The Synergistic Transition of China’s Manufacturing Industry Towards Digitalisation and Green Development: A Study on Level Measurement, Analysis of Influencing Factors and Interactive Effects

Abstract

Faced with industrial transformation and environmental constraints, exploring the synergistic transition towards digitalisation and greening in manufacturing is of great significance. Using panel data from 30 Chinese provinces (2011–2023), this paper employs the modified distance synergy model, XGBoost, and PVAR to measure the synergistic transition level, identify key characteristics, and reveal regional heterogeneity. The findings show that: (1) The synergistic level has risen steadily, forming a pattern of “East leading, Central catching up, West stable, Northeast slowing”. (2) Digital content resources, internet address resources, the share of high-tech new product revenue, and environmental protection expenditure are key characteristics. (3) Regional heterogeneity is evident: the Northeast exhibits bidirectional synergy; the East and West show short-term digitalisation promoting greening and long-term greening driving digitalisation; the Central region shows short-term suppression of digitalisation by greening, with no significant reverse effect. These findings provide empirical evidence for advancing digital-green synergy and regional high-quality development.

1. Introduction

In light of global climate change and the urgent need for sustainable development, the transition of industrial systems towards low-carbon and smart models is a matter of the utmost urgency. Integrating digital technologies with green objectives to promote their synergistic development has become a key pathway to achieving global climate goals (George et al., 2021) [1]. Digital technologies can provide support for the green transition through precise monitoring, intelligent decision-making and efficient management, while green objectives chart the course for the application of digital technologies and create a vast market. Countries around the world have prioritised the synergy between digitalisation and greening as a key national strategy. The European Union is advancing this through the parallel implementation of the ‘Green New Deal’ and the ‘Digital Europe Programme’, systematically driving the greening and digitalisation of industry (Green, 2024; Stefanis et al., 2024; Broeders et al., 2023; Heidebrecht, 2024) [2,3,4,5]. Germany’s ‘Industry 4.0’ strategy also emphasises the use of digital technologies to enhance resource efficiency (Ma et al., 2024; Ortega-Gras et al., 2025) [6,7]. Policies such as the US Inflation Reduction Act aim to incentivise clean technology innovation and digital upgrading, thereby accelerating the low-carbon transition of industry (Bistline et al., 2023; Bang, 2025) [8,9]. In-depth research into the underlying mechanisms, practical pathways and influencing factors of this transition process is of great significance for understanding global sustainable development trends and promoting international cooperation on climate governance.

Manufacturing forms the foundation of the global real economy. As a key driver of global economic growth and a major sector in terms of resource consumption, it plays a vital role in the transition towards digitalisation and greening. On the one hand, the manufacturing industry accounts for approximately a quarter of global greenhouse gas emissions, making it both a priority and a major challenge for carbon reduction (Sharma, 2024) [10]; on the other hand, it is also one of the sectors where technological innovation is most active and the application of digital technologies is most widespread. New technologies such as smart manufacturing, the Industrial Internet of Things and digital twins are profoundly transforming traditional manufacturing models, offering technical possibilities for reducing energy consumption, enhancing resource efficiency and realising a circular economy (Yang et al., 2022; Zhang et al., 2025; Bagherian et al., 2024) [11,12,13]. However, digital transformation may also give rise to new issues regarding energy consumption and electronic waste, while the transition towards sustainability requires technological innovation and financial investment. Consequently, research into how the manufacturing industry can achieve a virtuous cycle and synergistic advancement between digitalisation and sustainability is of decisive importance for the sustainable development of global industry.

As the country with the world’s largest manufacturing industry and most comprehensive industrial system, China’s transformation pathway serves as a global model and has far-reaching implications (Liu et al., 2022) [14]. The green transition of China’s manufacturing industry directly influences the achievement of global carbon emission targets, while its experience in digital transformation will also provide valuable insights for other developing nations. In recent years, China has coordinated the advancement of the ‘Manufacturing Powerhouse’, ‘Digital China’ and ‘Dual Carbon’ objectives, introducing a series of policies to promote the integration of smart manufacturing and green manufacturing. This has resulted in a practical pathway with Chinese characteristics (Hu and Jia, 2025; Jiang and Murmann, 2022; Zhou et al., 2021) [15,16,17], providing a wealth of scenarios for observing the synergistic application of digital technologies and green objectives at the industrial level. Furthermore, development across China’s regions is uneven, with significant disparities in industrial structure, resource endowments and digital infrastructure (Li et al., 2022; Li, 2025) [18,19]. These internal variations provide unique conditions for studying the diverse pathways and spatial differentiation of the synergistic digital and green transformation of manufacturing under different conditions.

While existing research has undertaken numerous valuable explorations in the field of industrial transformation, three key gaps remain to be addressed: firstly, past studies have largely focused on analysing transformation from a single dimension, such as digitalisation or greening (Verma et al., 2022; Kraus et al., 2022) [20,21], lacking systematic examination of the synergistic interactions between the two. This makes it difficult to fully reveal the multifaceted logic of complex industrial transformation. Furthermore, existing research is predominantly static in nature, failing to capture the dynamic evolution of transformation. Secondly, comprehensive empirical research on the synergistic transformation of digitalisation and greening in China’s manufacturing industry remains insufficient. Existing findings largely remain at the preliminary stage of exploring unidirectional influence mechanisms (Zhang et al., 2024) [22], lacking in-depth empirical explanations of the bidirectional interactions between the two. Compared with studies conducted from a global perspective, such as those by Topaloglu et al. (2025) [23], there is an urgent need to focus on the key sector of manufacturing to draw targeted conclusions. Thirdly, existing research methods largely rely on traditional econometric models, which struggle to effectively handle the complex non-linear relationships and multi-dimensional influencing factors inherent in the collaborative transformation process; consequently, the explanatory power and practical guidance offered regarding the transformation mechanisms require improvement.

This paper constructs a comprehensive evaluation framework for the digitalisation and greening of the manufacturing industry. By integrating the Modified Distance Synergy Model, XGBoost and the PVAR model, it systematically measures the level of synergy-driven transformation, identifies key characteristics and reveals the mechanisms of regional heterogeneity interactions, thereby providing empirical evidence to support the sustainable transformation of the manufacturing industry in emerging economies. The marginal contributions are manifested in three aspects: methodologically, by integrating the improved distance-based synergy model with XGBoost, we overcome the limitations of a single dimension and deconstruct the complex, non-linear landscape of synergistic transformation; theoretically, by utilising PVAR to reveal the long-term dynamic causal relationship between the two, the study expands the perspective from unidirectional empowerment to bidirectional interaction; empirical findings indicate that digital transformation significantly promotes green transformation, while the feedback effect of green transformation exhibits significant regional heterogeneity; and practically, the study deepens theoretical understanding of the synergy mechanism, emphasises differences in regional development stages, and provides empirical support and decision-making references for emerging economies to formulate differentiated industrial policies.

The remainder of this paper is structured as follows. Section 2 reviews the existing literature; Section 3 outlines the research design; Section 4 presents and analyses the empirical findings; Section 5 discusses the research; and Section 6 summarises the conclusions, proposes policy recommendations, and outlines the limitations and prospects for future research.

2. Literature Review

2.1. Theoretical Implications of the Coordinated Digital and Green Transformation of the Manufacturing Industry

The synergistic transformation of digitalisation and greening in manufacturing refers to a systematic evolutionary process in which the two mutually reinforce and coexist in terms of objectives, processes and outcomes. Its core lies in achieving improved resource efficiency and a systematic reduction in environmental impact through the deep integration of digital technologies and green objectives (Li et al., 2024; Yu et al., 2026) [24,25]. From a strategic management perspective, this process is essentially a strategic behaviour whereby enterprises utilise digital technologies to seek synergistic gains in both environmental and economic performance (Ansoff, 1965; Li et al., 2024) [26,27]. Theoretically, digital technologies are not merely tools for enhancing green efficiency; rather, they drive systemic restructuring through data intelligence, thereby achieving a dynamic integration of carbon reduction and value creation (Coll-Martínez et al., 2022) [28].

