Introduction

Industrial parks function as spatial hubs for production and innovation, driving economic growth, creating jobs and fostering technological advancements, as well as integrating industrial chains. As industries upgrade and urban functions diversity, the spatial vitality of these parks now encompasses interactions between businesses and communities, cross-sector collaborations, and social activities both within and beyond the parks1. Therefore, industrial park vitality (IPV) plays a critical role in creating more attractive and healthy environments for workers. Highly vibrant industrial parks are often associated with enhanced positive work emotions and improved physical and mental health for employees2,3, Furthermore, boosting IPV aligns with the Sustainable Development Goals (SDGs), particularly those focus on “Good Health and Well-being” and “Sustainable Cities and Communities”4,5.

Over the past 50 years, industrial parks worldwide have undergone continuous industrial spatial transformations to adapt to the ever-changing demands of cities6. China is currently in a transitional period from industrialization to post-industrialization. Taking Shenzhen, one of China’s fastest-growing cities in terms of land development and economic growth, as an example7, the industrial added value still accounts for 40\(\%\) of the city’s Gross Domestic Product (GDP). As land-use density in city centers increases and spatial resources become increasingly constrained8, the industrial vitality of approximately 254 million square meters of industrial buildings shows a spatial pattern of “sectoral differentiation”. On one hand, industrial parks in the city’s core areas are gradually transforming into hubs for advanced manufacturing and high-tech industries. On the other hand, many industrial parks in suburban areas remain in a “dark factory” state of low utilization, failing to fully realize their potential. Therefore, while suburban new towns of large cities need to maintain their attractiveness to both populations and industries, they also need to achieve high-quality development and high-quality living by continuously optimizing the jobs-housing relationship and supporting facilities, thus creating healthy and vibrant industrial parks. The first step is to gain a deep understanding of the temporal and spatial patterns of industrial park vitality and their associated factors, paving the way for a multi-dimensional synergy of “industry-city integration” in enhancing industrial park vitality.

With the advancement of information and communication technologies, we have entered the era of big data, where the volume of open data from diverse urban sources has reached unprecedented levels. Combined with big data analysis and machine learning techniques, this data is increasingly being applied to study urban vitality and its influencing factors, exploring the complex nonlinear relationships within it9,10,11. Simultaneously, urban vitality, as a complex system, and the creation of livable and vibrant cities have gained widespread recognition and become central to ongoing discussion3,12. Numerous studies have demonstrated that urban vitality is influenced by various factors, including morphology, function, transportation, population and economy3,13,14. Variables within the 5D framework-density, diversity, design, distance to transit, destination and accessibility- are considered important spatial factors affecting urban vitality15,16,17. However, previous studies have often treated the entire urban area as a whole, blending multiple urban functions and lacking in-depth analysis of specific functions. Recent studies5,18 highlight that urban function plays a pivotal role in driving vitality differences, suggesting that focusing on a specific functional area will yield more meaningful and insightful results.

Moreover, in the process of Shenzhen’s “industry-city integration”, particularly with regard to the organizational mechanisms that coordinate temporal and spatial vitality among “industry, city and people”, there remains a lack of a clear research framework. The multi-dimensional nature of urban vitality often leads to misunderstandings and ambiguities in its representation. Finally, the synergistic relationship between the physical environment of industrial parks and people’s perceptions in fostering multi-dimensional vitality within these parks warrants further exploration.

To address these research gaps, this study proposes a central research question: How can demand benchmarking and planning guidance improve the livability and workability of industrial parks, thereby fostering an “industry-city integration” spatial pattern that more efficiently meets the spatial and temporal needs of both industry parks and residents? This is crucial for the future healthy and high-quality development of industrial parks in Shenzhen. Based on this, we propose three sub-questions: First, how can the multi-dimensional vitality of industrial parks be quantified based on vitality quantification theory and the concept of “industry-city integration”? Second, from a multi-dimensional perspective, what is the relationship between the spatial and temporal vitality of industrial parks and the influencing factors, particularly in terms of the organizational mechanisms coordinating spatial and temporal vitality within the “industry, city, and people” synergy? To date, there is a lack of a clear research framework to comprehensively address these issues. Finally, what role do the physical environment and human perception play in shaping the multi-dimensional vitality of industrial parks?

This study utilizes various data sources to depict the spatial and temporal patterns of industrial park vitality in Shenzhen, a rapidly industrializing mega-city in southern China. We employ linear regression and spatial regression models to uncover the complex relationships between the multi-dimensional vitality of Shenzhen’s industrial parks and a range of influencing factors. The findings are expected to deepen our understanding of the spatial and temporal patterns of industrial park vitality and its associated factors, providing profound implications for targeted urban planning and renewal strategies for “industry-city integration”. The contributions of this study are as follows: (1) Redefining industrial park vitality by creating multi-dimensional vitality maps that integrate spatial dimensions of real and virtual environments, as well as weekdays and weekends; (2) Establishing spatial and temporal vitality patterns based on differentiated spatial and temporal needs, describing and comparing real, virtual, weekday and weekend patterns across different time periods; (3) Analyzing the dynamic interactions between industrial park vitality and related factors over time and space, expanding our understanding of how multi-dimensional vitality, physical space, and human perception interact; (4) Providing a fresh perspective on the development planning of “industry-city integration” for Shenzhen’s industrial parks following industrial upgrading, from a multi-dimensional, multi-model synergy approach.

Literature review

“Mixed-Use Urbanism” and the integrated functional development of industrial parks under the “Industry-City Integration” paradigm

The emergence of industrial parks in China dates back to the 1980s and has evolved in tandem with the nation’s urbanization strategy. In the early stages, China’s industrial park development borrowed extensively from the spatial and organizational models of developed Western countries. The introduction of coastal open-door policies, particularly the establishment of economic and technological development zones, served as a crucial driver. The Shekou Industrial Zone in Shenzhen, launched in 1985, was China’s first pilot initiative and marked the beginning of modern industrial park development. This model quickly proliferated across the country, catalyzing widespread spatial expansion of industrial zones19.

