A Study of the Non-Linear Relationship Between Urban Morphology and Vitality in Heritage Areas Based on Multi-Source Data and Machine Learning: A Case Study of Dalian

Abstract

The preservation of historic heritage not only fosters cultural significance and socio-economic development, but also enhances urban competitiveness. Investigating the vitality of historic urban areas is crucial for measuring their developmental attractiveness, contributing to more effective preservation and planning. However, existing research primarily focuses on urban areas, leaving the applicability of urban form elements to heritage sites and their influence mechanisms unclear. This study employs XGBoost and SHAP, utilizing geographic big data and deep learning techniques, to determine whether the urban form elements impacting the vitality of heritage and urban areas are the same or exhibit different spatial distributions and diurnal variations. Empirical analysis of Dalian reveals significant diurnal variations in the factors affecting vitality, along with distinct key elements for both heritage and urban areas. This study is innovative in being the first to apply deep learning methods to analyze the factors influencing the vitality of Dalian’s heritage areas at the district scale, providing theoretical support for enhancing vitality and promoting urban development.

1. Introduction

The protection of historical heritage is a crucial theme in urban planning, as it is widely recognized as an important means of preserving urban culture and promoting community economic development [1]. The global evolution of historical heritage protection concepts has seen shifts from individual buildings to entire environments, from elite-oriented approaches to public participation, from historical authenticity to living reality, and from static preservation to diverse utilization [2]. Among these, two evolutionary trends in protection concepts stand out. From one perspective, the early recognition of world historical heritage protection focused mainly on individual buildings. Over time, a holistic principle of heritage protection emerged, emphasizing the preservation of urban historical landscapes. Currently, the emphasis is on protecting entire historical heritage areas [2,3]. From another perspective, the concept of heritage protection has shifted from a museum-style approach to the diversified and sustainable use of heritage areas. This approach emphasizes “active protection” and “comprehensive revitalization” to maintain urban diversity and integrity. With the advancement of heritage protection concepts, the understanding of historical heritage has been strategically elevated from being seen as a “cultural resource” to a “civilizational asset”. Historical heritage protection is considered a driver and promoter of sustainable development [4]. The adaptive reuse of historical heritage can enhance urban vitality [5], promote community development and strengthen residents’ sense of identity [6,7], and spur economic growth and social revitalization [8,9].

Urban vitality is viewed as the fundamental energy and dynamism within a city [10], manifesting through the interaction between diverse citizen activities and surrounding entities [11]. Numerous qualitative and quantitative studies have indicated that elements of built environment morphology (e.g., building patterns, land use patterns) significantly impact vitality [12,13,14,15]. However, there is limited research on creating more vibrant heritage areas (historical districts) through planning [2], and the mechanisms of vitality remain unclear. The quantitative analysis of the spatial differentiation of morphology and vitality in heritage areas and the mechanisms influencing vitality is a crucial research topic in urban planning. There are several reasons for this focus. First, heritage areas often have unique morphological patterns subject to strict morphological planning controls. Urban morphology patterns are difficult to change, and strategies to enhance heritage area vitality must consider complex efficiency and equity issues [16]. Second, heritage areas are often located in the city’s core zones with high development intensity and land value [17,18]. High land values may pose risks of gentrification [19] and conflicts between heritage protection and development [20], with some “valueless historical buildings” suffering from low vitality and cultural value loss [21]. This affects the authenticity of heritage protection, influencing residents’ and tourists’ perceptions and experiences.

With the proliferation of big geospatial data and advancements in information technology, it has become possible to measure urban morphology elements and vitality in heritage areas more precisely [22,23]. Machine learning techniques address some limitations of traditional quantitative analysis methods by processing vast amounts of data more quickly and providing more accurate research results [24]. Additionally, these methods reveal the nonlinear relationships between urban morphology elements and vitality, identifying the threshold effects of each variable. This study also employs the SHAP interpretability model to address the black-box issue in machine learning, providing an accurate quantification of the relationship between urban morphology elements and vitality.

Based on the above analysis, this research employs geospatial multi-source data to construct a technical approach to analyze the impact mechanisms of urban morphology elements on vitality. This study addresses two primary research questions:

• What is the spatial distribution of urban morphology elements and vitality in Dalian’s heritage areas and urban districts?
• How do morphology elements in heritage areas and urban areas influence daytime and nighttime vitality, and what are the nonlinear effects?

The study quantifies block morphology elements and uses Points of Interest (POI) data to identify the main land use functions of blocks. Geo-tagged small food facilities and nighttime lighting data are used to measure daytime and nighttime urban vitality, respectively. XGBoost and SHAP analyses are then employed to assess the impacts of urban morphology elements on the daytime and nighttime vitality of heritage and urban areas, providing planning recommendations. Finally, the study presents corresponding conclusions and discussions.

2. Research Framework

This paper proposes a quantitative research method to explore the relationship between urban morphology elements and vitality at the block level, which is divided into three basic parts (Figure 1). First, big geospatial data are collected; second, indicators are extracted for the independent variable of urban morphology elements and the dependent variable of vitality; and lastly, machine learning algorithms are employed to examine the varying factors influencing vitality from multiple perspectives, such as heritage versus urban areas and differences between daytime and nighttime.

