Introduction

At the global level, environmental sustainability has emerged as a central concern of recent international agendas. The outcomes of COP27 and COP28 reaffirmed the urgent need to transition away from fossil fuels, triple renewable energy capacity, and double energy efficiency improvements by 2030, as concrete steps toward achieving Sustainable Development Goals (SDG 7 on affordable and clean energy and SDG 13 on climate action)1,2,3. Recent empirical evidence further suggests that renewable energy expansion, green finance instruments, and political stability play a significant role in reducing carbon emissions and advancing SDG 134,5. Within this broader context, air pollution has been recognized as a critical environmental challenge that intersects with energy use and climate change. In the context of rapid industrialization and urbanization, the grave threat that air pollution poses to socio-economic development, human health, and natural ecosystems has become an undeniable fact6,7. Several studies have highlighted the profound impacts of air pollutants, particularly PM2.5, on developing countries8,9,10. As Zhao et al. (2022) pointed out, these pollutants not only deteriorate air quality but also pose significant public health risks10. Despite efforts to combat air pollution, data from China’s Ministry of Ecology and Environment shows that the PM2.5 concentration in many cities remained above the national air quality standards in 2018, with regions such as the Beijing-Tianjin-Hebei area and the Yangtze River Delta—economically developed areas—suffering particularly notable impacts, demonstrating a significant spatial clustering characteristic of air pollution.

Another reality related to air pollution is China’s continuous efforts to optimize its investment environment to attract substantial international capital inflow. According to data from China’s Ministry of Commerce, the total amount of foreign capital attracted increased from USD 124.1 billion to USD 141.0 billion from 2012 to 2019, with an average annual growth rate of 2.99%. FDI has provided China with the capital, advanced technology, and management expertise necessary for economic growth. However, the relationship between FDI and air pollution is complex and multifaceted, becoming a broad topic of academic research. Recent studies have shown that the impact of FDI on the air quality of host countries can be positive or negative11,12, influenced by factors such as regional policies13, human capital14, and the capacity to absorb and implement GTI15.

FDI is considered to have the potential to promote economic growth and technological progress16. In the context of GTI, FDI can act as a catalyst, transferring advanced technology and management practices to developing countries, thereby improving local air quality17. This suggests that, under the right conditions, FDI can contribute to the development of GTI and processes that mitigate air pollution. Moreover, FDI significantly promotes GTI in a nonlinear effect, depending on whether the host country city’s absorptive capacity exceeds a certain threshold18. However, research also points to the complex dynamics between FDI and environmental sustainability, highlighting the potential negative impacts of FDI19. Without the innovation of green technologies and attention to sustainable practices, FDI may lead to increased pollution and resource depletion, especially in developing countries with lower technological levels20,21.

The interaction between GTI and air quality further complicates this relationship. The effectiveness of GTI in improving environmental outcomes is widely recognized12,22, and adopting GTIs is crucial for mitigating climate change and ecological modernization23,24. Specifically, Zhang and Liu (2022)revealed through a spatial econometric model that GTI can improve carbon productivity25, while studies by Yi et al. (2022)26 and Xu et al. (2021)27 verified that GTI can suppress air pollution in Chinese provinces and cities, respectively. However, although FDI may bring about technology transfer, the control of core technologies is often tightly held by foreign enterprises, and may not be easily transferred to local firms28. Wang et al. (2023)’s study on emerging European countries also shows that although technological innovation has mitigated the extent of environmental degradation, FDI has actually exacerbated environmental problems12. In summary, the relationship between FDI, GTI, and air pollution is complex, but promoting GTI is key to FDI improving air quality and achieving environmental sustainability. This provides a reference for our research.

Several unresolved questions remain. Firstly, it’s important to clarify that, unlike traditional technological innovations which primarily focus on production efficiency and economic growth29,30, GTIs aim to balance economic benefits and environmental protection within a country31,32,33,34. However, the potential environmental impacts of FDI may be hindered by the technological control of investors or the lack of strong technology absorption capability in the host country. Moreover, the emissions of PM2.5, which are closely related to the formulation of air quality management policies aimed at solving air quality issues, necessitate more detailed sample data analysis to evaluate their actual effects. Additionally, the significant spatial differences in PM2.5 levels across cities suggest that focusing solely on whether FDI alters the technological level, thereby indirectly affecting air pollution, may not clearly determine whether economic growth and air pollution are truly decoupled. This also indirectly explains why previous studies have provided inconsistent answers to this question. Furthermore, many studies have not considered the potential bidirectional causal relationship between overseas investment and air pollution, leading to biased estimation coefficients. We ask four questions. RQ1: Does FDI worsen urban PM2.5 in China? RQ2: Does GTI directly improve air quality? RQ3: Does GTI mitigate the pollution impact of FDI. RQ4: Is there a GTI threshold beyond which the FDI–pollution relationship changes?

To address these issues, we utilize urban panel data from 2008 to 2020 to determine whether and to what extent GTI at the city level in China, a rapidly developing economy, changes the functional relationship between FDI and air quality. Secondly, considering the spatial diffusion characteristics of air pollution, this study employs optical satellite remote sensing to obtain annual average PM2.5 concentration data for Chinese cities and uses GS2SLS to control for spatial correlation and the endogeneity of variables, ensuring the unbiasedness and validity of the empirical results. This method also enables temporal and spatial analysis of air quality differences across cities. Lastly, and importantly, we address the issue of using overly simplistic spatial weight matrices in related research by constructing traditional geographical distance spatial weight matrices, economic distance spatial weight matrices, and nested weight matrices of geographical and economic distances to enhance the accuracy and robustness of the analysis results. This paper contributes in three ways. (i) Methodological. We estimate a dynamic spatial specification and use GS2SLS to address simultaneity and spatial dependence, providing consistent estimation in the presence of spillovers and persistence. (ii) Mechanism. We bring a moderation-plus-threshold perspective to the role of GTI in the FDI–pollution nexus. (iii) Evidence. We assemble a city-level panel for 236 cities (2008–2020) and document strong spatial and temporal dependence in PM2.5, clarifying when and where FDI’s environmental impact is attenuated by green innovation.

