Shaping Sustainable Urban Development: Spatiotemporal Evolution and Drivers of Newly Established Digital Enterprises in Hangzhou, China

Abstract

As a key driver of sustainable urban development, the digital economy transforms urban spatial structures through novel organizational forms such as digital enterprises. Understanding the spatiotemporal distribution of these enterprises is crucial for fostering equitable and efficient urban growth. Focusing on Hangzhou, a leading digital city in China, this study applies kernel density estimation, the standard deviational ellipse, and the nearest neighbor index to analyze the evolution patterns of newly established digital enterprises (NDEs) from 2010 to 2020. It further integrates geodetector and multiscale geographically weighted regression (MGWR) to uncover the drivers behind their spatial differentiation. The results indicate that: (1) The spatial pattern of NDEs evolved from “single-core diffusion” to a “dual-core with multi-center and axial contiguous” structure, yet the density gap between cores and peripheral counties persisted. (2) NDEs exhibited increasing spatial agglomeration over time. (3) Global drivers: the nighttime light index exerts the strongest positive effect, while land costs and population density show negative effects, reflecting cost-squeeze and decentralized locational preferences. (4) Locally, bus accessibility, innovation level and science-education-culture level, display strong spatial heterogeneity; innovation level has very high positive coefficients in innovation poles but negative effects in ecologically sensitive or deindustrialized areas, revealing an “innovation multiplier effect” alongside resource misallocation risks. These findings provide empirical evidence of how digital economy actors spatially manifest, offering insights for urban planners and policymakers to leverage digital growth for guiding sustainable spatial restructuring, enhancing resource allocation efficiency, and promoting balanced regional development.

1. Introduction

The transition towards sustainable urban development is increasingly driven by the digital economy [1,2], which reshapes urban spatial structures and economic geography through novel organizational forms such as digital enterprises. Understanding the spatiotemporal logic of these enterprises is a prerequisite for fostering urban sustainability [3,4].

Digital enterprises represent a new business form born from digital technology integration [5]. They epitomize the “creative destruction” process [6], wherein technological innovation disrupts existing economic structures and generates new spatial agglomerations of entrepreneurial activity. Unlike traditional manufacturing firms, digital enterprises rely heavily on knowledge spillovers, network externalities, and iterative innovation, which are central to contemporary theories of entrepreneurial ecosystems [7,8]. Concretely, they are market entities characterized by deep digital technology penetration, data-driven business innovation, and platform-based or networked organizational structures [9,10]. Their spatial distribution directly indicates regional digital economic vitality and profoundly influences urban spatial evolution and industrial ecosystem construction through agglomeration and diffusion effects.

China has elevated the digital economy to a core driver of quality-oriented economic expansion, with policies promoting digital infrastructure, innovation, and cross-regional integration [11]. Within this national context, Zhejiang Province has emerged as a leading digital economy hub, and its capital, Hangzhou, has become a national benchmark city for digital economy development. Hangzhou offers an ideal sample for observing the spatiotemporal evolution patterns of digital enterprises, owing to the leading effects of head enterprises such as Alibaba and NetEase, its sophisticated digital infrastructure, and policy innovations as the “First City of Digital Governance.” According to the Zhejiang Province Digital Economy Development Comprehensive Evaluation Report (2025), Hangzhou has ranked first in the province’s digital economy comprehensive evaluation for eight consecutive years. Hangzhou’s core digital economy accounted for 27.1% of its total GDP in 2022. Industrial clusters not only exhibit differentiated spatial distribution characteristics but also show dynamic coupling relationships with urban renewal, technological iteration, and policy regulation over time [12,13].

Existing studies have predominantly focused on the location selection mechanisms of traditional manufacturing and service industries, whereas systematic theoretical explanations for the spatial behavior of digital enterprises, as emergent economic entities, remain limited. The spatial distribution of digital enterprises constitutes a complex system shaped by multiple drivers, including agglomeration economies, innovation externalities, land costs, and policy incentives [14,15]. New enterprises exhibit heightened environmental sensitivity and strategic flexibility [16]; their distribution pattern more directly reflects the reconfiguration of locational determinants in the digital economy era. Therefore, taking Hangzhou as a paradigmatic case, this study aims to: (1) decipher the spatiotemporal evolution patterns of newly established digital enterprises (NDEs); (2) uncover the global and local drivers behind their distribution; and (3) translate these empirical findings into actionable insights for planning sustainable urban spatial structures, optimizing resource allocation, and promoting balanced regional development.

The remainder of the study is structured as follows. Section 2 contains a literature review and establishes the theoretical framework. Section 3 describes the research area, data and methods. Section 4 reports the spatial distribution of NDEs and the results of the influencing factors. Section 5 presents the conclusions, policy implications and limitations.

2. Literature Review and Theoretical Framework

2.1. Literature Review

As micro-level actors in economic activities, enterprises’ location decisions and their determinants have long been central to economic geography and regional economics. Classical location theory, rooted in Weber’s least-cost model [17] and Lösch’s market-area analysis [18], emphasized transport costs, labor supply, and market access. Later, behavioral approaches [19] and institutional perspectives [20] enriched understanding of how uncertainty, information asymmetry, and agglomeration economies shape firm siting. Innovation and entrepreneurship theory has since been integrated into spatial analysis, highlighting that innovative enterprises do not merely minimize costs but actively seek localized knowledge spillovers and innovation milieus [21].