2.1. Measuring Urban Vitality

Researchers employ two methods to measure urban vitality. A traditional approach involves an on-site survey counting pedestrians [25]. However, this method is more difficult to measure on a large scale due to its labor-intensive nature. Another is multi-source big data, such as GPS data from cabs and buses [26], Wi-Fi access points [27], road traffic datasets [13], cell phone signaling data [28], and social media data [29], all of which provide a more nuanced picture of human activities and offer new possibilities for measuring urban vitality. However, all methods have their own advantages and disadvantages, and the above data suffer from the problem that they are not open to the public and are more difficult to access, while some of them fail to cover all demographics (

Table 1. Commonly used methods of measuring vitality.

Author (Year)	Data Source and Method	Advantages	Disadvantages
Jalaladdini, S.; Oktay, D. 2012 [25]	Counting the number of pedestrians on the street	The street vitality each hour can be recorded with high accuracy.	Requires significant resources and is challenging for large-scale measurements.
Kim., 2018 [27]	Density of Wi-Fi access points provided by government and network operators	Allows comparison between virtual and physical space vitality.	Does not account for private Wi-Fi access points, and may introduce errors for younger and older users.
Jiang et al., 2024 [13]	Demographic data and OD flow data derived from cell phone	Offers large amounts of accurate data, with mobile location records correlating well with population density.	The socioeconomic characteristics and activity types of users at a specific time and place cannot be identified.
Wu et al., 2022 [2]; Xia et al., 2020 [30]; Ye et al., 2018 [31]	Geotagged small food facilities	Reflects changes in human activities, representing urban economic vitality.	Only captures a specific aspect of urban vitality, limiting its overall representativeness.

).

Geo-tagged small food facilities are frequently used to represent urban vitality [2,30,31]. Typically found in areas with high density, accessibility, and diversity driven by population demands, the distribution of these establishments largely mirrors shifts in human activity [32]. Using these data to measure urban vitality is more efficient than traditional methods, as they are updated promptly and can more accurately represent urban vitality. However, these data primarily capture daytime vitality and cannot reflect nighttime activity.

The difference between daytime and nighttime urban vitality is significant [30], as cities are reshaped at night due to the variability in people’s activities. Nighttime lighting strongly correlates with economic activity and population density [33], making it a key indicator of nighttime vitality [13,34].

2.2. Measurement of Urban Morphology

Conzen first established a system of urban morphology analysis [35], which is a seminal theoretical framework in the field and serves as the foundation for subsequent research. This method takes the block as the fundamental research unit and conducts a comprehensive overlay analysis of three systematic elements: ground plan, building form, and land use. Based on this theoretical framework, this paper adopts the block as the basic scale, extracts the quantitative indicators of urban morphology, and constructs a quantitative research model for morphological analysis (

Table 2. Quantitative indicators of urban morphology.

Components		Indicators	Calculation Method
Ground plan	Street system	Public transportation convenience degree (PTCD)	Ratio of the number of transport stops (bus stops and metro stations) to the size of the block within 500 m ${PTCD}_i={\displaystyle \frac{{Subway}_i^{all}+{Bus}_i^{all}}{A_i}}$
		Road inter section quantities (RIQ)	$\mathrm{Number}\mathrm{of}\mathrm{road}\mathrm{junctions}\mathrm{within}500\mathrm{m}\mathrm{of}\mathrm{the}\mathrm{block}{RIQ}_i$
		Distance to nearest bus stop (DNBS)	Distance from the center of mass of the block to the nearest bus stop ${NBD}_i=\left|\sqrt{{\left(x_i-x_b\right)}^2{+\left(y_i-y_b\right)}^2}\right|$ $x_i$$\mathrm{is}\mathrm{the}\mathrm{block}\mathrm{center}\mathrm{of}\mathrm{mass}\mathrm{longitude}\mathrm{coordinates},y_i$$\mathrm{is}\mathrm{the}\mathrm{block}\mathrm{center}\mathrm{of}\mathrm{mass}\mathrm{latitude}\mathrm{coordinates},x_b$$\mathrm{is}\mathrm{the}\mathrm{nearest}\mathrm{bus}\mathrm{stop}\mathrm{longitude}\mathrm{coordinates},y_b$ is the nearest bus stop longitude coordinates
		Distance to the nearest metro station (DNMS)	The calculation formula is the same as for the distance to the nearest bus stop
	Block pattern	Area (A)	$\mathrm{Area}\mathrm{of}\mathrm{the}\mathrm{block}{A}_i$
		Fractal dimension (FD).	Complexity of block shape ${FD}_i={\displaystyle \frac{2ln\left(\frac{P_i}{4}\right)}{ln{A}_i}}$ $P_i$ is the perimeter of the block
		Spatial compact ratio (SCR)	Compactness of the block shape ${SCR}_i={\displaystyle \frac{2\sqrt{\pi A_i}}{P_i}}$
Building form pattern		Mean of building area (MBA)	Average of all building footprints in the block ${MBA}_i={\displaystyle \frac{{BA}_i}{B_i}}={\displaystyle \frac{\sum _{b=1}^{B_i}{BA}_b}{B_i}}$ ${BA}_i$ is the sum of building footprints, $B_i$ is the total number of buildings in the block.
		Mean of building height (MBH)	${MBH}_i={\displaystyle \frac{{BH}_i}{B_i}}={\displaystyle \frac{\sum _{b=1}^{B_i}{BH}_b}{B_i}}$ ${BH}_i$ is the sum of building heights.
		Building density (BD)	Ratio of total building footprint to total block area ${BD}_i={\displaystyle \frac{{BA}_i}{A_i}}={\displaystyle \frac{\sum _{b=1}^{B_i}{BA}_b}{A_i}}$
		Floor area ratio (FAR)	Ratio of the total volume of the building to the total area of the block ${BE}_i={\displaystyle \frac{{BV}_i}{A_i}}={\displaystyle \frac{\sum _{b=1}^{B_i}{BV}_b}{A_i}}$ ${BV}_i$ is the sum of building volume
Land use pattern		Richness (RPOI)	Ratio of total number of POIs to block size ${FUCD}_i={\displaystyle \frac{\sum _{m=1}^M{POI}_i^m}{A_i}}$ M is the number of POI categories
		Entropy (EPOI),	Information entropy values for all POI function types ${MIX}_i=-{\displaystyle \sum _{m=1}^M}{\displaystyle \frac{{POI}_i^m}{\sum _{m=1}^M{POI}_i^m}}·ln{\displaystyle \frac{{POI}_i^m}{\sum _{m=1}^M{POI}_i^m}}$
		Simpson (SPOI)	Simpson’s index of POI function types ${DIV}_i=1-\sum _{m=1}^M{\left({\displaystyle \frac{{POI}_i^m}{\sum _{m=1}^M{POI}_i^m}}\right)}^2$

