How Did Built Environment Affect Urban Vitality in Urban Waterfronts? A Case Study in Nanjing Reach of Yangtze River

Abstract

The potential of urban waterfronts as vibrant urban spaces has become a focus of urban studies in recent years. However, few studies have examined the relationships between urban vitality and built environment characteristics in urban waterfronts. This study takes advantage of emerging urban big data and adopts hourly Baidu heat map (BHM) data as a proxy for portraying urban vitality along the Yangtze River in Nanjing. The impact of built environment on urban vitality in urban waterfronts is revealed with the ordinary least squares (OLS) and geographically weighted regression (GWR) models. The results show that (1) the distribution of urban vitality in urban waterfronts shows similar agglomeration characteristics on weekdays and weekends, and the identified vibrant cores tend to be the important city and town centers; (2) the building density has the strongest positive associations with urban vitality in urban waterfronts, while the normalized difference vegetation index (NDVI) is negative; (3) the effects of the built environment on urban vitality in urban waterfronts have significant spatial variations. Our findings can provide meaningful guidance and implications for vitality-oriented urban waterfronts planning and redevelopment.

1. Introduction

Urban waterfronts, as the important part of a town or city adjoining water area (i.e., river, lake, sea and ocean, and harbor), have a unique spatial interface and attractive waterscape [1,2,3,4]. Moreover, urban waterfronts also have obvious advantages in terms of economic development, ecological environment, social interaction, and cultural heritage [5,6,7,8]. The redevelopment of urban waterfronts, which has become a well-established global trend, provides urban waterfronts with new functions (e.g., leisure, recreation, retail, and tourism) to satisfy both economic and social needs [9,10,11,12,13]. In recent decades, a growing body of studies has focused on the redevelopment of urban waterfronts as an important way for cities to improve their vitality, attraction, and international competitiveness [14,15,16]. In China, the development of urban waterfronts is now facing new opportunities for transformation and redevelopment [17,18,19]. Therefore, assessing urban vitality in urban waterfronts and deciphering its influencing mechanisms are crucially important for urban waterfronts planning and design.

The concept of urban vitality is brought into view by Jacobs [20] when it was noted that the presence of more active streets could encourage more people to engage in various activities, whether commercial or residential. Jacobs maintained that vibrant urban space was positive to create a diverse city life. Lynch [21] later defined urban vitality as to what extent vital functions and biological requirements of individual are buttressed by the capacity of the environment. Maas [22] described urban vitality as a representation of spatial quality involving the continuous presence of people, activities and opportunities, as well as the physical environment in which these activities occur. Montgomery [23] claimed that the characteristics of successful urban places tend to have a more vibrant public realm breeding rich human activities. While there is no consensus on the definition of urban vitality, human interactions as well as activities have commonly been the focus of urban vitality research.

In the era of big data, the availability of massive crowdsourced data has become a prominent part of characterizing urban vitality in geography and urban studies [24]. Generally, current research of urban vitality can be categorized into two streams: measuring urban vitality and examining its determinants. The first research stream applies various crowdsourced data to assess the spatiotemporal characteristics of urban vitality. The mobile phone data, social media data, GPS tracking data, as well as Baidu heat map (BHM) data serve as the most dominant proxies of urban vitality by reason that these data provide detailed information regarding people’s behavioral characteristics [25,26,27,28,29]. The second research stream delves into the relationship between urban vitality and its determinants. Scholars claimed that built environment characteristics, such as building density, development intensity, and transportation network, have significant effects on urban vitality [30,31,32,33,34]. These studies have enriched the literature of urban vitality and provided meaningful insights for creating vibrant urban space. For example, well-designed public spaces and small street blocks provide opportunities for more diverse human activities and interactions, and thus foster livable streets and vibrant neighborhoods [35,36,37]. However, most previous studies ignore the spatiotemporal analysis of urban vitality in specific areas, especially for urban waterfronts. Therefore, it is necessary to apply reliable crowdsourced data to effectively assess urban vitality in urban waterfronts.

In this paper, the BHM data collected in Nanjing is adopted to investigate the spatiotemporal analysis of urban vitality in urban waterfronts. Besides, both ordinary least squares (OLS) and geographically weighted regression (GWR) models are used to quantify the associations between urban vitality and built environment characteristics in urban waterfronts. This study is intended to (1) explore the spatiotemporal characteristics of urban vitality in urban waterfronts through the analysis of BHM data; (2) examine how built environment characteristics correlates with urban vitality in urban waterfronts; and (3) afford useful insights and references as to fostering urban vitality in urban waterfronts.

