Next Article in Journal
The Nonlinear Vibration Response of Umbrella-Shaped Membrane Structure Under Heavy Rainfall Loads
Previous Article in Journal
Liquefaction Response and Reinforcement Effect of Saturated Soil Treated by Dynamic Compaction Based on Hydro-Mechanically Coupled Explicit Analysis
Previous Article in Special Issue
Investigating the Structural Health of High-Rise Buildings and Its Influencing Factors Using Sentinel-1 Synthetic Aperture Radar Imagery: A Case Study of the Guangzhou–Foshan Metropolitan Area
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Buildings
Volume 15
Issue 14
10.3390/buildings15142528
buildings-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

The Impact of the Built Environment on Innovation Output in High-Density Urban Centres at the Micro-Scale: A Case Study of the G60 S&T Innovation Valley, China

by
Lie Wang
and
Lingyue Li
*
College of Architecture and Urban Planning, Tongji University, Shanghai 200092, China
*
Author to whom correspondence should be addressed.
Buildings 2025, 15(14), 2528; https://doi.org/10.3390/buildings15142528
Submission received: 9 June 2025 / Revised: 11 July 2025 / Accepted: 15 July 2025 / Published: 18 July 2025
(This article belongs to the Special Issue Planning for Sustainable Land Use and Built Environment in High-Density Cities)
Browse Figures
Versions Notes

Abstract

The micro-scale interplay between the built environment and innovation has attracted increasing scholarly attention. However, discussions on how such microdynamics operate and vary across high-density cities remain insufficient. This study focuses on nine high-density urban centres along the G60 S&T Innovation Valley and employs a fine-grained grid unit, viz. 1 km × 1 km, combined with the gradient boosting decision tree (GBDT) model to address these issues. Results show that urban construction density-related variables, including the building density, floor area ratio, and transportation network density, generally rank higher than the amenity density and proximity-related variables. The former contributes 50.90% of the total relative importance in predicting invention patent application density (IPAD), while the latter two contribute 13.64% and 35.46%, respectively. Threshold effect analysis identifies optimal levels for enhancing IPAD. Specifically, the optimal building density is approximately 20%, floor area ratio is 5, and transportation network density is 8 km/km2. Optimal distances to universities, city centres, and transportation hubs are around 1 km, 17 km, and 9 km, respectively. Furthermore, significant city-level heterogeneity was observed: most density-related variables consistently have an overall positive association with IPAD, with metropolitan cities (e.g., Hangzhou and Suzhou) exhibiting notably higher optimal values compared to medium and small cities (e.g., Xuancheng and Huzhou). In contrast, the threshold effects of proximity-related variables on IPAD are more complex and diverse. These findings offer empirical support for enhancing innovation in high-density urban environments.

1. Introduction

As a spatial organisation form of collaborative innovation, the innovation corridor promotes the agglomeration and diffusion of innovation elements by leveraging high-speed transportation networks [1]. Representative examples include the belt-shaped areas formed along California’s Route 101 (Silicon Valley) and Boston’s Route 128 in the mid-20th century [2]. In recent years, governments around the world have increasingly sought to replicate such successful practices through the implementation of innovation corridor initiatives, such as Malaysia’s Multimedia Super Corridor [3], the Cascadia Innovation Corridor [4], and the Toronto–Waterloo Innovation Corridor [5]. In China, since “collaborative innovation” was elevated to a national strategy in 2011, the development of innovation corridors has been actively pursued nationwide. Among them, the G60 S&T Innovation Valley in the Yangtze River Delta (YRD) stands out as one of the earliest and most representative examples [1], aiming to cultivate a world-class innovation cluster by integrating innovation resources across cities through expressways and high-speed railway networks. While innovation corridors have garnered extensive academic attention in recent years, for example, in terms of their network structure, collaborative patterns, and clustering dynamics [1,3,5], systematic investigation into the determinants of their innovative performance remains inadequate and warrants deeper exploration.
Innovation output, typically measured by indicators such as patents, scientific publications, and the emergence of new products or technologies [6,7], is widely recognised as a key indicator of economic growth and regional competitiveness [8]. Two critical research gaps persist in understanding the determinants of innovation output. First, existing studies examining innovation output are generally at a regional or national scale [6,7,8], seldom differentiating urban and rural areas. Indeed, urban centres, characterised by a high concentration of skilled labour, firms, knowledge resources, and robust infrastructure [9], as well as more diverse opportunities for interaction and collaboration [10], are often engines of innovation [11]. But, the nuanced features and dynamics of urban centres, where innovative activity is highly concentrated [12,13], are often overlooked. Second, there is limited understanding of the heterogeneity in innovation drivers among different types of cities [14]. Particularly in complex innovation networks, such as the Yangtze River Delta (YRD), substantial differences in factors such as urban scale (e.g., metropolitan vs. medium and small cities), geographical location, industrial structure, and policy orientation can lead to divergent innovation performance [15,16]. Uncovering these differences is crucial for formulating tailored and effective innovation policies [17].
Among the various factors influencing innovation output, the effects of the built environment, particularly on innovation spatial distribution and agglomeration patterns, are increasingly important [18]. The built environment normally refers to various spatial elements, such as land use, transportation systems, supporting infrastructure, service facilities, employment, and population density [19,20,21,22,23,24], which are constructed or modified by human activities [18]. These elements collectively shape the vitality and functional layout of cities, which not only provides a container for urban innovation but also facilitates the generation and diffusion of innovation activities by fostering interaction and efficient resource utilisation [23,25,26,27]. So far, comprehensive quantitative analyses of how the built environment associates with innovation outputs at finer micro-scales (e.g., at the level of street blocks, grid cells, or land parcels) are limited. Micro-scale features of the built environment, such as street compactness, walkability, commuting distance, functional diversity, and facility accessibility, are more relevant to human sensory experience and daily interaction [28,29], and impact the occurrence and development of innovation activities through agglomeration and proximity effects [23,30]. A few recent studies have adopted 1 km × 1 km grids [23,31,32,33] and tested their validity in capturing the nuanced interactions between innovation activities and the surrounding built environment [23]. This finer scale will be used in this study to identify the micro-scale impacts of the built environment on innovation outputs in the urban centres of the G60 S&T Innovation Valley.