Our research finds that FDI exacerbates air pollution in 236 Chinese cities. However, cities that adopt more GTI experience less negative impact. Notably, there is a turning point: once a city’s GTI reaches a certain level, the relationship between investment and pollution changes—it is no longer direct or linear. These insights are interesting to policymakers aiming to balance environmental and economic growth strategies, offering scientific references for precise policy implementation in countries facing similar challenges.

In the following sections, we first review the literature on the relationship between FDI, GTI, and air pollution, clarifying the theoretical assumptions and empirical background of this study. Then, we detail the methodology, data sources, and variable definitions employed in our research. On this basis, we present the results of the empirical analysis and discuss these findings. Finally, the study concludes with a summary of the main findings and recommendations for policymakers.

Literature review

FDI and air pollution

Prior studies report mixed evidence on FDI’s environmental consequences. The “pollution haven” view emphasizes relocation of pollution-intensive activities to jurisdictions with laxer regulation, whereas the “pollution halo” view stresses technology and management spillovers that may improve environmental performance. We position our analysis within this debate by focusing on city-level PM2.5 and explicitly modeling spatial spillovers that earlier work often abstracts from.

Since the late 20th century, the relationship between FDI and environmental pollution has been a focal point of research. The concentration of FDI in high-pollution industries such as heavy industry, chemical, and mining has raised widespread concerns about its environmental impact. Overall, there are several viewpoints.

First, studies supporting the “Pollution Haven Hypothesis” proposed by Copeland and Taylor (1994) argue that developed countries, facing strict environmental regulations and rising governance costs, tend to transfer high-pollution industries to developing countries with more lenient environmental laws, thereby worsening the host country’s air quality35. Wagner and Timmins (2009)found through an analysis of German manufacturing FDI that, under strict environmental regulations, the environmental impact of FDI in the chemical industry not only supports the “Pollution Haven Hypothesis” but also shows a clustering effect36. Bulus and Koc (2021) also found that FDI increased the emissions of carbon dioxide in South Korea37. Singhania and Saini (2021) examining 21 high-carbon-emitting countries in sub-Saharan Africa, found that FDI has a positive impact on environmental degradation, especially more evident in developing countries38. Wang et al. (2023) confirmed similar conclusions in an analysis of European emerging economies12. Additional empirical research supports the view that FDI leads to environmental degradation in host countries19,39,40. Recent global evidence also shows that trade openness and FDI are significant drivers of PM2.5 concentrations, reinforcing the pollution haven argument41.

However, another perspective exists, the “Pollution Halo Hypothesis,” which argues that FDI, by introducing advanced technology and management experience, can promote clean and green production in industries, thereby improving air quality. Zhang and Zhou (2016) using panel data from 30 provinces in China, found that FDI contributes to the reduction of CO2 emissions in China42. Saqib et al. (2023)through an ARDL (auto-regressive distributed lag) model analysis, concluded that FDI aids in improving the ecological environment in European countries43. Shao et al. (2019) used the VECM model to analyze the pressure of FDI inflows on the environment in BRICS countries, proving that the “Pollution Haven Hypothesis” does not exist44. Sung et al. (2018) confirmed from an industry perspective that FDI could reduce carbon emissions in China’s manufacturing sector by introducing clean technologies and expertise45.

Furthermore, research has also revealed regional heterogeneity and differences among pollutants in the impact of FDI on air pollution, indicating that the environmental effects of FDI are significantly influenced by the environmental policies of the host country. Zugravu-Soilita (2017) using panel data of manufacturing FDI from France, Germany, Sweden, and the UK between 1995 and 2008, showed that countries with lenient environmental policies are more likely to suffer from increased carbon emissions due to FDI17. Mert and Caglar (2020) after a co-integration analysis, found that FDI into low and middle-income countries supports the “Pollution Haven Hypothesis,” while FDI into high-income countries supports the “Pollution Halo Hypothesis.“46 Kivyiro and Arminen (2014) examined the impact of FDI on carbon emissions in six sub-Saharan African countries, finding that FDI led to increased carbon emissions in some countries but the opposite in others47. Liu et al. (2018) discovered that FDI had different impacts on different pollutants, improving China’s smoke and dust pollution levels but exacerbating wastewater and sulfur dioxide pollution48. Xu et al. (2021) concluded that there is an inverted U-shaped relationship between FDI and SO2 emissions49.

Overall, while related research is abundant, studies using structural models often face sensitivity issues with variable ordering and frequently overlook (or cannot address) spatial spillover effects. Although recent studies have employed spatial econometric methods to analyze the impact of FDI on environmental pollution50,51, they have only considered the spatial effects of environmental pollution in the regression process, ignoring the fact of air pollution’s temporal persistence, and the use of static models to estimate dynamic relationships has led to biased regression results. Additionally, existing studies often use provincial-level data in China9,52 and focus on pollutants such as CO2 and SO253,54 to some extent overlooking the fact that Chinese cities are densely distributed and PM2.5 is widespread at a more micro urban level. To test the applicability of the competing “Pollution Haven” and “Pollution Halo” hypotheses in the context of current urban air pollution issues in China, the first research hypothesis is as follows:

H1

FDI exacerbates air pollution in Chinese cities.

Green technology innovation and air pollution

A growing literature links GTI to improved environmental outcomes through cleaner production, energy efficiency, and diffusion of abatement technologies. Yet the strength of this link varies across places and over time, suggesting that GTI’s effect may interact with local industrial structure and external capital. Our setting allows us to quantify GTI’s direct association with PM2.5 while controlling for spatial dependence.

GTI, distinct from traditional technological innovation, significantly reduces resource consumption and pollution emissions, having an important impact on regional environmental improvement. This type of innovation is often defined as technologies and processes that reduce the use of raw materials and enhance air quality, thereby promoting sustainable development and the green transformation of industries55,56. Research indicates that technological innovation is one of the beneficial measures for reducing environmental degradation in countries like China, Africa, and Mercosur countries57,58.