As technology-intensive industries and advanced manufacturing sectors grew rapidly, the functional requirements of enterprises diversified. Industrial parks have since evolved from mono-functional production zones into integrated spaces that encompass research and development, office facilities, residential units, and public services. This transformation has led to a “production-city integration and innovation-driven” development paradigm, wherein technological innovation serves as the primary engine20. The model emphasizes a holistic integration of scientific functions and urban life, necessitating spatial frameworks that are efficient, adaptable, and human-centered.

Historically, the idea of functional integration in urban industrial spaces can be traced to Ebenezer Howard’s “Garden City” concept of 1898, which advocated for self-contained communities on city peripheries. These communities were designed to harmonize industrial and residential uses, facilitating a rational redistribution of population and urban functions. This laid the foundation for planning models such as “satellite cities” and “organic decentralization”21. Later, the Athens Charter of 1933 formalized the principle of functional zoning, promoting spatial separation and coordination of living, working, transport, and recreation to enhance urban efficiency and order22. In contrast, Jane Jacobs, in her 1961 work, redefined mixed-use development by highlighting the importance of diversity and permeability in public spaces. She emphasized that urban vitality emerges from active, multifunctional streetscapes-an insight that became foundational for contemporary mixed-use urban planning23,24.

In today’s urban and economic context, industrial parks serve as both engines of regional innovation and critical components of urban spatial restructuring. Unlike traditional public spaces, they are characterized by goal-driven functional configurations, integrated governance systems, and a concentrated user base. According to James F. Moore’s innovation ecosystem theory, technology parks should cultivate collaborative networks linking academia, industry, government, and capital. Such networks enhance innovation outcomes through knowledge exchange and resource synergy25. Faced with growing land constraints and fiscal pressure, Chinese industrial parks are shifting away from the traditional model of “scale-based replication and physical expansion” toward a strategy focused on “demand-responsive planning and high-quality development”. This new approach prioritizes spatial flexibility, efficient land use, and long-term sustainability.

The “on-demand customization” model seeks to more precisely allocate spatial resources to better meet the varied spatial and temporal needs of the new era, thus fostering the development of high-quality, worker-centered industrial parks. In contrast to traditional large-scale spatial production, this model emphasizes not only spatial flexibility and time efficiency time but also the dynamic interaction between space and its users, facilitating the sustainable development of industrial parks. Therefore, by reviewing the spatial collaboration framework of “industry, city and people” within the “on-demand” model (Fig. 1), this paper introduces a spatial model based on the differentiated layout of time and space, aiming to deepen our understanding of the vitality of industrial parks and explore its practical application.

Fig. 1
figure 1

Model of a spatial and temporal demand-setting framework for “industry-city integration”.

Definition and measurement of the multi-dimensional vitality of industrial parks

Urban vitality reflects the dynamic interaction between human activities and the spatial environment, serving as a measure of quality of life, social living standards, and urban competitiveness3,26. For industrial parks, vitality is not only essential for livability, productivity, and sustainability, but it is also closely related to their integration within the broader context of urban development. However, the concept of urban vitality is broad, and various methods exist to measure it. Current quantitative approaches can be categorized into two main types: static and dynamic. Static methods predominantly rely on fixed spatial data, such as urban morphology, structure, and functional distribution, to represent urban vitality.Common indicators include nighttime light data27, small restaurants28, housing prices29, cultural points of interest(POIs)30, and public facility accessibility31. These indicators are widely used for comparative analysis across different cities or regions, often using a single metric to assess urban vitality. However, relying on a single indicator may lead to biases and may fail to fully capture the multidimensional nature of vitality32.

On the other hand, with the acceleration of digital transformation, the development of digital media has freed people from the constraints of physical “places”, and interactions within virtual spaces have gradually become a central aspect of place-making33. In the digital era, physical proximity is no longer the sole determinant of vitality, as online interaction technologies are revolutionizing traditional face-to-face social activities34. Therefore, there is a pressing need for empirical research to quantify and verify the hybrid nature of physical and virtual vitality in urban spaces35. Evaluating urban vitality through big data has emerged as a key research focus. For instance, Baidu heat maps have been used to illustrate the spatial distribution of populations15, Weibo check-in data to measure app popularity3, mobile phone data36, and Wi-Fi spatial distribution Wi-Fi35 to assess virtual vitality. Additionally, emotional data can provide insights into shifts in urban attractiveness, emotional fluctuations among virtual network users, and measures of online popularity5. Research has also demonstrated that variations in light levels and emotional states at different times can influence changes in vitality37.

By combining multi-dimensional and multivariate data to evaluate spatial vitality, this approach yields more accurate and comprehensive results compared to relying on single data element. Based on the theoretical framework of “industrial-city integration”, this study quantifies the multidimensional vitality of industrial parks from both physical and virtual spatial perspectives. Additionally, it accounts for the vitality differences between weekdays and weekends, establishing four dimensions-physical, virtual, weekday, and weekend-to achieve precise measurements of the vitality associated with the daily work and life of industrial park workers. This multidimensional evaluation method not only enhances our understanding of the dynamic vitality of industrial parks but also provides more comprehensive data support for future urban planning and industrial development.

Influencing factors of industrial park environment on vitality

Influence of physical environment on industrial park vitality

The physical environment plays a significant role in shaping urban vitality, influenced by a variety of factors, such as density, diversity, transportation, and greenery, which interact in complex ways. Jacobs highlighted the profound influence of street design on vitality, particularly aspects like street width, length, corners, and the aesthetic appeal of street scenes38. These factors collectively shape spatial usage patterns and residents’ behaviors39. The “human-centered” approach to street design not only focuses on the physical layout but also addresses the social functions of public spaces40. However, limited research has focused on the distinctions between industrial parks and other urban functional zones in this context. Density and diversity are core factors of urban vitality and their interactive relationship demonstrates dynamic characteristics. Chen et al. noted that appropriate land-use diversity enhances the vitality-promoting effects of density41. For example, mixed-use zones that combine commercial, residential, and recreational functions help foster interaction and vitality42. An efficient public transportation system enhances regional accessibility, facilitates interaction and integration between different functional areas, and boosts the commercial appeal and vitality of the region43. These studies suggest that the physical environment’s relationship with urban vitality, as outlined in the 5D framework by Ewing and Handy44, is highly relevant. Our research indicates that similar positive correlations between urban vitality and other factors such as public transportation and greenery may also apply to industrial parks.