Although the digital industry shares fundamental agglomeration and diffusion mechanisms with traditional sectors, it exhibits distinctive spatial features [22,23,24]. At the regional level, China’s digital industry displays a gradient pattern declining from eastern coastal areas to the inland regions, with a clear positive spatial agglomeration [25]. Developed city clusters, such as the Yangtze River Delta, clearly exhibit this core-periphery structure [26]. At the city scale, fine-grained analyses using postal code zones and street-level administrative units reveal that digital enterprises are heavily concentrated in provincial capitals and municipalities directly under the central government, but sparse in outlying cities [27]. At the industry level, existing studies largely focus on specific subsectors such as software development, electronic information technology, and internet-based services rather than treating the digital industry as an integrated whole [28,29,30].

Scholars have identified multiple drivers of this spatial pattern. Classic location factors, such as regional economic development level [31,32], market potential [33,34], transportation accessibility [35], and industrial upgrading [36], have been robustly validated as foundational conditions for digital enterprises. Meanwhile, the technological environment has profoundly changed, reducing dependence on conventional production factors [37] and prompting strategic relocation or reconfiguration to align with emerging technologies and institutional contexts.

Complementing classical frameworks, new economic geography [38] and innovation systems theory highlight emergent location factors: knowledge spillovers, human capital endowments, regional technological innovation capacity, and institutional quality [39,40,41,42]. The influence of these factors is characterized by nonlinear relationships and pronounced spatial heterogeneity [43]. Furthermore, spatial externalities generated by the digital industry yield dual effects: they foster positive agglomeration synergies via knowledge spillovers, yet may trigger backflow effects that draw resources away from lagging regions, exacerbating interregional disparities [44].

Research methodologies have evolved from qualitative inquiry and conventional statistical toward the deep integration of spatial econometrics and geographic information science. Geographically weighted regression (GWR) and spatial panel models have been widely adopted to uncover spatial dependence and heterogeneity [45,46,47]. In the big data era, studies leveraging high-resolution micro-level data, such as business registration records, patent filings, and digital platform footprints, have proliferated, enabling fine-grained characterization of enterprise location choices and spatiotemporal evolution [48,49].

In summary, prior research has made valuable contributions, yet critical gaps remain. First, most studies rely on administrative units (postal zones, street-level boundaries), risking intra-unit heterogeneity and modifiable areal unit problem biases. To enhance measurement validity, we advocate adopting the equal-area spatial grid method [50], which partitions the study region into uniform geographic grids calibrated to total area and observation density, enabling scale-invariant analysis. Second, the extant literature predominantly examines individual digital sectors in isolation, whereas the digital economy functions as an integrated ecosystem; thus, a holistic view is essential for robust theoretical and policy insight.

2.2. Theoretical Framework

Digital enterprises operate with asset-light structures, high human capital density, and strong network externalities. They feature long industrial chains and pronounced cross-sector integration. Their spatial distribution is shaped by regional economic foundations, technological innovation resources, and agglomeration effects.

Early location theory stressed transport conditions, land costs, and market proximity. For digital enterprises, although physical transport costs have diminished, improved accessibility still lowers the implicit costs of accessing high-quality labor [51]. Despite requiring little production land, digital enterprises need high-quality space for headquarters offices, which often command high rents in urban cores or innovation hubs. Excessive land costs crowd out investment in technology and human capital, weakening long-term competitiveness [52]. Land prices differentials reflect spatial heterogeneity in expected returns and value capture capacity [53]. Consumer market size and potential form a basic attraction: user density and spending power directly determine demand for digital services. Hence, substantial market scale, reasonable land cost and convenient commuting networks are important prerequisites for digital enterprise location.

New economic geography further identifies internal economies of scale are a core agglomeration mechanism. Given the high tradability of digital services, cities close to large markets with dense upstream and downstream ecosystems offer lower matching costs and faster innovation feedback, generating strong regional attractiveness. Knowledge spillovers and specialized labor pools provide essential support [54]. Due to digital enterprises’ heavy reliance on technological innovation, localized economies in terms of technology ecosystems and talent pools enable more efficient collaborative innovation and create a “centripetal force” attracting new entrants [55]. Consequently, the spatial layout of the digital industry must prioritize knowledge spillovers and the innovation environment.

Sustainable city theory is emerging as a new site-selection criterion. Energy efficiency and carbon emission constraints drive digital enterprises to prefer sustainable urban areas with clean energy supply, smart grids, and waste recycling systems. Moreover, compact and polycentric urban structures alleviate congestion and land prices inflation caused by monocentric agglomeration, offering more balanced and sustainable benefits [56]. Incorporating sustainable city theory into the analytical framework not only helps explain the location preferences of emerging digital economy hubs but also provides a theoretical basis for policies that balance innovation-driven development with ecological protection.

Integrating these three theoretical dimensions, we construct an analytical framework shown in Figure 1. Location theory supplies basic constraints of costs and markets; new economic geography explains self-reinforcing agglomeration and innovation ecosystems; sustainable city theory adds spatial form and environmental performance evaluation.

This study uses Hangzhou’s digital economy firm-level data and GIS spatial analysis to examine the spatiotemporal evolution and drivers of digital enterprise distribution. The findings offer evidence-based policy recommendations for optimizing digital industry spatial configuration and advancing sustainable urban development.

3. Area, Data and Methods

3.1. Study Area

Hangzhou serves as the economic, cultural, and educational hub of Zhejiang Province located between 29°11′ to 30°34′ N and 118°20′ to 120°37′ E. According to the 2020 administrative division, Hangzhou comprises 10 urban districts (Shangcheng, Xiacheng, Jianggan, Gongshu, Xihu, Binjiang, Xiaoshan, Yuhang, Fuyang, and Lin’an), two counties (Tonglu and Chun’an), and one county-level city (Jiande) (Figure 2). In 2020, the city’s tertiary industry added value reached 1.096 trillion yuan, with the digital economy output totaling 429 billion yuan. Hangzhou has cultivated three leading digital sectors, cloud computing and big data, artificial intelligence, and information software, as well as signature industrial chains in intelligent computing, network communications, and intelligent equipment. Digital new towns, advanced technology industrial zones, and economic development areas underpin Hangzhou’s spatial development, and all 13 district and county-level administrative bodies execute this framework. This study selects Hangzhou due to its strong representativeness and practical policy relevance for examining the spatial evolution and driving forces of NDEs.