2. Literature Review

Previous empirical studies have consistently adopted the widely used “3Ds” or “5Ds” framework (including density, diversity, design, destination accessibility, and transportation distance) proposed by Cervero et al. [34] and Ewing et al. [24] to evaluate the impact of built environment on innovation, but there is still no unified paradigm for translating these indicators to the micro-scale. Some scholars emphasise that micro-scale built environment studies should focus more on pedestrians’ perceptions of their surroundings [21,28], such as the perceived intensity of human activity and the perceived distance to complementary resources or collaborators [35,36]. In this regard, we categorise the micro-scale built environmental factors influencing innovation into two pedestrian-perceived spatial dimensions: urban density and geographical proximity.
Urban density quantifies the intensity of spatial occupation and use [37]. It comprises two interrelated aspects: construction density, which can be measured by indicators such as land use intensity, building density, floor area ratio, and road density [18,21] and reflects the quantitative aspects of physical development in urban spaces, and amenity density, which can be measured by key living amenity density (e.g., coffee shops, restaurants, hotels, convenience stores, supermarkets, and bars) [18,19,20,23], which represents the qualitative aspects of various service provisions in urban spaces. For construction density, compact urban forms, mixed land uses, and high-quality public transportation systems collectively foster social interactions and knowledge spillovers, further driving innovation [25,26,27]. For instance, Hamidi et al. [25] emphasise that high-density urban environments promote interpersonal interactions and resource sharing, thereby boosting innovation. Similarly, Berkes et al. [10] argue that density serves as a crucial catalyst for knowledge diffusion, as informal interactions in high-density areas greatly facilitate unconventional innovation. For the amenity density, the layout of high-density urban amenities plays a crucial role in attracting skilled labour, improving quality of life and work efficiency, and indirectly fostering innovation [38], especially in urban centre areas [39] and more developed cities [40].
Geographical proximity captures the accessibility between innovation entities and transportation stations or key living and working places (such as city centres, parks, etc.) [18,19,22,27], and is regarded as a vital determinant influencing behavioural patterns, resource allocation, and spatial equity within the built environment [41,42]. For example, Credit et al. [43] found that proximity to light rail transit stations led to an 88% increase in Phoenix’s knowledge sector startups, which surpassed the service and retail sectors by 40% and 28%, respectively. In addition, the geographical proximity among various innovation entities (such as universities, enterprises, and institutions) promotes the flow, transfer, and sharing of knowledge [44] and is also considered crucial for inter-organisational collaboration and innovation [45].
While the positive effects of high urban density and geographical proximity on innovation are well established, growing evidence suggests that excessive density or proximity may negatively impact innovative performance [38]. For instance, Duranton et al. [46] highlight that excessive urban density may lead to negative externalities such as pollution and congestion, which may hinder innovation. Similarly, Boschma et al. [35] warn that excessive geographical proximity may hinder innovation due to “lock-in effects”, and emphasise that “too much and too little proximity are detrimental to innovation.” Boschma et al. [47] further argue that for each dimension of proximity, including cognitive, organisational, social, institutional, and geographical, there is an optimal level of distance at which interactive learning and innovation are maximised.
However, existing studies offer no clear guidance on the “optimal levels” of density and proximity for fostering innovation. One important reason is that traditional regression models have difficulty in precisely capturing such optimal levels [48]. Recently, with the development of new technology for big data analysis, machine learning methods like gradient boosting decision trees (GBDTs) have performed well in capturing complex threshold effects [49] and are increasingly being applied to explore the intricate dynamics that influence innovation [50].
Based on the above discussions, this study investigates nine high-density urban centres within the G60 S&T Innovation Valley of the Yangtze River Delta (YRD) in China, with 1 km × 1 km grids adopted as the micro-scale analytical unit. The GBDT model is employed to estimate the relative importance of built environment factors and to detect their non-linear effects, in general, and threshold effects, in particular, on innovation output. Specifically, this study aims to (1) identify the optimal values of density and proximity-related building environmental factors on innovation output at the micro scale and (2) reveal city-level heterogeneity in these optimal values across nine high-density urban centres.

3. Research Area, Data, and Method

3.1. Research Area

The G60 S&T Innovation Valley is located in the central area of the YRD in China, including nine prefecture-level cities along the G60 national expressway and the Shanghai–Suzhou–Huzhou–Hefei high-speed railway, as specified in the Development Plan for the G60 S&T Innovation Valley of YRD, which are Songjiang in Shanghai, Suzhou in Jiangsu, and Hangzhou, Huzhou, Jiaxing, and Jinhua in Zhejiang, as well as Hefei, Wuhu, and Xuancheng in Anhui, covering a total administrative area of 76,200 km2. By the end of 2020, the combined GDP of these nine cities made up one-fifteenth of the national total, and they hosted 36,500 high-tech enterprises, accounting for one-tenth of the national total.
To delineate the “high-density urban centres” of the above nine cities, we adopted the Statistical Area Codes and Urban-Rural Classification Codes (2023) issued by the National Bureau of Statistics as a unified standard. This coding system assigns unique codes to each administrative unit at various levels (province, city, county, township, and village), and distinguishes urban and rural areas by assigning codes to village-level units, including administrative villages (rural) and residential communities (urban). Based on this system, if more than 80% of a township’s village-level units are assigned code 111, the township is identified as part of the urban centre. All such identified township-level units are then aggregated to delineate the final urban centres of the nine prefecture-level cities (Figure 1). Then, the research area was further divided into 1 km × 1 km grids, with natural areas such as mountains and water bodies excluded. As a result, 2553 valid analytical grids were delineated, accounting for 3.13% of the total administrative area.

3.2. Research Framework

Patent applications reflect both the timeliness of new technological creation and innovation quality, and are widely regarded as the most definitive measures of a region’s innovation capacity [23,33]. Notably, they include detailed spatial information, such as the applicant’s address, making them an optimal indicator of innovation output in grid-level analysis [18,31], especially at 1 km × 1 km grids [23,33]. Thus, this study uses the number of invention patent applications within each grid to measure innovation output, defining the dependent variable as invention patent application density (IPAD).
For independent variables, we focus on two key dimensions: density and proximity, which are the most relevant to the micro-scale built environment. Drawing on grid-level literature and data availability, we selected nine key variables, categorised into three sub-dimensions: (1) urban construction density, represented by building density, floor area ratio, and transportation network density. Building density reflects building aggregation, neighbourhood compactness, community cohesion, and potential for social interaction and economic vitality [51]. In contrast, the floor area ratio shapes the 3D characteristics of urban form, serving as a key indicator of land use efficiency [52]. Transportation network density reflects the spatial distribution and connectivity of transportation infrastructure, influencing mobility and accessibility [18,33]. Together, they collectively shape the physical form of urban environments. (2) Amenity density, measured by amenity counts within each grid, includes coffee shops reflecting urban vitality and daily social interaction [20,38], convenience stores and supermarkets proxying urban living convenience [18], and public parking lots as mobility-supporting amenities for daily travel. Together, these amenities indirectly support innovation by enhancing the quality of life and fostering social interaction. (3) Proximity-related variables capture accessibility to key innovation-driving nodes, including city centres as internal cores for employment, resources, and markets [20], high-speed railway stations and airports as channels for external connectivity [43], and universities as sources of knowledge spillovers [53].
Based on these, this study proposes a framework to guide the empirical analysis (Figure 2), encompassing the dependent and independent variables outlined above.