Visual perception-based environmental measurements

In recent years, considerable efforts have been devoted to environmental measurement methods based on visual perception, particularly through the use of Street View Images (SVI) and semantic segmentation technologies. In dense urban environments, the strength of a sense of enclosure directly influences individuals’ psychological experiences and behavioral patterns. In contrast, streets that convey a stronger sense of openness tend to foster more active social interactions and economic activities45. Measuring factors such as greenness, enclosure, walkability, and visual accessibility have proved valuable for assessing street accessibility46. However, subjective visual measurements alone often fail to accurately capture residents’ real-life experiences. For instance, spatial characteristics like building continuity, openness, and enclosure may be differently interpreted across diverse cultural and social contexts. Research has shown that identical spatial features within a neighborhood can trigger varying psychological responses and behavioral patterns among different resident groups. Therefore, our study integrates objective planning data with subjective visual perception of environmental features to explore the impact of multidimensional environmental characteristics on the vitality of industrial parks.

Data and methodology

Study area

This study focuses on Shenzhen city, China, as the research area, with its geographical location shown in Fig. 2. The city is divided into core and non-core areas. The core areas include Nanshan, Futian, Luohu, and the eastern part of Bao’an, while the non-core areas include the western parts of Bao’an, Longhua, Longgang, Dapeng, and Yantian. Shenzhen is home to 425 industrial parks, categorized into first- and second-level industrial lands, with first-level industrial lands being predominant and no third-level industrial lands. The total industrial land area spans 274 \({\text {km}}^{2}\), accounting for 30\(\%\) of the built-up land. According to Shenzhen’s “14th Five-Year Plan”, the city is currently in a post-industrial era of industrial upgrading and functional transformation. The renewal of the industrial park environment aims to rejuvenate the vitality of the park, adapt to new industrial trends and promote the sustainability of industrial parks.

Fig. 2
figure 2

Study area. (a) Map of China. (b) Map of Guangdong Province. (c) Different industrial blocks in Shenzhen.

Research framework

Our study is divided into five phases (Fig. 3). The first step involves identifying the fundamental spatial differences between industrial parks and traditional urban public spaces. Conventional urban public spaces-such as plazas, parks, and commercial streets-are characterized by their openness to all user groups and their continuous temporal vitality. These spaces typically exhibit stable levels of activity throughout the day, with increased intensity during weekends and holidays.

In contrast, industrial parks serve predominantly functional purposes-namely, production, research, and office operations. Their spatial usage is primarily determined by the commuting patterns and operational schedules of employees working within the park. As a result, activity in these areas tends to be highly concentrated during weekday working hours, with noticeable declines during evenings and weekends. To capture these temporal dynamics and reveal potential spatial underutilization, we introduced a comparative framework comprising four vitality indicators: actual vitality, virtual vitality, weekday vitality, and weekend vitality. This approach allows for the nuanced assessment of how activity patterns differ across temporal scales and supports a deeper understanding of how to optimize land use and spatial livability under the paradigm of “industry-city integration”.

Actual vitality is quantified using a composite index incorporating GDP, nighttime light intensity, residential spot density, and Baidu heatmap data. Virtual vitality is derived from sentiment analysis of Weibo social media data. Weekday and weekend vitality are separately measured using Baidu thermal maps corresponding to those specific periods. By comparing these indicators, the study seeks to provide evidence-based strategies for activating underutilized industrial park spaces, particularly through the introduction of mixed-use functions such as commercial, service-oriented, and residential programs to extend active hours and enhance spatial efficiency.

The second phase involves quantifying the physical environment indicators of industrial parks, using the “5Ds” built environment classification framework47 for variable analysis. The third phase focuses on visual perception indicators of industrial parks. We design visual perception indicators corresponding to the physical environment of industrial parks based on semantic segmentation of street images. In the fourth phase, we first test the correlation between all variables and use Ordinary Least Squares (OLS) regression to filter them. Then, we test spatial autocorrelation using Moran’s I and explore spatio-temporal relationships using Geographically Weighted Regression (GWR) models. Next, to address the bias caused by prematurely assuming linear relationships in OLS48, we use XGBoost to train and predict all variables under different vitality conditions. Finally, this study proposes differentiated planning recommendations to support the transformation and enhancement of Shenzhen’s industrial parks for “integration of industry and city” development.

Fig. 3
figure 3

Analytical framework.

Calculation of industrial park real vitality, workday vitality and weekend vitality

We quantified real vitality using a composite index comprising GDP, Nighttime Lights, Residential Spots Density, and Baidu Heat Maps. Meanwhile, Baidu Heat Maps from weekdays and weekends are used to represent weekday and weekend vitality, respectively. These heat Maps leverage real-time geographical location data provided by smartphone users of Baidu products to calculate pedestrian activity across various industrial parks, offering insights into the dynamic population distribution within these areas41. For this study, Baidu Heat Map data for Shenzhen were collected on Monday, April 15, and Sunday, April 21, 2024. Data were recorded at 30-minutes intervals between 6:00 AM and 12:00 AM, resulting in a total of 152 heat maps were used, with a spatial resolution of 3.5 meters. The vitality values derived from Baidu Heat Maps represent the average activity levels across different time periods on weekdays and weekends.

