.011)	(0.011)	(0.010)	(0.011)	(0.011)
SA	−0.189 ***	−0.158 ***	−0.174 ***	−0.201 ***	−0.162 ***	−0.153 ***
	(0.011)	(0.012)	(0.011)	(0.011)	(0.012)	(0.012)
ABH	0.001	−0.013	−0.026	−0.036 *	−0.061 ***	−0.051 **
	(0.016)	(0.017)	(0.017)	(0.016)	(0.018)	(0.018)
FAR	0.311 ***	0.336 ***	0.272 ***	0.314 ***	0.335 ***	0.276 ***
	(0.019)	(0.020)	(0.020)	(0.019)	(0.021)	(0.021)
GVI	−0.105 ***	−0.112 ***	−0.128 ***	−0.115 ***	−0.116 ***	−0.121 ***
	(0.021)	(0.022)	(0.022)	(0.021)	(0.023)	(0.023)
EN	−0.114 ***	−0.144 ***	−0.071 **	−0.049 *	−0.087 **	−0.063 *
	(0.025)	(0.026)	(0.026)	(0.024)	(0.027)	(0.027)
SVI	−0.226 ***	−0.222 ***	−0.268 ***	−0.229 ***	−0.178 ***	−0.221 ***
	(0.022)	(0.023)	(0.022)	(0.021)	(0.024)	(0.024)
3D_MD	0.267 ***	0.284 ***	0.261 ***	0.248 ***	0.237 ***	0.228 ***
	(0.039)	(0.041)	(0.040)	(0.039)	(0.043)	(0.043)
R2	0.496	0.454	0.467	0.510	0.388	0.389

Note: * p < 0.05; ** p < 0.01; and *** p < 0.001.

shows the correlation coefficients. ABH demonstrates no statistically significant association with urban vitality on weekdays, and it exhibits a negative correlation on weekends. FAR shows a significant positive impact both on weekdays and weekends, and its impact is weakened at night. SA is negatively correlated with the vitality of the city, with the greatest impact in the morning. Urban vitality is significantly influenced by RD in the evening and morning on weekends, while it is insignificant in other periods. GVI exerts a negative impact on urban vitality, and its impact is amplified at night. EN is negatively correlated with urban vitality, with the strongest correlation in the afternoon. SVI shows temporally varied negative effects, with the strongest effect in the evening on weekdays and in the morning on weekends. BD is negatively correlated with urban vitality. In contrast, POI_MD emerges as a positive indicator with significant impact during the evening period on weekdays. Urban vitality is closely associated with MSD and BSD. RID is positively correlated with urban vitality, and the correlation is stronger on weekends. 3D_MD demonstrates a significant positive impact on urban vitality. NDVI generally correlates negatively with urban vitality.

3.3.2. GWR Model Results

We standardise the variables to control their range in [0, 1] to eliminate the effects of inconsistent measurement units and scales [4]. On weekdays, the values of R² in the morning, afternoon, and night are 0.686, 0.652, and 0.642, respectively. On weekends, the values of R² in the three periods are 0.686, 0.572, and 0.564, respectively (

Table 4. GWR parameters on weekdays and weekends.

Diagnostic Index	Weekday				Weekend
	Morning	Afternoon	Night	All Day	Morning	Afternoon	Night	All Day
Residual Squares	1799.985	1993.542	2048.099	1803.190	1795.147	2452.490	2495.854	425,039.370
AICc	10,082.584	10,667.611	10,822.261	10,092.771	10,067.166	11,854.398	11,954.794	41,382.685
R2	0.686	0.652	0.642	0.685	0.686	0.572	0.564	0.626
Adjusted R2	0.669	0.633	0.623	0.668	0.670	0.549	0.541	0.606
Bandwidth	955.844	955.844	955.844	955.844	955.844	955.844	955.844	955.844

). The abovementioned results show that GWR exhibits a superior explanatory power and model fitting effect compared to OLS.

In terms of all time periods, indicators including POI_MD, MSD, BSD, RID, ABH, FAR, and 3D_MD are positively correlated. Meanwhile, NDVI, BD, RD, SA, GVI, EN, and SVI are negatively correlated with urban vitality. The five most significant independent variables related to the dependent variable are SA, 3D_MD, SVI, FAR, and POI_MD. By comparing the values of the regression coefficients, we find that 3D indicators generally reveal stronger associations with urban vitality compared to 2D indicators, both in terms of improving and reducing urban vitality. Specifically, regarding ecological space, the regression coefficient of GVI is higher than that of NDVI; regarding road space, EN is higher than that of RD; regarding building space, FAR and ABH is higher than that of BD (

Table 5. GWR results on weekdays and weekends.