2. Literature Review

2.1. The Measurements of Urban Vitality

Urban vitality, as a vital element for achieving a higher quality of human life, describes the attractiveness, diversity, and competitiveness of public spaces [25,38]. However, how to accurately measure urban vitality remains a challenging issue. Traditional data collection methods, such as field survey and on-site observation, provide detailed human activities information that includes gender, age, characteristics, and activities and behavior of users [39,40]. Such methods, however, are usually costly and time-consuming, and thus may not be suitable for investigating urban vitality at a large scale [41].

Fortunately, in recent years, the rise of crowdsourced data sources, notably data derived from mobile phones as well as social media, offer massive opportunities for observing various human activities and interactions [24]. For example, Nadai et al. [42] proposed a method to measure urban vitality with mobile phone records and examined the association between urban vitality and diversity in the Italian context. Yue et al. [26] quantified neighborhood vitality based on mobile phone data and investigated the association between the point of interest (POI)-based mixed use and neighborhood vitality. Wu et al. [25] suggested that social media check-in data can be used as a proxy for characterizing spatiotemporal patterns of urban vitality in Shenzhen. Recently, BHM data, as a kind of crowdsourced data regarding human activity, provide a new angle to portray population distribution and urban dynamics [43,44]. Numerous novel studies have tapped into the BHM data as a crucial tool in the research of green spaces and parks [43,45,46], urban population aggregation characteristics [47,48], and urban structure and land use [49]. In contrast to social media data and other traditional datasets, BHM data can provide real-time analysis for the dynamics of human activities on daily or hourly intervals [43]. However, comprehensive studies using BHM data to characterize urban vitality are rare, thus this study attempts to adopt the BHM data as the proxy for describing the spatiotemporal characteristics of urban vitality in urban waterfronts.

2.2. The Relationship between Built Environment and Urban Vitality

According to classical theories in urban planning and design [20,23,50], the built environment proves to produce significant effects on the creation of urban vitality in urban spaces. Many existing attempts to link urban analytics and design have been less well-received by urban planners and designers [32], while Salingaros [51] and Alexander [52] have called for new analytical processes, which are derived from principles in urban structure and complexity, are applied in urban design. More recently, substantial efforts have been devoted to examining the relationship between built environment and urban vitality integrated with quantitative analysis [27,32,33,41]. For example, Jacobs-Crisioni et al. [53] investigated the impact of dense and mixed land use on urban activity intensity in Amsterdam, and verified that higher densities and mixed land use contribute to higher urban vitality. Sung et al. [54] attempted to apply Jacobs’s urban design theory to study the urban vitality of Seoul, and the empirical findings point to the significant role of mixed use, small-scale blocks, as well as density in improving the urban vitality. Conducted a study for five megacities in China, Xia et al. [55] found a remarkable positive spatial autocorrelation that connects urban land use intensity with urban vitality based on a local indicator of spatial association (LISA) method. Mouratidis and Poortinga [37] provided evidence that neighborhood density and land use mix are positively associated with urban vitality, whereas green space is found to be associated with lower urban vitality.

Furthermore, the traditional global regression model (e.g., OLS regression model) has been verified as an effective method in unearthing the impact of various built environment characteristics on urban vitality [27,31,41]. However, the global regression model may not be able to adequately exhibit the spatial nonstationarity and actual phenomena, as the obtained global relationships are constant within the entire study area and can only reflect the average conditions [56,57]. In this context, the geographically weighted regression (GWR) model, which overcomes the limitations of the global regression model (i.e., OLS) and can effectively solve the problem of spatial nonstationarity, was introduced to explore the geographical varying relationship by direct simulation of local nonstationary data [58,59,60]. Although GWR has been widely applied in many fields of applied geography [58,61,62], the efforts are still insufficient to uncover the local correlations between built environment characteristics and urban vitality in urban waterfronts.

From the above review, it is apparent that urban vitality has close associations with the built environment variables. Nevertheless, researches that put their focus on the impact of built environment characteristics on urban vitality in urban waterfronts are still limited. Furthermore, few studies have applied the GWR model to investigate the spatial variations in the influence of built environment on urban vitality. Therefore, this article serves as an attempt to extend and expand the previous research by quantifying the associations that connect built environment characteristics and urban vitality in urban waterfronts based on the BHM data and the application of OLS and GWR models.