Calculation of industrial park virtual vitality

To further describe virtual vitality, we collected Weibo text data from April to June 2024 (https://open.weibo.com/wiki/API). Mobile users sharing their location information on the Weibo platform generated a large number of check-in records49. Based on the provided geographical data, we filtered out records outside industrial parks, resulting in a dataset of 222, 477 text samples from 425 industrial parks in Shenzhen. To process sentiment data and drive emotional vitality values, we selected the Bidirectional Encoder Representations from Transformers (BERT) model,developed by Devlin et al. in 201850, The BERT model fine-tunes on specific text task, accommodating various input types, including individual text sentences or pairs of concatenated sentences represented as token sequences51. These inputs are encoded with tokens, segments, or specific sentences and positional embeddings. For sentiment analysis, a logistic regression classifier was trained on the first 100 dimensions derived from Principal Component Analysis (PCA) applied to the BERT model’s outputs to classify the sentiment scores of social media posts. The resulting sentiment scores provide a measure of the virtual vitality of the workers in the industrial parks.

Calculation of physical environment of industrial parks

We adopted the ‘5Ds’ built environment classification framework and incorporated insights from previous research. Ultimately, we selected fifteen predictor variables (Table 1), which can be grouped into the following five categories.

  1. (1)

    Density variables include population density, road density, building density and building acreage.

  2. (2)

    Diversity variables include \(\mathrm {POI_{mix}}\), POID, NDVI and NDWI.

  3. (3)

    Design variables include intersection density and street segment length.

  4. (4)

    Distance variables include distance to the metro station and the number of bus stops.

  5. (5)

    Destination variables include include BtA500, TPBt500 and NQPT500.

Table 1 Definitions and calculations of 5D variables.

Eye-level features industrial park environment

First, we conducted road point sampling at 50-meter intervals, integrating Open Street Map road data to create a comprehensive visualization of the streets in the study area. Second, we utilized Baidu Maps API to batch download SVI (Street View Imagery) data, collecting 119,622 SVI images from Baidu Maps (https://map.baidu.com/) between 2021 and 2024. Each sampling point included a panorama (Fig. 4). After filtering out invalid images, 5251 SVIs were retained, providing panoramic street views for each sampling point. These images, captured with camera equipment at a resolutions of 600\(\times\)400 pixels, were designed to reflect the visual experience of workers on industrial park streets. Finally, the Segnet and Cityscapes models, both trained on the Cityscapes dataset (https://www.cityscapes-dataset.com/), were used to segment landscape elements in the SVI images54. Based on the studies in3,46, we selected seven indices to enhance the understanding of landscape patterns54. Table 2 shows their corresponding explanations and calculation formulas.

Fig. 4
figure 4

Collection of street view images.

Table 2 Formulas and explanations for scores of indices.

Happiness, life satisfaction

OLS, Spatial autocorrelation and Moran’s I

To analyze the impact of variables on vitality, we employed an ordinary least squares (OLS) regression model to demonstrate the problem of fitting the relationship between the dependent and independent variables59, as represented by the following formula:

$$\begin{aligned} y=\beta _0+\sum _{i=1}^{n}X_iY_i+\varepsilon \end{aligned}$$
(1)

Formulas: y is the dependent variable, \(X_i\) and \(Y_i\) are the independent variables, and \(\varepsilon\) is the residual.

The global Moran’s I and the local Moran’s I are used to determine whether the spatial distribution of a variable is spatially autocorrelated or not. \(I_1\) is the global Moran’s I and \(I_2\) is the local Moran’s I. The formulas can be written as follows:

$$\begin{aligned} & I_1=\frac{n}{\sum _{i=1}^n\left( X_i-\overline{X}\right) ^2}\frac{\sum _{i=1}^nX_i\sum _{j=1}^nX_iw_{ij}(x_i-\overline{X})(x_j-\overline{X})}{\sum _{i=1}^nX_i\sum _{j=1}^nX_iw_{ij}} \end{aligned}$$
(2)
$$\begin{aligned} & I_2=\frac{n(X_i-\overline{X})\sum _{j=1}^n(X_i-\overline{X})}{\sum _{i=1}^n(X_i-\overline{X})^2} \end{aligned}$$
(3)

Formulas: n is the number of industrial parks, x is the strength of the industrial parks, \(W_{i,j}\) is the spatial weight. The values range from [−1,1] for \(I_1\) and \(I_2\), if the value is > 0, the spatial distribution of the variables is positively correlated. If the value is < 0, the spatial distribution of the variables is negatively correlated. If the value = 0, the spatial distribution of the variables is random.

MGWR model

The MGWR model captures the spatial heterogeneity of how landscape environmental characteristics and visual perception features influence vitality within industrial parks. Unlike traditional regression methods, MGWR allows the relationship between independent and dependent variables to vary across both time and space60. Due to its advantages in local spatial data analysis, it has attracted more attention from researchers worldwide compared to other methods, such as the spatial expansion method, spatially adaptive filtering, and the Bayesian spatially varying coefficients (BSVC) model61. The following introduction to MGWR is primarily based on62. The mathematical expression of the MGWR model is as follows:

$$\begin{aligned} & y_{i}=\beta _{0}({u}_{i},{v}_{i})+\sum _{\dot{\kappa }}\beta _{bw\dot{\kappa }}({u}_{i},{v}_{i})\textrm{x}_{i\dot{\kappa }}+\varepsilon _{i} \end{aligned}$$
(4)
$$\begin{aligned} & \beta _{bw\dot{\kappa }}({u}_{i},{v}_{i})=(\textrm{X}^{T}\textrm{W}({u}_{i},{v}_{i})\textrm{X})^{-1}\textrm{X}^{T}W({a}_{i},{b}_{i})\textrm{Y} \end{aligned}$$
(5)

where \(W(a_i,b_i)\) denotes the matrix with the following equation:

$$\begin{aligned} W_{i,j}=e^{-\frac{d_{ij}^2}{b^2}} \end{aligned}$$
(6)

Formulas: \(u_i\) and \(v_i\) are the coordinates of each independent variable position i. \(\beta _0(u_i,v_i)\) is the intercept value of position i. \(\beta _{bwk}(u_i,v_i)\) is the n-th variable mile at position k of the slope coefficient i. \(\varepsilon _i\) the calibration error value. \(d_{ij}\) expresses the Euclidean distance of point i and point j, value \(W_{i,j}\) decreases as d increases. \(W_{i,j}\) will be set to 0 when the distance is greater than the base bandwidth of the kernel function|number. b is the bandwidth, i.e., the coverage of the kernel function, which reflects the difference in the degree of influence of the independent variable on the dependent variable. The weight values decrease from the centre to the edges within the bandwidth, gradually converging to zero.