).

Specifically, the ground plan comprises two subsets of elements: the street system and the block pattern. The building form model primarily selects indicators from both two-dimensional and three-dimensional perspectives. The land use pattern emphasizes the heterogeneity of urban morphology and selects indicators based on a comprehensive view of land use diversity [28].

According to the Guidelines for Land Use and Sea Use Classification for Land Space Survey, Planning, and Use Control, six primary categories of urban functions were selected and further subdivided into 12 types. Subsequently, based on the research method proposed by Chi et al. [27,36], the POI data were redefined to identify the main functions of the blocks (

Table 3. Types of functions.

Types of Function		Explanation
Residential services		Residential and related services;
Administration and public services	Administration and office	Offices of government, social groups, institutions, etc., and their related facilities
	Cultural facilities	Facilities for cultural public activities such as books and exhibitions
	Education and research facilities	Educational facilities such as higher education, secondary vocational education, etc., and scientific research institutions and their research facilities
	Sports facilities	Facilities such as stadiums and sports training bases
	Health and hygiene	Medical, preventive, health, nursing, rehabilitation, first aid, hospice and other facilities
Commercial and business facilities	Commercial facilities	Retail and wholesale markets, catering, hotels and other services
	Business facilities	Comprehensive office facilities for finance, insurance, art and media
	Recreation facilities	Various recreational and sports facilities
Green and open space		Public open spaces such as parks, protected green spaces, squares, etc.
Industrial		Production workshops, warehouses and ancillary facilities of industrial and mining enterprises
Transportation facilities		Railway, road and other transport facilities and their ancillary facilities

). Additionally, we have taken into account the impacts of socioeconomic variables, such as population size and housing prices, on urban vitality.

2.3. Research Methodology

Currently, the most common model for exploring urban form and vitality is the linear regression model based on least squares and maximum likelihood estimation [37]. However, there are two problems with the existing studies: First, due to the limitations in sample size and the existence of common trends among variables, datasets may have multiple covariance relationships, potentially leading to errors in the analysis of conclusions [2,38]. Second, previous studies have primarily focused on the positive and negative influences and the degree of influence of independent variables on vitality. This approach may obscure locally existing nonlinear relationships and threshold effects [39,40], hindering the accurate and reasonable regulation of the various elements of urban form to enhance urban vitality.

Therefore, this paper introduces the gradient boosting decision tree (XGBoost) model, which has higher accuracy than algorithms such as the support vector machine and random forest, and has been validated as highly effective in urban vitality research [41,42,43]. In this paper, the model is used to analyze the nonlinear relationship between the independent variables and the dependent variable, and to analyze the threshold value of the role of each independent variable. Here is the objective function formula for XGBoost:

$$obj=\sum _{i=1}^{n}l\left(y_i,{\widehat{y}}_i\right)+\sum _{i=1}^t\Omega \left(f_i\right)$$

(1)

The objective function consists of a loss function L of the model with a regular term $\Omega$ that suppresses the complexity of the model, where the loss function consists of the predicted value ${\widehat{y}}_i$ and the true value $y_i$, and $n$ is the number of samples. $\sum _{i=1}^t\Omega \left(f_i\right)$ denotes that the complexity of all $t$ trees is summed up.