3. Data and Methods

3.1. Study Area

Nanjing, a historical city located in the Yangtze River Delta region of eastern China, is the capital of Jiangsu Province. The Yangtze River runs through the city and divides it into two regions. The development strategy of Nanjing city centers on creating a humanistic, green and innovative modern city that enjoys international reputation and global influence [63]. For the development of Nanjing city, urban waterfronts along the Yangtze River have a great development potential, the important urban landscape and tourist resource, as well as the symbol of geography, history and culture. Combining with neighborhoods and road networks along the Yangtze River, this study focused on the urban waterfronts with a range of 3–6 km from the shoreline along the Yangtze River in Nanjing (Figure 1).

The analytic unit for studying urban vitality is important [26]. Based on the previous research that focused on the spatial analysis of urban vitality with big data [25,33], this study applied a grid-based method (spatial resolution 1 km * 1 km) as the neighborhood-scale unit to measure spatiotemporal urban vitality in urban waterfronts, and then to explore how built environment variables is related to urban vitality. As shown in Figure 1, the study area can be divided into 1239 spatial units.

3.2. Data

3.2.1. BHM Data

BHM, as a common type of crowdsourced data in China, provides a powerful tool to describe the real-time distribution, density, and dynamics of population. This crowdsourced data gathers the geolocated locations provided by the mobile phone users who use application products provided by Baidu (e.g., Baidu search, Baidu map, and Baidu cloud, etc.), and then displays the relative population distinguished by colors, where red represents high density, and blue represents low density (Figure 2). Recent studies have verified that BHM data could be used as a reasonable proxy for measuring the dynamics of human activities in different areas [29,43]. In this study, therefore, the BHM data at hourly intervals from 6:00 to 22:00 across the Nanjing city were collected on a weekday (October 14th in 2020, Wednesday) and a weekend (October 17th in 2020, Saturday) (Figure 2). In total, 34 BHMs (spatial resolution 3.5 m * 3.5 m) were adopted for analysis.

3.2.2. Other Complementary Data

In this research, multi-source data were employed to quantify built environment characteristics, including point of interest (POI), building footprints, bus and subway stations, polygon of the Yangtze River, and the normalized difference vegetation index (NDVI) data. POI datasets were collected from Baidu map (available from https://map.baidu.com/ (accessed on 25 August 2021)), which provides free API interfaces and detailed location information on geographic entities, such as commercial facilities, traffic facilities, and green spaces and squares. These data have substantially assisted many earlier related studies to reflect land use [25,27]. In our study, therefore, as many as 264,001 POIs were adopted to measure functional density and mixed use. The detailed building footprint data was also acquired from Baidu map (available from https://map.baidu.com/), and served the effort to evaluate the building density and floor area ratio. The information as to transit stations/stops were derived from Nanjing public transportation website (available from http://nanjing.gongjiao.com/ (accessed on 25 August 2021)). Such data, which provides useful traffic information, were utilized to measure the distance to public transport stations. Polygon data of the Yangtze River, which was derived from the Nanjing Master Planning (2011–2020) [63], was used to measure the distance to shoreline. The NDVI data (spatial resolution 30 m ∗ 30 m) was calculated based on the Landsat 8 OLI image, which was downloaded from the Geospatial Data Cloud (available from http://www.gscloud.cn/ (accessed on 25 August 2021)). This vegetation data was used to measure the degree of vegetation coverage for urban waterfronts.

3.3. Methods

3.3.1. Evaluating Urban Vitality in Urban Waterfronts

The BHM data, as an important population-oriented visualization product, could directly indicate real-time population density as distinguished by colors on the map [45,64]. This data ranges from 0 to 7 and the larger value means more human activities. Fan et al. [43] proposed a method to assess the vitality of urban parks through BHM data. This study adopted the BHM data-based method for measuring urban vitality and calculated the average urban vitality value of each spatial unit. The calculation was performed using a so-called “Zonal Statistics” in ArcGIS 10.5, which referenced the BHM data-based method described by Fan et al. [43]. The average urban vitality value can be quantified as follows:

$$Q_i=\frac{{\displaystyle {\sum }_{i=1}^n{A}_i}}{n\ast S_i}$$

(1)

where $Q_i$ is the average urban vitality of spatial unit $i$ of per day, $A_i$ is the urban vitality of spatial unit $i$ at a given time, $S_i$ is the area of spatial unit $i$, $n$ = 6:00, 7:00, 8:00, … 22:00 (17 time slots).