3.2. Data Sources and Processing

According to the Statistical Classification Directory of Core Industries in Zhejiang Province’s Digital Economy (2018), NDEs are defined as entities involved in: (1) computer communications and other electronic equipment manufacturing; (2) telecommunications, radio, television, and satellite transmission services; (3) internet and related services; (4) software and information technology services; and (5) cultural digital content and related services”. An enterprise is considered an NDE if it does not exist in the previous period (t−1) but exists in the current period (t) [57].

The dataset comprises two complementary components. First, NDE data were drawn from the official registration database of the Administration for Industry and Commerce (AIC), covering 2010–2020. This source contains core registration attributes: enterprise name, registered address, and incorporation date. Only enterprises classified under nationally recognized digital economy sectors were retained. Raw records underwent systematic quality control: duplicates, entries with missing or inconsistent registration fields, or implausible temporal attributes were excluded. Additionally, enterprises that lacked any social insurance contribution record over the whole observation window, a stringent indicator of zero actual employee presence, or that they have zero registered capital were removed. The analytical samples comprised 1876 enterprises in 2010, 9888 in 2015, and 17,530 in 2020. Geographic coordinates were assigned via batch geocoding using the AutoNavi Maps API, with address standardization applied prior to geocoding to minimize positional error. The decoding success rates were 99.52%, 99.91%, and 97.11% respectively, yielding final valid counts of 1867 (2010), 9879 (2015), and 17,023 (2020). All coordinates were imported into ArcGIS 10.8, projected into the WGS 1984 geographic coordinate system, and integrated into a unified spatial database.

Second, spatial reference data consist of the administrative boundary map of Hangzhou’s urban districts. Using this as the base layer, the study area was tessellated into a regular grid of 2 km × 2 km cells, generating 4505 non-overlapping units. Attribute values for each cell were derived through spatial overlay analysis with enterprise point data and auxiliary layers, enabling spatial modeling and detection of contextual drivers.

OpenStreetMap (https://www.openstreetmap.org/) provided road network data. Point-of-interest (POI) data for bus stops and science, education, and cultural services were sourced from the Gaode Map API. Land information statistics came from the China Land Market Network (https://www.landchina.com). The Resource and Environmental Sciences Data Platform (http://www.resdc.cn) supplied population density and land use data. Patent data were acquired from the China National Intellectual Property Administration (https://www.cnipa.gov.cn/). Carbon emission data originated from the Emissions Database for Global Atmospheric Research (EDGAR). Nighttime light remote sensing imagery was derived from the global 500 m resolution “NPP-VIIRS-like” nighttime light dataset (https://doi.org/10.7910/DVN/YGIVCD).

3.3. Research Methods

(1)

Kernel Density Estimation

Kernel density estimation provides a non-parametric estimate of the spatial intensity surface generated by point events. It assigns a smoothed, continuous density value to each location based on the proximity and weighting of neighboring sample points, thereby revealing both the spatial distribution pattern and aggregation centers [58]. The calculation formula is:

$$\mathrm{f}\left(\mathrm{x},\mathrm{y}\right)={\displaystyle \frac{3}{\mathrm{n}{\mathrm{h}}^2\mathsf{\pi }}}{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}}{\left[1-{\displaystyle \frac{{\left(\mathrm{x}-{\mathrm{x}}_{\mathrm{i}}\right)}^2-{\left(\mathrm{y}-{\mathrm{y}}_{\mathrm{i}}\right)}^2}{{\mathrm{h}}^2}}\right]}^2$$

(1)

where f(x,y) is the kernel density value; h is the bandwidth; (x_i,y_i) are the coordinates of feature i within the bandwidth; and n is the count of NDEs within the bandwidth. Higher values of f(x,y) indicate greater local point concentration. According to Silverman’s rule, the bandwidth approximates 1507 m. Considering the typical spatial scale of urban functional units (e.g., a coverage radius of 1500 m) and the conciseness and replicability of results, we rounded the bandwidth h to 1500 m. Sensitivity analyses using 0.5 h and 1.5 h are provided in Appendix A.1.

(2)

Standard Deviation Ellipse

The standard deviation ellipse characterizes the directional orientation, dispersion, and central tendency of point distributions [59]. Its geometric parameters include the major axis, minor axis, centroid, and rotation angle. These parameters quantify key spatial properties: the major and minor axes indicate the principal and secondary spread directions; their lengths reflect the standard deviation of point coordinates along those directions. The ellipse area corresponds to the overall dispersion. The centroid represents the mean center of the point distribution. The orientation angle specifies the clockwise rotation from north to the major axis, thereby identifying the dominant directional trend. The calculation formulas are:

$${\mathrm{SDE}}_{\mathrm{x}}=\sqrt{{\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{\left({\mathrm{x}}_{\mathrm{i}}-\overline{\mathrm{X}}\right)}^2}{\mathrm{n}}}}$$

(2)

$${\mathrm{SDE}}_{\mathrm{y}}=\sqrt{{\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{\left({\mathrm{y}}_{\mathrm{i}}-\overline{\mathrm{Y}}\right)}^2}{\mathrm{n}}}}$$