Meanwhile, this paper introduces a SHAP interpretable model to solve the black-box problem of machine learning and visualize the relationship between urban form elements and circadian vitality with the following formula:

$$g\left(\acute{Z}\right)={\phi }_0+\sum _{j=1}^M{\phi }_j{\acute{Z}}_j \approx f\left(x\right)$$

(2)

where $g$ is the explanatory model and $f\left(x\right)$ is the original machine learning model. ${\phi }_0$ is a constant for the explanatory model, ${\phi }_j$ is the imputed value for each feature, $M$ is the number of input features, and ${\acute{Z}}_j\in {\left\{\mathrm{0,1}\right\}}^M$ denotes whether or not the corresponding features are observed.

Meanwhile, this paper introduces the SHAP interpretable model to solve the black-box issue of machine learning and visualize the relationship between urban form elements and circadian vitality.

To ensure the transparency and reproducibility of our data processing, we have provided detailed documentation of the preprocessing and modeling procedures. Specifically, we normalized the vitality data and assessed multicollinearity among independent variables, ensuring that all variance inflation factors (VIF) remained below 10 to minimize collinearity effects. To enhance model robustness and prevent overfitting, we split the dataset into an 8:2 training–test ratio and employed 10-fold cross-validation. The hyperparameters of the XGBoost model were optimized using grid search to achieve optimal performance. Regularization techniques were applied to mitigate overfitting risks. Additionally, we experimented with OLS and RF models to determine the approach that yields the highest predictive accuracy for our specific application. To balance the runtime and prediction performance,

Table 4. Performance of OLS, RF and XGBoost.

Model	Indicator	Heritage Areas		Urban Areas
		Day Vitality	Night Vitality	Day Vitality	Night Vitality
OLS	R2	0.418	0.355	0.392	0.390
	MAE	0.035	0.114	0.063	0.051
	RMSE	0.067	0.157	0.094	0.075
RF	R2	0.579	0.572	0.585	0.630
	MAE	0.071	0.091	0.046	0.036
	RMSE	0.029	0.126	0.087	0.061
XGBoost	R2	0.694	0.595	0.629	0.740
	MAE	0.024	0.090	0.041	0.031
	RMSE	0.061	0.122	0.082	0.051

summarizes the research results. Overall, the XGBoost model exhibits the best performance across all predictions for vitality scores.

All analyses were conducted in a Python 3.6 environment, and we have documented key code and parameter selection criteria in

Table 5. Comparison of model hyperparameter settings.

Indicator	Heritage Areas		Urban Areas
	Day Vitality	Night Vitality	Day Vitality	Night Vitality
learning_rate	0.065	0.084	0.05	0.074
max_depth	3	4	3	3
n_estimators	646	316	497	222

to facilitate reproducibility and potential adaptation of our approach to other contexts.

We conducted a cross-validation analysis and obtained the mean R² value along with the 95% confidence interval (

Table 6. Cross-validation results.

Indicator	Heritage Areas		Urban Areas
	Day Vitality	Night Vitality	Day Vitality	Night Vitality
R2	0.778	0.605	0.676	0.653
95%CI	(0.716, 0.851)	(0.510, 0.655)	(0.620, 0.732)	(0.598, 0.721)

). This process allowed us to assess the stability and robustness of the model’s performance, ensuring that the results are not overly sensitive to specific data splits and providing a more reliable measure of the model’s predictive accuracy.

3. Research Area and Dataset

3.1. Research Area

Dalian, also known as Bin Cheng and Romantic City, is a prefecture-level city, a sub-provincial city, a listed city, and a megacity under the jurisdiction of Liaoning Province, located between 120°58′ and 123°31′ east longitude and 38°43′ and 40°10′ north latitude. The historical city of Dalian features a mountain–sea pattern and a “dual-form juxtaposition” that combines a radial road layout with a square grid road pattern, alongside numerous historical parks, squares, and eight designated historical areas.

To ensure a dense residential population within the study area and facilitate data access, this paper selects the largest construction land patches in downtown Dalian rather than the central city area based on administrative divisions, ensuring that the analyzed blocks are built-up spaces. The boundaries of the heritage area were obtained from the Dalian Natural Resources Bureau. This study uses the block as the smallest research unit, defined as a continuous built-up area bounded by roads and water [44]. The blocks are carved up according to road classifications, with corresponding buffer zones established based on these classifications. A total of 2851 blocks were identified, of which 713 are located in the heritage area and 2138 in the urban area (Figure 2).

3.2. Research Dataset

Open data sources are used in this study, i.e., POI data, building data, street data, nighttime lighting data, and small catering business data. POI data were collected for 2022 to 2024 from the Gaode Map (https://www.amap.com/, accessed on 5 May 2024) and Baidu Map (https://map.baidu.com/, accessed on 5 May 2024), which represent food, hotel, shopping, and other spatial and attribute information such as names, spatial coordinates, and categories of real geographic entities, reflecting the fundamental functions of the city such as living, working, and leisure [45]. Building and street data include building spatial coordinates, base contours, and floors.