Digitalisation is underpinned by digital technologies such as the Internet of Things, cloud computing, big data and artificial intelligence. Its essence lies in achieving intelligent, flexible and networked production and management through data-driven approaches. At the level of value creation, digital transformation primarily follows three key pathways—process optimisation, innovation empowerment and experience re-engineering—to significantly enhance enterprises’ production efficiency, innovation capacity and agility in responding to customers, thereby building sustainable long-term competitive advantages (Burström et al., 2021; Yaqub & Alsabban, 2023; Liu et al., 2024; Xing et al., 2026) [29,30,31,32]. However, the realisation of these positive effects relies heavily on the synergistic support of soft factors such as organisational structure, strategic leadership and data culture (Leso et al., 2023) [33]. At the same time, digital transformation has also driven the evolution of organisational structures towards flatter, more networked forms (Shahzad et al., 2025) [34], and has had a profound impact on the structure of human capital. These profound changes at the organisational and managerial levels provide the foundational conditions for the systematic restructuring required to achieve greening objectives.

The greening approach aims to significantly reduce environmental impacts across the entire life cycle and enhance resource efficiency by integrating methods such as cleaner production, the circular economy and eco-design (Yang et al., 2025; Ma et al., 2025; Shu et al., 2025) [35,36,37]. According to Porter’s hypothesis, well-designed environmental regulations can effectively stimulate firms to engage in process and product innovation. This innovation-offset effect not only offsets compliance costs but can also translate into significant productivity gains and long-term competitive advantages (Zhu et al., 2025) [38]. At the same time, by developing green products, firms can meet growing demand for environmentally friendly consumption and build differentiated brand equity and market reputation (Qiu et al., 2020; Liang et al., 2025) [39,40]. Green transition is therefore regarded as a key driver of technological progress in fields such as clean technology, energy-efficient processes and circular materials (Demirel et al., 2025) [41]. Digital technology, in turn, provides the supporting capabilities of data integration and intelligent decision-making for this process, enabling the green transition to evolve from incremental improvements towards systemic integration.

Furthermore, the synergistic transition between digitalisation and greening transcends the mere juxtaposition of technologies; rather, it operates through differentiated and dual innovation pathways. Recent evidence underscores that green innovation serves as a critical mediator, translating digital advancements into high-quality economic development (Chen and Xing, 2025) [42]. Within this nexus, digital technologies empower manufacturing enterprises to pursue both substantial and strategic green innovation. Although digitalisation inherently optimises resource allocation and mitigates information asymmetry, it is substantial green innovation—characterised by core technological breakthroughs—that generates the most robust synergistic momentum, thereby structurally driving the manufacturing system toward long-term sustainability.

2.2. Measurement Methods and Testing of Interactions in the Coordinated Digital and Green Transformation of the Manufacturing Industry

Existing research generally constructs multi-dimensional evaluation indicator systems and employs coupling coordination models to assess the level of synergistic development between the two systems. Studies have found that the synergy between the digital economy and green industrial development at the provincial level in China exhibits significant regional imbalances (Liu et al., 2022) [43], while the coupling coordination between the digital economy and green technological innovation at the urban agglomeration scale shows a marked upward trend (Zhong et al., 2024) [44]. Regarding indicator construction methods, some studies have taken the degree of coupling and coordination between digital productivity and green productivity as a measure of the momentum of new-quality productivity, employing a combination of the entropy-weighted TOPSIS method, coupling and coordination models, and spatial Markov chains for measurement and spatio-temporal evolution analysis (Yu and Zhang, 2024) [45]. In the context of developing countries, principal component analysis has been employed to construct a multidimensional digitalisation index encompassing digital infrastructure, industrialisation and innovation capacity (Kaushal and Dwivedi, 2026) [46]. At the micro level, research has examined the impact of corporate digital transformation on pollution emissions and its underlying mechanisms (Zhu et al., 2023) [47], while other studies have focused on the role of government digitalisation levels in promoting corporate green investment (Wang and Li, 2025) [48]. Furthermore, some research has examined the institutional role of industry–academia–research collaboration, employing data envelopment analysis to measure the innovation efficiency of such collaborations. It was found that technological progress is the dominant driver of this growth, and that this efficiency indirectly influences carbon emissions through green innovation (Song et al., 2020) [49]. At the level of policy discourse, a study conducted an interpretative analysis of the EU’s ‘Twin Transition’ policy texts, exploring how green and digital transitions are discursively reconfigured as synergies (Kovacic et al., 2024) [50]. Other research has employed index decomposition analysis to examine the relationship between digitalisation and energy demand in high-income countries, finding that digitalisation is closely linked to value-added growth in high-digital-intensity industries and amplifies energy use (Hambye-Verbrugghen et al., 2026) [51].

With regard to the examination of interactive relationships, existing research has primarily focused on analysing the unidirectional impact of digitalisation on greening. In terms of unidirectional effects, studies have found that improvements in the level of government digitalisation can significantly promote corporate green investment by enhancing the efficiency of environmental regulation (Wang and Li, 2025) [48], digital transformation also plays a significant role in collaborative emission reductions among heavily polluting enterprises (Mai et al., 2025) [52], and digital transformation at the enterprise level can reduce pollution emissions through mechanisms such as green technological innovation, factor allocation efficiency and environmental information disclosure (Zhu et al., 2023) [47]. The generalised method of moments estimation has further confirmed the unidirectional promotional effect of digitalisation on green development (Dou and Gao, 2022) [53]. Some studies have begun to focus on non-linear characteristics; analyses based on data from developing countries have found a U-shaped relationship between digitalisation and carbon emissions, while the synergistic emission reduction effects of digitalisation and human capital have not yet been fully realised (Kaushal and Dwivedi, 2026) [46]. However, existing research remains insufficient in systematically examining how the demand for greening drives digital innovation in reverse and how the two achieve a bidirectional co-evolution. Although a small number of studies have identified influencing factors in the synergy between the digital economy and low-carbon development in the construction sector (Lan et al., 2025) [54], empirical exploration of the bidirectional interaction mechanism as a whole still requires further deepening.

2.3. Literature Review

Existing research has made significant progress in theoretical frameworks, measurement methods and mechanism testing, but the following shortcomings remain: firstly, the bidirectional interaction mechanisms from a systemic synergy perspective lack sufficient empirical testing, with most studies confining themselves to analysing the unidirectional impact of digitalisation on greening; secondly, measurement methods are predominantly static in nature, failing to adequately capture the dynamic nature of the co-evolutionary process; thirdly, integrated analysis of the macro-level policy environment and micro-level corporate behaviour needs to be strengthened. Furthermore, the debate regarding the interactive relationship between digitalisation and greening has yet to be rigorously examined.

3. Research Design

3.1. Development of the Indicator System

In recent years, China has continued to advance the transformation and upgrading of its traditional manufacturing industry, integrating the concept of green development with advanced and applicable technologies throughout the entire industrial chain. The country is accelerating the establishment of an economic system characterised by green, low-carbon and circular development to promote new industrialisation and support the building of a manufacturing powerhouse. The coordinated development of digitalisation and greening in manufacturing entails fostering new growth engines—such as next-generation information technology and high-end equipment—to generate new products, business models and operational paradigms. This will drive profound transformations in manufacturing production methods and industrial structures, thereby enhancing total factor productivity. To systematically evaluate the synergistic digital–green transformation of manufacturing, this study constructs an evaluation indicator system encompassing four primary indicators: Digital Transformation, Digital Investment, Digital Applications, and Green Transformation. The conceptual and empirical foundations for Digital Transformation and Digital Applications are drawn from Hao et al. (2023) [55] and Dian et al. (2024) [56]; Digital Investment is informed by Yang et al. (2024) [57]; Green Transformation is informed by Liu et al. (2025) [58]. Specific indicator descriptions are provided in

Table 1. Evaluation indicator system for the coordinated digital and green transformation of the manufacturing industry.