(3)

$$\tan \hspace{0.17em}\mathsf{\theta }={\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{{\overline{\mathrm{x}}}_{\mathrm{i}}}^2-{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{{\overline{\mathrm{y}}}_{\mathrm{i}}}^2+\sqrt{{\left({\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{{\overline{\mathrm{x}}}_{\mathrm{i}}}^2-{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{{\overline{\mathrm{y}}}_{\mathrm{i}}}^2\right)}^2+4{\left(\sum {\overline{\mathrm{x}}}_{\mathrm{i}}{\overline{\mathrm{y}}}_{\mathrm{i}}\right)}^2}}{2{\left(\sum {\overline{\mathrm{x}}}_{\mathrm{i}}{\overline{\mathrm{y}}}_{\mathrm{i}}\right)}^2}}$$

(4)

$${\mathsf{\sigma }}_{\mathrm{x}}=\sqrt{2}\sqrt{{\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{\left({\overline{\mathrm{x}}}_{\mathrm{i}}\cos \hspace{0.17em}\mathsf{\theta }-{\overline{\mathrm{y}}}_{\mathrm{i}}\sin \hspace{0.17em}\mathsf{\theta }\right)}^2}{\mathrm{n}}}}$$

(5)

$${\mathsf{\sigma }}_{\mathrm{y}}=\sqrt{2}\sqrt{{\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{\left({\overline{\mathrm{x}}}_{\mathrm{i}}\sin \hspace{0.17em}\mathsf{\theta }+{\overline{\mathrm{y}}}_{\mathrm{i}}\cos \hspace{0.17em}\mathsf{\theta }\right)}^2}{\mathrm{n}}}}$$

(6)

where SDE_x and SDE_y are the ellipse center coordinates; (x_i, y_i) are the planar coordinates of point i; $\left(\overline{\mathrm{X}},\overline{\mathrm{Y}}\right)$ is the mean center of all points, with ${\overline{\mathrm{x}}}_{\mathrm{i}}$ and ${\overline{\mathrm{y}}}_{\mathrm{i}}$ indicating the coordinate deviations of point i from the centroid; θ is the rotation angle; and ${\mathsf{\sigma }}_{\mathrm{x}}$ and ${\mathsf{\sigma }}_{\mathrm{y}}$ are standard deviations of the x and y coordinates.

(3)

Nearest Neighbor Index

The nearest neighbor index quantifies the degree of clustering or dispersion in a point pattern relative to complete spatial randomness:

$$R = {\displaystyle \frac{\overline{D}}{{\overline{D}}_e}}$$

(7)

$$\overline{D}={\displaystyle \frac{{\sum }_{i=1}^n{d}_i}{n}}$$

(8)

$${\overline{D}}_e={\displaystyle \frac{0.5}{\sqrt{\raisebox{1ex}{$n$}\!\left/ \!\raisebox{-1ex}{$A$}\right.}}}$$

(9)

where R stands for the nearest neighbor index; $\overline{D}$ is the observed mean nearest neighbor distance; ${\overline{D}}_e$ is the expected mean nearest neighbor distance under complete spatial randomness; d_i is the actual nearest neighbor distance for enterprise i; n is the total number of NDEs in the study area; and A denotes the area of that region. R < 1 implies significant spatial clustering, R > 1 implies significant spatial dispersion, and R = 1 suggests a random pattern.

(4)

Geodetector

The Geodetector is a spatial statistical method grounded in geographic information system overlay analysis and set theory. This metric captures how well a geographical factor explains spatial heterogeneity by assessing how much stratifying the data according to that factor lowers the variance of the dependent variable. It also assesses interaction effects between factors [60]. The q-statistic is defined as:

$$\mathrm{q}=1-{\displaystyle \frac{1}{\mathrm{N}{\mathsf{\sigma }}^2}}{\sum }_{\mathrm{i}=1}^{\mathrm{L}}{\mathrm{N}}_{\mathrm{i}}{\mathsf{\sigma }}_{\mathrm{i}}^2$$

(10)

where N_i and N are the observation counts for stratum i of the explanatory factor and for the whole study zone, respectively; L is the stratum count; ${\mathsf{\sigma }}_{\mathrm{i}}^2$ and σ² stand for the variance of the dependent variable within the stratum i and over the entire area, respectively; and q ranges from 0 to 1. A higher q-value reflects a more powerful explanatory capacity.

(5)

Multiscale geographically weighted regression (MGWR)

Spatial nonstationarity is common in geographic relationships, i.e., attributes change with location [61]. By incorporating geographical location, the GWR model and MGWR can capture spatial heterogeneity in explanatory variables, outperforming OLS in explanatory power [62,63,64]. A key advantage of MGWR over GWR is its ability to assign distinct bandwidths to different variables. The formal specification of the MGWR model is as follows:

$${\mathrm{y}}_{\mathrm{i}}={\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}}{\mathsf{\beta }}_{{\mathrm{b}\mathrm{w}}_{\mathrm{j}}}\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right){\mathrm{x}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(11)

where (u_i,v_i) are the spatial coordinates of sample i; bw_j is the bandwidth for the j-th variable. ${\mathsf{\beta }}_{{\mathrm{b}\mathrm{w}}_{\mathrm{j}}}\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$ is the locally estimated regression coefficient for variable j; x_ij is the j-th independent variable; y_i is the dependent variable; ε_i is the random error term. Variation in bandwidth magnitude reflects the spatial scale at which each variable exerts its effect. The MGWR model uses the quadratic kernel function and the corrected Akaike information criterion (AICc) for bandwidth selection, and the golden section algorithm finds the optimal bandwidth.