The building data were crawled from Gaode Map (https://www.amap.com/, accessed on 5 May 2024) after 2022, and the road data are from the OSM (https://www.openstreetmap.org/, accessed on May 5 2024) website, which is crucial for measuring urban space. Nighttime lighting data are available as a free download from the Lakeland Data and Applications Centre’s High Resolution Earth Observation System of Systems (http://www.hbeos.org.cn/, accessed on 10 March 2025). Geo-tagged small food facilities data were crawled in 2022 on the Volkswagen website (https://www.dianping.com/, accessed on 5 May 2024). Housing price data were sourced from Beike (https://m.ke.com/bj/, accessed on 10 March 2025). Population data were acquired from East View Cartographic (https://geospatial.com/, accessed on 10 March 2025). These open data were collected within a reasonable period and the obtained data were normalized.

4. Research Results

4.1. Analysis of the Spatial Distribution of Vitality

The results of the study show (Figure 3) that the daytime and nighttime vitality in the main urban area of Dalian City shows a consistent trend; from the heritage areas in the center of Dalian City to the peripheral urban areas, the vitality of the city gradually decreases in a “core–edge” distribution. In contrast, nighttime urban vitality is more evenly distributed, with relatively high levels observed in the heritage areas. During the daytime, the number of high-vitality areas increases, primarily located in business districts such as Xi’an Road, Olympia 66, Qingdiawa Bridge, and South China Plaza, as well as transport hubs like Dalian Station. Notably, some residential districts in Ganjingzi District, such as BaoYa District and QuanShui District, also exhibit higher vitality.

We also analyzed the monthly variations in nighttime lights for 2024 (Figure 4). The results indicate that the brightest areas at night remain largely unchanged, with a concentration around Zhongshan Square, the Xi’an Road business district, and Dalian Airport. Additionally, the intensity of nighttime lights in winter is noticeably higher than in summer.

4.2. Analysis of the Spatial Distribution of Urban Form

As shown in Figure 5, for the street system in the ground plan, the accessibility of the heritage area is generally higher than that of the urban area, and the PTCD and RIQ indicate that the density of the road network is unevenly distributed, showing significant heterogeneity. Buses and metro services cover most of the city, except for some peripheral areas. For the block pattern in the ground plan, the blocks within the heritage area are more homogeneous in size, and some of the blocks in the urban area are larger in size, particularly those containing pit parks and Dalian Airport. Additionally, there is no evident spatial distribution pattern for FD. The SCR indicates the land-use efficiency of a block. SCR is higher in most parts of Dalian, particularly in the heritage area. In contrast, blocks with lower SCR values are primarily located in geographic areas prone to natural hazards and risks, such as along railway lines, near coastal factories, and adjacent to green parks. The land use functions of these blocks often contradict the principles of compact development.

Regarding building form patterns, the MBA, MHA, BD, and FAR in the heritage area are higher than those in the urban area, indicating a higher degree of spatial development in the heritage area. In terms of land use pattern, the RPOI is high in the heritage area, forming several hotspots in Xi’an Road, Qingnaiwa, and Zhongshan Square. Meanwhile, the EPOI and SPOI form several high agglomeration hotspots in different parts of the city, such as the science, technology, and culture cluster in Qixianding-Lingshui, the integrated service cluster in Qingnaiwa-Renmin Road-Donggang, the integrated service cluster in the eastern coast of Ganjizi, and the industrial cluster on the eastern coast of Ganjingzi. These patterns reflect the city’s emphasis on functional diversity.

Figure 6 lists the top ten functions by share, revealing that residential land use is the primary characteristic of urban and heritage areas. Among the public service functions, education, healthcare, and business are often mixed with other functions and are less likely to dominate completely; in contrast, culture and sports usually play a subordinate role in the blocks and rarely become dominant functions. The heritage areas contain more blocks with a higher proportion of residential, commercial, and institutional land use, while the urban areas have more blocks with predominantly residential functions.

As shown in

Table 7. Functional mix patterns of heritage areas and urban areas.

Heritage Areas		Urban Areas
Patterns of Functional Mixing	Percentages	Patterns of Functional Mixing	Percentages
Administration	10.24%	Residence	17.63%
Commercial	8.56%	Residence and Commercial	10.50%
Residence	7.01%	Commercial	8.96%
Residence and Commercial	5.33%	Education	3.71%
Business	3.65%	Administration	3.54%
Health and hygiene	3.09%	Residence and Administration	3.24%
Administration and Commercial	2.81%	Residence and Health and hygiene	2.95%
Residence and Health and hygiene	2.66%	Health and hygiene	2.36%
Health and hygiene and Education	2.38%	Commercial and Business	2.30%
Business and Administration	2.24%	Health and hygiene and Commercial	2.06%

, the percentage of single-function blocks in heritage areas is 37.7%. With 111 functional types identified in the heritage area, the mix of block functions is more diverse. In addition, there are more functionally mixed blocks between residential, commercial, medical, and institutional, reflecting the contribution of the mix of block functions to vitality in the heritage area. Typical heritage areas, such as Zhongshan Square and Shengliqiao North Historical and Cultural District, are mixed-function blocks between residential, commercial, institutional, and green spaces. The urban area identifies 121 functional types, with 38.5% of single-function blocks, a higher proportion of single-function residential sites and blocks with a mix of residential and the rest of the functions, and a relatively small proportion of sites with a mix of functions.