3.3. Data Sources

Data on invention patent applications for 2020 were retrieved from the Patent Search and Analysis System of the Shanghai Intellectual Property Information Service Platform. In total, 102,884 patent applications were collected, accounting for 55.91% of the total administrative area. Data on built environment variables for the same year were compiled from different sources. Specifically, indicators of building density and floor area ratio were derived from the latest building footprint dataset developed by Che et al. [54]. Transportation network density was calculated using road data filtered from the open-source platform OpenStreetMap (OSM). Points of interest (POI) data, including coffee shops, convenience stores, supermarkets, and public parking lots, as well as key urban nodes in prefecture-level city centres, China Railway High-Speed (CRH) stations, airports, and universities, were all sourced from the Amap platform and processed into corresponding density and proximity indicators. Table 1 summarises these variables along with their descriptive statistics, while Figure 3 depicts their spatial distribution.

3.4. Method

The GBDT model is applied in this study to detect the non-linear and threshold effects of the selected built environment variables on innovation output. GBDT, as a machine learning method based on classification and regression decision trees, works by integrating multiple decision trees in an iterative process, where each tree is trained to minimise the residuals. This process gradually reduces the error between predicted and actual values, thereby optimising the overall accuracy [22]. Recently, GBDT has been widely applied to detecting driving factors in fields such as transportation geography [55,56] and economic geography [57,58], yet it has been scarcely applied to detecting innovation-related drivers. In this study, we employed the GBDT method for two main reasons: (1) the method can evaluate the relative importance of each independent variables [59], thereby helping to identify critical built environmental factors; (2) the method does not predetermine any linear assumptions between predictors and the target [55], allowing partial dependence plots (PDPs) to illustrate the non-linear associations between built environment variables and innovation output [60].
Despite its advantages, GBDT has certain inherent limitations, such as the inability to directly perform significance testing or calculate confidence intervals for coefficients [55]. Following previous research [55,57,58], this study employed a five-fold cross-validation approach and optimised the model parameters to enhance overall stability and predictive performance. As depicted in Figure 4, the dataset was split into five mutually exclusive subsets, each containing 20% of the data. In each iteration, one subset served as the test set, while the remaining four were used for training the model. Moreover, in this study, the original data of patent applications exhibited a severe right-skewed distribution. Based on existing research [61], to mitigate the effect of data skewness on model performance, we applied a logarithmic transformation to the predicted variable (IPAD). Regarding parameter configuration, the final model was configured by referencing previous studies [55,60,62], with a maximum of 10,000 trees, a learning rate of 0.001, and a tree depth of 35.

4. Results

4.1. The Relative Importance and Threshold Effects of Independent Variables Across All Urban Centres

The GBDT model converged after 3130 iterations with a Pseudo R2 of 0.611 for the sample across all urban centres. Table 2 reports the relative importance and rankings of built environment variables in predicting IPAD. The total relative influence of all independent variables adds up to 100% [62].
Urban construction density is the most important sub-dimension in predicting IPAD, accounting for a total relative importance of 50.90%, indicating that the physical form and spatial structural characteristics of urban environments play a crucial role in fostering innovation outputs. Although all three constituent variables are highly contributing, there are notable differences between them. Building density (BD) accounts for 22.68% of the relative importance, significantly higher than floor area ratio (FAR) and transportation network density (TND), at 14.84% and 13.38%, respectively. Figure 5 shows that the three variables exhibit consistent positive associations with IPAD across various effective ranges. The difference is that BD exhibits a steeper positive association with IPAD, characterised by a substantial increase in IPAD when BD rises from 10% to 20%, and then undergoes slight fluctuations before stabilising. In contrast, IPAD surges sharply as FAR increases from 0 to 1, exhibits a gradual fluctuating upward trend across FAR values from 1 to 5, and then undergoes a pronounced up–down fluctuation before levelling off. As for TND, IPAD increases noticeably as TND rises from 0 to 8 km/km2, then slightly declines, and the association plateaus as TND approaches 15 km/km2, indicating no further effect [55,62].
The collective relative importance of amenity density is limited, with an overall contribution of only 13.64% to IPAD. The three associated variables, including coffee shop density (CSD), convenience store and supermarket density (CSSD), and public parking density (PPD), each exhibit a relative importance below 6%, placing them at the bottom of the ranking. Figure 6 shows that the three variables exhibit a significant difference in their non-linear associations with IPAD. The effects of CSD and PPD are similar: both are positively correlated with an increase in IPAD before reaching their respective plateaus of 11 shops/km2 and 80 lots/km2. In sharp contrast, as CSSD gradually increased from 0 to 40 shops/km2, IPAD unexpectedly suffered an atypical decline. A review of grids with high CSSD reveals that retail services are mainly clustered in residential areas, contrasting with commercial districts or high-tech parks where patents are most concentrated. Particularly in Hangzhou, the Binjiang High-tech Zone and Future Science City produce over 20% of the city’s patents but account for only 5% of its supermarkets and convenience stores.
Variables related to proximity to key urban nodes collectively contribute 35.46% of the total relative importance. Three constituent variables (proximity to prefecture-level city centres (PPCC), proximity to CRH stations/airports (PCA), and proximity to universities (PU)) exhibit nearly identical relative importance, at 11.35%, 10.17%, and 13.94%, respectively. This may indicate that multiple urban core nodes, including internal resources and markets, external connections, and knowledge production, exert consistent and indispensable importance in driving innovation output. However, the threshold effects of the three variables on IPAD differ significantly, as reported in Figure 7. PPCC shows an overall positive association with IPAD in the range of 0–17 km. As this distance continues to increase toward 20 km, IPAD experiences a decline, followed by a gradual plateau. The non-linear association between PCA and IPAD resembles a peak-shaped curve, with IPAD reaching its peak at around 9 km. In comparison, PU is negatively associated with IPAD, which drops sharply within a range of 1 to 5 km from universities. This may suggest that urban innovation’s spatial reliance on resources and connectivity manifests within specific distance thresholds, while its spatial stickiness toward higher education depends more on localised knowledge spillovers.