4. Results

4.1. Spatiotemporal Distribution Pattern of NDEs

4.1.1. Current Spatial Distribution of NDEs

Figure 3 presents the annual count of NDEs in Hangzhou from 2010 to 2020. Registrations declined from 1867 in 2010 to a trough of 1573 in 2012, then rose steadily to a peak of 25,457 in 2019. Although registrations fell to 17,023 in 2020, this remained over eight times the 2010 figure. Growth was heterogeneous: core urban districts (Jianggan, Binjiang, Yuhang) experienced rapid expansion, whereas peripheral counties such as Jiande and Fuyang had persistently low growth rates. Consequently, while the citywide total increased substantially, inter-district spatial inequality in NDE distribution widened over the decade.

Spatially, a clear northeast–southwest gradient characterized NDE counts across districts and counties, with the northeast exhibiting higher values and the southwest lower ones (Figure 4). Core urban districts, such as Jianggan, Xihu, Binjiang, and Yuhang, consistently maintained high NDE counts. In contrast, peripheral jurisdictions such as Tonglu County, Fuyang City, and Lin’an remained low-value but showed steady upward trends. Two interrelated drivers explain these patterns: the advanced socioeconomic conditions and infrastructure in core districts, and heterogeneous initial endowments (natural resources, industrial base, policy capacity), across jurisdictions.

4.1.2. Spatiotemporal Evolution Characteristics of NDEs

We imported NDE counts for 2010, 2015, and 2020 into ArcGIS 10.8 for kernel density estimation, generating spatial density maps (Figure 5, Figure 6 and Figure 7). Comparison of the maps reveals that from 2010 to 2020, the spatial pattern of NDEs transitioned from “single core diffusion” to a “dual core emergence”, and finally to “dual cores with multiple centers and axial contiguous development.”

In 2010, the spatial agglomeration centers were mainly located in the core urban districts (Shangcheng, Gongshu, Xihu, etc.), forming a clear “single core diffusion” pattern. This area, featuring the Wulin Central Business District, creative industry parks, and technology entrepreneurship centers, saw NDE density gradually weaken outward from the core.

In 2015, the pattern shifted from single core to dual cores, with Yuhang District emerging as the second growth pole. The original core districts expanded their agglomeration scope and intensity, and the primary core remained dominant. Hangzhou advanced the “No. 1 Project” for the information economy, establishing major industrial zones (China (Hangzhou) Smart Information Industrial Park, Cloud Town, Future Sci Tech City, etc.) and hosting leading digital enterprises (Alibaba, Huoshi Creation, Yunxiang Technology, etc.).

By 2020, spatial contiguity had been achieved between the main urban area and Yuhang District, reflecting the powerful agglomeration economies of the digital economy. However, the density gap continued to widen. Suburban areas such as Tonglu and Chun’an remained locked in a low-density state, indicating that polycentric development did not automatically translate into a balanced and sustainable development pattern.

4.1.3. Spatiotemporal Evolution Direction of NDEs

We generated standard deviation ellipses for 2010, 2015, and 2020 in ArcGIS 10.8 (Figure 8) and extracted geometric parameters to quantify spatiotemporal changes (

Table 1. Parameters of the standard deviation ellipse for Hangzhou’s NDEs.

Standard Deviation Ellipse Parameters	2010	2015	2020
Rotation angle θ	49.046	53.755	54.207
Area	1291.157	1372.543	2105.632
Major axis (km)	29.309	29.857	39.316
Minor axis (km)	14.024	14.635	17.050
Eccentricity	0.522	0.510	0.566

). The ellipse area expanded continuously from 2010 to 2020, with accelerated growth after 2015, confirming progressive spatial expansion of NDEs. Rotation angle θ increased from 49.046° in 2010 to 53.755° in 2015 and to 54.207° in 2020, reflecting a persistent eastward and northward shift in the main agglomeration axis. Eccentricity declined from 2010 to 2015 (trend toward isotropic dispersion) but rose steadily from 2015 to 2020 (renewed anisotropy and alignment along a northeast–southwest orientation). This directional change aligns with Hangzhou’s strategic development framework: core functional zones, such as the Wulin and Hubin historic centers, Qianjiang New City, Yuncheng, and Future Science and Technology City, are concentrated in eastern, northeastern, and northern corridors, directly shaping the observed clustering trajectory.

4.1.4. Spatiotemporal Agglomeration Characteristics of NDEs

We used the nearest neighbor index to measure spatial agglomeration of NDEs in Hangzhou. To eliminate edge effect, we created a 5 km inner buffer zone extending inward from the administrative boundary of Hangzhou.

Table 2. Nearest neighbor index of NDEs in Hangzhou across periods.

Year	Observed Mean Distance/m	Expected Mean Distance/m	NNI Ratio	Z Value	p Value
2010	308.754	1249.039	0.247	−57.499	0.000
2015	90.350	808.800	0.112	−160.632	0.000
2020	52.939	615.166	0.086	−213.195	0.000

Note: p < 0.05 indicates a statistically significant spatial pattern.

reports the results. The observed mean distances were 308.754 m, 90.35 m, and 52.939 m for 2010, 2015, and 2020, respectively, each substantially lower than the expected distances under complete spatial randomness (1249.039 m, 808.8 m, and 615.166 m). All p values were < 0.01, confirming statistically significant spatial clustering in all three periods. The Z value declined from −57.499 in 2010 to −160.632 in 2015 and further to −213.195 in 2020, indicating progressively stronger deviation from randomness and intensifying agglomeration. Concurrently, the nearest neighbor index ratio decreased from 0.247 in 2010 to 0.086 in 2020, reflecting a time-dependent increase in spatial concentration.

4.2. Influencing Factors of the Spatial Distribution of NDEs

4.2.1. Variable Selection

The dependent variable corresponds to the count of NDEs inside each 2 km $\times$ 2 km grid cell. Drawing upon location theory, new economic geography, and sustainable city theory, we constructed ten indicators spanning eight dimensions (

Table 3. Variable definition.