XGBoost

XGBoost, originally introduced by Tianqi Chen63 is an optimized machine learning algorithm that has gained widespread adoption among many developers. It employs a machine boosting approach based on decision trees, leveraging Classification and Regression Trees (CART) for data classification and prediction. XGBoost supports three main gradient boosting techniques: gradient boosting, regularized boosting, and stochastic boosting. Its ability to incorporate and fine-tune regularization parameters sets it apart from other libraries. The algorithm is highly efficient, minimizing computation time and optimizing memory usage64. In this study, XGBoost is primarily used for training and predicting independent and dependent variables.

Results

Spatial temporal patterns of different vitality values of industrial parks

The results (as shown in Fig. 5) reveal notable spatial heterogeneity in the multidimensional vitality performance of industrial parks, with distinct spatial demands of industrial chains across different regions. All models indicate that most areas with higher vitality are concentrated in the southern part. Interestingly, the spatial distribution of real vitality and weekend day vitality are strikingly similar, with the highest values centered in regions such as the Donggualing area of Futian District, (known for financial technology services), the Hi-tech Park in Nanshan District, the Convention Center area in Bao’an District, the Foxconn manufacturing area in Longhua District, the Henggang eyewear industrial area in Longgang District, and the Guangming Science City area. These regions have experienced industrial upgrading, suggesting successful integration of industry and urban development. Both real vitality and weekend vitality follow a multi-center clustering pattern, while weekday vitality is more evenly distributed. In contrast, virtual vitality shows the lowest values, expanding outward from the highest-value core areas, indicating that social media expressions by workers in industrial parks tend to be more negative.

Fig. 5
figure 5

Distribution of different industry park vitality.

Simultaneously, according to the study in41, this research divides the analysis into five time periods along the temporal dimension. By segmenting these different time periods, we can gain deeper insights into the geographical and spatial variations of real vitality in industrial parks. Fig. 6 illustrates the temporal and spatial changes in social vitality during weekdays and weekends across Shenzhen’s industrial parks. Overall, the spatial patterns reveal that the distribution of community vitality in the industrial parks across the ten districts of Shenzhen follows a multi-center clustering pattern. The temporal patterns indicate that, on weekdays, vitality peaks in the early morning, with a significant difference compared to other times of day. On weekends, vitality is more evenly distributed, with the peak occurring in the afternoon. The vitality range of vitality on weekdays spans from 0 to 114.442, while the vitality range on weekends extends from 0 to 31.

Fig. 6
figure 6

Distribution of social vitality value during weekdays and weekends at different time periods.

Fig. 7
figure 7

Box plot of different vitality in the working day and weekend.

Figures 6 and 7 reveal distinct spatial distribution trends and temporal variations in the vitality of industrial parks. The temporal distribution of real vitality shows that overall vitality is higher on weekdays than on weekends. On weekdays, the peak vitality occurs between 7:00 and 8:00 a.m., with a secondary peak between 12:00 and 2:00 p.m., aligning with typical working hours. In contrast, on weekends, the highest vitality values are observed between 9:00 and 10:00 a.m., 4:00 and 5:00 p.m., and 8:00 and 9:00 p.m. The distribution of urban vitality during the two night-time periods (18:00–24:00 and 00:00–06:00) is more uniform compared to that in the morning and afternoon. Interestingly, the night-time vitality trend on weekdays mirrors that of weekend nights.

Regression model results

In this study, we calculated the Variance Inflation Factor (VIF) before interpreting the modeling results to detect the presence of multicollinearity65. The results showed that the VIF values for the variables “Street segment length”, “Road density”, “Mixture”, and “Greenery” were all greater than 10, indicating significant multicollinearity among these variables. To address this issue, we could either simply remove the independent variables exhibiting high multicollinearity or apply techniques such as ridge regression and stepwise regression tor adjust for the collinearity.

Tables 3 and 4 respectively show the linear regression results for two different sets of predictor variables: real vitality versus virtual vitality and weekday vitality versus weekend vitality. Most of the predictor factors performed as expected. Additionally, the \(R^{2}\) values for real vitality and virtual vitality are 0.288 and 0.12, respectively, while the \(R^{2}\) values for weekday and weekend models are 0.312 and 0.221, respectively. These results indicate that the models can explain 28.8\(\%\), 12\(\%\), 31.2\(\%\), and 22.1\(\%\) of the variance in the vitality of industrial parks for real and virtual vitality,as well as for weekday and weekend vitality. Real vitality performs better than virtual vitality, and weekday vitality performs better than weekend vitality. This suggests that real vitality values are more predictable and exhibit less randomness compared to virtual vitality values. In other words, the relationship between the selected predictor factors and weekday urban vitality is less influenced by unobserved factors, making it a more reliable predictor. Meanwhile, this findings also indicates that human activity patterns are more regular on weekdays than on weekends, consistent with the findings of66.

Table 3 Linear regression results on Real vitality and Virtual vitality.
Table 4 Linear regression results on weekdays vitality and weekends vitality.

Spatial cluster of industrial park vitality

Table 5 presents the Global Moran’s I index for industrial park vitality during five different time periods on both weekdays and weekends. The results show that the vitality of industrial parks in each time period exhibits spatial autocorrelation, indicating a significant positive correlation in the spatial distribution of vitality across the industrial park.

Table 5 Global Moran’s I for industrial park vitality on weekdays and weekends.