4.2. The Relative Importance and Threshold Effects of Independent Variables Across Respective Urban Centres

Table 3 reveals significant city-level heterogeneity in the relative importance of built environment variables in predicting IPAD, and Figure 8 presents their distinct threshold effects.
In Wuhu and Songjiang, the overall relative importance of urban construction density exceeds 60%. In Hefei, Suzhou, and Jinhua, this aggregate metric ranges between 40% and 50%. In Xuancheng and Huzhou, however, the results are both below 30%. Specifically, BD exhibits the highest relative importance in Hefei, Wuhu, and Songjiang, at 25.47%, 24.62%, and 28.46%, respectively. While BD is positively associated with IPAD in most cities, the threshold effects vary considerably. The optimal BD value is approximately 30% in Hefei and Wuhu, around 20% in Suzhou, Songjiang, Jiaxing, and Jinhua, and only about 15% in Xuancheng. Furthermore, FAR’s relative importance exceeds 20% in Wuhu, Suzhou, and Songjiang. Similar to BD, FAR is positively associated with IPAD in most cities. The optimal FAR value is approximately 8 in Hangzhou and Jiaxing, around 6 in Songjiang, about 4 in Suzhou and Jinhua, and less than 2 in Wuhu. Regarding TND, its relative importance exceeds 10% in Hefei, Suzhou, Hangzhou, and Jinhua. The optimal TND value in these cities is consistently around 8 km/km2, aligning with the value observed across all urban centre samples.
The total relative importance of amenity density is less than 20% in most cities, except for Xuancheng and Hangzhou, where it is 37.55% and 24.16%, respectively. Notably, despite having low relative importance in most cities, CSD is positively associated with IPAD across all cities, suggesting that enhancing spaces for daily interaction may universally support innovation, regardless of urban context. For CSSD, its relative importance is below 10% in all cities, and the threshold effects vary significantly across cities. In Hefei, Wuhu, Suzhou, and Hangzhou, CSSD is negatively associated with IPAD. In contrast, it shows a positive association in Xuancheng, Jiaxing, and Jinhua. Meanwhile, in Songjiang and Huzhou, this association follows a U-shaped curve. PPD has a positive association with IPAD in all cities except Songjiang and Jiaxing. Its relative importance exceeds 10% only in Xuancheng and Hangzhou, with optimal values of 5 lots/km2 and 25 lots/km2, respectively.
The contribution of proximity to key urban nodes to predicting IPAD mostly ranges between 30% and 50%. Specifically, the relative importance of PPCC exceeds 10% in Hefei, Wuhu, Suzhou, Jiaxing, Hangzhou, and Jinhua, and the threshold effects differ significantly across these cities. In Hefei and Jinhua, IPAD remains steadily high within 2 km of the city centre but declines continuously beyond this threshold. This may suggest that innovation output in these cities relies more heavily on the radiative influence of the city centre. While Jiaxing and Hangzhou also exhibit a similar reliance on central proximity, as distance continues to increase, IPAD rebounds, peaking at approximately 8 km and 16 km, respectively. This U-shaped curve suggests a dual reliance of innovation on both central urban resources and peripheral innovation hubs. In contrast, in Wuhu and Suzhou, the non-linear associations between PPCC and IPAD resemble a peak-shaped curve, with IPAD reaching its peak at 11 km and 17 km, respectively, implying a single dependence on peripheral innovation hubs. Moreover, the non-linear association between PCA and IPAD also exhibits similar complexity. In metropolitan cities such as Hefei, Suzhou, and Hangzhou, IPAD consistently reaches its optimal distance approximately 10 km from external transportation hubs, consistent with the value observed across all urban centre samples. In contrast, for medium and small cities like Songjiang, Jiaxing, and Jinhua, this optimal distance typically falls within the 5–7 km range. PU is positively associated with IPAD in Wuhu, Suzhou, Songjiang, and Jinhua, with the optimal values being approximately 25 km, 17 km, 7 km, and 8 km, respectively. In contrast, in Jiaxing, Huzhou, and Hangzhou, PU shows a negative association with IPAD, consistent with the trend in all urban centre samples, where IPAD continues to decline within the 1–5 km range from universities. Hefei and Xuancheng exhibit a more complex fluctuating pattern, with IPAD reaching a distinct trough around 5–6 km from universities, forming a U-shaped curve.

5. Discussion and Policy Implications

The built environment provides the necessary physical space and resource support for urban innovation activities, serving as an essential condition for the generation and diffusion of innovation [23,26,27]. While previous studies have pointed out that only “moderate” built environment conditions are conducive to innovation [38,46,47,63], there remain substantial limitations in capturing these “optimal levels” due to the constraints of conventional linear regression methods [48]. In this study, we innovatively applied a GBDT model to detect the non-linear and threshold effects of density-related and proximity-related built environment variables on innovation output at the micro-scale.
Consistent with previous studies [12,18,38,39], our study confirms that moderate levels of density and proximity facilitate knowledge spillovers and interactions, which, in turn, enhance innovation output. As emphasised by Fitjar et al. [45] and Hua et al. [64], the relationship between density and proximity should be “neither too high nor too low” and “neither too close nor too far.” Notably, exceeding prior studies, our study investigates threshold effects and city-level heterogeneity in built environment variables using both aggregated and city-specific samples.
Across all urban centres, urban construction density is the most important built environment sub-dimension, with BD, FAR, and TND having respective optimal values of approximately 20%, 5, and 8 km/km2, all of which are approximately 1.5 times higher than the current average levels of 15.29%, 3.57, and 5.14 km/km2. This finding aligns with previous studies by Hamidi et al. [26] and Berkes et al. [10], which suggest that compact, high-density urban areas encourage social interactions and provide more convenient facilities, thereby promoting innovation. The three amenity density variables, CSD, CSSD, and PPD, each exhibit a relative importance of less than 6% in predicting IPAD. This may suggest that although these amenities improve urban livability and facilitate social interaction [19], their predictive contribution to patents remains relatively weak within the G60 S&T Innovation Valley. The “optimal distance” to city centres and external transportation hubs is around 17 km and 9 km, reflecting a moderate distance that enhances innovation output. Additionally, proximity to universities exerts the strongest influence on innovation output within a relatively close radius (e.g., 1 or 2 km), which is consistent with findings from Bereitschaft et al. [27] regarding Omaha’s downtown cluster.
Across respective urban centres, we confirm that there is significant city-level heterogeneity in the optimal values of built environment variables. As noted by Fan et al. [16], the impact of variables on innovation shows heterogeneity across cities due to specific city characteristics such as urban scale, industrial base, and economic structure. In our study, except for CSSD, all five other density-related variables consistently show an overall positive association with IPAD, indicating that increasing urban construction density and amenity density generally support innovation output [10,25,38]. However, the threshold effects of these variables vary significantly across different urban contexts. For instance, in Hefei, Wuhu, Suzhou, and Songjiang, urban construction density plays a more important role, with optimal values generally higher than those observed in smaller cities like Xuancheng and Jinhua. This may suggest that larger cities with more diverse economic activities may benefit from higher density [65]. In contrast, the non-linear associations between proximity-related variables and IPAD tend to be more complex. For example, in Hangzhou, proximity to the city centre and universities is crucial for innovation, with optimal distances generally within 2 km. However, in Suzhou, the situation is quite the opposite, with optimal distances to these key nodes being around 16 km. This suggests that even cities of similar urban scale may exhibit differences in the role of proximity on urban innovation due to variations in their industrial bases, urban spatial layouts, and other factors. These findings highlight the complexity of proximity effects on innovation [66], emphasising that urban innovation strategies should be tailored to the unique city context.
The G60 S&T Innovation Valley represents a crucial innovation cluster within the YRD, serving as a pioneering region for scientific and technological advancement in China. This study focuses on the nine cities along the G60 S&T Innovation Valley, aiming to identify the zones within these cities that are most suitable for the development of innovation industries, as well as to detect the optimal built environment conditions for fostering innovation in specific urban zones. Based on the empirical findings, we propose the following suggestions for this specific case:
First, in Hefei, Wuhu, Suzhou, Songjiang, Hangzhou, and Jinhua, urban construction density is the most important driver of innovation output. Although the optimal BD, FAR, and TND values are slightly higher than their respective current average levels in these cities, they tend to be greater in larger cities compared to smaller ones. For example, in metropolitan cities like Hangzhou, the optimal BD, FAR, and TND values are 35%, 8, and 11 km/km2, compared to current average levels of 21.94%, 5.45, and 5.78 km/km2. Similarly, Hefei’s optimal BD, FAR, and TND values are 30%, 6, and 8 km/km2, against current average levels of 14.32%, 2.95, and 6.52 km/km2. By contrast, in medium and small cities like Jinhua, the optimal BD, FAR, and TND values are only 20%, 4, and 7 km/km2, respectively, even though they remain slightly higher than the current average levels of 16.47%, 3.01, and 5.12 km/km2. Thus, it is advisable to moderately raise the upper limits of construction intensity in urban renewal and new development areas for cities of different urban scales: 20–30% for BD, 4–6 for FAR, and 8–10 km/km2 for TND in metropolitan cities, and 15–20%, 3–4, and 6–8 km/km2, respectively, in medium and small cities.
Second, the results show that amenity density, particularly CSD, is positively associated with innovation output in most cities. Thus, local governments are encouraged to develop third spaces such as cafés, co-working spaces, and social venues, with a target density of approximately 5–11 shops/km2, to strengthen informal interactions and foster a vibrant atmosphere conducive to innovation. However, in metropolitan cities like Hefei, Suzhou, and Hangzhou, CSSD is negatively associated with innovation output. This may stem from the “encroachment effect” of the urban service industry or indicate spatial separation between innovation zones and leisure spaces. Local governments should therefore prioritise policy guidance to promote coordinated development between innovation activities and supporting facilities.
Third, in Hangzhou and Hefei, innovation output shows a clear proximity dependence on city centres and universities, typically within a 2 km radius. Therefore, it is recommended that focus be placed on developing innovation clusters within central urban areas, anchored by universities. In contrast, Suzhou exhibits innovative responsiveness to city centres, transportation hubs, and universities primarily within a 10–20 km radius. Accordingly, innovation resources should be directed toward suburban high-tech zones, with improved connectivity to high-speed rail stations, university towns, and other key nodes, fostering a multi-node innovation spatial structure. Medium and small cities, however, exhibit more fragmented proximity patterns, with no consistent optimal distance or stable centre-periphery structure. This may reflect the lack of a mature core-driven system in these cities, with innovation more reliant on coordination among local functional zones. To address this, such cities should promote mechanisms like the “15 min innovation circle” to facilitate the efficient flow and interaction of local innovation elements.