Variable	Weekday								Weekend
	Morning		Afternoon		Night		All Day		Morning		Afternoon		Night		All Day
	Mean	STD	Mean	STD	Mean	STD	Mean	STD	Mean	STD	Mean	STD	Mean	STD	Mean	STD
POI_MD	0.151	0.079	0.119	0.098	0.187	0.082	0.153	0.085	0.167	0.078	0.097	0.112	0.162	0.083	1.999	1.271
NDVI	−0.095	0.116	−0.089	0.113	−0.084	0.092	−0.092	0.108	−0.100	0.097	−0.084	0.089	−0.080	0.073	−1.268	1.224
BD	−0.004	0.100	0.005	0.103	0.024	0.142	0.008	0.112	0.009	0.120	0.012	0.123	0.040	0.148	0.287	1.822
RD	−0.034	0.095	−0.019	0.100	−0.006	0.103	−0.021	0.098	−0.054	0.081	−0.033	0.089	−0.003	0.114	−0.435	1.250
RID	0.103	0.055	0.101	0.060	0.065	0.070	0.074	0.056	0.119	0.066	0.117	0.079	0.076	0.082	1.236	0.797
MSD	0.074	0.058	0.081	0.061	0.057	0.045	0.100	0.065	0.069	0.052	0.103	0.061	0.075	0.053	1.417	0.819
BSD	0.103	0.065	0.106	0.066	0.077	0.059	0.094	0.061	0.102	0.060	0.108	0.057	0.079	0.055	1.527	1.078
SA	−0.586	0.492	−0.600	0.547	−0.415	0.401	−0.560	0.495	−0.526	0.487	−0.588	0.616	−0.421	0.448	−7.578	7.609
ABH	0.067	0.099	0.052	0.094	0.039	0.081	0.055	0.091	0.049	0.084	0.023	0.079	0.019	0.084	0.427	1.143
FAR	0.146	0.132	0.172	0.128	0.134	0.137	0.157	0.132	0.137	0.142	0.169	0.124	0.140	0.124	2.203	1.808
GVI	−0.097	0.112	−0.104	0.121	−0.101	0.089	−0.104	0.110	−0.100	0.097	−0.106	0.111	−0.093	0.083	−1.462	1.358
EN	−0.095	0.097	−0.117	0.115	−0.033	0.074	−0.089	0.090	−0.034	0.078	−0.067	0.104	−0.023	0.086	−0.646	1.244
SVI	−0.183	0.148	−0.169	0.168	−0.193	0.140	−0.186	0.152	−0.177	0.145	−0.118	0.157	−0.140	0.154	−2.063	2.129
3D_MD	0.234	0.190	0.245	0.222	0.192	0.131	0.233	0.183	0.204	0.151	0.194	0.180	0.154	0.120	2.686	2.101

).

Through further visualization of coefficients for each independent variable in the GWR model, we conclude that all variables exert spatially heterogeneous effects on urban vitality at different locations (Figure 9, Figure A1 and Figure A2).

The built environment exerts significantly distinct temporal effects on urban vitality between weekdays and weekends. In general, for most built environment indicators, urban vitality on weekdays is more significant than that on weekends. This situation may be due to the fact that people on weekdays mainly travel for work and study, while on weekends they mainly travel for entertainment. Compared with entertainment activities, work and study are more limited by time. Therefore, the built environment shows less evident effects on the latter.

The spatial distribution of built environment impacts also exhibits significant heterogeneity. For instance, the impact of FAR shows a distinct spatial difference, demonstrating positive effects within the 2nd Ring Road but negative correlations beyond this boundary. This result may be due to the fact that the areas within the 2nd Ring Road mainly consist of historical blocks and old communities. In these areas, building heights are under control and FAR is limited within the appropriate range, which result in the enhancement in urban vitality and attractiveness. In addition, the effect of POI_MD on urban vitality shows significant spatial differences. For the university town in Yanta District and the administrative office cluster in Jingkai District, POI_MD exerts a significant positive impact on urban vitality, while it shows a negative impact in Beilin District, where the block function is highly composite.

4. Discussion

Baidu heat map data, which are a form of big data and are capable of effectively reflecting the spatial and temporal characteristics of human behaviour, are collected to quantify urban vitality in this study [39,40]. To better understand the connection between the built environment and urban vitality, we present 3D built environment indicators from the ‘human perspective’, which are based on the 2D built environment indicators. We discover that the 3D indicators promote urban vitality much more than their 2D counterparts, whether that impact is positive or negative, and a strong spatial heterogeneity exists. Thus, we address the following four topics: green space, road space, building space, and mixed function. The reason for the abovementioned selection is that humans perceive 3D space more intensely, such that high-density, compact land uses can attract more people and activities.

4.1. Green Space: NDVI and Green View Index

Previous studies show that GVI and NDVI are significantly negatively correlated with urban vitality [1,41]. One explanation is that residents are unable to use these green spaces during the day due to work commitments [42]. Furthermore, areas with high GVI are mostly open spaces such as parks and squares, where the population density is high in the day but low in the evening. Thus, urban vitality fluctuates greatly with time, and the overall vitality is low. Moreover, areas with high GVI are characterised by low FAR and low build density. The space in these areas is limited for human activities, which results in low urban vitality. Compared with NDVI, the negative effect of GVI is higher, indicating that green space in the vertical direction brings a stronger perception to people and thus influences their behaviour more. Notably, the impact of GVI and NDVI on urban vitality has spatial heterogeneity. First, from the perspective of 3D green spaces, areas with positive correlation between GVI and urban vitality, regardless of morning, afternoon, or night, are mainly concentrated in the 1st Ring and the north of the 2nd Ring. These areas are mostly historical blocks, old communities, and traditional business districts. Building height is limited to protect the historical fabric. The plot scale in these areas is more walking friendly. The street activity is mainly leisure walking, and people here are more sensitive to the perception of 3D green spaces. Li et al. found a similar view based on the study of Beijing’s historical blocks [43]. Second, from the perspective of 2D green spaces, the planar green spaces with large area and block distribution (e.g., the area with high NDVI value) do not have strong attraction to residents’ activities. The influence of NDVI in these areas (e.g., Daming Palace National Heritage Park) shows significant spatial heterogeneity. On the contrary, in the small area and fragmented area of NDVI in the north of the study area (e.g., industrial district), NDVI is positively correlated with urban vitality. Therefore, the impact of NDVI on urban vitality is significantly associated with the size and structure of regional green space and the function of the surrounding land.

4.2. Road Space: Road Density and Enclosure

Existing research on urban vitality mainly focuses on 2D spatial indicators, such as road density, on urban vitality [18,29]. Recently, the importance of the 3D space of roads has been emphasised, that is, the enclosure of streets. The areas with the most significant negative correlation between EN and urban vitality are distributed differently on weekdays and weekends, and they are mainly reflected in daytime. First, on weekdays, the most influential area is the southwest, and the inhibiting effect on vitality reaches the highest in the afternoon. Second, on weekends, EN significantly shows a negative impact on urban vitality in the northwest of the study area. This result is associated with the various activity arrangements of residents during different periods. On weekdays, residents are concentrated in industrial-intensive high-tech zones due to work-related activities. On weekends, residents will choose places with cultural, entertainment, and commercial functions for leisure and entertainment (e.g., Han Chang’an City Site and Zhonglou Business District). The promoting effect of EN on urban vitality may be related to the appropriate aspect ratio of the street. For example, within the 1st Ring Road area, high building density coexists with low building height, generating favorable street width-to-height ratios that enhance the appeal to citizens. The areas where RD is negatively correlated with urban vitality are mainly distributed along the north-south direction, but the inhibiting effect in the southern region decreases in the afternoon and even changes to a promoting effect in the evening. The reason may be the presence of scenic spots with distinctive night scenes such as Grand Tang Mall and Tang Paradise, and people are attracted here at night. In addition, we find that RD is negatively correlated with urban vitality within the 1st Ring Road and positively correlated between the 2nd and 3rd Ring Roads in the south. The reason is that the ancient city area is contained within the 1st Ring Road. Although the road density here is high, many T-junctions with low road network accessibility are present. Some narrow walkways exist, where pedestrian safety and experience are generally poor. Although the road density is low due to the relatively late construction sequence within the 2nd and 3rd Ring Roads, the non-motorised lanes and sidewalks on both sides of the road are rather complete and suitable for residents to walk. This observation is consistent with that reported in Yang et al. [33].

4.3. Building Space: Building Density and FAR

FAR and BD are important indicators of urban development intensity [5]. A strong correlation exists between these variables and the increase in social and economic activities (e.g., residence and work). For example, a higher FAR means that a region can accommodate more people and the urban vitality is high [5,41,44]. Previous research shows that FAR and BD exert a positive impact on enhancing urban vitality. The conclusion is consistent with urban planning theories, particularly new urbanism and compact cities, namely that high-density land use patterns benefit sustainability. Moreover, the correlation between FAR and urban vitality is more obvious than BD. We speculate that the areas with high FAR are mostly developed areas, with enterprises or business agglomeration that can provide many jobs and entertainment opportunities. Thus, these areas can attract large crowds. Regarding spatial distribution, the promoting effect of BD on urban vitality shows that the central area is higher than the surrounding areas, which is generally consistent with the distribution of BD. This effect is especially obvious in the southwest region of the city, which is densely populated and consists of universities and businesses, and where increasing BD can significantly promote urban vitality. Interestingly, within the 1st Ring Road, BD is the highest, but the positive effect is insignificant. The reason is the existence of old residential buildings characterized by substandard living conditions, including small house sizes and poor livability. These factors result in low urban vitality. In addition, areas showing a negative correlation between BD and urban vitality are predominantly distributed in the peripheral areas, such as the area near the 3rd Ring Road in the east with high-rise resettlement communities. The function of these plots is single, and the basic public service facilities are deficient, which results in a weak population agglomeration effect [45]. Therefore, implementing urban renewal projects in old communities and promoting the construction of mixed-function blocks in suburbs are important to enhance vitality.

4.4. Mixed Function: POI Mixing Degree and 3D Mixing Degree

POI_MD and 3D_MD are positively correlated with urban vitality, and the correlation of 3D_MD is more significant. Therefore, the enrichment in street environment influences urban vitality more than the diversity of urban functions. In other words, people in outdoor spaces will pay more attention to their own perception of 3D space in addition to the basic service functions of the area. A similar point was made in a recent study, which showed that 3D spatial hybridization of streets would increase pedestrian flow [46]. Therefore, planners should plan and design from a ‘people-oriented’ perspective to promote urban vitality and optimise the quality of urban areas. In terms of spatial distribution, areas where POI_MD is negatively correlated with urban vitality are distributed within the 1st Ring Road and are mostly distributed in the afternoon. By contrast, areas that are strongly positively correlated with urban vitality are distributed in the north, southwest, and southeast of the city outside the 2nd Ring Road. The results show that, within the 1st Ring Road, all kinds of facilities are distributed along the street and building space size is limited. The high-dense layout squeezes the ground spaces, which inhibits social activities. Outside the 2nd Ring Road for the city’s new areas, neighbourhoods and buildings have a larger spatial scale. High buildings in the vertical space can carry a wider range of functions and can therefore accommodate a larger number of users. This condition is conducive to the improvement in urban vitality. Similar to POI_MD, 3D_MD is also negatively correlated with urban vitality within the 1st Ring Road on weekends, but it shows a promoting effect on weekdays.

5. Conclusions

A critical challenge in urban planning is to quantitatively identify the geospatial difference phenomenon and influence mechanisms of urban built environment and urban vitality. Huge amounts of research focuses on traditional built environment factors at the overhead level while seldom measuring them from the eye-level view. Moreover, the spatial heterogeneity in the relationship between the built environment and urban vitality is ignored. The major findings of the present study are enumerated as follows. First, urban vitality is closely correlated with the built environment. Moreover, GWR performs better than OLS in capturing localized variations. The effect of each variable varies with time and location. Notably, this study introduces 3D spatial built environment-related data and simulates and measures the objective environment of the street from a human perspective via street view data. The results show that factors at the eye level show more connections than the factors at the overhead level in stimulating urban vitality. Specifically, in terms of green space, the negative impact of GVI is greater than that of NDVI. For the road space, RD and EN inhibit urban vitality, and the impact of EN is more obvious. From the building space perspective, FAR demonstrates a more pronounced positive effect on urban vitality than that of BD. Regarding mixed function, 3D_MD promotes urban vitality much more than POI_MD. Among all the indicators, 3D_MD has the most significant positive impact on urban vitality.

Therefore, with regard to enhancing urban vitality, researchers should not only consider traditional factors (e.g., distance from traffic stations, density of traffic station distribution, and POI_MD) but also pay attention to human scale and perspective (e.g., 3D_MD, sky view index, GVI, and other 3D spatial elements). In addition, the influence of the built environment on urban vitality is spatio-temporally heterogeneous. Urban designers and policymakers are supposed to analyse the degree of influence and the relationships of built environment factors affecting urban vitality from the temporal and spatial dimensions. This enables accurate and rational regulation of various elements and stimulation of urban vitality. Based on these conclusions, we proposed the following suggestions with the aim of improving urban vitality. First, in addition to considering the functional mixing on the plane, more attention should be paid to the three-dimensional organization in the vertical space. Within a reasonable range, we should improve the functional density of facilities and enrich the functional formats of plots. Second, urban planners must account for the psychological impact of the 3D spatial built environment from the perspective of human perception. In the plots dominated by historical blocks and old communities, the quantities of plants and shading area should be controlled to minimize the interference of trees.

However, this study has some limitations. First, we only collected Baidu heat map data for measurement due to the difficulty of data acquisition. As a result, data sources are relatively homogeneous. In future research, data from social media, such as Dazhong Dianping, Meituan, Weibo, and other applications, should be collected to comprehensively represent urban vitality. Second, a nonlinear relationship may exist between some built environment factors and urban vitality given that cities are complex systems. We only used a linear model, which may not perfectly reflect the actual situation. Third, streetscape is only discussed as an objective environmental factor in this study. However, Wu et al. found that streetscape also affected people’s subjective perception and thus influence their behaviour [1]. Therefore, in the following studies, subjective perception and objective environment can be integrated to obtain more reasonable streetscape variables.