Figure 8 illustrates the variation in the local Moran’s I index for vitality levels within the industrial park throughout the day, both on weekdays and weekends. Specifically, the local Moran’s I indices for weekdays are 0.114, 0.127, 0.099, 0.093, and 0.062, while for weekends, they are 0.067, 0.090, 0.098, 0.108, and 0.068. These results highlight the positive spatial dependence and clustering effects of vitality within the area at different times of the day. The maps show the hot spots and cold spots of vitality within the industrial park during weekdays and weekends. The hot spots (high-high clusters) in Nanshan District remainstable across all time periods, concentrated in the city center and suburban areas such as southern Bao’an, southern Longhua, and southwestern and central Longgang. In contrast, the cold spots (low-low clusters) are mainly located in the suburbs, with northern Bao’an consistently showing low values across all time periods. This indicates a positive correlation between Bao’an and the daily vitality levels of workers in the area.

Fig. 8
figure 8

The LISA cluster maps displayed local Moran’s I value in relation to the neighborhood vitality values.

XGBootst results

XGBoost is regarded as one of the top six machine learning models for handling datasets64. Figure 9 presents the ranking of various factors affecting vitality in industrial parks. Notably, “subway station” and “building acreage” have higher influence, suggesting that workers’ daily commutes are heavily dependent on subway transportation. For real vitality, the top four influential factors are “building acreage”, “intersection density”, “bus stop”, and “subway station”, highlighting that the size of the area and public transportation infrastructure are crucial in boosting actual vitality. In terms of virtual vitality, two variables-NDVI and NQPT 500- differ from those influencing actual vitality, indicating that street greenery and ease of access have a stronger impact on workers’ emotional responses. For work vitality, the variables that differ from the first three types of vitality are POID and \(\mathrm {POI_{mix}}\). Numerous studies have confirmed that land-use mix positively impacts urban vitality67,68. However, this study further refines these findings by demonstrating that this factor is particularly significant for work vitality during weekdays. For weekend vitality, the more influential factors are primarily derived from visual perception data, and the overall impact of variables tends to be more balanced.

Fig. 9
figure 9

Feature importance indifferent vitality.

MGWR results

Tables 6 and 7 present the MGWR modeling results for actual and virtual vitality, as well as for weekday and weekend vitality in the industrial park. The \(R^2\) values for actual and virtual vitality, as well as for weekday and weekend vitality, are higher than those obtained from the linear regression model. These results indicate that the MGWR model has greater explanatory power, allowing it to account for more variance in the predictor variables.

Table 6 MGWR results on real vitality and virtual vitality.
Table 7 MGWR results on weekday and weekend.

Joint comparative analysis

Comparison of the impact of the objective physical environment on the multidimensional viability of industrial parks

As shown in Fig. 10, the spatial distribution of high and low values for “building acreage” and “subway stop” in weekend vitality is more well-distributed compared to other vitality measures, and they exhibit a negative correlation. This indicates that, on weekends, when workers engage in a broader range of activities, the influence of “building acreage” and “subway stop” on the vitality of industrial parks is diminished.

Fig. 10
figure 10

MGWR result of the influence of objective physical environment on different vitality of industrial park.

Subway stop and intersection density both show a negative correlation across the four types of vitality in industrial parks. The vitality values for subway stop in terms of real, virtual, and weekday vitality follow a similar overall pattern, with the lowest values in the central area, medium in the west, and highest values in the east. The high-tech industrial park in the core area shows a weak negative correlation with multidimensional vitality. Research has indicated that the commuting distance for the migrant population in industrial parks tends to be longer than that for local residents69. Generally, public transportation facilities in core areas are more adequately distributed compared to suburban areas, offering workers in these regions more diverse commuting options or shorter commuting routes.

The correlation between subway stop and the vitality of industrial parks is the weakest in the eastern suburbs (Longgang, Pingshan, Dapeng) and the western suburbs (Guangming, Shajing). In addition, due to Shenzhen’s unique elongated geographic shape, workers in Pingshan and Dapeng districts face long cross-city commutes to the core areas. The increasing commuting distance between industrial parks and residential areas implies an imbalance between job and housing opportunities, leading to long-distance commutes70, and more transfers required for subway travel. To address these issues, the recent Shenzhen City Master Plan (2020–2035) (Shenzhen Municipal Planning and Natural Resources Bureau, 2021) proposes a “multi-center, networked, cluster-based” overall layout model to strengthen the guiding role of rail transit in urban vitality and spatial development5. This includes improving subway connections between the northwestern and eastern areas and the core areas to address the difficulties faced by residents in these regions regarding subway access. This in turn validates the effectiveness of Shenzhen’s overall planning. Both POID and NDVI show a positive correlation with different types of vitality, indicating that urban NDVI and POI density contribute to enhancing the IPV, consistent with previous research71. At the same time, efficient and diverse building functions and environments enable workers to access a variety of work-life supporting facilities in a shorter time, aligning with the goal of creating a “15-minute life circle”72. However, the spatial distribution of vitality across the four dimensions differs significantly between POID and NDVI. Actual vitality distribution gradually increases from west to east, while virtual vitality shows an eastward increase. This suggests that the positive impact of green spaces within industrial parks on actual vitality, weekday vitality, and weekend vitality is more pronounced for high-income workers in the core area. In contrast, suburban greening has a greater influence on virtual vitality, implying that workers in the core area experience more stress than those in suburban areas. Workers in suburban areas with higher greening rates tend to provide more positive and optimistic evaluations on social media.

Population density shows a negative correlation with weekday vitality but a positive correlation with other vitality values. This finding suggests that population density tends to stimulate more vitality on weekends than on weekdays. The multidimensional vitality exhibits a spatial pattern that strengthens from east to west. Significant differences in population structure and industrial chain composition exist across different regions. Based on China’s seventh population census data (as shown in Fig. 11), Shenzhen has the highest proportion of young people, with Bao’an District having the largest population, while Nanshan District boasts the highest education level. Among all vitality metrics, the spatial distribution of vitality in Nanshan and Bao’an districts shows the strongest correlation. These areas are often supported by the development of urban villages and talent-oriented affordable housing, which provide low-wage jobs and affordable housing for the migrant population73. The combination of industrial park and urban villages has created an e-commerce economic model that enhances the area’s attractiveness, stimulates regional consumption demand, and fosters a positive development cycle of industrial-city integration.