Type	Variable	Description
Transportation Infrastructure Level	X1: Road network density	Total length per unit area of road (km/km2)
	X2: Bus accessibility	Bus stop total per grid cell (Nos)
Land Cost	X3: Industrial land price	Grid-level average industrial land price per square meter (yuan/m2)
	X4: Commercial and service land price	Grid-level average commercial and service land price per square meter (yuan/m2)
Scale of the Consumer Market	X5: Population density	Population density within the grid (persons/km2)
Knowledge Spillover	X6: Science, education, and cultural level	Number of science, education, and cultural service facilities within the grid (Nos)
Technological Innovation	X7: Innovation level	Number of granted patents within the grid (Nos)
Environmental Pollution	X8: Carbon emission	Total carbon emissions within the grid (ton)
Land Use	X9: Land use intensity	Land use intensity index within the grid
Economic Vitality	X10: Nighttime light index	Nighttime light index within the grid

) to capture the drivers behind spatial disparities in NDE distribution.

Transportation infrastructure is captured by two indicators. Road network density reflects the connectivity and accessibility of the spatial unit. Bus accessibility, quantified as the number of bus stops, indicates the availability of public transit. Land cost is expressed through industrial land price as well as commercial and service land price, both expressed as the average price per square meter, which captures the economic barriers to land acquisition.

Consumer market scale is approximated by population density, a classic proxy for market size and demand potential. Knowledge spillover is indicated by the science, education, and cultural level, reflecting the local knowledge-intensive amenities and human capital. To measure technological innovation, we adopt innovation level, operationalized as the grid-level count of granted patents, which captures both innovative production and technological momentum.

To capture environmental pollution, we adopt carbon emissions, with total emissions denoting the environmental burden and regulatory pressures. Land use intensity serves as an indicator of urban development intensity and land utilization levels. The nighttime light index is a widely used remote sensing proxy for overall economic activity and urban vibrancy.

4.2.2. Factor Detection Results

All indicator data were processed in ArcGIS 10.8. Each variable was reclassified using the natural breaks (Jenks) method. We then applied geographic detector analysis, including factor detection (q-statistic) and interaction detection, to evaluate individual and joint effects on NDE spatial distribution.

Table 4. Factor detection results for NDEs in Hangzhou.

	X1	X2	X3	X4	X5	X6	X7	X8	X9	X10
q value	0.261	0.336	0.115	0.091	0.330	0.390	0.411	0.229	0.172	0.335
p value	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01	0.01

Note: The q value measures explanatory power; all variables are significant at the 1% level.

reports the q-statistics for all ten explanatory variables (ranging from 0.091 to 0.411; all p < 0.01), revealing a clear hierarchy of influences.

Four factors show relatively high explanatory power: innovation level (q = 0.411), science, education, and cultural level (q = 0.390), bus accessibility (q = 0.336), and nighttime light index (q = 0.335). Population density (q = 0.330) is only marginally lower. Thus, NDEs concentrate primarily in areas with strong knowledge spillovers, good public transit connectivity, high economic vitality, and dense markets, consistent with the knowledge-based ecosystem of digital firms, where access to talent and innovation networks outweighs traditional industrial location factors.

Road network density (q = 0.261) and carbon emissions (q = 0.229) play secondary roles. The modest effect of road density suggests that basic road infrastructure is no longer a binding constraint in Hangzhou’s well-served urban area. The moderate q-value for carbon emissions may proxy for legacy industrial zones where digital enterprises are attracted by lower land rents or policy-driven redevelopment, rather than a direct causal link to pollution.

Land use intensity (q = 0.172), industrial land price (q = 0.115), and commercial land price (q = 0.091) have the lowest explanatory power. This low influence of land factors likely stems from digital startups’ small physical footprint, their frequent use of shared or repurposed spaces, and their access to preferential land policies in innovation districts.

4.2.3. Factor Interaction Detection Results

Table 5. Interaction q-values for NDEs in Hangzhou.

Variable	X1	X2	X3	X4	X5	X6	X7	X8	X9	X10
X1	0.261
X2	0.399	0.336
X3	0.332	0.380	0.115
X4	0.288	0.390	0.168	0.091
X5	0.367	0.386	0.376	0.346	0.330
X6	0.431	0.411	0.421	0.418	0.475	0.390
X7	0.482	0.501	0.488	0.434	0.500	0.514	0.411
X8	0.306	0.378	0.271	0.236	0.343	0.414	0.473	0.229
X9	0.275	0.365	0.252	0.188	0.341	0.410	0.433	0.261	0.172
X10	0.372	0.411	0.364	0.339	0.388	0.442	0.481	0.342	0.342	0.335

Note: All pairwise factor interactions displayed bivariate enhancement, indicating that the explanatory power of two factors combined surpasses that of either factor individually.

presents the interaction effects among the ten driving factors. All significant interaction pairs show bivariate enhancement: the joint explanatory power of two factors surpasses that of the stronger factor alone. Hence, NDE spatial patterning stems from synergistic, multiplicative processes. The strongest interaction is between innovation level and science, education, and cultural level (q = 0.514). Land cost factors are weak even in interactions, with X3∩X4 being the lowest at 0.168, reinforcing their marginal role.

In summary, the spatial concentration of NDEs in Hangzhou is best explained by the joint effects of innovation, knowledge infrastructure, market density, and transit connectivity, while land costs remain ineffective even synergistically.

4.3. Spatial Heterogeneity of Influencing Factors Based on MGWR

4.3.1. Comparison of Local Regression Models

Prior to MGWR, we performed OLS regression to examine how the core explanatory variables correlate with the dependent variable. Multicollinearity was tested using variance inflation factors (VIFs), and all VIF values were below 7 (see Appendix A.2), indicating no multicollinearity.