6. Conclusions

This study explores the micro-scale conditions through which the built environment fosters innovation output in high-density urban areas, specifically focusing on nine urban centres along the G60 S&T Innovation Valley. By using 1 km × 1 km grids and the gradient boosting decision tree (GBDT) model, the research estimates the relative importance and threshold effects of built environment factors on innovation output.
Results show that urban construction density exerts the most significant impact on innovation output, followed by proximity to key urban nodes, while amenity density plays a relatively minor role. Specifically, building density (22.68%), floor area ratio (14.84%), and transportation network density (13.38%), ranking first, second, and fourth, collectively contributing 50.90% to the prediction of invention patent application density (IPAD), with optimal values identified at approximately 20%, 5, and 8 km/km2. Proximity to universities (13.94%), prefecture-level city centres (11.35%), and CRH stations/airports (10.17%), ranking third, fifth, and sixth, collectively contribute 35.46% to IPAD. Threshold effect analysis identifies their optimal distances as approximately 17 km, 9 km, and 1 km, respectively. Amenity density, encompassing coffee shop density (2.46%), convenience store and supermarket density (5.43%), and public parking density (5.75%), has a total relative importance of only 13.64%. Notably, significant city-level heterogeneity is observed in the impacts of built environment factors. Most density-related variables exhibit an overall positive association with IPAD. Metropolitan cities, such as Hangzhou and Suzhou, exhibit generally higher optimal density values compared to medium and small cities like Xuancheng and Huzhou. In contrast, the threshold effects of proximity-related variables on IPAD are more complex and diverse.
Notably, this study has potential limitations. First, the empirical analysis is based on cross-sectional data from a single year, which may limit the temporal generalizability and robustness of the results. Second, innovation output is measured solely by the number of patent applications. While patents are a widely used and accessible indicator, they may not fully reflect regional innovation output, especially in the context of high-density urban centres with more complex spatial structures and highly mixed functions, where the forms of innovation are more diverse. Future research could employ multi-period panel data to explore the dynamic interplay between built environment factors and innovation over time, and incorporate multiple indicators such as academic publications and new product releases to capture the multifaceted nature of innovation in high-density urban contexts.

Author Contributions

Conceptualisation, L.L. and L.W.; methodology, L.W.; software, L.W.; validation, L.L. and L.W.; formal analysis, L.W.; investigation, L.L. and L.W.; resources, L.L.; data curation, L.W.; writing—original draft preparation, L.W.; writing—review and editing, L.L. and L.W.; visualisation, L.W.; supervision, L.L.; project administration, L.L.; funding acquisition, L.L. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the China National Key R&D Program, grant number 2023YFC3804001.

Data Availability Statement

Data are available on request due to privacy restrictions. The data presented in this study are available on request from the corresponding author.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

The following abbreviations are used in this manuscript:
S&TScience and technology
GBDTGradient boosting decision tree
YRDYangtze River Delta
CRHChina Railway High-Speed
IPADInvention patent application density
BDBuilding density
FARFloor area ratio
TNDTransportation network density
CSDCoffee shop density
CSSDConvenience store and supermarket density
PPDPublic parking density
PPCCProximity to prefecture-level city centres
PCAProximity to CRH stations/airports
PUProximity to universities