4.3. Influences on Heritage Areas and Urban Areas

4.3.1. Importance of SHAP Variables

In Figure 7, the importance of urban form elements is ranked in ascending order from top to bottom. This study calculates and visualizes the SHAP values for each urban form element’s impact on daytime vitality and nighttime vitality values. The analysis examines the spatio-temporal effects of these urban form elements on vitality by comparing the absolute SHAP values with changes in their rankings.

Daytime vitality in heritage areas is more influenced by land use patterns, with urban areas being more affected by building forms. In contrast, nighttime vitality in heritage areas is significantly influenced by spatial planning. Building forms, particularly building density and floor area ratio (FAR), have a notable positive impact on nighttime vitality.

For urban areas, REC, RES, and COM are the most important factors influencing vitality, whereas BUS, HEL, and CUL functions exert a lesser influence (Figure 8). In heritage areas, urban vitality is primarily driven by functions such as COM, REC, and BUS, while SPO and HEL also contribute significantly. In contrast, education EDU and green spaces GRE have a limited capacity to enhance vitality. Cultural facilities play a significant role in the vitality of heritage districts in Dalian. Public perception influences the identity and utilization of heritage sites. The Victory Bridge North district in Dalian, with its unique cultural imagery due to its Russian-style architecture and industrial heritage, faces challenges due to uneven preservation, resulting in decay in certain areas and leading to cognitive fragmentation. This has caused a weakening of cultural functions and a decline in vitality. It is essential to promote culture-driven reuse, such as historical exhibitions and cultural festivals, to enhance spatial complexity. Additionally, introducing community co-building models, such as guided tours and cultural experiences, can engage residents as active participants in heritage preservation. Strengthening local features and a sense of place will reinforce local identity, restore historical fabric, and drive cultural revitalization projects, thereby increasing the appeal of heritage districts and achieving sustainable development.

4.3.2. Non-Linear Effects of Variables

Figure 9 illustrates the nonlinear relationship between urban form and vitality, meticulously depicting the relationship between the mean vitality (dependent variable) and the three most important independent variables. Fitted curves are included in the figure to enhance the clarity of the scatterplot trend.

In heritage areas, RPOI shows a positive trend, with a gradual increase that transitions into a sharp rise. This pattern aligns with Jacobs’ theory of “diversity” [46], which posits that high-density commercial and service facilities—particularly those offering food and beverage, shopping, and accommodation—are essential for meeting people’s needs. The concentration of these facilities attracts more visitors, encourages residents to engage in various activities, and increases the likelihood of travel. RIQ shows an impact on the vitality of historic districts that initially rises, then declines, and eventually levels off, indicating the importance of well-planned road networks.

Larger blocks provide more venues for activities, reduce population density and congestion, and thus foster more positive emotions. As shown in Figure 8, when the area (A) ranges from 0 to 2 hm², vitality exhibits a sharp upward trend with increasing A; however, for areas exceeding 4 hm², the local effects tend to stabilize. In this study’s sample, the average block size in Dalian’s heritage areas is 2.37 hm², with 72% classified as small blocks, aligning with the planning requirement for “small blocks and dense road networks”. For the RIQ, an excessive number of intersections show a declining trend, indicating that an overabundance of traffic intersections may weaken the continuity and safety of the transportation system. Excessively high housing prices lead to a reduction in vitality, with 12,000 being the critical threshold for housing prices.

For urban areas, MBA indicates the size and proportion of buildings per unit of space, providing a two-dimensional perspective for analyzing buildings, which has a much greater impact on vitality than MBH. As MBA increases, vitality shows a tendency to decrease and then level off, suggesting that as building footprints decrease, blocks provide more accessible space that is more able to accommodate all types of activities and stimulate vitality. The effect of A on vitality is also more significant, with vitality showing a sharp increase with increasing A when A is between 0 and 4 hm², but compared to the heritage areas, the urban areas have larger blocks, with only 49% of the blocks being small.

DNMS shows a sharp decline followed by a leveling off in the effect on urban vitality, and in general, the effect of public transport on urban vitality is mainly within 500 m of the metro station’s service area. When DNMS is less than 500 m, urban vitality decreases rapidly with distance, and this threshold is much lower than the nearest metro distance of 1500 m in previous studies; it is thus reasonably assumed that the metro stations in Dalian have been planned to be relatively few in number, and the system is not sufficiently linked to the rest of the transport to influence vitality to a significant degree.