3.3.2. Associated Built Environment Variables

Good built environment features tend to promote the development of vibrant streets, neighborhoods, and of course, urban waterfronts [23,31,33]. Previous studies have suggested that many built environment characteristics are associated with urban vitality, include building density, road intersections, functional density, mixed land use, greenspace, and accessibility [26,27,32,33,37,65]. Based on previous studies and data availability, we established built environment variables from four major dimensions, namely density, diversity, accessibility, and vegetation. The density dimension has three variables, namely the building density, floor area ratio, road intersections, and functional density. The diversity dimension is mainly measured by mixed use. The accessibility dimension including two variables, namely the distance to public transport stations and the distance to shoreline. The vegetation dimension is mainly measured by the NDVI. The detailed quantification variables are listed in

Table 1. Description of built environment variables.

Dimensions	Variables	Abbr.	Descriptions	Data Source
Density	Building density	BD	The building density of each square kilometer grid	map.baidu.com (2020)
	Floor area ratio	FAR	The floor area ratio of each square kilometer grid	map.baidu.com (2020)
	Road intersections	RI	The number of road intersections of each square kilometer grid	Open Street Map (2020)
	Functional density	FD	The number of POI of each square kilometer grid	map.baidu.com (2020)
Diversity	Mixed use	MU	The Shannon entropy is used to calculate the mixed use [27,41], $D=-{\displaystyle \sum _1^n{p}_i\ln (p_i)}$, where D is mixed use index, $p_i$ is the proportions of each of the POI types (residential POI, commercial POI, traffic POI, office POI, science, education and health POI, and green space and square POI), and $n$ is the number of the POI types, in this case $n$ = 6.	map.baidu.com (2020)
Accessibility	Distance to public transport stations	DPTS	The distance to the nearest bus or subway stations of each square kilometer grid	nanjing.gongjiao.com (2020)
	Distance to shoreline	DS	The distance to the nearest shoreline of each square kilometer grid	Nanjing master planning (2011–2020)
Vegetation	Normalized difference vegetation index	NDVI	The average value of NDVI of each square kilometer grid, $NDVI=\frac{NIR-Red}{NIR+Red}$, where $NIR$ denotes near-infrared band, and $Red$ is the red band.	Landsat 8 OLI, spatial resolution 30 m × 30 m (2020)

.

3.3.3. Global and Local Regression Models

Initially, this article explored the relations that connect urban vitality and built environment in urban waterfronts from a global perspective. The global regression model is conducted by the OLS regression model, which serves as a commonplace and effective statistical model for research concerning urban vitality [33,66]. The OLS regression is expressed as thus:

$$y={\beta }_0+{\displaystyle \sum _{j=1}^m{\beta }_j{x}_j+\epsilon }$$

(2)

where $y$ stands for the dependent variable, $x_j$ for the $j$th built environment indicators, ${\beta }_j$ for the corresponding estimated coefficient, $\epsilon$ and for the residual.

Second, this study investigated the spatial heterogeneity in the effect of built environment on urban vitality in urban waterfronts from a local perspective. The local regression model is conducted by the GWR model, which is a location-dependent method to characterize the spatial nonstationarity by fitting a regression model at each local observation, weighting nearby observations around each subject point based on a distance decay function [67]. The GWR model is formulated as follows:

$$y_i={\beta }_0\left(u_i,v_i\right)+{\displaystyle \sum _{i=1}^m{\beta }_j\left(u_i,v_i\right)x_{ji}+{\epsilon }_i}$$

(3)

where $i$ represents the spatial unit $i$, $y_i$ denotes the value of urban vitality of the spatial unit $i$, $x_{ji}$ is the $j$th built environment indicators of the spatial unit $i$, $m$ stands for the total number of spatial units, ${\epsilon }_i$ denotes the random error term of the spatial unit $i$, $\left(u_i,v_i\right)$ signifies the location of spatial unit $i$, ${\beta }_0\left(u_i,v_i\right)$ stands for the intercept at the location $i$, and ${\beta }_j\left(u_i,v_i\right)$ represents the local estimated coefficient of the built environment variable $x_{ji}$.