References

  1. Guo, Y.; Chen, P.; Wan, Y.; Zhu, Y.; Cao, X.; Zeng, G. Technology transfer in asymmetric innovation corridors: Theory and empirical evidence from China. Appl. Geogr. 2025, 178, 103599. [Google Scholar] [CrossRef]
  2. Saxenian, A. Regional Advantage: Culture and Competition in Silicon Valley and Route 128; Harvard University Press: Cambridge, MA, USA, 1994. [Google Scholar]
  3. Lepawsky, J. Clustering as Anti-politics Machine? Situating the Politics of Regional Economic Development and Malaysia’s Multimedia Super Corridor. Reg. Stud. 2009, 43, 463–478. [Google Scholar] [CrossRef]
  4. Cappellano, F.; Richardson, K.; Trautman, L. Cross border regional planning: Insights from Cascadia. Int. Plan. Stud. 2021, 26, 182–197. [Google Scholar] [CrossRef]
  5. Wachsmuth, D.; Kilfoil, P. Two logics of regionalism: The development of a regional imaginary in the Toronto–Waterloo Innovation Corridor. Reg. Stud. 2021, 55, 63–76. [Google Scholar] [CrossRef]
  6. Schwartz, M.; Peglow, F.; Fritsch, M.; Günther, J. What drives innovation output from subsidized R&D cooperation?—Project-level evidence from Germany. Technovation 2012, 32, 358–369. [Google Scholar] [CrossRef]
  7. Block, J.H.; Spiegel, F. Family firm density and regional innovation output: An exploratory analysis. J. Fam. Bus. Strategy 2013, 4, 270–280. [Google Scholar] [CrossRef]
  8. Hasan, I.; Tucci, C.L. The innovation–economic growth nexus: Global evidence. Res. Policy 2010, 39, 1264–1276. [Google Scholar] [CrossRef]
  9. Florida, R.; Adler, P.; Mellander, C. The city as innovation machine. Reg. Stud. 2017, 51, 86–96. [Google Scholar] [CrossRef]
  10. Berkes, E.; Gaetani, R. The Geography of Unconventional Innovation. Econ. J. 2021, 131, 1466–1514. [Google Scholar] [CrossRef]
  11. Dvir, R.; Pasher, E. Innovation engines for knowledge cities: An innovation ecology perspective. J. Knowl. Manag. 2004, 8, 16–27. [Google Scholar] [CrossRef]
  12. Jaffe, A.B.; Trajtenberg, M.; Henderson, R. Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations. Q. J. Econ. 1993, 108, 577–598. [Google Scholar] [CrossRef]
  13. Mascarini, S.; Garcia, R.; Quatraro, F. Local knowledge spillovers and the effects of related and unrelated variety on the novelty of innovation. Reg. Stud. 2023, 57, 1666–1680. [Google Scholar] [CrossRef]
  14. Zhao, K.; Li, W. High-Speed Railway Network, City Heterogeneity, and City Innovation. Sustainability 2023, 15, 15648. [Google Scholar] [CrossRef]
  15. Fritsch, M.; Wyrwich, M. Does Successful Innovation Require Large Urban Areas? Germany as a Counterexample. Econ. Geogr. 2021, 97, 284–308. [Google Scholar] [CrossRef]
  16. Fan, X.; Xu, Y. Does high-speed railway promote urban innovation? Evidence from China. Socio-Econ. Plan. Sci. 2023, 86, 101464. [Google Scholar] [CrossRef]
  17. Duan, J.; Zhao, Z.; Xu, Y.; You, X.; Yang, F.; Chen, G. Spatial Distribution Characteristics and Driving Factors of Little Giant Enterprises in China’s Megacity Clusters Based on Random Forest and MGWR. Land 2024, 13, 1105. [Google Scholar] [CrossRef]
  18. Wu, K.; Wang, Y.; Zhang, H.; Liu, Y.; Ye, Y. Impact of the Built Environment on the Spatial Heterogeneity of Regional Innovation Productivity: Evidence from the Pearl River Delta, China. Chin. Geogr. Sci. 2021, 31, 413–428. [Google Scholar] [CrossRef]
  19. Mouratidis, K.; Poortinga, W. Built environment, urban vitality and social cohesion: Do vibrant neighborhoods foster strong communities? Landsc. Urban Plan. 2020, 204, 103951. [Google Scholar] [CrossRef]
  20. Mouratidis, K. Built environment and social well-being: How does urban form affect social life and personal relationships? Cities 2018, 74, 7–20. [Google Scholar] [CrossRef]
  21. Chen, L.; Lu, Y.; Ye, Y.; Xiao, Y.; Yang, L. Examining the association between the built environment and pedestrian volume using street view images. Cities 2022, 127, 103734. [Google Scholar] [CrossRef]
  22. Yang, W.; Fei, J.; Li, Y.; Chen, H.; Liu, Y. Unraveling nonlinear and interaction effects of multilevel built environment features on outdoor jogging with explainable machine learning. Cities 2024, 147, 104813. [Google Scholar] [CrossRef]
  23. Jiang, H.; Xiong, W. The Impact of Land-Use Mix on Technological Innovation: Evidence from a Grid-Cell-Level Analysis of Shanghai, China. Land 2024, 13, 462. [Google Scholar] [CrossRef]
  24. Ewing, R.; Cervero, R. Travel and the Built Environment: A meta-analysis. J. Am. Plan. Assoc. 2010, 76, 265–294. [Google Scholar] [CrossRef]
  25. Hamidi, S.; Zandiatashbar, A. Does urban form matter for innovation productivity? A national multi-level study of the association between neighbourhood innovation capacity and urban sprawl. Urban Stud. 2018, 56, 1576–1594. [Google Scholar] [CrossRef]
  26. Hamidi, S.; Zandiatashbar, A.; Bonakdar, A. The relationship between regional compactness and regional innovation capacity (RIC): Empirical evidence from a national study. Technol. Forecast. Soc. Change 2019, 142, 394–402. [Google Scholar] [CrossRef]
  27. Bereitschaft, B. Are walkable places tech incubators? Evidence from Nebraska’s ‘Silicon Prairie’. Reg. Stud. Reg. Sci. 2019, 6, 339–356. [Google Scholar] [CrossRef]
  28. Xu, G.; Zhong, L.; Wu, F.; Zhang, Y.; Zhang, Z. Impacts of Micro-Scale Built Environment Features on Tourists’ Walking Behaviors in Historic Streets: Insights from Wudaoying Hutong, China. Buildings 2022, 12, 2248. [Google Scholar] [CrossRef]
  29. Steinmetz-Wood, M.; El-Geneidy, A.; Ross, N.A. Moving to policy-amenable options for built environment research: The role of micro-scale neighborhood environment in promoting walking. Health Place 2020, 66, 102462. [Google Scholar] [CrossRef] [PubMed]
  30. Wood, S.; Dovey, K. Creative Multiplicities: Urban Morphologies of Creative Clustering. J. Urban Des. 2015, 20, 52–74. [Google Scholar] [CrossRef]
  31. Li, L.; Zhang, X. Spatial Evolution and Critical Factors of Urban Innovation: Evidence from Shanghai, China. Sustainability 2020, 12, 938. [Google Scholar] [CrossRef]