In areas with low building density (0–0.34), urban vitality gradually increases, suggesting that a moderate building density can promote pedestrian flow, commercial activities, and social interactions, thereby enhancing urban vitality. However, in ** high-density areas (above 0.34), urban vitality tends to decline, possibly due to excessive density leading to congestion and traffic issues, which may inhibit vitality. Notably, Mean Building Height (MBH) plays a crucial role in shaping urban vitality. In low-rise (one to three floors) and mid-rise (four to six floors) districts, the impact on urban vitality shows a positive trend, whereas high-rise areas (seven floors and above) exhibit only a weak correlation. Compared to high-rise districts, mid-rise areas provide more active public spaces and better sunlight exposure, which help avoid the negative psychological effects associated with high-rise environments [47].

4.3.3. Interaction Effects Between Key Variables

Figure 10 illustrates the interaction effects between two independent variables, with color changes indicating the high or low values of the interacting variable.

In heritage areas, the land use pattern exhibits a strong interaction with the other variables. Among them, A shows a strong interaction with PTCD, which is in the same direction, and block vitality is basically unchanged when A is 4 hm² and the number of PTCD is 1.5. A negative correlation between MBA and EPOI was found, and the block vitality value was higher when MBA was 0–1000 m² and EPOI was between 0.4 and 1. Both the number of RIQ and the EPOI can promote block vitality. Still, the block vitality value decreases when the number of RIQ is 75–100. When the number of RIQ is greater than 100, the block vitality is unchanged, which suggests that too many traffic intersections may weaken the continuity and safety of the traffic system and reduce access efficiency, which is not conducive to the improvement of block vitality.

In urban areas, the interaction between the ground plan and the rest of the variables was the strongest. EPOI interacts positively with SCR, and block vitality rises sharply when EPOI is greater than 1, suggesting that different levels of merchandise shopping spaces exhibit hierarchical circle structures, and that municipal and district shopping districts such as Xi’an Road and South China Plaza in Dalian are more likely to attract clusters of people. There is a positive interaction between A and MBH, and the “small block–mid-rise” block has the greatest impact on urban vitality. DNMS is negatively related to SCR, and the presence of building near transportation hubs with high compactness ratios is in line with urban TOD development [48], but the metro station has relatively little impact on urban vitality.

4.4. Planning Recommendations

4.4.1. Planning Strategies

The above analysis aims to identify existing issues affecting the vitality of historic and cultural blocks in Dalian’s typical heritage areas. Zhongshan Square, due to its diverse land use functions, demonstrates high vitality, as shown in Figure 11. This functional diversity brings considerable foot traffic and a variety of activities, establishing Zhongshan Square as a vibrant core in the area, reflecting high land use efficiency and activity levels. In contrast, Dongguan Street, Nanshan Street, and Fengming Street exhibit lower vitality due to their more monotonous land use patterns, failing to attract sufficient visitors or residents for daily engagement, thus classifying them as “low vitality-functional monotonous types”. The Zhongshan Road Historic and Cultural District in Qingdao, with its similar colonial architectural background, also faced issues of low vitality, delayed industrial settlement, insufficient urban vitality, and challenges such as a lack of social vitality and the aging of physical spaces. The district adheres to the principle of preserving the urban form and landscape characterized by “red tiles, green trees, and blue sea”, while actively introducing innovative economies and cultural industries. The district has made valuable attempts at blending the old with the new. Through a variety of activities, it has successfully attracted people and continuously enhanced the district’s spatial appeal. In terms of international experience, the revitalization of the Marais district in Paris offers valuable insights. By optimizing functional diversity and enhancing spatial adaptability, the district has significantly improved its urban vitality. This case demonstrates that a well-planned mix of land uses and the adaptive reuse of historical and cultural resources are crucial strategies for enhancing the vitality of historic districts. The practice of Zhongshan Road in Qingdao, when integrated with international experiences, provides a useful reference for the revitalization of similar historic and cultural districts.

Under the policy framework of land spatial planning, policies are often delayed due to issues such as administrative interests, jurisdiction, institutions, and technical challenges. Relying solely on administrative documents to regulate and manage the preservation of historic cities is insufficient and cannot provide feasible heritage protection practices or in-depth theoretical research for non-historic cities. During implementation, issues such as the disconnection between preservation policies and urban development planning, as well as incomplete regulatory mechanisms, still persist. To address these challenges, consideration could be given to establishing a specialized agency, similar to France’s ANRU, responsible for the review, funding support, and supervision of historical district renewal projects. At the local level, tools like “urban contracts” could be introduced to clarify the responsibilities, contributions, and performance indicators of all parties involved, granting local autonomy and flexibility while ensuring the implementation of national policies. By introducing a rational POI theme access mechanism, commercial development can be promoted. However, unilateral decision-making may lead to conflicts among different interest groups. Therefore, a consultation mechanism involving users and residents should be incorporated into the spatial development process. This co-governance model, through meticulous management, aims to “allow everyone to realize their aspirations in a shared environment”, ensuring project compatibility and sustainability.