For the geographical weighting function, a fixed Gaussian distance decay function [56], which assumes that things in closer proximity give rise to more robust influence, is adopted in our study. The bandwidth defines the scope of the spatial weighting function and in the meantime bears on the degree of the local regression model’s calibration [67]. The weighting function along with the best bandwidth size in the model adopted is determined through the corrected Akaike information criterion (AICc), which demonstrates the extent to which the model is consistent with actual phenomena, and the numbers of degrees of freedom in the varied models are considered too [68].

4. Results

4.1. Characteristics of Urban Vitality in Urban Waterfronts

According to the BHM data, urban vitality values were classified into as many as five grades according to natural breaks in ArcGIS 10.5. In general, Figure 3 illustrates that urban vitality in urban waterfronts displays similar spatial distributions on weekdays and weekends, although local differences can be observed. Specifically, as shown in

Table 2. The statistical characteristics of urban vitality index in urban waterfronts on weekdays and weekends.

	Maximum	Minimum	Mean	Median	SD
weekdays	4.660	0.000	0.380	0.041	0.715
weekends	4.803	0.000	0.382	0.029	0.737

, the maximum and mean of urban vitality in urban waterfronts on weekends (maximum of 4.803 and mean of 0.382) are slightly higher than that on weekdays (maximum of 4.660 and mean of 0.380).

In addition, the distribution of urban vitality in urban waterfronts shows obvious agglomeration characteristics on both weekdays and weekends. Figure 3 demonstrates that the identified vibrant core in urban waterfronts located in the southern Yangtze River regions are the urban central areas (Hexi), whereas the northern regions are mainly town centers of Nanjing, including Zhujiang (A in Figure 3), Qiaobei (B in Figure 3), and Dachang (C in Figure 3). These identified vibrant cores are basically consistent with the urban growth of Nanjing.

4.2. OLS Regression Analysis and Global Relationships

The OLS regression model was deployed to examine the global relationship between urban vitality on weekdays and weekends and built environment characteristics in urban waterfronts. As the regression results are reported in

Table 3. Regression results of urban vitality by OLS model on weekdays.

Variable	Coefficient	t-Statistic	Std.	VIF
BD	1.021	5.315 **	0.192	4.874
FAR	0.182	5.309 *	0.034	5.784
RI	0.020	7.20 **	0.003	1.983
FD	0.002	29.812 **	0.000	3.085
MU	0.065	5.106 **	0.012	1.813
DPTS	−0.014	−1.484 **	0.009	1.447
DS	0.032	6.493 **	0.005	1.362
NDVI	−0.290	−3.717 **	0.078	1.383
AICs = 346.960
Adjusted R2 = 0.850

* significant at the 0.05 level, ** significant at the 0.001 level.

and

Table 4. Regression results of urban vitality by OLS model on weekends.

Variable	Coefficients	t-Statistic	Std.	VIF
BD	0.690	3.407 **	0.203	4.874
FAR	0.291	8.069 **	0.036	5.784
RI	0.015	5.143 **	0.003	1.983
FD	0.002	28.709 **	0.000	3.085
MU	0.053	3.993 **	0.013	1.813
DPTS	−0.016	−1.681 **	0.010	1.447
DS	0.033	6.309 **	0.005	1.362
NDVI	−0.269	−3.276 **	0.082	1.383
AICs = 479.678
Adjusted R2 = 0.843

** significant at the 0.001 level.

, BD, FAR, RI, FD, MU, DPTS, DS, and NDVI are significantly associated with urban vitality on weekdays and weekends in urban waterfronts along the Yangtze River in Nanjing. All of these variables are significant at the 0.05 confidence level. The adjusted R square is 0.850 for weekdays and 0.843 for weekends, thus verifying that the independent variables determined are able to explain 85.0% and 84.3% of the urban vitality on weekdays and weekends in urban waterfronts. In addition, this study examined the variance inflation factor (VIF) values of each built environment variable, which are all far less than 10, indicating that no multi-collinearity exists between the independent variables.