Combining the convergence principle of residual sum of squares (RSS), the AICc criterion, the goodness-of-fit, and the adjusted R², we compared the MGWR and traditional GWR models.

Table 6. Model comparison between GWR and MGWR.

Model Indicators	MGWR	GWR
RSS	1609.640	2075.116
AICc	8572.344	9438.825
R2	0.643	0.539
Adj. R2	0.626	0.530

shows that MGWR has higher goodness-of-fit (R²) both before and after adjustment, lower AICc value, and substantially reduced RSS, indicating improved robustness. The comparison further validates the reliability of MGWR.

4.3.2. Analysis of Spatial Heterogeneity of Influencing Factors

We mapped the MGWR-derived regression coefficients in ArcGIS 10.8. Figure 9 illustrates the resulting spatial patterns, whereas

Table 7. MGWR coefficient descriptive summary.

Variable	Mean	Standard Deviation	Minimum	Median	Maximum	Bandwidth
Intercept	−0.031	0.001	−0.032	−0.031	−0.030	275,688.23
Road network density	0.124	0.000	0.123	0.123	0.124	275,688.23
Bus accessibility	0.039	0.157	−0.581	0.041	1.184	12,165.53
Industrial land price	−0.080	0.001	−0.084	−0.080	−0.077	275,688.23
Commercial and service land price	−0.035	0.001	−0.037	−0.035	−0.034	275,688.23
Population density	−0.112	0.000	−0.112	−0.112	−0.112	275,688.23
Science, education, and cultural level	0.183	0.164	−0.050	0.158	0.495	34,603.16
Innovation level	0.347	0.431	−0.564	0.333	5.266	12,165.53
Carbon emission	0.088	0.003	0.069	0.088	0.092	213,478.96
Land use intensity	−0.098	0.001	−0.100	−0.098	−0.094	275,688.23
Nighttime light index	0.205	0.001	0.202	0.205	0.208	275,688.23

provides the coefficient estimates. Based on bandwidth proportion relative to the global sample, the scale effects are categorized as global scale (ratio > 50%) and local scale (ratio ≤ 50%).

The global drivers influencing the spatial distribution of NDEs include seven indicators: road network density, industrial land price, commercial and service land price, population density, carbon emissions, land use intensity, and the nighttime light index. These factors exhibit relatively strong influence and low spatial heterogeneity.

Across the study area, the nighttime light index yields a significantly positive (coefficients 0.202–0.208, standard deviation 0.001). It is the strongest positively associated driving factor among the global variables, indicating that regions with higher economic vitality are more conducive to the agglomeration of NDEs. Road network density coefficients are significantly positive and highly consistent between the core urban area and peripheral districts. Carbon emissions are significantly positive (coefficients 0.069–0.092, standard deviation 0.003), suggesting that high carbon emissions (e.g., industrial parks, logistics bases) are potential locations where digital enterprises seek collaborative ecosystems.

The regression coefficients of industrial land price and commercial and service land price are significantly negative across the whole region. The coefficient of industrial land price ranges from −0.084 to −0.077, while that of commercial and service land price ranges from −0.037 to −0.034 (standard deviation 0.001). This indicates that higher land prices are less favorable for the establishment of nascent digital enterprises, reflecting a notable cost-squeeze effect. The population density coefficient is significantly negative, confirming a “decentralized” layout pattern of digital enterprises; that is, they tend to avoid overly dense population grids. Land use intensity coefficients are significantly negative; intensively mixed-use areas with rigid spatial functions are less conducive to trial-and-error entry of emerging digital business forms.

Local drivers of NDEs in Hangzhou include bus accessibility, science, education, and cultural level, and innovation level. The influence of bus accessibility on the distribution of NDEs exhibits significant spatial heterogeneity (coefficients −0.581–1.184, standard deviation 0.157). Positive effects occur mainly in the main urban area of Hangzhou (e.g., Shangcheng, Gongshu, eastern Xihu) as well as around Yuhang Future Sci-Tech City and transport hubs such as Xiaoshan Airport, where coefficient values reach 0.634–1.184. This indicates that improved bus systems in these regions effectively promote the agglomeration of digital enterprises. The underlying mechanism is that convenient public transport reduces commuting costs for talent and enhances the attractiveness of young technical workers to firms. Negative effects concentrate in the western mountainous regions including Lin’an, Tonglu, and Chun’an, as well as some peripheral agricultural towns, with coefficient values as low as −0.581 to −0.251. This suggests that in sparsely populated areas with low bus coverage, improvements in bus accessibility may instead be accompanied by ecological protection constraints or inefficient investment, failing to generate positive pull for digital enterprises. Only 6.22% of grids have significant coefficients indicating limited overall explanatory power; statistical significance is restricted to the core urban area and a few transport nodes.

Science, education, and culture level coefficients range from −0.050 to 0.495 (standard deviation 0.164), with a westward-decreasing gradient. Positive effects (coefficients 0.246–0.495) concentrate in the Xiasha Higher Education Zone, areas surrounding Zijingang Campus, and the Binjiang High-tech Zone—all densely endowed with science and education resources. This demonstrates a significant proximity effect of universities and research institutions, where spillovers of high-quality talent and industry–university–research collaboration effectively foster digital startups. Medium-value areas (coefficients 0.102–0.246) cover most of Xihu District, Gongshu District, and some subdistricts of Xiaoshan, reflecting a transitional zone of science–education cultural radiation. Negative or weak effect areas (coefficients −0.050–0.102) occur in Lin’an, western Fuyang, Tonglu, and other outer suburbs, where higher education institutions and research platforms are scarce; marginal increases in the level of science, education, and culture barely translate into locational advantages for digital enterprises. The proportion of significant grids is 29.12%, indicating reliable local influence in the core metropolitan area but decay outward.