Extensive literature supports the positive role of GTI in curbing air pollution and enhancing ecological sustainability12,14,32,59. Recent works further document that technological innovation exerts asymmetric environmental effects60, that the role of green technologies depends on policy uncertainty61, and that environmental technology can even mitigate carbon-emission inequality across countries62. Evidence from BRICS economies further supports this view, showing that GTI significantly reduces ecological footprints and strengthens environmental sustainability, particularly when supported by political stability5. Furthermore, Chen et al. (2022)found that technological innovation can significantly reduce firms’ pollution emissions63, and He et al. (2022)64 study verified the spatial spillover effect of technological innovation on mitigating urban PM2.5. Johnstone et al. (2017) discovered that advancements in environmental technologies in production help reduce pollutant emissions and improve pollution control efficiency65. However, technological progress is not always beneficial for the environment66; while it promotes the efficient use of resources, it may also stimulate more production and consumption, leading to increased resource consumption and waste generation67. Similarly, previous research has mainly focused on the differences in GTI between provinces in China, which might not be conducive to inter-city economic development and GTI68,69. Therefore, to test whether the improvement in GTI levels is beneficial for mitigating current air pollution in Chinese cities, we propose the second hypothesis:

H2

The elevation of GTI levels can effectively mitigate current air pollution in Chinese cities.

Moreover, Hao et al. (2020) analyzed data from Chinese provinces and found that FDI could enhance technological innovation levels, thereby reducing SO2 and dust emissions and effectively mitigating local air pollution70. In contrast, Rafique et al. (2020) based on data from BRICS countries, showed that the entry of FDI did not promote local technological innovation levels but increased CO2 emissions due to technological barriers71. Clearly, technological innovation may be one of the conduits between FDI and air pollution, but is this conduit efficient for city-level data in China? And are the environmental consequences of the possible interactions between FDI and GTI positive? To address these questions, we propose the third research hypothesis.

H3

GTI can alleviate the impact of FDI on air pollution in Chinese cities.

Further roles of GTI

Beyond its direct effect, GTI can shape how FDI translates into pollution. If local green capabilities facilitate adoption of cleaner processes by foreign-invested firms, GTI should dampen the FDI–pollution linkage. We therefore test both an interaction between FDI and GTI and a threshold in GTI, providing evidence on whether sufficiently strong GTI conditions can convert FDI’s impact from more harmful to less harmful.

The literature related to the mechanisms analyzed above shows that FDI can improve local air quality through GTI. However, it does not consider the changes that could come with the continuous improvement of the host country’s own levels of GTI. For example, as the level of GTI in cities increases, local firms may reduce their reliance on high-pollution foreign firms for technology, attracting high-quality, environmentally friendly foreign investment72, thus further positively impacting the local ecological environment. FDI can directly affect air pollution through industrial activities and play a key role in promoting GTI39, which in turn can mitigate air pollution. This resonates with evidence from the circular economy literature showing that higher innovation capacity can shift the FDI–environment nexus toward sustainability, and from OECD countries where strong governance offsets the negative ecological effects of resource overuse73,74. Therefore, to further explore the impact of the development level of GTI in Chinese cities on the relationship between FDI and air pollution, we propose the fourth research hypothesis.

Within the Environmental Kuznets Curve (EKC) perspective, environmental pressure reflects the joint influence of scale, composition, and technique. FDI chiefly expands scale and reshapes industrial composition, tending to raise emissions at lower development levels, whereas GTI operates through the technique channel by improving process efficiency and abatement. As local green capabilities deepen, technique effects can outweigh scale and composition effects, so the pollution consequence of a marginal increase in FDI weakens. Multinational entry may also generate knowledge spillovers—through demonstration, labor mobility, and supplier linkages—that cities with stronger GTI are better able to absorb and translate into cleaner production. This theoretical logic motivates our empirical focus on the FDI×GTI interaction and the existence of a GTI threshold within a spatial–dynamic setting.

H4

When the level of GTI surpasses a threshold value, the negative impact of FDI on air pollution begins to diminish.

Spatial dependence and spillovers

Air pollution and economic activities exhibit spatial clustering and cross-boundary propagation. Ignoring spatial dependence can bias estimates and understate policy externalities. We adopt a spatial–dynamic framework that captures both cross-city spillovers and temporal persistence in PM2.5, aligning identification and inference with the geography of the problem. At the regional scale, analyses of the Chengdu–Chongqing Economic Circle also reveal heterogeneous ecological outcomes driven by the interaction of natural and human activities, highlighting the spatial heterogeneity of carbon sinks and the importance of regional governance75.

In summary, on one hand, against the backdrop of rapid economic growth in China, the inflow of foreign capital plays a decisive role in economic development. However, the inflow of “dirty industries” from developed countries has intensified China’s air pollution, creating a so-called pollution haven effect76,77. On the other hand, the effectiveness of air pollution control depends on the level of GTI among cities. The continuous improvement of technological innovation levels in cities (regardless of the method) will gradually reduce their economic development’s dependence on FDI with lower technology levels, reduce resource consumption, and thus improve the ecological environment78. Therefore, in cities with higher levels of GTI, the exacerbating effect of FDI on air pollution will weaken. To clarify, our research hypothesis process is illustrated in Fig. 1.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Theoretical Hypothesis on the Relationship Among FDI, GTI and Air Pollution.

Data and methodology

Econometric model

Previous research has identified population, affluence, and technological levels as primary contributors to environmental issues, as suggested by the IPAT model79. The IPAT model decomposes environmental pressure into three components, expressed as follows:

$$\:I=P\times\:A\times\:T$$
(1)

Here, \(\:\:I\:\:\)represents environmental pressure, and \(\:P,\) \(\:A,\) and \(\:T\) respectively denote population, level of economic development, and technology level. However, a major limitation of the IPAT model is its assumption of unitary elasticity for each independent variable. To address this, Dietz and Rosa (1994) proposed the STIRPAT model, which allows for non-proportional impacts of these factors on environmental outcomes80. The STIRPAT model is generally expressed as:

$$\:I=\alpha\:{P}^{b}{A}^{c}{T}^{d}\epsilon\:$$
(2)

In Eq. (2), \(\:\alpha\:\:\)is the model coefficient, \(\:b,\:\:c,\:\)and \(\:d\) are the estimated coefficients for the three variables, and \(\:\epsilon\:\) is the error term. The equation in logarithmic form is as follows:

$$\:lnI=\alpha\:+b\left(lnP\right)+c\left(lnA\right)+d\left(lnT\right)+\epsilon\:$$
(3)

The STIRPAT model permits the inclusion of additional factors, providing a more nuanced framework for environmental research. For this study, we replace population(\(\:P\)) with population density, affluence(\(\:A\)) with economic development level, and technology (\(\:T\)) with foreign direct investment (FDI). To reflect the Environmental Kuznets Curve (EKC) hypothesis and the potential inverted U-shaped relationship between FDI and air pollution39, we also incorporate the squared terms of economic development level and FDI. The resulting relationship is specified as:

$$\:{PM}_{it}=f({FDI}_{it},{FDI}_{it}^{2},{GDP}_{it},\:{GDP}_{it}^{2},{Url}_{it})$$
(4)

Here, \(\:i\) and \(\:t\) represent the city and year, respectively; \(\:f\) the specific functional form, \(\:{PM}_{it}\) the dependent variable indicating air pollution concentration, \(\:{FDI}_{it}\) foreign direct investment, \(\:{GDP}_{it}\)the economic development level, and \(\:{Url}_{it}\) the population density.