Moreover, Table 3 and Table 4 show that the BD, FAR, RI, FD, MU, and DS have positive associations with the urban vitality in urban waterfronts. Among all built environment variables, the BD (1.021 for weekdays and 0.690 for weekends) has the strongest positive associations with urban vitality in urban waterfronts, followed by another density variable, FAR (0.182 for weekdays and 0.291 for weekends), demonstrating that high-density is significantly correlated with sustained urban vitality. Besides, our results show that the MU (0.065 for weekdays and 0.053 for weekends) has a positive correlation with urban vitality in urban waterfronts, which overlaps with Jacobs’s view that a diversity of urban functions motivate to spend more time around urban spaces and undertake varied activities. The DS (0.032 for weekdays and 0.033 for weekends) has positive associations with urban vitality in urban waterfronts, demonstrating that the more proximity to the shoreline, the lower urban vitality. As shown in Table 3 and Table 4, the DPTS (−0.014 for weekdays and −0.016 for weekends) is found to have a negative association with urban vitality, thus indicating that convenient public transport can generate more urban vitality in urban waterfronts. Notably, the NDVI (−0.290 for weekdays and −0.269 for weekends) has a significant negative association with urban vitality in urban waterfronts.

4.3. GWR Analysis and Spatial Variations

A further examination with GWR models was conducted to provide an insightful understanding of the local correlations between urban vitality and built environment in urban waterfronts. As shown in

Table 5. Regression results of urban vitality by GWR model on weekdays.

Variable	Mean	Std.	Min	Lower Quartile	Median	Upper Quartile	Max
BD	0.734	1.564	−1.676	−0.065	0.761	1.104	13.239
FAR	0.322	0.308	−0.609	0.194	0.539	1.084	2.646
RI	0.016	0.009	−0.008	0.011	0.014	0.023	0.037
FD	0.002	0.001	−0.000	0.001	0.002	0.002	0.004
MU	0.049	0.042	−0.041	0.021	0.035	0.079	0.183
DPTS	−0.040	0.040	−0.195	−0.059	−0.025	−0.010	−0.001
DS	0.037	0.035	−0.001	0.016	0.020	0.050	0.137
NDVI	−0.223	0.149	−0.894	−0.281	−0.188	−0.114	−0.006
AICs = 111.032
Adjusted R2 = 0.880

and

Table 6. Regression results of urban vitality by GWR model on weekends.

Variable	Mean	Std.	Min	Lower Quartile	Median	Upper Quartile	Max
BD	0.392	1.516	−2.913	−0.506	0.398	0.925	11.171
FAR	0.377	0.268	−0.445	0.264	0.355	0.409	2.266
RI	0.012	0.010	−0.014	0.006	0.009	0.018	0.035
FD	0.002	0.001	0.000	0.002	0.002	0.002	0.004
MU	0.030	0.035	−0.059	0.017	0.030	0.065	0.130
DPTS	−0.024	0.047	−0.214	−0.077	−0.024	−0.009	−0.000
DS	0.038	0.034	−0.001	0.017	0.026	0.048	0.141
NDVI	−0.211	0.132	−0.732	−0.269	−0.177	−0.112	−0.000
AICs = 269.201
Adjusted R2 = 0.871

, the adjusted R-squared values of GWR models (0.880 for weekdays and 0.871 for weekends) have increased in contrast with those derived from OLS regression models (0.850 for weekdays and 0.843 for weekends). Furthermore, the AICs values of GWR models (111.032 for weekdays and 269.201 for weekends) are remarkably lower than those of OLS regression models (346.960 for weekdays and 479.678 for weekends). The improvement of adjusted R-squared values and the significant decrease in AICs values indicate that the GWR model has a better capability to interpret the correlations between built environment characteristics and urban vitality in urban waterfronts.

Figure 4 demonstrates that under the GWR models, the spatial characteristics of local R-squared values are similar on weekdays and weekends. The local R-squared values exhibit notable spatial variations, demonstrating that the explanatory capacity of the GWR models is different for the spatial location of urban waterfronts. It is obvious that the central regions of urban waterfronts (Hexi and Jiangbei) have a relatively higher explanatory capacity with the GWR models. In addition, it is detected that the spatial characteristics of the local R-squared value display a decay tendency from central regions to peripheral areas.

Figure 5 and Figure 6 make it clear that with respect to the spatial variations of local estimated coefficients of all the built environment variables on weekdays and weekends, the deeper color represents larger coefficients. For variables positively related, darker colors stand for a stronger positive effect, while for variables negatively related, such as DPTS and NDVI, paler colors denote a stronger negative effect.