Innovation-level coefficients range from −0.564 to 5.266 (standard deviation of 0.431), with extreme spatial heterogeneity. High-value coefficients of 3.215–5.266 appear in patches across innovation-active zones: Future Sci-Tech City (around Alibaba), Binjiang Internet of Things Town, and Xiaoshan Information Port. In mature innovation ecosystems, an increase in innovation level greatly promotes explosive growth of digital enterprises, with elasticity coefficients far exceeding those of other variables, reflecting an “innovation multiplier effect” characteristic of the digital economy era. Medium-value areas (coefficients 0.496–3.215) cover most science and technology parks in the main urban area as well as industrial transition zones in Yuhang and Xiaoshan. Notably, negative-value areas (coefficients −0.564–0.038) appear in the western mountainous area of Lin’an, the Chun’an Qiandao Lake ecological protection zone, and some traditionally declining industrial areas. This suggests that in these regions, “innovation input” lacking a solid innovation foundation may fail to generate actual firm attraction effects and may even produce negative consequences due to resource misallocation. Only 7.19% of grids have significant coefficients, indicating that significant influence is highly concentrated in a few innovation poles and statistically insignificant in peripheral areas.

5. Conclusions and Policy Implications

5.1. Conclusions

This study analyzed the spatiotemporal evolution and driving factors of newly established digital enterprises in Hangzhou from 2010 to 2020, integrating location theory, new economic geography, and sustainable urban theory. The main findings are threefold.

(1)

The spatial pattern of NDEs transformed significantly. It evolved from a “single-core diffusion” pattern in 2010 to a “dual-core emergence” pattern in 2015, and finally to a “dual-core with multiple centers and axial contiguous development” pattern by 2020. Yuhang District emerged as a second growth pole alongside the traditional core. However, spatial inequality widened: the density gap between urban cores and peripheral counties (e.g., Tonglu, Chun’an) persisted, indicating that polycentric development does not automatically ensure balanced sustainability.

(2)

The spatial expansion of NDEs followed a northeast–southwest orientation, with growth intensity decreasing toward the southwest periphery. Over the decade, the overall agglomeration intensity of NDEs in the city increased steadily.

(3)

The spatial distribution of NDEs is shaped by both global and local drivers with distinct scale effects. Global drivers include road network density, land prices, population density, carbon emissions, land use intensity, and the nighttime light index. Among these, economic vitality (the nighttime light index) exerts the strongest positive effect, while land costs and population density show negative effects, reflecting cost-squeeze and decentralized locational preferences.

Local drivers—including bus accessibility, the level of science, education, and culture, and the level of innovation—display strong spatial heterogeneity. The level of innovation has the most pronounced effect, with extremely high positive coefficients in innovation poles such as Future Sci-Tech City and Binjiang IoT Town, but negative effects in ecologically sensitive or deindustrialized areas, revealing an “innovation multiplier effect” alongside risks of resource misallocation.

5.2. Policy Implications

Based on the empirical findings, four policy directions are proposed to foster a balanced and sustainable spatial distribution of digital enterprises.

First, strengthen axial development to convert the dual-core pattern into a networked polycentric structure. Policy should prioritize transport and digital infrastructure investments along the corridor connecting the main urban area and Yuhang District, facilitating the spatial spillover of agglomeration benefits to intermediate zones. This can prevent the dual cores from becoming isolated growth poles and promote inclusive development.

Second, adopt a “decentralized concentration” strategy rather than further densifying the hyper-dense core. Given that digital enterprises tend to avoid high-density areas while thriving in economically vibrant but moderately dense subcenters, zoning policies and tax incentives should encourage cluster formation in well-connected peripheral innovation nodes. Shared office spaces, innovation vouchers, and flexible land use regulations can support such decentralized agglomeration.

Third, target land and infrastructure policies to innovation hotspots instead of applying uniform interventions. The strong local effects of bus accessibility, science–education–culture facilities, and the level of innovation imply that place-based investments yield higher returns. Public transit upgrades should be concentrated around university districts and technology parks. Land subsidies should be selectively offered in areas with demonstrated innovation potential to avoid wasteful spending in regions lacking foundational innovation ecosystems.

Fourth, integrate environmental constraints into location planning for digital enterprises. The positive association between carbon emissions and NDE density in some zones suggests that redeveloping brownfields and legacy industrial sites can attract digital startups through lower rents and existing industrial symbiosis. However, to prevent lock-in of polluting legacies, such redevelopment must be coupled with decarbonization mandates and green infrastructure requirements, aligning digital agglomeration with low-carbon urban transition.

5.3. Research Limitations and Prospects

This study has several limitations that point to future research directions.

First, while our explanatory variables are grounded in location theory, new economic geography, and sustainable urban theory, several important sustainability-related dimensions are omitted. These include job-housing balance, green innovation dynamics, social equity considerations, housing affordability pressures, ecological footprint assessment, and urban resilience to shocks. Incorporating these dimensions would provide a more holistic assessment of how digital enterprise agglomeration interacts with broader sustainability goals.

Second, the study is confined to Hangzhou, a leading digital economy hub in China. The generalizability of our findings to other cities or national contexts requires further comparative research. Cross-case studies involving cities with different economic structures, policy regimes, and spatial forms would help distinguish context-specific from universal drivers.

Third, the MGWR model, though capturing spatial heterogeneity, assumes a linear functional form and predefined bandwidth selection. Future work could apply machine learning methods to accommodate nonlinearities and complex interactions without losing interpretability.