To further explore the moderating effect of green technology innovation (GTI) on the FDI-air pollution relationship, we include an interaction term, drawing on Cheng et al. (2020)81 and Du et al. (2021)82:

$$\:{PM}_{it}={\alpha\:}_{0}+{\alpha\:}_{1}{FDI}_{it}+{\alpha\:}_{2}{GTI}_{it}+{\alpha\:}_{3}{FDI}_{it}\times\:{GTI}_{it}+{\alpha\:}_{4}{X}_{it}+{\epsilon\:}_{it}$$
(5)

In this equation \(\:{PM}_{it}\) represents the annual average concentration of PM2.5, \(\:{FDI}_{it}\) is foreign direct investment, and \(\:{GTI}_{it}\)is GTI. \(\:{FDI}_{it}\times\:{GTI}_{it}\) is the interaction term, capturing the moderating effect. \(\:{X}_{it}\) are control variables, \(\:{\alpha\:}_{0}-{\alpha\:}_{4}\:\)are coefficients to be estimated, and \(\:{\epsilon\:}_{\:i,t\:}\)represents the random disturbance term.

Building on this, we incorporate the spatial lag term of PM2.5 (\(\:Wln{PM}_{it}\)) and the time-lagged term(\(\:{lnPM}_{i,t-1}\)) to construct a dynamic spatial econometric model:

$$\:{lnPM}_{it}={\alpha\:}_{0}+\rho\:Wln{PM}_{it}+{\beta\:lnPM}_{i,t-1}+{{\alpha\:}_{1}FDI}_{it}{+{\alpha\:}_{2}GI}_{it}{+{{\alpha\:}_{3}FDI}_{it}{\times\:GTI}_{it}+\alpha\:}_{4}{X}_{it}+{\epsilon\:}_{it}$$
(6)

Here \(\:W\) is the spatial weight matrix, and \(\:\rho\:\) represents the regression coefficient for the spatial lag term, capturing the spatial spillover effects of air pollution. The coefficient \(\:\beta\:\) for the time-lagged term reflects the cumulative effect of air pollution over time.

Addressing the bidirectional causality between FDI and GTI poses a challenge. Pollution-intensive FDI inflows may worsen air quality, while severe air pollution may prompt stricter environmental regulations, potentially reducing FDI attractiveness. This endogeneity issue between explanatory and dependent variables makes traditional ordinary least squares (OLS) estimation methods unsuitable. To address this, we follow Boubacar (2016) and use the explanatory variables and their spatial lag terms as instrumental variables, accounting for geographic spatial correlations and ensuring robust estimation of FDI’s environmental impact83.

The choice of generalized spatial two-stage least squares (GS2SLS) represent a methodological innovation within the context of this research theme, particularly in addressing the endogeneity and spatial dependence inherent in environmental econometric models. This strategy exploits the spatial structure of the data and delivers consistent estimates under common regularity conditions in the presence of spillovers and persistence. It simultaneously tackles endogeneity and spatial dependencies more effectively than traditional methods such as OLS, MLE, SEM, and SDM. By employing spatially lagged variables as instrumental variables in a two-stage estimation process, GS2SLS ensures robust and efficient parameter estimates, capturing feedback loops and dynamic spatial interactions that provide a useful tool for understanding the relationship between FDI, air pollution, and green technology innovation.

Variable descriptions

Dependent Variable: This study utilizes panel data from 236 prefecture-level and above cities in China spanning 2008 to 2020 to analyze the mechanisms through which FDI and GTI influence PM2.5 levels. The dependent variable is the annual average concentration of global PM2.5, derived from gridded data published by the Socioeconomic Data and Applications Center (SEDAC) at Columbia University. Using ArcGIS software, we processed the raw data to calculate the annual average PM2.5 concentration for each city, which serves as a measure of air pollution.

Core Explanatory Variable: FDI is a key explanatory variable affecting air pollution. It is measured as the ratio of the actual amount of FDI to the city’s gross domestic product (GDP). This indicator captures the relative scale of foreign investment within the local economy.

Moderating and Threshold Variable: The level of Green Technology Innovation (GTI) serves both as a moderating and a threshold variable in this study. We measure GTI using the number of green patent applications in each prefecture-level city. Patents are a reliable indicator of GTI for two main reasons: They provide a direct, quantifiable measure of innovation output and impact, enabling regional comparisons. Secondly, through International Patent Classification (IPC) information, patents allow precise identification of green innovations, distinguishing them from other types of technological advancements. In line with common practice in the literature and recommendations by international organizations such as the OECD and WIPO, patent applications are particularly suitable because they offer a timely and comprehensive reflection of innovation activity. To ensure robustness, we further replace applications with granted patents, and the results remain consistent. Nevertheless, we acknowledge that patent indicators cannot capture all forms of green innovation (e.g., process improvements or non-patented innovations) and that patent quality may vary. For these reasons, patents should be regarded as a widely accepted but approximate measure of GTI.

Control Variables: We also introduce several control variables that may affect PM2.5, which are Industrial Structure (IS), Population Density (Url), Government R&D Investment (Gov), Economic Development Level (GDP), Infrastructure Construction (Inf), and Environmental Regulation (ER). Table 1 provides descriptive statistics for all key variables. Additionally, we define the variables and list the data sources in Supplementary Appendix A.

Table 1 Descriptive statistics of main Variables.
Full size table

Empirical analysis

Spatial autocorrelation Moran’s I test

Before analyzing spatial effects, it is necessary to verify the spatial correlation of variables. \(\:Mora{n}^{{\prime\:}}s\:I\) is a widely used measure to identify spatial correlations among regional samples. The formula is as follows:

$$\:Mora{n}^{{\prime\:}}s\:I=\frac{n\sum\:_{i=1}^{n}\sum\:_{j=1}^{n}{W}_{ij}({x}_{i}-\stackrel{-}{x})({x}_{j}-\stackrel{-}{x})}{\sum\:_{i=1}^{n}\sum\:_{j=1}^{n}{W}_{ij}\sum\:_{i=1}^{n}({x}_{i}-\stackrel{-}{x}{)}^{2}}$$
(7)

where \(\:\stackrel{-}{x}\:\)is the sample mean, and \(\:{W}_{ij}\) is the spatial weight matrix. The Moran’s Index ranges from − 1 to 1, where positive values indicate a positive (clustered) spatial pattern, negative values indicate a negative (dispersed) spatial pattern, and values close to 0 indicate no spatial correlation. The closer the Moran’s Index is to 1 or −1, the stronger the spatial correlation.

Calculating \(\:Mora{n}^{{\prime\:}}s\:I\) depends on spatial weights, and the estimation results of spatial econometric models are often influenced by the spatial weight matrix. To ensure the robustness of the estimation results and considering the geographic and socio-economic linkages between cities, this study selects the geographical adjacency spatial weight matrix (W1), economic distance spatial weight matrix (W2), and geo-economic nested spatial weight matrix (W3) to test for spatial autocorrelation effects and for subsequent spatial econometric analysis. The descriptions of these matrices are as follows:

Geographical adjacency spatial weight matrix

The spatial weight matrix expresses the degree of interdependence between geographical or economic attributes in different spatial regions, which is an essential component of spatial econometric analysis. The adjacency weight matrix considers the similarity between spatial units, more intuitively representing spatial adjacency relationships and dependencies.

$$\:{W}_{ij}=\left\{\begin{array}{c}1,\:\:\:\:i\ne\:j\\\:0,\:\:\:\:i=j\end{array}\right.$$
(8)

Cities\(\:\:i\) and \(\:j\) are considered adjacent if they share a common boundary or point, where \(\:{W}_{ij}=1\); otherwise, \(\:{W}_{ij}=\)0.

Economic distance spatial weight matrix

Considering the similarities in economic development levels among regions, a spatial weight matrix can be constructed by calculating the absolute difference of an economic indicator between spatial units to reflect the spatial effects between different spatial units. The economic distance weight matrix is constructed as follows:

$$\:{W}_{ij}=\left\{\begin{array}{c}\frac{1}{{|\theta\:}_{i}-{\theta\:}_{j}|},\:\:\:i\ne\:j\\\:\:0\:,\:\:\:\:\:i=j\:\:\:\:\:\:\:\:\end{array}\right.$$
(9)

Where \(\:{\theta\:}_{i}\) and \(\:{\theta\:}_{j}\) represent the difference in per capita GDP between cities \(\:i\) and \(\:j\) from 2008 to 2020. In the spatial matrix, the smaller the difference in per capita GDP, the larger the matrix element, indicating closer economic “proximity” between the two cities.

Geo-economic nested spatial weights matrix. The geo-economic nested matrix comprehensively considers the effects of geographic distance and economic factors between cities. According to the study by Case et al. (1993),84 the nested weight matrix is constructed as follows:

$$\:{W}_{3}=\rho\:{W}_{1}+(1-\rho\:){W}_{2}$$
(10)

Where \(\:\rho\:\) represents the weight of the geographical distance spatial weight matrix, ranging between 0 and 1. In this study, we equally consider the impacts of geographical and economic distances, setting\(\:\:\rho\:\:\) to 0.5.

This study tested the spatial correlation of air pollution among Chinese cities from 2008 to 2020 using prefecture-level spatial weight matrices. To facilitate the examination of the geographic and economic clustering characteristics of air pollution, tests were conducted using the W1, W2, and W3. The results of spatial correlations are presented in Supplementary Appendix B. The trend graph of Moran’s I index in Fig. 2 also shows a significant spatial autocorrelation relationship in the PM2.5 of Chinese cities.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Trend Chart of Global Moran’s I for PM2.5 Concentration in China, 2008–2020.

Given the potential for local spatial autocorrelation trends to diverge from global trends85, this study employs the Local Indicators of Spatial Association (LISA) to further analyze the local spatial relationships of PM2.5 pollution in cities. Using ArcGIS software, LISA cluster maps for PM2.5 in the years 2008, 2014, and 2020 were created, as detailed in Fig. 3 (2008), Fig. 4 (2014), and Fig. 5 (2020).

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

LISA agglomeration map of PM2.5 in 2008.

Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

LISA agglomeration map of PM2.5 in 2014.

Fig. 5
Fig. 5The alternative text for this image may have been generated using AI.
Full size image

LISA agglomeration map of PM2.5 in 2020.

The LISA clustering maps for 2008, 2014, and 2020 reveal that the concentration of PM2.5 in Chinese cities shows high clustering characteristics in regions such as Beijing, Tianjin, Hebei, Shandong, and Henan. This indicates that these areas have higher levels of PM2.5 pollution and stronger spatial correlations. Conversely, low-low clusters of PM2.5 are primarily found in Guangdong, Fujian, Yunnan, Inner Mongolia, and Heilongjiang, suggesting lower levels of PM2.5 pollution and weaker spatial correlations among cities in these regions. Thus, it is evident that the concentration of PM2.5 across Chinese cities exhibits significant positive spatial autocorrelation effects, indicating that cities with high pollution levels are often adjacent to each other, as are cities with low pollution levels. This observation underscores the validity of employing spatial econometric models for in-depth analysis in subsequent research.

Estimation of the full sample model

The comprehensive analysis demonstrates significant spatial correlations in PM2.5 concentration among Chinese cities, influenced by geographic and economic factors. This suggests that traditional regression models, which overlook spatial effects, may yield biased estimates. Therefore, following the approach proposed by Lesage and pace86, this study integrates the time lag of the dependent variable into a static spatial panel model framework, employing the GS2SLS method to simultaneously consider the spatial and dynamic effects of PM2.5 in the regression analysis.

Table 2 presents the results of the dynamic spatial econometric model analysis under three different spatial weight matrices. Columns (1) and (2) use W1, columns (3) and (4) employ the W2, and columns (5) and (6) utilize the W3. Based on Hausman test results, all estimated models opt for a fixed effects model. To ensure comparability, columns (1), (3), and (5) fix city-specific effects, while columns (2), (4), and (6) control for both time and city fixed effects.

Table 2 Results of the full sample baseline model estimation.
Full size table

Focusing on the spatial weight matrix W3, the regression results in Table 2, under the dual fixed effects of time and city, indicate that the first-order regression coefficient of the core explanatory variable FDI is 0.056, significant at the 1% level. This outcome validates the pollution haven hypothesis in Chinese cities54,87, suggesting the need for stricter regulation and oversight of air pollutant emissions from foreign investments, reassessment of current investment policies, and support for environmentally friendly projects conducive to urban sustainable development.

To examine the potential nonlinear relationship between FDI and air pollution, the square of FDI was included in the regression model. The negative coefficient of the squared term indicates a significant inverted U-shaped relationship between FDI and air pollution in Chinese cities. Initially, FDI may exacerbate PM2.5 concentration, but over time, factors such as technology transfer, regulatory strengthening, and technological upgrades will reduce air pollution, ultimately improving air quality, consistent with findings by Xu et al. (2021)54. This underscores the importance of creating an environment favorable to the absorption of green technologies and practices from foreign investments.

In the regression analysis of column (6) of our study, we discovered a significant negative impact of GTI on PM2.5 concentration, with a regression coefficient of −0.033, significant at the 1% statistical level. This finding strongly supports Hypothesis H2, suggesting that enhancing the level of GTI can significantly reduce urban air pollution. The effectiveness of GTI policies implemented in various cities in China demonstrates that these policies not only directly combat PM2.5 pollution but also help to regulate the environmental effects of FDI. Given these findings, policymakers should place greater emphasis on promoting the research and development (R&D) and application of green technologies.

Additionally, the coefficient of the interaction term between FDI and GTI is significant at the 1% level at −0.528. This indicates that the impact of an increase in the stock of FDI capital on air pollution varies with the level of GTI71,88, meaning that the level of GTI weakens the exacerbation of air pollution by FDI, confirming Hypothesis 3. This result underscores the critical role of R&D investment in green technologies and the importance of integrating these innovations into the broader economic and industrial structures of cities. It also suggests the need to continue advocating for GTI as a cornerstone for achieving air quality goals, emphasizing the necessity of encouraging domestic and foreign investment in the green sector.

Spatially, the estimated results of the spatial lag variable coefficients under three spatial weights are positive and significant at the 1% level, indicating significant spatial spillover effects of air pollution among Chinese cities. Air pollution is influenced by both atmospheric circulation diffusion and frequent economic exchanges between cities, making production activities related to one area easily spread to neighboring cities, thereby exacerbating air pollution across urban areas. This finding highlights the necessity for regional cooperation and coordinated policy efforts. It also calls for cities to adopt holistic air quality management approaches that transcend urban boundaries, advocating for shared standards, joint monitoring initiatives, and cooperative pollution control strategies.

From a temporal perspective, the time-lagged term of air pollution, lnPMi, t-1, is significantly positive across columns (1) to (6) in Table 2, indicating a cumulative effect of air pollution over time in Chinese cities. In other words, if a region experienced severe air pollution in the past and did not implement timely environmental policy interventions, air pollution would further worsen. The use of dynamic panels increases the accuracy of this coefficient, providing a reference for future research.

Turning our focus to the coefficients of control variables, the impact of industrial structure on PM2.5 in Chinese cities shows a significant negative correlation, indicating that a high proportion of the secondary industry plays a significant role in suppressing PM2.5. With China’s recent industrial structure upgrade and the implementation of low-carbon transition goals, the internal structure of the secondary industry has started to adjust, gradually transitioning from high energy consumption to low-carbon production. This shift reflects China’s commitment to upgrading its industrial sector and actively participating in the global value chain, indicating that industrial modernization and efficiency improvements are crucial for reducing environmental footprints.

The impact of urban population density on PM2.5 is not significant, suggesting that the density of urban areas may not be a determinant of air pollution levels by itself. Even in densely populated cities, urban planning and management practices can effectively mitigate pollution. However, Gov has, to some extent, suppressed the exacerbation of air pollution, indicating that greater government R&D spending leads to more pronounced effects on environmental pollution control. This supports the argument that government-driven technological advancements in the environment are crucial for addressing pollution and underscores the role of policy in guiding efforts to achieve sustainable development goals. Additionally, the estimated coefficients for the level of economic development (GDP), with the first term being positively related and the second term negatively related, indicate an inverted U-shaped relationship between economic development levels and PM2.5. This provides empirical support for the Environmental Kuznets Curve hypothesis89, suggesting that economic growth initially leads to increased pollution, which decreases as income levels rise and investments in clean technologies and practices increase.

The positive coefficient of infrastructure construction indicates its lagged impact on air quality, reflecting the complexity between new infrastructure projects (including those aimed at low-carbon urban development) and their short-term and long-term environmental impacts. This finding suggests the necessity of comprehensive environmental assessments during the planning and implementation phases of infrastructure projects to ensure they align with sustainable development goals. Regarding environmental regulation, the regression coefficients under three spatial weight matrices are negative, highlighting the effectiveness of stringent environmental policies in attracting high-quality FDI that complies with environmental standards, which helps to reduce PM2.5 levels. This supports the view that a strong regulatory framework can motivate foreign investors to adopt clean production methods and technologies, thereby enhancing the environmental performance of the host country90.

Using the same instrument set as in the GS2SLS specification, the Kleibergen-Paap rk LM test rejects under identification (p = 0.000), the Kleibergen-Paap rk Wald F statistic is 20.8, and the Hansen J test does not reject the over-identifying restrictions (p = 0.31), supporting instrument relevance and validity. Diagnostics computed in an auxiliary first stage (2SLS) using the same endogenous set and instruments as the GS2SLS.

These results provide insights into the factors affecting air quality in Chinese cities and emphasize the multi-faceted strategies needed to address air pollution. By placing these findings within the broader context of urban air quality management and decision-making, the analysis highlights the importance of integrated multi-sector strategies in achieving air quality improvements and sustainable urban development.

It is worth noting that, from the overall regression results, the role of the W1 and the W3 is significantly superior to that of the W2. This indicates that using economic distance information alone is not sufficient to make the results convincing. Therefore, subsequent research will be based on empirical analysis using the W1 and the W3.

Robustness tests

This study also conducted robustness tests on the empirical results in the following ways. First, by replacing the dependent variable. The concentration of PM2.5 in various cities was replaced with the amount of industrial sulfur dioxide emissions (ISD) to re-estimate and analyze the regression model. Second, by replacing the explanatory variable. The amount of green patent grants (GTI_a) in each city was used to replace the green patent applications for re-estimation. The estimated results are shown in Table 3.

Table 3 Results of robustness Tests.
Full size table

Columns (1), (2) and (3), (4) in Table 3 present the estimated results after replacing the dependent variable PM2.5 concentration and the level of GTI indicator, respectively. From the results, it can be observed that the signs of the estimated coefficients for the main explanatory variables, FDI and GTI, remain consistent overall with those in Table 2. This indicates that the research results of this paper are robust and reliable. That is, at the current stage, FDI has exacerbated air pollution in Chinese cities, while the level of GTI has had a significant positive impact on improving air quality in various cities. Additional robustness results—winsorization, placebo timing, and subsamples by fossil-fuel reliance—are reported in Supplementary Appendix C and confirm the stability of our findings.

Further analysis: threshold effect and regional spatial heterogeneity

So far, to a certain extent, we have demonstrated that the functional relationship between FDI and air pollution varies with the level of GTI. However, the actual turning point is not clear, which may not provide precise policy references for government investment attraction. Therefore, this paper adopts the threshold regression model proposed by Hansen (1999),91 setting GTI as the threshold variable, to further discuss the nonlinear impact mechanism of FDI on PM2.5. The model constructed is as follows:

$$\begin{aligned} \:lnPM_{{i,t}} = & \alpha \:_{0} + \gamma \:_{1} FDI_{{i,t}} \cdot \:I\left( {GTI_{{i,t}} \le \:\mu \:_{1} } \right) + \gamma \:_{2} FDI_{{i,t}} \cdot \:I\left( {\mu \:_{1} < GTI_{{i,t}} \le \:\mu \:_{2} } \right) + \ldots \: \\ & + \gamma \:_{n} FDI \cdot \:I\left( {\mu \:_{{n - 1}} < GTI_{{i,t}} \le \:\mu \:_{n} } \right) + \gamma \:_{{n + 1}} FDI \cdot \:I\left( {GTI_{{i,t}} > \mu \:_{n} } \right) \\ & + \alpha \:_{1} X_{{i,t}} + {\varepsilon}_{{i,t}} \\ \end{aligned}$$
(11)

Where the threshold variable \(\:{GTI}_{i,t}\) represents the level of green technological innovation in cities, \(\:{\gamma\:}_{1},\:{\gamma\:}_{2}\dots\:{\gamma\:}_{n},\:{\gamma\:}_{n+1\:}\)represent the stage effects of FDI on air pollution under different levels of GTI, respectively. \(\:I(\bullet\:)\:\)is an indicator function that takes the value of 0 or 1, \(\:{\mu\:}_{i}\:\) represents specific threshold values, and other variable symbols are the same as in Eq. (1).

Table 4 Regression results of the threshold panel model.
Full size table

After 500 Bootstrap resampling iterations, GTI passed the single threshold test but failed the multiple threshold tests. According to the estimation results in Table 4, we observe a significant threshold effect at the 1% significance level. Specifically, when the level of GTI is below the threshold value of 2.773, the contribution coefficient of FDI to air pollution is 1.245. However, once the GI indicator exceeds this threshold value, the contribution coefficient of FDI to air pollution decreases to 1.078. This threshold effect implies that the ability of cities to utilize FDI to improve air quality is not static but changes with their technological innovation patterns. Cities with advanced GTI capabilities can better attract high-quality FDI that aligns with environmental sustainability goals, thereby reducing dependence on pollution-intensive foreign investments. Conversely, cities where the green technology sector is just emerging may suffer adverse environmental impacts from FDI, highlighting the need for targeted policies to enhance their level of GTI.

Furthermore, the threshold effect is also influenced by external factors such as government policies, market demand, and the global technological environmen75,92. Policies that encourage green technology research and development, provide incentives for clean energy investments, and enforce strict environmental regulations can raise the threshold level92, making cities more likely to achieve positive environmental outcomes from FDI. Because the threshold is identified as a ratio of coefficients in the interaction specification, it is scale-dependent. Within the GTI range observed in our sample, GTI delivers a sustained mitigating effect on the FDI–PM2.5 linkage, but it does not reach an in-sample cut-off that would fully offset or reverse FDI’s marginal pollution effect.

Considering the regional heterogeneity of cities, this study divides 236 Chinese sample cities into two sub-samples based on their location in the eastern, central, and western parts of the country. According to the classification by the National Bureau of Statistics of China, the eastern provinces include Liaoning, Hebei, Beijing, Tianjin, Shandong, Jiangsu, Shanghai, Zhejiang, Fujian, Guangdong, and Hainan, with all other provinces categorized as central or western regions. Table 5 presents the regression results for these sub-samples.

Table 5 Results of the regional heterogeneity test divided by region.
Full size table

The regression outcomes in Table 5 indicate that the coefficients for FDI are significantly positive for both eastern and central-western cities, validating the pollution haven hypothesis. This suggests that cities across China, regardless of their geographical location, tend to attract foreign investments that could potentially exacerbate local air pollution levels. This phenomenon is more pronounced in central-western cities than in eastern cities, indicating that FDI in these areas might lead to higher vulnerability or lower environmental degradation thresholds.

The estimated coefficients for the level of GTI are significantly negative under both the W1 and the W3, demonstrating that the enhancement of GTI levels has contributed to the improvement of urban PM2.5 concentrations across Chinese cities. Notably, the mitigating impact of GTI on air pollution is more pronounced in eastern cities compared to central-western cities. This difference clearly reflects the influence of factors such as the level of economic development, industrial structure, and existing technological infrastructure, which enable eastern cities to more effectively utilize green technologies to improve the environment. These findings highlight the critical importance of targeted regional strategies in addressing environmental challenges. For central-western cities, policies that attract cleaner FDI and foster local green innovation capabilities are essential. For eastern cities, the research underscores the significance of continuing to advance green technology and innovation to further reduce air pollution levels.

Discussion

Several key points can be further interpreted according to our findings. Our study reveals that the inflow of FDI exhibits characteristics of the “pollution haven” hypothesis, as suggested by Xu et al. (2021),49 Dhrifi et al. (2020),87 and Naz et al. (2019)93. These studies indicate that FDI tends to flow into developing countries with relatively lax environmental regulations, leading to a deterioration in local air quality. However, previous research has predominantly focused on the impact of FDI on pollutants like CO2 and SO253,54, whereas our study extends the examination to PM2.5, providing a significant addition to the existing literature. This focus on PM2.5 is crucial as it directly impacts public health, making our findings highly relevant for policymakers seeking to mitigate adverse health outcomes associated with air pollution. Moreover, green technology innovation (GTI), as a novel technology and process, plays an indispensable role in mitigating the environmental impact of FDI. GTI aims to reduce environmental burdens and is closely aligned with sustainable development principles. Our study shows that the implementation of GTI policies in various Chinese cities has been effective, significantly reducing resource consumption and pollutant emissions, and positively impacting PM2.5 concentrations. This demonstrates the potential of GTI in balancing economic development with environmental sustainability. Behera and Sethi (2022) found that government regulations and environmental policies often encourage the adoption of GTI through incentives16. This finding complements our own, collectively emphasizing that policymakers should consider integrating GTI initiatives into broader economic and environmental strategies to ensure a more sustainable future.

Furthermore, the advanced environmental technologies and stringent regulations in the eastern region of China have been crucial in effectively controlling the environmental impact of FDI, thereby reducing air pollution94. In contrast, the central and western regions, with lower levels of economic development, have been more inclined to attract heavy industries and energy-intensive FDI to spur rapid economic growth. This phenomenon supports the “pollution haven” hypothesis and underscores the need for flexible and adaptive policy measures tailored to specific regional economic and environmental contexts95. This regional heterogeneity highlights the necessity of localized policy interventions to address the distinct challenges faced by different areas.

We also observed the spillover and cumulative effects of PM2.5 over space and time, indicating that pollution not only affects local cities but can also spread to surrounding areas. This finding aligns with Jiang et al. (2018)96. Enhancing the level of GTI across cities can help mitigate this issue. Specifically, through industrial restructuring, increased government investment in R&D, and stronger environmental regulations, significant improvements in urban air quality can be achieved. Additionally, policymakers should focus on increasing investments in green technology R&D97, and formulating effective environmental policies to promote sustainable urban development and improve air quality. Such measures are not only vital for China but also provide valuable insights for other developing countries facing similar challenges16.

Conclusions and policy recommendations

Conclusions

Using the GS2SLS dynamic spatial econometric model, this study analyzes data from 236 Chinese cities between 2008 and 2020 to explore the relationship between FDI, GTI, and air pollution. Three findings stand out. First, on average, higher FDI is associated with higher PM2.5, consistent with a pollution-haven mechanism under heterogeneous regulatory stringency and industrial composition. Second, GTI is directly linked to lower PM2.5, suggesting that green capabilities help reduce pollution through cleaner production, abatement, and efficiency gains. Third, GTI conditions how FDI translates into pollution: the FDI×GTI interaction is negative, and threshold analysis indicates that beyond a GTI cut-off the marginal pollution effect of FDI attenuates. We also document strong spatial spillovers and temporal persistence in PM2.5, underscoring the need to model cross-city propagation and dynamics. Heterogeneity analyses indicate that the adverse association between FDI and PM2.5 is stronger where green capabilities are weak and industrial structures are more fossil dependent, whereas cities with stronger GTI show a muted FDI–pollution link. Robustness checks using alternative measures and specifications support these patterns. Overall, the results highlight that strengthening local green innovation capacity is pivotal to reconciling external capital with cleaner urban air.

Policy recommendations

The evidence points to a joint approach: steer the composition of FDI while lifting GTI to—ideally beyond—the operative threshold so that external capital arrives with cleaner processes and lower marginal pollution. In practice, authorities can strengthen environmental screening at entry and expansion, align fiscal incentives with verifiable abatement outcomes, and create “green lanes” for projects deploying best-available technologies while tightening scrutiny for pollution-intensive investments. At the same time, expand city-level GTI and absorptive capacity through mission-oriented green R&D, targeted support for low-emission process upgrading, university–industry collaboration, pilot/testing infrastructure, and green finance (e.g., credit enhancements and transition bonds) to accelerate retrofits along supplier networks. Translate the estimated GTI threshold into operational targets—such as green patents per capita or verified adoption rates of energy-efficient equipment—and track progress annually.

Given spatial spillovers, coordination across airsheds is essential. Regional platforms for monitoring, data sharing, siting, and joint enforcement can reduce cross-boundary leakages. Policies should also reflect development stages: in more advanced coastal cities, prioritize upgrading existing FDI supply chains and attracting clean-tech-intensive projects; in less-developed or fossil-dependent cities, first build GTI and regulatory capacity, tighten entry standards for high-emission projects, and support process retrofits in anchor industries. International cooperation that facilitates the diffusion of advanced foreign green technologies and engagement in standards setting can further raise the quality of incoming FDI.

Limitations and future research directions

This study has several limitations that provide opportunities for further research. First, this study measures green technology innovation using green-patent indicators, primarily patent applications. This measure is widely used in the literature and provides a timely, quantifiable reflection of innovation output across cities. To enhance robustness, we also replace applications with granted patents, and the results remain consistent. Nevertheless, patent-based indicators have inherent limitations: they cannot fully capture technology quality, adoption intensity, or abatement effectiveness, and they may overlook non-patented forms of innovation such as process improvements or managerial practices. Future research could refine the measurement of GTI by distinguishing among different types of green innovation and by incorporating additional proxies such as R&D expenditures or firm-level environmental performance.

Second, although a spatial–dynamic specification and instrumental variables are employed, residual endogeneity arising from unobserved city policies or firm strategies cannot be fully ruled out. Future research could report additional diagnostic tests and translate the GTI threshold into more policy-relevant metrics (e.g., percentiles or adoption rates) to improve interpretability and practical application. Finally, linking firm-level FDI attributes to plant-level emissions offers a promising avenue to uncover the specific channels through which GTI moderates the FDI–pollution nexus.