Overall Indicator	Primary Indicators	Secondary Indicators	Explanation of Indicators	Unit	Direction
Digital Transformation in Manufacturing	Digital Fundamentals	Level of mobile device penetration	Number of mobile phones per 100 people	Number of units per 100 people	+
		Carrying capacity of communication backbones	Length of long-distance optical fibre cables	kilometres	+
		Number of broadband access ports	Number of internet access ports	ten thousand	+
		Scale of digital content resources	Number of regional webpages	ten thousand	+
		The scale of Internet address resources	Number of IPv4 addresses	ten thousand	+
		Wireless access point density	Number of mobile phone base stations	ten thousand	+
	Digital investment	Intensity of investment in technological upgrading by enterprises above a certain scale	Expenditure on technological upgrading by enterprises above a certain scale	ten thousand yuan	+
		Intensity of investment in technology introduction by enterprises above a certain scale	Expenditure on technology introduction by enterprises above a certain scale	ten thousand yuan	+
		R&D intensity of industrial enterprises above a certain scale	R&D expenditure of industrial enterprises above a certain size	ten thousand yuan	+
		Density of digital talent in the manufacturing industry	Number of R&D personnel in high-tech industries/Number of employees in the manufacturing industry	%	+
		Digital R&D Innovation Efficiency	The ratio of granted national patent applications to the full-time equivalent of R&D staff	%	+
	Digital Applications	Industrial robot density	Industrial robot installation density	one	+
		Concentration of artificial intelligence companies	Number of artificial intelligence companies	one	+
		Concentration of R&D institutions	Number of enterprises with R&D facilities	one	+
		Proportion of IT professionals	Employees in urban enterprises in the information transmission, software and IT services sector/Employees in the manufacturing industry	%	+
		Digital labour productivity	Industrial value added/Average number of employees in the manufacturing industry	%	+
	Digital output	Proportion of revenue from embedded software	Revenue from embedded systems/Main business revenue of enterprises above a certain scale	%	+
		Proportion of revenue from information technology services	Revenue from information technology services/Main business revenue of enterprises above a specified size	%	+
		Proportion of revenue from software products	Revenue from software products/Main business revenue of enterprises above a certain scale	%	+
		Share of revenue from high-tech industries	Revenue from high-tech industries/Revenue from industrial enterprises above a specified scale	%	+
		Market share in smart manufacturing	The proportion of the company’s operating revenue relative to the total operating revenue of all smart manufacturing enterprises nationwide	%	+
		Proportion of revenue from high-tech new products	Revenue from sales of new products in high-tech industries/Operating revenue of industrial enterprises above designated size	%	+
		Trading activity in the technology market	Turnover in the technology market	hundreds of millions	+
The green transition in manufacturing	Green R&D	Proportion of green utility model patents	Number of green utility model patents granted/Number of green patents granted	%	+
		Proportion of green patents	Number of green invention patents granted/Total number of green patents granted	%	+
	Environmental governance pressure	Proportion of environmental protection expenditure in the budget	Environmental protection expenditure/General government expenditure	%	−
		Proportion of expenditure on energy conservation and environmental protection	Fiscal expenditure on the energy-saving and environmental protection sector/Total fiscal expenditure	%	−
	Resource and energy consumption	Carbon emissions intensity per unit of value added	Carbon emissions from manufacturing/Industrial value added	million tonnes per 10,000 yuan	−
		Water intensity per unit of value added	Industrial water consumption per unit of industrial value added	billion cubic metres per 10,000 yuan	−
		Energy intensity per unit of value added	Physical consumption of coal/Industrial value added	tonnes per 10,000 yuan	−
	Ecological governance	Sulphur dioxide emission intensity per unit of value added	Sulphur dioxide emissions per unit of industrial value added	tonnes per 10,000 yuan	−
		Intensity of solid waste generation per unit of value added	Volume of industrial solid waste utilised/Industrial value added	tonnes per 10,000 yuan	−
		COD emission intensity per unit of value added	Chemical oxygen demand (COD) in wastewater/Industrial value added	tonnes per 10,000 yuan	−
		Intensity of investment in pollution control per unit of value added	Intensity of investment in pollution control per unit of value added	tonnes per 10,000 yuan	−

.

It should be noted that some indicators in the digital application dimension, such as industrial robot density and the number of artificial intelligence enterprises, are used as proxy variables rather than exhaustive measures of digital transformation across all manufacturing subsectors. Due to the availability and comparability of provincial panel data, it is difficult to directly observe the digital transformation level of every manufacturing subsector. Therefore, this study constructs a multidimensional indicator system covering digital infrastructure, digital investment, digital application, digital output, innovation capacity and talent structure, so as to reduce the potential bias caused by relying on any single proxy variable.

3.2. Research Methods

3.2.1. Improving the Distance-Coordination Model

Compared with the coupling coordination model, the improved distance synergy model not only incorporates ideal state values but also takes into account the grey information associated with the indicators. Drawing on relevant research (Rao and Gao, 2022; Dong et al., 2022) [59,60], the improved distance synergy model is used to determine the level of coordinated development between digitalisation and greening in the manufacturing industry.

$${\mathrm{S}}_{\mathrm{ij}}={\displaystyle \frac{{\mathrm{x}}_{\mathrm{ij}}-{\mathrm{x}}_{\mathrm{j}}^{\min }}{{\mathrm{x}}_{\mathrm{j}}^{\max }-{\mathrm{x}}_{\mathrm{j}}^{\min }}}$$

(1)

On this basis, the Euclidean distance is used to measure the overall degree of divergence between the two subsystems across each indicator dimension. A larger distance value indicates a greater deviation in their developmental status and a weaker foundation for collaboration.

$${\mathrm{d}}_{\mathrm{ij}}=\sqrt{{\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{n}}}{({\mathrm{S}}_{\mathrm{ik}}-{\mathrm{S}}_{\mathrm{jk}})}^2}$$

(2)

The grey correlation correction coefficient $r_j$ is integrated by multiplying it with the entropy weight $w_j$ to obtain the corrected weight:

$$w_j^{*}=r_j\times w_j$$

(3)

By calculating the objective weights of indicators using the entropy weighting method, and determining the synergy correlation correction coefficients based on grey correlation, the original distances are weighted, corrected and normalised to construct an improved distance synergy model, thereby enabling a quantitative measurement of the level of synergy in development among the evaluation objects.

$${\mathrm{C}}_{\mathrm{i}}={\displaystyle \frac{1}{\mathrm{m}}}{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{m}}}\left[w_j^{*}\cdot (1-{\displaystyle \frac{{\mathrm{d}}_{\mathrm{ij}}}{{\mathrm{d}}_{\mathrm{j}}^{\max }}})\right]$$

(4)

Here, $\mathrm{n}$ represents the total number of evaluation criteria, $\mathrm{m}$ represents the total number of samples or items under comparison, ${\mathrm{S}}_{\mathrm{ij}}$ represents the standardised value of the i-th sample for the j-th criterion, ${\mathrm{d}}_{\mathrm{ij}}$ represents the Euclidean distance between the i-th and j-th samples, and ${\mathrm{C}}_{\mathrm{i}}$ represents the final value of the improved distance synergy for the i-th sample.

3.2.2. The XGboost-SHAP Model

XGBoost is an ensemble learning algorithm based on gradient boosting. Its core mechanism involves combining multiple weak learners in an additive manner to progressively optimise the objective function, thereby constructing a high-precision predictive model. SHAP, on the other hand, is a game-theory-based model interpretation method capable of consistently quantifying the contribution of each feature to the model’s output. Based on relevant research (Liu et al., 2025; Li et al., 2025) [61,62], we have constructed an XGBoost-SHAP model to investigate the feature importance in the collaborative digital and green transformation of the manufacturing industry. Unlike traditional gradient boosting methods, XGBoost explicitly incorporates a regularisation term ${\mathrm{L}}^{\left(\mathrm{t}\right)}$ into the objective function to penalise model complexity and mitigate the risk of overfitting. Nevertheless, to address the “black-box” interpretability limitations characteristic of complex machine learning algorithms, the SHAP (Shapley Additive exPlanations) method is integrated into the analytical framework. Rooted in cooperative game theory, SHAP consistently and equitably quantifies the marginal contribution of each driving factor to the model’s overall output. As demonstrated by Zhao et al. (2025) [63], this hybrid approach not only accurately ranks the global importance of various transformation characteristics but also uniquely elucidates the non-linear threshold effects and local interactive synergies between digital and green variables at the micro level.

$${\mathrm{L}}^{\left(\mathrm{t}\right)}={\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}}\mathrm{l}({{\mathrm{y}}_{\mathrm{i}}}_{,}{{\widehat{\mathrm{y}}}_{\mathrm{i}}}^{\left(\mathrm{t}\right)})+{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{T}}}\mathsf{\Omega }\left({ƒ}_{\mathrm{k}}\right)$$

(5)

In Equation (5), $\mathrm{l}$ denotes the loss function; ${\mathrm{y}}_{\mathrm{i}}$ denotes the true value of the i-th sample; ${{\widehat{\mathrm{y}}}_{\mathrm{i}}}^{\left(\mathrm{t}\right)}$ denotes the predicted value of the i-th sample at the t-th iteration; $\mathrm{n}$ denotes the number of samples; $\mathrm{T}$ denotes the number of trees; and $\mathsf{\Omega }({ƒ}_{\mathrm{k}})$ denotes the complexity penalty term for the k-th tree.