  32. Hellwig, V. Tech hubs within cities—Determinants and dynamics. Ann. Reg. Sci. 2024, 73, 1885–1916. [Google Scholar] [CrossRef]
  33. Li, J.; Li, Y.; Tu, M.; Liu, X. Third places as catalysts for technological innovation? Evidence from a grid cell level analysis of Nanjing, China. Int. J. Urban Sci. 2024, 28, 105–123. [Google Scholar] [CrossRef]
  34. Cervero, R.; Kockelman, K. Travel demand and the 3Ds: Density, diversity, and design. Transp. Res. Part D Transp. Environ. 1997, 2, 199–219. [Google Scholar] [CrossRef]
  35. Boschma, R. Proximity and Innovation: A Critical Assessment. Reg. Stud. 2005, 39, 61–74. [Google Scholar] [CrossRef]
  36. Balland, P.-A.; Boschma, R.; Frenken, K. Proximity and Innovation: From Statics to Dynamics. Reg. Stud. 2015, 49, 907–920. [Google Scholar] [CrossRef]
  37. Gordon, I. Density and the built environment. Energy Policy 2008, 36, 4652–4656. [Google Scholar] [CrossRef]
  38. Chen, Z.; Dong, B.; Pei, Q.; Zhang, Z. The impacts of urban vitality and urban density on innovation: Evidence from China’s Greater Bay Area. Habitat Int. 2022, 119, 102490. [Google Scholar] [CrossRef]
  39. Wu, Y.; Wei, Y.D.; Li, H.; Liu, M. Amenity, firm agglomeration, and local creativity of producer services in Shanghai. Cities 2022, 120, 103421. [Google Scholar] [CrossRef]
  40. Zhang, M.; Partridge, M.D.; Song, H. Amenities and the geography of innovation: Evidence from Chinese cities. Ann. Reg. Sci. 2020, 65, 105–145. [Google Scholar] [CrossRef]
  41. Tao, T.; Cao, J. Exploring nonlinear and collective influences of regional and local built environment characteristics on travel distances by mode. J. Transp. Geogr. 2023, 109, 103599. [Google Scholar] [CrossRef]
  42. Jin, M.; Guo, J.; Xie, S.; Lin, Y.; Gong, Y.; Liu, Y. Understanding the Effect of Built Environment on Travel Behavior at Multiple Geographic Scales. Transp. Res. Rec. J. Transp. Res. Board 2023, 2677, 1062–1076. [Google Scholar] [CrossRef]
  43. Credit, K. Transit-oriented economic development: The impact of light rail on new business starts in the Phoenix, AZ Region, USA. Urban Stud. 2018, 55, 2838–2862. [Google Scholar] [CrossRef]
  44. Audretsch, D.B.; Feldman, M.P. Chapter 61—Knowledge Spillovers and the Geography of Innovation. In Handbook of Regional and Urban Economics; Henderson, J.V., Thisse, J.-F., Eds.; Elsevier: Amsterdam, The Netherlands, 2004; Volume 4, pp. 2713–2739. [Google Scholar]
  45. Fitjar, R.D.; Huber, F.; Rodríguez-Pose, A. Not too close, not too far: Testing the Goldilocks principle of ‘optimal’ distance in innovation networks. Ind. Innov. 2016, 23, 465–487. [Google Scholar] [CrossRef]
  46. Duranton, G.; Puga, D. The Economics of Urban Density. J. Econ. Perspect. 2020, 34, 3–26. [Google Scholar] [CrossRef]
  47. Boschma, R.; Frenken, K. Chapter 5: The Spatial Evolution of Innovation Networks: A Proximity Perspective. In The Handbook of Evolutionary Economic Geography; Edward Elgar Publishing: Cheltenham, UK, 2010. [Google Scholar]
  48. Dong, W.; Cao, X.; Wu, X.; Dong, Y. Examining pedestrian satisfaction in gated and open communities: An integration of gradient boosting decision trees and impact-asymmetry analysis. Landsc. Urban Plan. 2019, 185, 246–257. [Google Scholar] [CrossRef]
  49. Zhang, W.; Lu, D.; Zhao, Y.; Luo, X.; Yin, J. Incorporating polycentric development and neighborhood life-circle planning for reducing driving in Beijing: Nonlinear and threshold analysis. Cities 2022, 121, 103488. [Google Scholar] [CrossRef]
  50. Wang, R.; Wang, X.; Zhang, Z.; Zhang, S.; Li, K. Research on the Nonlinear and Interactive Effects of Multidimensional Influencing Factors on Urban Innovation Cooperation: A Method Based on an Explainable Machine Learning Model. Systems 2025, 13, 187. [Google Scholar] [CrossRef]
  51. Yang, J.; Shi, B. Scale or size? An analysis of the factors that affect building density: Evidence from high-density central urban zones in Asia. Urban Des. Int. 2021, 26, 220–233. [Google Scholar] [CrossRef]
  52. Barr, J.; Cohen, J.P. The floor area ratio gradient: New York City, 1890–2009. Reg. Sci. Urban Econ. 2014, 48, 110–119. [Google Scholar] [CrossRef]
  53. Fritsch, M.; Slavtchev, V. Universities and Innovation in Space. Ind. Innov. 2007, 14, 201–218. [Google Scholar] [CrossRef]
  54. Che, Y.; Li, X.; Liu, X.; Wang, Y.; Liao, W.; Zheng, X.; Zhang, X.; Xu, X.; Shi, Q.; Zhu, J.; et al. 3D-GloBFP: The first global three-dimensional building footprint dataset. Earth Syst. Sci. Data 2024, 16, 5357–5374. [Google Scholar] [CrossRef]
  55. Zhang, W.; Zhao, Y.; Cao, X.J.; Lu, D.; Chai, Y. Nonlinear effect of accessibility on car ownership in Beijing: Pedestrian-scale neighborhood planning. Transp. Res. Part D Transp. Environ. 2020, 86, 102445. [Google Scholar] [CrossRef]
  56. Ding, C.; Cao, X.; Næss, P. Applying gradient boosting decision trees to examine non-linear effects of the built environment on driving distance in Oslo. Transp. Res. Part A Policy Pract. 2018, 110, 107–117. [Google Scholar] [CrossRef]
  57. Ma, J.; Cheng, J.C.P.; Jiang, F.; Chen, W.; Zhang, J. Analyzing driving factors of land values in urban scale based on big data and non-linear machine learning techniques. Land Use Policy 2020, 94, 104537. [Google Scholar] [CrossRef]
  58. Wang, K.; Wang, X. Generational differences in automobility: Comparing America’s Millennials and Gen Xers using gradient boosting decision trees. Cities 2021, 114, 103204. [Google Scholar] [CrossRef]
  59. Jun, M.-J. A comparison of a gradient boosting decision tree, random forests, and artificial neural networks to model urban land use changes: The case of the Seoul metropolitan area. Int. J. Geogr. Inf. Sci. 2021, 35, 2149–2167. [Google Scholar] [CrossRef]
  60. Tao, T.; Wang, J.; Cao, X. Exploring the non-linear associations between spatial attributes and walking distance to transit. J. Transp. Geogr. 2020, 82, 102560. [Google Scholar] [CrossRef]
  61. Chao, L.; Zerun, L.; Mi, D. Study on the Impacts of Urban Density on Common Chronic Diseases in American Metropolitan Areas. Urban Plan. Int. 2024, 40, 1–22. (In Chinese) [Google Scholar] [CrossRef]