As shown in Figure 5, the BD, FAR, RI, and FD were found to have positive impacts on the urban vitality in most urban waterfronts on weekdays and weekends. Figure 5a,b illustrate that BD had a more significant positive impact on the urban vitality in the central and western regions of urban waterfronts (Hexi, Jiangbei, and Longtan). Figure 5c,d demonstrate that FAR presents a greater positive driving on the urban vitality in the eastern and western regions of urban waterfronts (Longtan and Binjiang). This influence dwindled by degrees from the outer to the center. Figure 5e,f show that RI had a slight positive driving on the urban vitality in the various urban waterfronts. Figure 5g,h depict that FD had a slight positive driving on the urban vitality in the western regions of urban waterfronts (Longtan, and Longpao), whereas a slight negative driving in the eastern regions of urban waterfronts (Binjiang).

As shown in Figure 6, the MU and DS were found to have positive impacts on the urban vitality in most urban waterfronts on weekdays and weekends, while DPTS and NDVI show significant negative influences on the urban vitality. For MU, a remarkable positive effect took place in the central regions of urban waterfronts (Hexi and Jiangbei), whereas a negative effect was detected in the northern area (Figure 6a,b). For DS, the spatial characteristics of the correlation coefficients were higher in the central regions of urban waterfronts (Hexi and Jiangbei), and the impacts gradually decreased from the core area to the suburbs (Figure 6e,f). Figure 6c,d demonstrate that DPTS has a significant negative effect in the central regions of urban waterfronts (Hexi and Jiangbei). With regard to NDVI, a strong negative effect can be observed in various regions of urban waterfronts, and the central regions of urban waterfronts located in the southern Yangtze River regions (Hexi) have the strongest negative effect (Figure 6g,h).

In a word, the variables in relation to built environment characteristics and urban vitality present significant spatial variation within the whole area studied, demonstrating the spatial nonstationary relations between these variables and urban vitality in urban waterfronts. The OLS model provides the global associations that connect built environment characteristics and urban vitality in urban waterfronts, and the GWR model further discovers some distinctive differences in various regions by taking account of spatial autocorrelation and spatial heterogeneity.

5. Discussion

5.1. Towards Establishing a BHM Data-Based Method for Assessing Urban Vitality in Urban Waterfronts

Emerging crowdsourced data enable urban planners and policymakers to assess urban vitality with less expense but more efficiency [24]. In particular, real-time crowdsourced data provides a dynamic perspective for urban space. However, how to develop an effective and accurate method to assess spatiotemporal urban vitality remains a challenging issue. Therefore, this study aims to propose a BHM data-based approach to assess spatiotemporal urban vitality in urban waterfronts, which is a response to the increasing interest in adopting emerging crowdsourced data and new analytical methods into urban vitality studies [25,32]. The results suggest that the distribution of urban vitality in urban waterfronts shows similar agglomeration characteristics on weekdays and weekends. Furthermore, the identified vibrant cores based on the BHM data tended to be the important city and town centers, which is largely consistent with the urban growth of Nanjing city. It is also supported by an earlier study pointing out that the identified vibrant cores are mainly town centers in Shanghai city [33]. This study suggests that the BHM data-based method can be extended to other rapidly urbanizing areas.

5.2. The Influencing of Built Environment Characteristics on Urban Vitality in Urban Waterfronts

Understanding the relationship between built environment characteristics and urban vitality could provide significant implications for cultivating more vibrant urban spaces and enhancing the urban quality of life. In this study, OLS and GWR models are employed to quantify the association between urban vitality and built environment characteristics in urban waterfronts. The OLS regression results indicate that the BD (1.021 for weekdays and 0.690 for weekends), FAR (0.182 for weekdays and 0.291 for weekends), and MU (0.065 for weekdays and 0.053 for weekends) have a remarkable influence on urban vitality in urban waterfronts. This finding can further support the importance of density and diversity in urban spaces, which was empirically evidenced by recent studies [31,32]. Dovey and Pafka [69] also provided convincing evidence that density concentrates more people and places within walkable distances and diversity produces a greater range of walkable destinations. Besides, the results show that the NDVI (−0.290 for weekdays and −0.269 for weekends) has a significant negative association with urban vitality in urban waterfronts. This suggests that more vegetation coverage may restrain the sense of liveliness. A study by Mouratidis and Poortinga [37] in Oslo metropolitan area, also found that a negative relationship exists between green space and urban vitality. In some sense, the low urban vitality can be said to partly account for the calming and restorative effects of green spaces [70].

The GWR model also proves effective to analyze the local associations between variables under the category of built environment and urban vitality in urban waterfronts. As shown in Figure 5 and Figure 6, the correlations of the built environment variables with urban vitality exhibit considerable spatial variations within the whole area studied. It reflects the impact of built environment variables on urban vitality has apparent spatial heterogeneity. Specifically, taking variables such as BD and MU as examples, more significant positive impacts on urban vitality in urban waterfronts were found in the central regions (Hexi and Jiangbei), whereas negative impacts were found in the peripheral area. For DPTS, a strong negative effect can be observed in the central regions (Hexi and Jiangbei), demonstrating that these regions have more convenient public transport, and then higher urban vitality. A study by Liu et al. [71] also showed that traffic access and land use mix have strong positive correlations with urban vitality in the central area rather than peripheral area. Besides, our findings suggest that the DS has positive associations with urban vitality in urban waterfronts, especially in the central regions (Hexi and Jiangbei), demonstrating that the more proximity to the shoreline, the lower urban vitality. Similarly, a study by Liu et al. [18] in Shanghai city, showed that traffic accessibility has a negative effect on the urban vitality in urban waterfronts. This could be attributable to that several urban waterfronts near the central regions in our study are characterize by less connectivity to the shoreline and monotonous landscape. Therefore, it is important to realize that the openness of shoreline and diversity landscape of urban waterfronts will enhance the vitality of urban waterfronts.

5.3. Limitations and Future Studies

This study also has several limitations. First, even though the BHM data can serve as a reliable proxy for human activities and interactions by reason that it reflects the dynamic distribution of population in urban space, this data is unlikely to represent all age and social groups (especially the older adults) and the different types of activities. Future research could attempt to combine multiple data sources (e.g., social media and mobile phone) and traditional surveys to extract more representative information for urban vitality [72]. Second, this study adopted a grid with the 1 km * 1 km spatial resolution as the spatial unit to investigate the effect of built environment characteristics on urban vitality in urban waterfronts. It is likely that more granular data can help us divide urban areas into more fine-scale spatial units, in contrast to the 1 km * 1 km spatial unit, and hence permit more accurate examination. Finally, this study has primarily focused on the built environment variables that affect urban vitality in urban waterfronts, while certain other variables (e.g., social economy, landscape quality, and historic culture) also have influences on urban vitality in urban waterfronts [5,8,73,74]. Researches could further explore the relationship between other variables and urban vitality in urban waterfronts under the proposed method in the future.

6. Conclusions

The increasing demands of urban residents for high-quality urban life have triggered substantial attention concerning urban vitality and built environment, as well as their relationships. However, questions about how the built environment characteristics influence urban vitality in urban waterfronts have not been thoroughly answered. Accordingly, this study proposed a method to uncover the spatiotemporal traits of urban vitality in urban waterfronts involving BHM data. Moreover, the effect of built environment characteristics on urban vitality in urban waterfronts is revealed with the OLS and GWR models. The results prove that (1) the identified vibrant cores based on the BHM data tended to be the important city and town centers, which is largely consistent with the urban growth of Nanjing city; (2) the BD has the strongest positive associations with urban vitality in urban waterfronts, while the NDVI is negative; (3) the influences of the built environment characteristics on urban vitality in urban waterfronts have significant spatial variations.

The major contributions of this article are threefold: First, this study develops a practical approach for policymakers and urban planners to deepen the understanding of the spatiotemporal characteristics of urban vitality in urban waterfronts with the BHM data. This study demonstrates that the BHM data can serve as a reliable proxy for human activities and interactions. For future work, it may be useful to apply the BHM data in different cities for assessing urban vitality. Second, this study presents a way to examine the local relationship between the built environment characteristics and urban vitality in urban waterfronts by the GWR model. Compared with the OLS regression model, the GWR model fits the regression model at each local observation point and can greatly capture the spatial variation in the local relationship between built environment characteristics and urban vitality in urban waterfronts. Third, the study in Nanjing city proves the feasibility of the approach. Our findings suggest that high urban vitality in urban waterfronts has a strong positive correlation with the BD, FAR, as well as MU, thus appropriately increasing density and diversity may be an effective way for improving urban vitality in urban waterfronts. These findings can provide policymakers and urban planners a comprehensive overview of urban vitality, which has significant implications for vitality-oriented urban waterfronts planning and redevelopment.