Calculating the SHAP values for each influencing factor not only identifies key features but also reveals the interactions between them. The formula is as follows:

$${\phi }_{\mathrm{i}}=\sum _{\mathrm{S}\subseteq \mathrm{H}/\left\{\mathrm{i}\right\}}{\displaystyle \frac{\left|\mathrm{S}\right|!(\left|\mathrm{H}\right|-\left|\mathrm{S}\right|-1)!}{\left|\mathrm{H}\right|!}}\left(\mathrm{t}\right(\mathrm{S}\left\{\mathrm{i}\right\}\left)\right)-\mathrm{t}\left(\mathrm{S}\right)$$

(6)

In the formula, ${\phi }_{\mathrm{i}}$ represents the SHAP value of influencing factor $\mathrm{i}$; $\mathrm{H}$ denotes the set of all influencing factors; $\mathrm{S}$ denotes a subset of features that does not include influencing factor $\mathrm{i}$; and t denotes the model’s predicted result corresponding to the combination of influencing factors in subset $\mathrm{S}$.

3.2.3. PVAR Model

The panel vector autoregression (PVAR) model builds upon the vector autoregression (VAR) model by incorporating individual and time effects to account for unobservable individual heterogeneity and analyse the dynamic responses of variables to various shocks. Building on the research by Xu et al. (2023) [64], we utilise the PVAR model to place the digital transformation of the manufacturing industry and its green transformation within the same analytical framework, conducting an empirical study of the interactive relationship between these two variables.

$${\mathrm{Y}}_{\mathrm{it}}={\mathsf{\alpha }}_0+{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}}{\mathrm{A}}_{\mathrm{j}}{\mathrm{Y}}_{\mathrm{i},\mathrm{t}-\mathrm{j}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\delta }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{it}}$$

(7)

$${\mathrm{Y}}_{\mathrm{it}}=\left\{\begin{array}{c}\mathrm{D}\mathrm{T}\mathrm{M}\\ \mathrm{G}\mathrm{T}\mathrm{M}\end{array}\right\}$$

(8)

${\mathrm{Y}}_{\mathrm{it}}$ represents the dependent variable, comprising digital transformation in manufacturing and green transformation in manufacturing; ${\mathrm{A}}_{\mathrm{j}}$ is the coefficient corresponding to the dependent variable lagged by $\mathrm{j}$ periods; a represents the constant term; ${\mathsf{\alpha }}_0$ represents the provincial fixed effects; ${\mathsf{\mu }}_{\mathrm{i}}$ represents the time fixed effects; and ${\mathsf{\delta }}_{\mathrm{t}}$ represents the random error term.

3.3. Data Sources

This paper utilises panel data from 2011 to 2023 relating to the coordinated digital and green transformation of the manufacturing industry across 30 provinces (autonomous regions and municipalities) in China. Due to data availability, the sample excludes Taiwan, Hong Kong, Macao and Tibet. All data are primarily sourced from the annual China Statistical Yearbook, China Science and Technology Statistical Yearbook, China Industrial Statistical Yearbook, and the statistical yearbooks of various provinces and municipalities.

4. Empirical Results

4.1. Measurement of Coordination Levels and Analysis of Spatio-Temporal Evolution

Using an improved distance-based synergy model, we assess the level of collaborative digital and green transformation across different tiers of the manufacturing industry, thereby providing a detailed analysis of its evolutionary characteristics.

At the regional level (Figure 1), the level of synergistic transformation towards digitalisation and greening in China’s manufacturing industry exhibits a distinct spatial gradient. The Eastern region is in a leading position; supported by a robust economic foundation and dynamic technological innovation, its level of synergy has consistently remained significantly ahead. Rising steadily from 0.292 in 2011 to 0.454 in 2023, it has stabilised at a stage of good synergy and continues to advance towards higher levels. The Central and Western regions are steadily catching up, demonstrating strong growth potential: both entered the ‘good synergy’ stage in 2019, reaching levels of 0.297 and 0.204 respectively by 2023. The Central region’s growth rate has been particularly notable, with a significant surge after 2021 that has gradually narrowed the gap with the Eastern region; the Western region, meanwhile, has maintained steady growth; although its absolute value remains slightly lower than that of the Central region, the development trends are converging. By contrast, the development trajectory of the Northeast has been relatively moderate. The synergy level fluctuated from 0.152 in 2011 to 0.202 in 2023; although growth has continued, the pace has slowed, and the gap with other regions has widened, reflecting the challenges it faces in industrial restructuring and the transition between old and new growth drivers. Overall, the four major regions show marked divergence, forming a differentiated development pattern characterised by ‘the East leading, the Central catching up, the West remaining stable, and the Northeast slowing down’.

At the provincial level (Figure 2), the spatio-temporal evolution of the level of synergistic transformation towards digitalisation and greening in China’s manufacturing industry exhibits distinct phased characteristics and shifts in spatial patterns. As shown in Figure 2a, the overall level of synergistic transformation was relatively low in 2011. Regions with high levels of synergy had not yet emerged, with most provinces concentrated in the lower range, and only isolated areas along the eastern coast showing the first signs of synergistic transformation. By 2015 (Figure 2b), the number of provinces with low levels of synergy had decreased, and some provinces in the Central region began to move into the low-to-medium synergy range, marking the entry of the synergy-driven transformation into an initial diffusion phase. The year 2019 (Figure 2c) became a significant turning point in the transformation process, with a marked increase in the number of provinces at a medium level of synergy. A contiguous cluster of high-synergy areas formed in the Eastern region, while the Central region demonstrated an accelerating catch-up trend. By 2023 (Figure 2d), provinces with high levels of synergy had further concentrated in the Beijing–Tianjin–Hebei, Yangtze River Delta and Pearl River Delta regions. Many provinces in the Central region had stabilised within the medium to high synergy range; the Western region, represented by Sichuan and Shaanxi, had formed local growth poles, while the Northeast region remained at a relatively low level overall. The overall spatial pattern closely aligned with the regional economic development gradient. Overall, the level of collaborative transformation exhibits a fundamental trend of gradual diffusion from the eastern coast to the central interior, and an evolution from isolated breakthroughs to contiguous clusters. The Eastern region continues to consolidate its leading position, the Central region demonstrates strong momentum for catching up, the Western region has achieved local breakthroughs, while the transformation process in the Northeastern region remains relatively sluggish.

4.2. Analysis of Influencing Factors

Based on the XGBoost-SHAP model, a feature importance analysis was conducted to identify the factors influencing the level of synergistic digital and green transformation in the manufacturing sector. Standardised hyperparameters were applied to the XGBoost model, with a learning rate of 0.05, a maximum tree depth of 6, the number of decision trees set at 150, row and column sampling ratios both at 0.8, L1 regularisation of 0.1, L2 regularisation of 1.0, and a fixed random seed of 42. The study utilised panel data from 30 Chinese provinces covering the period 2011–2023, initially comprising 34 features. Following feature selection, the top 10 core explanatory variables ranked by SHAP values were retained for interpretability analysis.

Table 2. Model accuracy.

Sample	${\mathbf{R}}^2$	RMSE	MAE	MAPE
Training set	0.9699	0.0153	0.0115	4.23%
Test set	0.8955	0.0420	0.0247	9.85%

shows that the model performs excellently on the test set, with a goodness-of-fit of 0.8955, a root mean square error (RMSE) of 0.0420, a mean absolute error (MAE) of 0.0247, and a mean absolute percentage error (MAPE) of 9.85%. The model is generally stable, with 89.7% of prediction errors falling within ±20%. It should be noted that the above metrics reflect the explanatory power of the features rather than predictive accuracy. This paper further conducted leave-one-out cross-validation, and the results revealed significant regional heterogeneity; due to differences in regional endowments, the fitting performance varied across provinces. The overall average RMSE was 0.0384, and the average MAE was 0.0323.

The importance and direction of influence of the 10 explanatory variables on the level of coordinated digital and green transformation in the manufacturing industry are shown in Figure 3. In terms of importance, the variables include the scale of digital content resources (x4), the scale of internet address resources (x5), the proportion of sales revenue from high-tech new products (x20), the proportion of expenditure on energy conservation and environmental protection (x28), the market share of smart manufacturing (x21), the concentration of R&D institutions (x15), the intensity of investment in technological upgrading by enterprises above designated size (x8), the proportion of IT personnel (x14), the proportion of green utility model patents (x22), and the intensity of investment in technology introduction by enterprises above a designated size (x9)—their influence on the level of collaborative digital and green transformation in the manufacturing industry diminishes in that order from highest to lowest. The combined contribution of the first four features exceeds 66%, constituting the core set of characteristics for synergistic transformation. The scale of digital content resources (x4), the scale of internet address resources (x5), the proportion of sales revenue from high-tech new products (x20), the market share of smart manufacturing (x21), the proportion of IT personnel (x14) and the proportion of green utility model patents (x22) all exhibit a consistent positive driving effect; that is, the higher the value of these characteristics, the stronger their promotional effect on the level of coordinated digital and green development. It is worth noting that the proportion of expenditure on energy conservation and environmental protection (x28) is a reverse indicator; a larger standardised value implies a smaller actual burden of environmental expenditure. Consequently, the positive impact of this indicator essentially represents the beneficial effect of alleviating the burden of environmental expenditure on collaborative transformation. This indicates that the improvement of digital infrastructure, the enhancement of green innovation output, the strengthening of fiscal support, and the optimisation of the digital talent structure can all significantly promote the deep integration of digitalisation and greening, as well as collaborative transformation. Conversely, the concentration of R&D institutions (x15), the intensity of technical renovation investment by enterprises above designated size (x8), and the intensity of technology introduction investment by such enterprises (x9) exhibit certain non-linear or threshold effects in the figure. When these variables are at medium to low levels, their improvement has a clear positive effect on coordinated development; however, once they exceed a certain threshold, their marginal contribution gradually levels off or even declines slightly. This may reflect the fact that once innovation hubs become overly concentrated and enterprise technical upgrades or technology imports reach a certain scale, the synergistic promotional effects of further investment will weaken in the absence of systematic integration and the commercialisation of results. Digital infrastructure, green innovation output and policy support are currently the most critical drivers of the coordinated development of digitalisation and greening in the manufacturing industry. Efforts should be made to further strengthen the development of digital infrastructure, incentivise the market-driven application of green technologies, and focus on enhancing systematic integration and results-oriented approaches in the planning of R&D institutions, as well as in the processes of enterprise technological upgrading and technology introduction, in order to comprehensively elevate the overall level of coordinated development in digitalisation and greening.

This study selected the top four influencing factors by contribution rate (Figure 4) and found a clear non-linear synergistic mechanism between the scale of digital content resources (x4) and other key characteristics. When the scale of digital content resources (x4) is at a low level, increasing the scale of internet infrastructure resources (x5) can significantly amplify the positive impact of x4 on synergistic transformation; this gain becomes particularly pronounced once x5 exceeds 1.0. This suggests that in regions with a weak digital content ecosystem, prioritising the strengthening of network infrastructure can effectively activate the synergistic efficacy of content resources. Once the scale of digital content resources (x4) reaches a moderate level, an increase in the proportion of revenue from high-tech new products (x20) significantly enhances the contribution to collaborative development. This indicates that the accumulation of digital content resources helps to facilitate the commercialisation of green and innovative products, creating a virtuous cycle between the two. When both the scale of digital content resources (x4) and the concentration of R&D institutions (x15) are at high levels, the marginal contribution of their combined improvement to collaborative development tends to level off. This reflects that in regions where both digital content and R&D resources are already abundant, the synergistic effects of simply continuing to increase factor inputs may diminish, necessitating a shift towards systemic integration and the commercialisation of research outcomes. When the scale of digital content resources (x4) is at a medium to high level, an increase in the proportion of green utility model patents (x22) has a significant positive reinforcing effect on the collaborative digital and green transformation of the manufacturing industry. This indicates that in regions with a relatively well-developed digital content ecosystem, the accumulation of green utility model patents can more effectively facilitate technology diffusion and application through digital platforms, creating an acceleration effect from the integration of digital platforms and green technologies. Conversely, when the scale of digital content resources (x4) is low, the increase in the proportion of green utility model patents makes a relatively limited contribution to synergistic development. This suggests that in the absence of digital content and information support, the conversion efficiency and systemic synergy of green technologies remain constrained, even if they possess practical utility.

4.3. Analysis of the Synergistic Effects of Digital and Green Transformation in the Manufacturing Industry

To explore in depth the dynamic interactions between variables in the synergistic digital and green transformation of the manufacturing industry, this study constructs a panel vector autoregression (PVAR) model. The model treats both the level of digital transformation and the level of green transformation in the manufacturing industry as endogenous variables, systematically examining the interaction mechanisms between the two and their respective impacts on each other’s development. Given the regional heterogeneity in the levels of digital and green transformation within the manufacturing industry, this study further analyses the dynamic interactions between digital and green transformation in different regions, using samples at the national level and across the Eastern, Central, Western and Northeastern regions.

4.3.1. Stability Testing

To ensure the stationarity of all variables, the manufacturing digital transformation index (DTM) and the green transformation index (GTM) were first log-transformed to eliminate heteroscedasticity. Unit root tests were then conducted using the homogeneous LLC test and the heterogeneous IPS test, respectively, to avoid the issue of spurious regression that may arise from non-stationary panel data.

Table 3. Results of the stationarity test.

Sample	Test Methods	Variables				Determination
		lnDTM	dlnDTM	lnGTM	dlnGTM
Nationwide	LLC	−4.7282 ***	−12.0402 ***	−6.7520 ***	−9.1974 ***	First-order linear
	IPS	−1.0012	−7.2192 ***	−3.2776 ***	−8.9125 ***
East	LLC	−5.4960 ***	−8.2661 ***	−2.9626 ***	−4.4541 ***	First-order linear
	IPS	−1.0114	−4.4937 ***	−1.2072	−4.8383 ***
Central	LLC	−1.1114	−3.7850 ***	−0.4551	−2.3802 ***	First-order linear
	IPS	0.6243	−2.7442 ***	0.3625	−4.2300 ***
West	LLC	−3.1401 ***	−4.5004 ***	−5.9097 ***	−2.8413 ***	First-order linear
	IPS	−1.9314 **	−4.7458 ***	−2.9811 ***	−5.2363 ***
Northeast	LLC	−2.8331 **	−4.5491 ***	−1.3989 *	−4.1891 ***	First-order linear
	IPS	−2.3499 ***	−2.6005 ***	−1.5803 *	−2.8967 ***

Note: *, **, *** represent the significant level of 10%, 5%, and 1% respectively; the value in brackets below the coefficient is the robust standard error t value.

shows that, following first-order differencing, both the manufacturing digital transformation index and the green transformation index passed the unit root test, indicating that the two variables are first-order unitary at both the national and sub-regional levels, and that a PVAR model can be established. Cointegration tests revealed a long-term cointegration relationship between the manufacturing digital transformation index and the manufacturing green transformation index.

4.3.2. Determination of the Order of Lag

The optimal lag order was determined by selecting the order corresponding to the minimum value of the MAIC, MBIC and MQIC statistics. The results indicate that the optimal lag order for the PVAR model is first-order for the whole country as well as for the four major regions: East, Central, West and Northeast. Based on this, a first-order PVAR model was constructed; the specific results are shown in

Table 4. Determining optimal lag order for models.

Sample	Order of Lag	MBIC	MAIC	MQIC
Nationwide	1	−45.32026	8.233457	−13.41633
	2	−36.41632	3.748969	−12.48837
	3	−24.09629	2.680572	−8.144321
East	1	−46.08029	−10.10436	−24.39444
	2	−32.15096	−5.169021	−15.88658
	3	−20.29019	−2.302226	−9.447264
Central	1	−41.00726	−13.20455	−23.39534
	2	−31.28661	−10.43458	−18.07767
	3	−22.12881	−8.227458	−13.32285
West	1	−47.47505	−9.97416	−24.97419
	2	−36.18422	−8.058555	−19.30857
	3	−23.73668	−4.986238	−12.48625
Northeast	1	−33.15404	−16.44168	−20.06869
	2	−25.2083	−12.67403	−15.39429
	3	−6.468139	−2.29005	−3.196802

.

4.3.3. GMM Regression

The results of the GMM regression are shown in

Table 5. GMM regression estimates.

Sample of Grouping Variables	Group 1					Group 2
	lnDTM					lnGTM
	Nationwide	East	Central	West	Northeast	Nationwide	East	Central	West	Northeast
lnDTMit-1	0.879 *** [0.824, 0.933]	0.900 *** [0.755, 1.045]	0.971 *** [0.923, 1.019]	0.844 *** [0.785, 0.902]	0.793 *** [0.685, 0.901]	0.036 *** [0.010, 0.062]	0.031 * [−0.004, 0.067]	0.018 [−0.042, 0.079]	0.044 * [−0.005, 0.093]	0.066 * [−0.000, 0.133]
lnGTMit-1	−0.011 [−0.231, 0.209]	−0.144 [−0.751, 0.462]	−0.473 *** [−0.738, −0.208]	0.069 [−0.194, 0.333]	0.365 * [−0.011, 0.740]	0.728 *** [0.608, 0.849]	0.741 *** [0.534, 0.948]	0.779 *** [0.360, 1.198]	0.738 *** [0.562, 0.914]	0.532 *** [0.378, 0.685]

Note: * and *** represent the significant level of 10% and 1% respectively; the value in brackets below the coefficient is the robust standard error t value.

. Both the digital transformation (DTM) and green transformation (GTM) of China’s manufacturing sector exhibit significant path dependence. At both the national level and across the four major economic regions, the coefficients of the first-order lagged terms for both DTM and GTM are significantly positive. This implies that the level of transformation in previous periods exerts a significant positive influence on current development, and that the transformation process demonstrates a clear self-reinforcing effect. Specifically, in the digital transformation dimension, the Central region has the highest lag coefficient at 0.971, indicating that its digital transformation exhibits the most pronounced path dependence; the Northeast Region’s coefficient stands at 0.793, which is relatively weaker. In the green transition dimension, the lag coefficients for both the national level and all regions were similarly significantly positive at the 1% level, reflecting the continuity and cumulative nature of green transition. Among these, the Central region had a coefficient of 0.779, indicating strong continuity, while the Northeast region had a coefficient of 0.532, with the weakest self-reinforcing effect.

In terms of mutual influence, the interactive relationship between digital and green transitions exhibits significant regional heterogeneity. At the national level, the coefficient for the impact of prior digital transition on current green transition is 0.036, which is significantly positive at the 1% level, indicating that digital transition has a significant promotional effect on green transition. Conversely, the impact of prior green transition on current digital transition is not significant. It is worth noting that this insignificant conditional marginal effect does not preclude the possibility that green transition carries predictive information for digital transition in a reduced-form sense; rather, it indicates that once the strong autoregressive dynamics of digitalisation are controlled for, an isolated impulse from past GTM does not yield a direct partial effect that is statistically distinguishable from zero. This distinction becomes relevant when interpreting the Granger causality results that follow. Examining the data by region, in the Northeast, the coefficient of digital transformation on green transformation is 0.066, which is significant at the 10% level, while the coefficient of green transformation on digital transformation is 0.365, which is also significant at the 10% level, revealing a pattern of mutually reinforcing, positive interaction. In the Eastern region, past digital transformation has a significant driving effect on current green transformation, with a coefficient of 0.031, which is significant at the 10% level; however, the impact of green transformation on digital transformation is not significant. In the Western region, past digital transformation exhibits a weak promoting effect on current green transformation, with a coefficient of 0.044, which is significant at the 10% level. Conversely, the effect of green transformation on digital transformation is not significant. It is worth noting that a special situation emerged in the Central region, where prior green transformation exerted a significant inhibitory effect on current digital transformation, with a coefficient of −0.473, significant at the 1% level, while the direct promotional effect of digital transformation on green transformation was not significant. This result indicates that the synergistic mechanism between digital and green transformation in China’s manufacturing industry exhibits significant regional heterogeneity. The Northeast region exhibits a positive two-way interaction, which may stem from its status as a traditional heavy industrial base, where the pressure to transform has compelled the synergistic advancement of digitalisation and greening. By contrast, the significant negative effect in the Central region may reflect short-term resource crowding-out or policy coordination frictions, under which environmental investment may compete for factors allocated to digital transformation. As the PVAR model captures dynamic temporal correlations rather than structural causal paths, the above interpretation represents an economically meaningful inference rather than a statistically confirmed causal mechanism.

4.3.4. Granger Causality Test

To further elucidate the dynamic causal relationship between digitalisation and green transition, a Granger causality test was employed.

Table 6. Results of the Granger causality test.

Initial Assumption	Chi-Squared Statistic
	Nationwide	East	Central	West	Northeast
DTM is not the Granger cause of GTM	0.00938 (0.923)	0.21775 (0.641)	12.228 *** (0.000)	0.26492 (0.607)	3.6197 * (0.057)
GTM is not the Granger cause of DTM	7.4241 *** (0.006)	3.0046 * (0.083)	0.35152 (0.553)	3.105 * (0.078)	3.8003 * (0.051)

Note: * and *** represent the significant level of 10% and 1% respectively; the value in brackets below the coefficient is the robust standard error t value.

clearly illustrates the regional variations in the causal pathways between the two. At the national level, the green transition is a Granger cause of the digital transition, with a chi-square statistic of 7.4241 and a p-value of 0.006, which is significant at the 1% level. Conversely, the Granger causality from digital transformation to green transformation is not significant, indicating that, nationwide, green transformation is a key driver of digital transformation. The Eastern and Western regions similarly exhibit a unidirectional Granger causality from green transformation to digital transformation, both significant at the 10% level. This suggests that, in these regions, the demands and pressures of green development have effectively driven the digital upgrading of enterprises. In the Central region, the direction of causality is reversed, with only a unidirectional Granger causality from digital transformation to green transformation. The chi-square statistic is 12.228, with a p-value of 0.000, highly significant at the 1% level. This indicates that in the Central region, the application of digital technologies is the key driver of green transformation. In the Northeast region, there is a bidirectional Granger causality between digital transformation and green transformation, with chi-square statistics of 3.6197 and 3.8003 respectively, both significant at the 10% level. This corroborates the findings of the GMM regression, indicating that digital and green transformations in the Northeast region have formed a mutually reinforcing virtuous cycle. The results of the Granger causality test further validate the findings of the GMM regression analysis, namely that the mechanism of the synergistic digital and green transformation in China’s manufacturing industry exhibits significant regional heterogeneity, with marked differences in the drivers of transformation and patterns of interaction across different regions. At the national level, an apparent tension emerges between the Granger and GMM results: green transition significantly Granger-causes digital transition, yet its lagged GMM coefficient in the digitalisation equation is insignificant. This pattern is, however, methodologically consistent. Granger causality tests evaluate reduced-form predictability, which may operate through indirect, systemic, or time-distributed channels without requiring a direct structural link. By contrast, the GMM coefficient captures the conditional partial effect of a discrete impulse in past GTM on current DTM, holding constant the system’s own autoregressive dynamics. The significant Granger finding therefore suggests that green transition shapes the broader developmental context in which digitalisation unfolds, even though it does not serve as an immediate, independent driver in the structural specification. Moreover, the bivariate Granger test cannot fully rule out omitted variables (e.g., environmental regulation intensity, fiscal pressure, or technology diffusion climate) that might drive spurious predictability. In contrast, the GMM specification imposes a more stringent threshold by controlling for richer dynamics and fixed effects. Non-linear or threshold effects could also explain why Granger causality is significant while the linear GMM coefficient is not.

4.3.5. Impulse Response

The impulse response analyses the dynamic interaction between two variables while holding other variables constant, thereby characterising the long-term equilibrium relationship between them. Based on a first-order vector autoregressive model, an orthogonalised impulse response function (IRF) is employed to analyse the dynamic interaction between variables over the next 10 periods, in order to measure how the response variable changes over time—both immediately and in the future—following an external shock equivalent to one standard deviation of the shock variable.

(1)

The impact of digital transformation in the manufacturing industry on its own development

The impulse response results in Figure 5, Figure 6, Figure 7, Figure 8 and Figure 9 show that, across the national sample and the samples for the Eastern, Central and Western regions, the impact of digital transformation in the manufacturing industry on the sector itself is significantly positive. Furthermore, this positive effect gradually diminishes and tends towards zero, indicating that digital transformation in the manufacturing industry has a sustained promotional effect on the sector’s own level of transformation. In the samples from the Northeast region, this positive response is equally significant, but the rate of convergence is relatively slower. This suggests the existence of significant cumulative investment effects and learning reinforcement mechanisms during the digital transformation process. On the one hand, enterprises’ initial investments in digital infrastructure, technology platforms and talent reserves generate sustained returns through economies of scale and network effects; on the other hand, the knowledge and experience accumulated by organisations during the application of digital technologies continuously optimise the transformation pathway, creating a virtuous cycle of self-improvement. The slower rate of response attenuation in the Northeast region may stem from the fact that, given its heavy industrial structure, digital transformation involves deeper-level restructuring of production processes and institutional reforms, resulting in stronger inertia in the transformation process. This also indicates that the long-term benefits of digital investment in this region are realised more gradually, requiring more sustained policy support and patience on the part of enterprises.

(2)

The impact of digital transformation in manufacturing on green transformation

The impulse response results in Figure 5, Figure 6, Figure 7, Figure 8 and Figure 9 indicate that in the national and Eastern samples, digital transformation in manufacturing exerts a sustained positive effect on green transformation, with response curves showing an initial rise followed by a gradual decline. The Western region also exhibits a positive response, albeit of relatively weaker intensity. The Northeast region demonstrates the most significant and enduring positive response. In contrast, the impulse response results for the Central region exhibit fluctuations in certain intervals, and the intensity and persistence of the positive effect are relatively weak. This suggests that resource allocation frictions may exist in this region during the transformation process; that is, under resource constraints, there may be competitive allocation of corporate investment between digitalisation and greening. Furthermore, if local industrial policies fail to effectively guide the coordinated advancement of these two types of transformation, it may also make it difficult for enterprises to balance short- and long-term development objectives, thereby hindering the full realisation of transformation synergy.

(3)

The impact of the green transition in manufacturing on the development of digital transformation

The impulse response results in Figure 5, Figure 6, Figure 7, Figure 8 and Figure 9 show that in the national and Eastern samples, the green transition has a positive impact on digital transformation, albeit of relatively limited intensity. In the Central region sample, the response is positive in the short term but gradually weakens thereafter. In the Western region sample, the response curve fluctuates relatively gently, and the overall impact is not significant. The Northeast, however, exhibits a significant and sustained positive response. This suggests that the region’s green transition may, through stringent environmental regulations, compel enterprises to enhance the transparency of production processes and resource utilisation efficiency, thereby generating a systematic and rigid demand for digital monitoring, management and optimisation technologies, which in turn drives digital transformation. The weaker response in other regions may reflect that their green transition remains at the end-of-pipe governance stage, failing to effectively stimulate the endogenous motivation for enterprises to undertake deep digital transformation. At the same time, the mismatch between environmental regulations and digital infrastructure, as well as the tendency to view greening as a compliance cost rather than a strategic opportunity when facing multiple transformation pressures, may all inhibit the synergistic effects between the two.

(4)

The impact of the green transition in manufacturing on its own development

The results in Figure 5, Figure 6, Figure 7, Figure 8 and Figure 9 indicate that, for the national sample as well as the Eastern and Central region samples, the response to endogenous shocks from the green transition in manufacturing is significantly positive, and this positive impact gradually diminishes towards zero, suggesting that the green transition in manufacturing has a sustained promotional effect on its own level of transition. In the samples from the Northeast, the positive response of the green transition to its own impact is significant and sustained over a longer period, suggesting that its self-reinforcing mechanism is relatively stable. In contrast, the green transition of the manufacturing industry in the West exhibits a negative response to its own impact from the second period onwards, gradually diminishing towards zero. This indicates that the inertial effect of the green transition in the West is of shorter duration and exhibits weaker reliance on earlier stages, which may stem from the fact that green transformation in the Western region remains in the early stages of policy-driven and externally supported development. A virtuous cycle of market-driven momentum and voluntary corporate investment has not yet fully taken shape, resulting in relatively insufficient continuity and self-accumulation capacity in the transformation process.

4.3.6. Analysis of Variance

Based on GMM regression estimates and impulse response analysis, we further assess the long-term interactive relationship between digital transformation and green transformation in the manufacturing industry, as well as the extent of mutual influence between the two variables, through variance decomposition. The results of the variance decomposition are shown in

Table 7. Results of variance decomposition.

Variables	Episode Number	lnDTM					lnGTM
		Nationwide	East	Central	West	Northeast	Nationwide	East	Central	West	Northeast
lnDTM	10	0.868	0.945	0.790	0.785	0.863	0.074	0.173	0.210	0.045	0.071
	20	0.859	0.941	0.778	0.772	0.854	0.082	0.190	0.052	0.050	0.079
lnGTM	10	0.132	0.055	0.047	0.215	0.137	0.926	0.827	0.953	0.955	0.929
	20	0.141	0.059	0.222	0.228	0.146	0.918	0.810	0.948	0.950	0.921

. The decomposition results for the 10th and 20th periods are largely stable, indicating that the contributions of the various shocks to the fluctuations in the variable tend to balance out over the medium term. In terms of self-contribution, both digital transformation and green transformation exert a dominant influence on their own variance, with self-contribution rates generally exceeding 70% at both the national and inter-regional levels. In terms of interactive effects, significant heterogeneity is observed across regions: the Central region exhibits a relatively strong influence of digital transformation on green transformation, at 21%, while the Western region shows the lowest at merely 4.5%; the influence of green transformation on digital transformation is generally low at both the national and regional levels, indicating that the feedback effect of green transformation on digital transformation is, on the whole, limited. Overall, although digital transformation exerts a certain degree of driving force on green transformation in the Eastern and Central regions, the explanatory power of the interaction between the two remains significantly lower than their individual contributions. This reflects that, in the process of coordinated digital and green transformation within China’s manufacturing industry, a system of deep integration and mutually reinforcing mechanisms has yet to be established. There are clear shortcomings in the pathways through which green transformation feeds back into digital transformation, and regional coordination models and the intensity of interaction require further optimisation.

5. Discussion

This study found that the level of synergistic digital and green transformation in China’s manufacturing industry exhibits a distinct spatial gradient pattern: the Eastern region continues to lead, the Central region is catching up rapidly, the Western region is developing steadily, while the Northeast region is growing at a relatively slower pace. These findings are consistent with existing research on regional development imbalances (Li et al., 2022) [18]. An analysis of interaction effects using the XGBoost-SHAP model further revealed non-linear characteristics. When the scale of digital content resources is at a low level, an increase in the scale of internet address resources can significantly enhance the positive impact of the former; however, when both the scale of digital content resources and the concentration of R&D institutions (x15) are at high levels, the marginal contribution of factor inputs tends to level off, exhibiting a pattern of diminishing returns to scale.