4.4.2. Practical Applications

In the formulation of mandatory indicators for the Urban Design Guidelines, the key morphological indicators extracted and implemented in this study can be utilized to precisely control urban spatial morphology. Specifically, the Maximum Building Height (MBH) should be controlled at around five stories, the building density (BD) should not exceed 0.34, and a “precise bus coverage” plan should be promoted to ensure that the coverage of Neighborhood Bus Density (DNBS) within a 300-m radius reaches 100%, while simultaneously improving metro coverage to optimize public transportation accessibility. Based on the findings of this study, the Entropy Value (EPOI) exhibits optimal vitality when within the range of 0.8 to 1.4, thus precise regulation through land use compatibility provisions (e.g., allowing 20% office space in commercial zones) is necessary to ensure efficient land use. Furthermore, to promote more refined land use regulation, a dynamic control mechanism should be established, selecting three to five typical blocks as “policy laboratories” to compare different morphological intervention schemes. A “Morphological Vitality Improvement Task Force” should be established, led by the Natural Resources Bureau, in collaboration with the Urban Management Committee, the Culture and Tourism Bureau, and other relevant departments. To enhance decision-making transparency and public participation, a “Vitality Diagnosis” app should be developed, allowing citizens to upload street photos to obtain model evaluation scores and improvement suggestions. This approach will further enhance the scientific and operational nature of urban planning, while also increasing public recognition and engagement in the optimization of urban morphology.

5. Conclusions and Discussion

This study systematically analyzes the spatial distribution of morphological elements in Dalian heritage and urban areas and their non-linear relationship with vitality, and the results help to comprehensively understand the complex interactions between various urban morphological indicators and vitality so that urban planning can adopt a prioritized design strategy based on the results of the relative importance of the key influencing factors.

This empirical study finds that the morphological elements of Dalian heritage and urban areas significantly influence the distribution of vitality, and their non-linear effects and threshold effects are evident. For heritage areas, the local effect of each element on vitality is the most influential indicator control value for vitality creation when A is 0–4 hm², DNMS is within 500 m, EPOI is between 0.4 and 1, and RIQ is 75–100. For urban areas, proximity to transport hubs, high compactness ratios, and a diverse functional mix are more likely to increase block vitality [49]. In contrast to previous research [50], this paper finds that the street system significantly influences the vitality of urban areas rather than heritage areas. Therefore, urban areas should implement the concept of public transport-oriented development, and a good interface between metro transport and large shopping districts is an effective measure to enhance urban vitality [48].

This paper further emphasizes the time-driven effects of certain variables on vitality. During the daytime, RPOI, EPOI, and SPOI attract more citizens and travelers, thereby increasing block vitality. However, at night, vitality is less influenced by RPOI, EPOI, and SPOI. To enhance vitality, the mix and diversity of block functions can be improved by integrating typical heritage areas with cultural and green spaces and organizing cultural events [51].

The empirical findings of this paper reveal new insights, indicating that the key elements affecting the vitality of heritage and urban areas differ, aligning with previous theoretical studies [2,31]. Three clusters of urban form elements exert a strong, focused influence on the vitality of both heritage and urban areas. Land use patterns have the strongest impact on the vitality of heritage areas, whereas the ground plan plays a crucial role in shaping the vitality of urban areas. The reasons for this difference may be twofold. First, more substantial commercial enterprises can enhance the functionality of heritage areas [52] and increase their attractiveness, thereby contributing to vibrancy. Second, heritage areas are frequently governed by stricter planning regulations set by authorities, enforcing strict conservation standards. Consequently, the form of buildings and streets in heritage areas differs significantly from those in urban areas [53], resulting in a relatively weak influence of ground plans and building form patterns on the vitality of heritage areas.

This study constructs a more comprehensive framework to depict urban vitality based on big geographic data, and empirically analyzes the differences in the influences of urban form elements on diurnal vitality between heritage areas and urban areas using Dalian as an empirical example, which helps to understand the morphology of heritage areas, and provides planning theoretical support for heritage preservation, urban construction, and economic development in Dalian. Urban zoning should be developed to focus on different areas of the city and adopt targeted strategies.

Finally, this study has certain limitations. First, vitality is a multi-dimensional and abstract concept. As a result, the vitality evaluation indicators selected in this study may lead to incomplete results. Therefore, the findings may exhibit bias or differences due to data characteristics. Furthermore, while this study primarily uses machine learning methods to analyze urban vitality, we acknowledge the potential of deep learning methods in urban perception research. In the future, we plan to explore deep learning models such as CNN or GNN to more accurately extract complex features from street view images and spatial data. Additionally, the integration of real-time data streams, such as social media data or mobile signal data, will enhance the timeliness of the research by capturing the dynamic changes in urban vitality. Regarding predictive models, we aim to further optimize time-series modeling methods, such as LSTM or Transformer, to improve the prediction capabilities for long-term trends. Meanwhile, agent-based modeling can be employed to simulate the impacts of different policies or interventions on district vitality, assisting in the development of more targeted urban renewal strategies. These explorations will further expand the methodology of this study and enhance its practical value. However, this study is the first to apply deep learning at the district scale to analyze factors affecting the vitality of heritage areas in Dalian, offering valuable insights for improving heritage district vitality and promoting urban development.