Fig. 11
figure 11

Population structure of Shenzhen. (a) population age structure. (b) Percentage of population with higher education. (c) Population size. (Data source: The seventh national census of Shenzhen. (http://tjj.sz.gov.cn/zwgk/zfxxgkml/tjsj/tjgb/content/post_8771972.html)).

By analyzing the spatial distribution of population density, building density, and building acreage, along with the comparative analysis in Table 8, we found that in industrial parks with low population density but high building density, workers tend to engage in more frequent social interactions. This results in greater weekday vitality compared to industrial parks with high population density but low building density.

Table 8 Ranking of the total area of industrial parks in different regions of Shenzhen.

When studying the impact of NQPT500 on spatial vitality, we found that the spatial distribution patterns of weekday vitality differ significantly from other types of vitality. There is a positive correlation between road connectivity and weekday vitality, with this relationship being most pronounced in the geographic central area and decreasing toward the peripheral regions. This suggests that central areas exhibit higher levels of activity due to better road connectivity, likely related to weekday commercial activities, commuting needs, and the reliance of office environments on traffic flow. In contrast, other types of spatial vitality show less dependence on NQPT500 and are negatively correlated. Therefore, urban planning and traffic management should take into account the needs of different types of vitality to optimize road connectivity. In the geographic central area, it is recommended to encourage non-motorized travel and implement shared bike systems to improve accessibility and traffic vitality in industrial parks.

Comparison of the influence of the subjective perception of the street environment on the multidimensional vitality of industrial parks

As shown in Fig. 12, the spatial distribution of the high and low values of the four human-eye perception indicators in weekend vitality is more evenly distributed compared to other vitality dimensions. This indicates that weekday vitality is influenced by travel behavior patterns, which tend to form regular and distinct peaks74. However, on weekends, when workers’ activities diversify, more factors impact human-eye perception indicators. Moreover, enclosure and openness show a negative correlation in weekend vitality.

The spatial distribution of the four types of vitality related to motor vehicle diversity is generally similar, with higher levels observed in the core area compared to the suburbs, all exhibiting a positive correlation. Regarding enclosure, except for weekend vitality, all other types of vitality show a positive correlation, with their spatial layouts resembling each other: the highest levels in the southwest and the lowest in the east. This indicates that in core industrial park areas, regions with higher openness and more diverse, rich designs facilitate communication and interaction among workers, thereby enhancing the vitality of the industrial parks. Openness shows a negative correlation across all types of vitality, which differs from previous findings regarding residential areas75. Therefore, in industrial parks, merely having open spaces does not necessarily increasing vitality; instead, designs should incorporate street diversity and a higher degree of spatial integration to foster greater interaction and vitality.

Fig. 12
figure 12

MGWR result of the influence of subjective perception of street environment on different vitality of industrial park.

Discussion

Implications for shaping industrial park vitality

The comparative analysis of actual vitality, virtual vitality, weekday vitality, and weekend vitality reveals notable temporal discrepancies in spatial activation within industrial parks. Environmental interventions that are effective in enhancing weekday or actual vitality may not necessarily translate into improved spatial efficiency during weekends. This finding underscores the importance for urban planners and policymakers to acknowledge the temporal heterogeneity of vitality patterns and to anticipate potential conflicts arising from uniform policy applications. To address these discrepancies, context-sensitive strategies tailored to specific temporal characteristics must be adopted, especially under the paradigm of “industry-city integration”, in order to enhance multidimensional vitality and spatial performance within industrial parks.

The empirical outcomes of this study offer theoretical guidance for the spatial reconfiguration of Shenzhen’s industrial parks, supporting the organic integration of industrial functions and urban space amid ongoing urban renewal. Moreover, the proposed framework serves as a practical reference for economies undergoing similar development transitions. In the context of urban planning and regeneration, addressing the complex spatial demands imposed by the industry-city convergence has emerged as a critical challenge. To this end, the study proposes four strategic recommendations:

  1. (1)

    Flexible Working Hours and Staggered Commuting Schedules: Enterprises within industrial parks are encouraged to implement flexible work arrangements for non-essential positions, thereby mitigating peak-hour congestion (particularly between 7:00–8:00) and improving traffic flow efficiency at park entrances and exits.

  2. (2)

    Midday Space Activation: Between 12:00–14:00, parks should provide easily accessible leisure and wellness facilities-such as pop-up plazas, mobile coffee carts, and compact fitness stations-to enhance the micro-rest experience for employees and increase midday spatial utilization.

  3. (3)

    Nighttime Economy Promotion: Leveraging the relatively stable vitality during weekends and evenings, industrial parks can introduce nighttime districts or 24-hour co-working and social spaces6. These additions are intended to boost space usage efficiency, activate adjacent residential zones, and foster the growth of a local nighttime economy.

  4. (4)

    Integration of Mixed Functions: In response to the weekend demand for family-oriented and lifestyle activities, parks should incorporate diverse functions such as weekend markets, parent-child zones, cultural exhibitions, and community cinemas. These features would extend operating hours and increase both the diversity and attractiveness of park spaces76.

Comparison of “industry-city integration” and multi-dimensional industrial park vitality

We found that the multi-dimensional vitality of industrial parks shows significant differentiation, requiring a multi-dimensional perspective to examine urban vitality77. Moreover, the factors influencing the vitality of industrial parks differ from those of residential areas. Therefore, constructing a multi-dimensional urban vitality with a single urban function presents a promising direction for future research. There is both synergy and significant variation in the factors affecting the multi-dimensional vitality of industrial parks, which can be summarized as follows: First, intersection density, subway stops and openness are negatively correlated with IPV, while POID, NDVI, motor vehicle diversity, enclosure and imageability are positively correlated. This indicates that higher values of these indicators attract more human activity. Among these, land-use mix aligns with one of the fundamental conditions for vitality in Jacobs’ theory78, and corresponds to the current discussions around the concept and spatial policy of “industry-city integration” within the transformation of industrial parks. For SVI (Street View Image-based Index), important factors include motor vehicle diversity, which ranks highly in the XGBoost model for its impact on weekend vitality. Second, diversity in transportation modes and accessibility to pedestrian paths are key factors influencing workers’ daily activities. Third, weekend vitality shows the greatest difference from other types of vitality. The spatial distribution of subway stops, building acreage, motor vehicle diversity, and enclosure is more uniform, and their impact on weekends is weaker. Lastly, some factors display opposing relationships in the spatial distribution and correlation with different types of vitality. For example, NQPT500 shows a negative correlation with weekend vitality but a positive correlation with other types of vitality. This indicates that stronger road connectivity on weekdays may lead to traffic congestion, weakening weekend vitality. Similarly, population density and building density are negatively correlated with industrial park vitality on weekdays but positively correlated with other types of vitality. Building density determines spatial carrying capacity, and the more activities taking place in a space, the more likely they are to spread and interact within the network space.

Planning proposals for the development of “industry-city integration” in industrial parks with different dimensions of vitality

As a globalized city predominantly populated by young and middle-aged residents79, Shenzhen’s “industry-city integration” planning strategy must prioritize higher-level planning guidance to address its time and spatial needs. By refining the spatially alignment and fostering differentiated collaboration within the “industry-city-people” relationship, a time-space combination strategy can systematically tackle challenges such as long-distance commuting, job-housing imbalances , and insufficient industrial support services.

Urban planning should further enhance the development and accessibility of public transportation facilities in the eastern and western suburban areas. In the industrial parks located in these suburbs, it is crucial to ensure alignment with surrounding urbanization processes and foster the synergistic development of educational zones, industrial zones, and urban communities. This appraoch aims to create spatial clusters characterized by “R&D (Research&Development)+services” and “production+supporting facilities”. Population data from Nanshan and Bao’an Districts indicate an increasingly integrated interaction between “industry, city and people” in these areas. By leveraging the region’s strengths in technological innovation and modern services, effective spatial linkages and development can be achieved across the entire value chain- from R&D to manufacturing and services. At the same time, since the overall virtual vitality of industrial parks remains lower than other types of vitality, planning efforts should prioritize strengthening digital infrastructure within these parks. This includes expanding high-speed internet, implementing smart transportation systems, and integrating Internet of Things (IoT) technologies to enhance the parks’ digital service capabilities.

In terms of urban renewal in industrial parks, first, development should be guided by the principle of “industry-city integration”. This involves thoroughly exploring mixed-use land models, optimizing land use structures, and flexibly allocating land functions to achieve the integration of “promoting the city through industry, fostering industry through the city, and uniting industry, city and people”. Second, given the relatively low building density in most industrial parks, incorporating more green spaces and street landscaping can enhance aesthetics while encouraging worker interaction and activity80. The planning process should prioritize the inclusion of adequate green belts, public recreational spaces and pedestrian pathways. Lastly, redeveloping street spaces to optimize land use and promote functional diversity can enhance the diversity and vibrancy of these areas. Additionally, since industrial parks exhibit varying levels of vitality across different times, their operations should adopt targeted management and operational strategies tailored to specific needs during weekdays and weekends, as well as during daytime and nighttime periods.

Limitation and future work

Despite yielding some valuable insights, this study has several limitations that warrant further investigation. First, the results, which focus exclusively on industrial parks, may not be applicable to other functional areas of cities. Shenzhen, as a unique city dominated by first-tier industries, with minimal secondary industries and no tertiary industries. Presents a distinctive context. Secondly, caution must be exercised when interpreting virtual vitality81. Potential user group biases inherent in social media data may distort the representativeness of observed patterns. Therefore, overinterpreting these virtual vitality trends should be avoided. Thirdly, the vitality values in this study are based on single-day measurements, which presents a notable limitation. Future research is encouraged to extend the observation period to improve temporal accuracy and robustness of the findings. Lastly, the study lacks detailed information on the commuting patterns of different population segments within industrial parks, such as the distinction between white-collar and blue-collar workers, or high-income and low-income earners across various regions. Future research could delve deeper into the interactions between specific demographic groups and spatial vitality considering different cultural, industrial, and socio-economic contexts.

Conclusion

Guided by the concept of multi-system coordination in “industry-city integration”, this study aims to comprehensively reveal the distribution characteristics of the vitality across multidimensional industrial parks in Shenzhen based on differentiated spatial and temporal demands. Using linear regression and spatial regression, the study examines the relationship between urban vitality-analyzed across spatial and temporal dimensions, as well as at different times of the day, and its influencing factors. The global regression results can be summarized as follows: First, the spatial distribution of real vitality and weekend vitality shows a multi-centered clustering pattern, while weekday vitality is relatively evenly distributed. Virtual vitality has the lowest values, following a diffusion pattern radiating from the core area with the highest values towards the periphery. Weekday vitality peaks in the morning, while weekend vitality is more evenly distributed throughout the day. Second, factors such as NDVI, POID, motor vehicle diversity, enclosure, and imageability positively influence the vitality of industrial parks, while intersection density, subway stops, and openness negatively impact vitality across all IPV. Third, in terms of weekend vitality, the influence of building acreage, subway stops, enclosure and openness weakens, resulting in increased randomness. Population density negatively correlates with weekday vitality but positively correlates with other vitality measures. Lastly, under the multi-system spatial coordination framework of “industry, city and people”, addressing differentiating spatial and temporal differentiated demands offers innovative approaches for planning strategies aimed at “industry-city integration” in the new era. This approach contributes to enhancing industrial park vitality and promoting urban sustainability.