Estimation results from the panel vector autoregression model reveal significant regional heterogeneity in the interactive relationship between digital transformation and green transformation in the manufacturing industry. In the Northeast region, the two exhibit a pattern of bidirectional positive interaction; this finding validates that environmental regulatory pressures prompt enterprises to utilise digital means to improve resource utilisation efficiency, thereby forming a positive feedback loop where green transformation drives digital transformation and digital transformation empowers green transformation. In the Eastern and Western regions, digital transformation exerts a unidirectional promotional effect on green transformation, although the intensity of this effect is relatively weaker in the Western region. While this result corroborates the theory of technology empowerment (Yang et al., 2022) [35], it also reflects the constraining effect of disparities in digital infrastructure levels on transmission efficiency. A unique short-term inhibitory effect has emerged in the Central region, wherein green transformation exerts a negative impact on digital transformation in the short term.

Compared with existing research, the innovation of this study lies primarily in the following three aspects. First, by integrating the improved distance-based synergy model, the XGBoost-SHAP method and the panel vector autoregression model within a single analytical framework, we have achieved a logical leap from static level measurement to dynamic causal identification. Second, we have identified three distinct patterns of regional interaction—two-way positive interaction, one-way promotion, and short-term inhibition—thereby correcting the linear assumption that digitalisation and greening always exhibit a positive synergistic relationship. Third, we have revealed the non-linear threshold effects of key driving factors, providing empirical evidence for the formulation of differentiated regional industrial policies.

This study found that the level of synergistic digital and green transformation in China’s manufacturing industry exhibits a distinct spatial gradient pattern: the Eastern region continues to lead, the Central region is catching up rapidly, the Western region is developing steadily, and the Northeast region is growing at a relatively slower pace. These findings are consistent with existing research on regional development imbalances (Li et al., 2022) [18]. An analysis of interaction effects using the XGBoost-SHAP model revealed non-linear characteristics. When digital content resources are at a low level, increasing the scale of internet address resources significantly enhances the contribution of the former; however, when both digital content resources and the concentration of R&D institutions (x15) are high, the marginal contribution tends to level off, indicating diminishing returns.

Estimation results from the panel vector autoregression and GMM models reveal significant regional heterogeneity in the interactive relationships between digital transformation and green transformation. In the Northeast region, the two variables exhibit a bidirectional predictive relationship; that is, past changes in one variable provide information useful for predicting future changes in the other, reflecting predictive precedence rather than definitive causality. In the Eastern and Western regions, digital transformation shows a unidirectional predictive influence on green transformation, although the effect is weaker in the Western region. Similarly, a short-term inhibitory effect is observed in the Central region, wherein past values of green transformation provide predictive information about digital transformation outcomes. This result also provides theoretical support for the technology empowerment perspective, corroborating the argument that digital city construction can facilitate improvements in environmental efficiency (Yang et al., 2022) [35].

Compared with existing research, the innovation of this study lies in: (1) integrating the improved distance-based synergy model, XGBoost-SHAP, and panel vector autoregression into a single framework to transition from static level measurement to dynamic predictive analysis; (2) identifying three distinct patterns of regional interaction—bidirectional predictive association, unidirectional promotion, and short-term inhibition—correcting the linear assumption that digitalisation and greening always have a positive synergistic relationship; (3) revealing non-linear threshold effects of key drivers, providing empirical support for differentiated regional industrial policies.

6. Conclusions and Implications

6.1. Conclusions

Based on panel data from 30 Chinese provinces covering the period 2011–2023, this paper constructs a framework for comprehensive evaluation and mechanism analysis to examine the level of synergistic transformation between digitalisation and greening in the manufacturing industry. The main conclusions are as follows:

(1) the overall level of synergistic transformation shows a steady upward trend with notable regional disparities, forming a spatial pattern of “East leading, Central catching up, West stable, and Northeast slower”; (2) XGBoost-SHAP analysis identifies the scale of digital content resources, internet address resources, high-tech product sales, and expenditure on energy conservation and environmental protection as core drivers, with cumulative contribution exceeding 66%; digital infrastructure, green innovation output, and fiscal support show significant positive effects, while some innovation platforms and enterprise investment variables exhibit diminishing returns at higher levels; (3) GMM regression results indicate significant regional heterogeneity in the interaction between digital and green transformation, with bidirectional predictive associations in the Northeast, unidirectional predictive promotion in the East and West, and short-term inhibitory effects in the Central region.

6.2. Practical Significance

Based on the aforementioned theoretical analysis and empirical findings, we hereby propose the following four policy recommendations.

Firstly, a differentiated regional coordination strategy should be implemented to optimise the spatial development pattern. The Eastern regions should focus on breakthroughs in cutting-edge technologies and, building on clusters such as the Yangtze River Delta and the Pearl River Delta, establish demonstration zones for the integration of digital and green technologies. The Central and Western regions need to strengthen the introduction of digital infrastructure and green technologies, prioritising the deployment of integrated projects such as the Industrial Internet and new energy. In response to the short-term dampening effect of the green transition on digitalisation in the Central regions, the direction of fiscal guidance should be optimised, and transitional support policies explored. The Northeast regions should accelerate the transformation of traditional manufacturing and promote the transition of heavy industry towards smart and green manufacturing.

Secondly, strengthen investment in key factors to enhance systemic synergy. Extend the coverage of new infrastructure, such as data centres and the Internet of Things, to Central, Western and Northeastern regions to bridge the ‘digital divide’. Optimise the structure of green innovation R&D subsidies, redirecting support towards digital-green convergence technologies and establishing a dynamic adjustment mechanism based on output efficiency. Encourage enterprises, universities and research institutes to form ‘industry–academia–research–application’ joint innovation consortia to accelerate the commercialisation of green patents and digital technologies.

Thirdly, establish a mechanism for mutual empowerment between digitalisation and greening to break down barriers to synergy. Overcome departmental and regional barriers to promote the deep integration of digitalisation and greening in strategic planning, industrial policy and environmental regulation. When formulating digital economy development plans, clearly define green and low-carbon constraint indicators; when introducing environmental protection policies, provide financial support for accompanying digital solutions. For provinces lagging in synergy levels, strengthen cross-departmental policy coordination to prevent green investment and digital investment from crowding each other out.

Fourthly, improve the dynamic monitoring and forecasting system to enhance policy responsiveness. Leveraging big data and artificial intelligence technologies, establish a unified national platform for monitoring collaborative transformation to enable real-time assessment of provincial-level collaboration, trend forecasting and risk early warning. Through policy simulation and impact assessment, provide governments at all levels with a tailored toolkit of policy measures, and establish robust mechanisms for performance evaluation and dynamic adjustment to accelerate the high-quality collaborative transformation of the manufacturing industry.

6.3. Limitations and Future Prospects

This study is subject to certain limitations. Firstly, regarding the construction of indicators, due to constraints on the availability of inter-provincial data, micro-level enterprise indicators could not be fully incorporated, which may have affected the comprehensiveness of the measurement to some extent. Secondly, in the analysis of feature correlations, although the XGBoost model identified key influencing factors, the underlying mechanisms behind the non-linearity and threshold effects exhibited by certain variables require further exploration through a more extensive range of case studies or micro-level data. Future research could be expanded in the following directions: firstly, by refining the scale of analysis from the provincial level down to the urban agglomeration or enterprise level, in order to capture more micro-level dynamics of collaborative transformation; secondly, by exploring the spillover effects of digital and green transformation on the resilience of manufacturing industrial chains, employment structures and international competitiveness, thereby broadening the scope of the research. Thirdly, although the data covered by this study spans the COVID-19 pandemic period (2020–2022), no specific robustness tests were conducted for this period (such as excluding the pandemic years or introducing dummy variables). Whether the pandemic caused a structural break in the long-term interaction between the digital and green variables remains to be verified by future research.