  62. Ding, C.; Cao, X.; Liu, C. How does the station-area built environment influence Metrorail ridership? Using gradient boosting decision trees to identify non-linear thresholds. J. Transp. Geogr. 2019, 77, 70–78. [Google Scholar] [CrossRef]
  63. Ben Letaifa, S.; Rabeau, Y. Too close to collaborate? How geographic proximity could impede entrepreneurship and innovation. J. Bus. Res. 2013, 66, 2071–2078. [Google Scholar] [CrossRef]
  64. Hua, L.; Yang, Z.; Shao, J. Impact of network density on the efficiency of innovation networks: An agent-based simulation study. PLoS ONE 2022, 17, e0270087. [Google Scholar] [CrossRef] [PubMed]
  65. Balland, P.-A.; Jara-Figueroa, C.; Petralia, S.G.; Steijn, M.P.A.; Rigby, D.L.; Hidalgo, C.A. Complex economic activities concentrate in large cities. Nat. Hum. Behav. 2020, 4, 248–254. [Google Scholar] [CrossRef] [PubMed]
  66. Breschi, S.; Lissoni, F. Knowledge Spillovers and Local Innovation Systems: A Critical Survey. Ind. Corp. Change 2001, 10, 975–1005. [Google Scholar] [CrossRef]
Buildings 15 02528 g001
Figure 1. Location of the G60 S&T Innovation Valley.
Figure 1. Location of the G60 S&T Innovation Valley.
Buildings 15 02528 g001
Buildings 15 02528 g002
Figure 2. Research framework.
Figure 2. Research framework.
Buildings 15 02528 g002
Buildings 15 02528 g003
Figure 3. Spatial distribution of the variables.
Figure 3. Spatial distribution of the variables.
Buildings 15 02528 g003
Buildings 15 02528 g004
Figure 4. Construction of the GBDT model with five-fold cross-validation.
Figure 4. Construction of the GBDT model with five-fold cross-validation.
Buildings 15 02528 g004
Buildings 15 02528 g005
Figure 5. Threshold effects of urban construction density on IPAD across all urban centres.
Figure 5. Threshold effects of urban construction density on IPAD across all urban centres.
Buildings 15 02528 g005
Buildings 15 02528 g006
Figure 6. Threshold effects of amenity density on IPAD across all urban centres.
Figure 6. Threshold effects of amenity density on IPAD across all urban centres.
Buildings 15 02528 g006
Buildings 15 02528 g007
Figure 7. Threshold effects of proximity to key urban nodes on IPAD across all urban centres.
Figure 7. Threshold effects of proximity to key urban nodes on IPAD across all urban centres.
Buildings 15 02528 g007
Buildings 15 02528 g008
Figure 8. Threshold effects of built environment variables on IPAD across respective urban centres.
Figure 8. Threshold effects of built environment variables on IPAD across respective urban centres.
Buildings 15 02528 g008
Table 1. Descriptive statistics of the variables.
Table 1. Descriptive statistics of the variables.
DimensionsVariablesSymbolUnitMinMaxMeanSD
Dependent variable
Innovation outputInvention patent application densityIPADapplications/km20.004046.0040.30165.10
Independent variables
DensityUrban construction densityBuilding densityBD%0.0080.7215.2910.46
Floor area ratioFAR-0.0017.553.572.68
Transportation network densityTNDkm/km20.0022.245.143.29
Amenity densityCoffee shop densityCSDshops/km20.0024.000.812.10
Convenience store and supermarket densityCSSDshops/km20.0076.008.5211.49
Public parking densityPPDlots/km20.00140.0010.2216.48
ProximityProximity to key urban nodesProximity to prefecture-level city centresPPCCkm0.2227.709.935.76
Proximity to CRH stations/airportsPCAkm0.1429.957.605.15
Proximity to universitiesPUkm0.1528.046.744.67
Table 2. The relative importance of built environment variables across all urban centres.
Table 2. The relative importance of built environment variables across all urban centres.
Sub-DimensionsVariablesOverall RankingRelative Importance (%)Total (%)
Urban construction densityBD122.6850.90
FAR214.84
TND413.38
Amenity densityCSD92.4613.64
CSSD85.43
PPD75.75
Proximity to key urban nodesPPCC511.3535.46
PCA610.17
PU313.94
Table 3. The relative importance of built environment variables across respective urban centres.
Table 3. The relative importance of built environment variables across respective urban centres.
VariablesRelative Importance (%)
HefeiTotalWuhuTotalXuanchengTotalSuzhouTotalSongjiangTotalJiaxingTotalHuzhouTotalHangzhouTotalJinhuaTotal
BD25.4745.6324.6261.0716.4827.1316.1447.9728.4661.8014.4436.7911.7826.957.5739.6713.9744.05
FAR9.9326.513.9421.3124.0015.886.7419.8215.15
TND10.239.946.7110.529.346.478.4312.2814.93
CSD0.6412.610.136.0410.0537.551.4015.111.089.364.4618.442.4614.375.9324.163.5518.37
CSSD6.373.2010.866.944.626.257.135.919.60
PPD5.602.7116.646.773.657.734.7812.325.22
PPCC14.1641.7612.9132.897.0835.3212.1136.926.8328.8425.0344.775.4458.6812.5536.1711.1237.58
PCA10.3111.2015.4413.8510.147.978.8910.6012.85
PU17.298.7812.8010.9611.8711.7744.3513.0213.61
Best iteration23341371688185319141528290030239797
Pseudo R20.6990.5210.5900.6540.6270.5150.7620.7530.685
Obs.426262297513551144950562
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Wang, L.; Li, L. The Impact of the Built Environment on Innovation Output in High-Density Urban Centres at the Micro-Scale: A Case Study of the G60 S&T Innovation Valley, China. Buildings 2025, 15, 2528. https://doi.org/10.3390/buildings15142528

AMA Style

Wang L, Li L. The Impact of the Built Environment on Innovation Output in High-Density Urban Centres at the Micro-Scale: A Case Study of the G60 S&T Innovation Valley, China. Buildings. 2025; 15(14):2528. https://doi.org/10.3390/buildings15142528

Chicago/Turabian Style

Wang, Lie, and Lingyue Li. 2025. "The Impact of the Built Environment on Innovation Output in High-Density Urban Centres at the Micro-Scale: A Case Study of the G60 S&T Innovation Valley, China" Buildings 15, no. 14: 2528. https://doi.org/10.3390/buildings15142528

APA Style

Wang, L., & Li, L. (2025). The Impact of the Built Environment on Innovation Output in High-Density Urban Centres at the Micro-Scale: A Case Study of the G60 S&T Innovation Valley, China. Buildings, 15(14), 2528. https://doi.org/10.3390/buildings15142528

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop