The Impact of Smart City Construction on PM_2.5 Concentrations: Empirical Analysis from Chinese Counties

Abstract

Fine particulate matter (PM_2.5) pollution poses a major threat to human physical and mental health. Smart cities (SCs) provide innovative paths for PM_2.5 pollution prevention and control through Internet of Things (IoT) monitoring, intelligent transportation optimization, and other technological means. Based on the panel data of 2,141 counties in China between 2006 and 2021, this paper constructs a difference-in-differences with multiple time periods (MDID) to systematically assess the impact of SC on PM_2.5 concentration and analyze its mechanism of action by combining the satellite remote sensing PM_2.5 concentration (PM_2.5C) and the list of smart city pilots. This study finds the following: (1) SC significantly reduced the PM_2.5 concentration in the test area by about 3.58%. This conclusion was verified through rigorous robustness testing; (2) SC can effectively reduce PM_2.5C through the innovation effect; (3) High-quality economic development can strengthen the emission reduction effect of SC on PM_2.5C; (4) The environmental benefits of SC show significant spatial heterogeneity, with the largest PM_2.5 reductions occurring in the western regions (4.3% reduction), followed by regions with mature digital infrastructure and cities in high administrative level cities. The results of this study provide a reference for the regional differentiated implementation of the “14th Five-Year Plan for the Development of Innovative Smarter Cities”, and make targeted recommendations for the synergistic management of air quality under the “dual-carbon” goal.

1. Introduction

Fine particulate matter (PM_2.5) has attracted global attention in recent years due to its serious threat to public health [1,2,3]. PM_2.5 is defined as suspended particulate matter with an aerodynamic diameter of ≤2.5 μm, which contributes to respiratory disease through physical penetration and chemical toxicity [4,5], cardiovascular disease [6,7], and elevated risk of depression and anxiety in the case of prolonged exposure to low PM_2.5 [8]. By focusing on PM_2.5, our study aligns with national and global priorities to address its disproportionate health burden.

Serious haze pollution events occurred in China in 2013, especially in the Beijing Tianjin Hebei region [9,10]. The reduction of PM_2.5, a major component of haze [11], can help to reduce the frequency of haze and improve air quality [12]. For this reason, China has adopted two main types of attempts to manage PM_2.5. First, monitoring and early warning of PM_2.5. By incorporating PM_2.5 into the haze warning indicators of the meteorological department, setting up the Beijing Tianjin Wing Environmental Meteorological Forecasting and Warning Center, and releasing a heavy pollution weather monitoring and warning program to provide a scientific basis for PM_2.5 management [13]. Secondly, the implementation of policies and regulations to provide legal protection for PM_2.5 management. For instance, the 8-h concentration limit of PM_2.5 fine particulate matter was incorporated into the monitoring system of the Ambient Air Quality Standards, the “Air Pollution Prevention and Control Action Plan” was issued, clearly prioritizing the reduction of PM_2.5 as a core objective [14], and regulatory enhancements under the new environmental law to reduce PM_2.5C [15].

At the same time, academics are actively exploring the factors that influence PM_2.5 pollution from a multi-dimensional perspective, which can be summarized into three major driving directions: socioeconomic activities, natural environmental factors, and the impact of policies and regulations. The socioeconomic factors include the level of economic development [16], the urbanization process [17], the expansion of industrialization [18], the change of land use [19], ammonia due to agricultural activities [20], etc. As the spatial carrier of the above-mentioned social and economic activities, cities have become a key source of PM_2.5 emissions. Natural factors include meteorological conditions [21], natural disasters [22], biogeochemical processes, etc. [23]. The role of policies and regulations is also key. For example, the pilot policy of low-carbon cities [24] reduces major sources of PM_2.5 by improving residents’ green attitudes, promoting green transportation and sustainable energy use; Air Pollution Prevention and Control Action Plan [14] reduces PM_2.5 emissions through source control, emission process management and end-of-line treatment; the fiscal policies for energy conservation and emission reduction [25] reduces PM_2.5C through fiscal incentives to promote the upgrading of production capacity and transportation system, promote the development of new energy and increases the share of tertiary services.

However, even though policy-led regional management actions have significantly improved PM_2.5 pollution levels. For example, clean air actions resulted in a national population-weighted PM_2.5 concentration decline of 19.8 μg/m³ and 10.9 μg/m³ from 2013 to 2017 vs. 2018 to 2020, respectively [26], and a 48% reduction in PM_2.5 exposure nationwide [27], delivering a sizable PM_2.5 management effect. However, the current regional governance model has faced the problem of diminishing marginal returns. According to the China Ecological and Environmental Status Bulletin 2023, 31% of cities (105/339) in China still fail to meet the annual average PM_2.5 standard. This suggests that the effectiveness of city-based PM_2.5C management needs to be scrutinized. Against this backdrop, China launched its smart city policy in 2012 and has integrated PM_2.5 monitoring and management into its environmental governance framework through the following: IoT-based air quality sensors for real-time tracking of PM_2.5; intelligent transportation systems to reduce vehicle emissions; and energy management platforms to optimize industrial and residential fossil fuel use. Such governance innovations can provide long-term impetus to reduce urban PM_2.5. In this paper, we wish to verify through a quasi-natural experiment whether SC plays an important role in PM_2.5C emission reduction.

Smart city policy (SC) is an important innovative path proposed by China to address the problems of sloppy and inefficient traditional urban governance models and to promote high-quality urban development by digital and intelligent means. After 2016, China’s SC entered a deepening phase, the Innovative Smarter City (ISC) phase, the core of which is to promote the transition of SC policy from “construction” to “precise optimization”. ISC reduces PM_2.5C by formulating precise policy measures in the ecological field, transportation field, industrial field, etc. Specifically, firstly, SC formulates the general requirements for ecological environment detection, which promotes the urban meteorological detection system, which facilitates the government to target PM_2.5 emission sources at the source. Second, SC emphasizes the intelligent transformation of transportation infrastructure, which promotes the construction of urban transportation intelligent systems to alleviate the high concentration of PM_2.5 emissions brought about by the status quo of urban traffic congestion. Thirdly, SC attaches great importance to the enhancement of resource utilization through technological empowerment and promotes the construction of an intelligent energy management system, realizing real-time optimization of regional energy supply and demand matching, enabling energy-consuming enterprises and urban energy networks to improve their energy-saving capabilities and reduce PM_2.5 emissions in the industrial and living sectors. Fourth, SC supports the establishment of green smart cities, promotes the replacement of fossil energy with renewable energy, and facilitates the development of green infrastructure, which reduces unnecessary energy consumption in production and life and further reduces PM_2.5 emissions. In addition, the government has introduced a innovative smarter city evaluation process, in which smart environmental indicators (including air quality) are formulated in the evaluation of eco-livability areas to ensure the realization and optimization of SC in PM_2.5 management. Through these mechanisms, not only the concentration of PM_2.5 is reduced at the source of emission and in the emission process, but also the emission reduction effect of the policy is maintained and optimized, which effectively improves air quality.

Established studies have extensively explored the environmental benefits of SC [28,29,30,31], but their focus is more on the area of carbon emission reduction. Zhang and Zhong based on county data found that the SC pilot counties on average reduced 1.378 units of carbon emissions [32]; Guo et al. further emphasize that the reduction effect of SC in terms of per capita CO2 emissions is about 18.42 logarithmic percentage points [33], which shows its effectiveness in treating pollution; Shu et al. argue that SC systematically reduces the level of carbon emissions by promoting the transformation of high-polluting industries and integrating natural resource data to achieve precise resource management [34]; An et al. based on city-level data, SC can reduce urban carbon emissions by 11.4% on average, supplemented from the perspective of green technology progress [35]. These empirical studies emphasize the important contribution of SC concerning carbon emission reduction. However, the implications of SC policies on PM_2.5, another key air pollutant, have not been sufficiently emphasized. Although Cui and Cao found that SC can reduce PM_2.5 concentration [36]. However, there are two major limitations in their study: first, the data stops at 2018, failing to capture the policy effects of ISC, the deepening phase of SC initiated in 2016, which promotes more precise and controllable urban governance of SC through a dynamic evaluation system and a cross-sectoral data platform, and whose emission reduction efficiency may be significantly higher than that of the earlier phases of SC, and whose omission at this stage might have underestimated the combined effects of SC on PM_2.5C. Second, studies based on city-level data make it difficult to identify the differentiated policy effects of SC at the county level, while China’s SC contains a large number of county-level units, and its governance experience may not be conducive to its generalized use at the county level. This paper aims to break through the above limitations and provide new insights into the broader impacts of SC on PM_2.5C.

In this regard, based on the panel data spanning 2006–2021 from 2169 Chinese counties, this paper applies a multi-temporal double difference model (MDID) to systematically examine the relationship between SC and PM_2.5C. Compared with the existing literature, the innovative contributions of this paper are as follows: first, for the first time, the period of SC study is extended to the deepening stage (2016–2021), revealing the PM_2.5 emission reduction effect after policy iteration; second, county panel data are used to capture the micro governance effectiveness of SC pilots, providing new insights for optimizing SC policies.

The rest of the paper is as follows: the second part is the policy background and research hypotheses; the third part is the research design; the fourth part is the empirical results; the fifth part is the analysis of heterogeneity; and the last part is to summarize the empirical results and to give relevant policy recommendations.

2. Policy Context and Research Hypotheses

2.1. Policy Context

The concept of a “smart city” was first proposed by IBM in 2008, emphasizing the optimization of urban management and resource allocation through the Internet of Things (IoT), big data, and other technologies. Facing the dual challenges of management effectiveness and service provision in the process of new urbanization, China launched the construction of smart cities in 2012 and gradually built a territory-wide promotion pattern through three phases of pilot projects. The first phase is the pilot exploration period (2012–2014), in which 193 national-level pilots (90 in the first batch and 103 in the second batch) were set up in two batches in 2012 and 2013. In the field of environmental protection, the focus is on the deployment of IoT monitoring networks and the establishment of pollution source databases, but the technology application has not yet matured, and the effect on air pollution reduction is limited. The second stage is the period of policy systematization (2014–2016). In 2014, the National New Urbanization Plan was promulgated, confirming that smart cities are one of the three pillars of new city construction. In August of the same year, eight departments, including the National Development and Reform Commission, formulated the Guiding Opinions on the Healthy Development of Smart Cities and the Smart City Evaluation Indicators, which included “air quality” as one of the indicators to be examined, forcing local governments to integrate environmental protection data and promoting cross-sectoral efforts together to combat air pollutants, such as PM_2.5, etc. The third stage is the deepening period (2016–2021). After 2016, SC enters the deepening stage, i.e., Innovative Smarter City (ISC), the core of which is to promote the transformation of SC policy from “construction” to “precision” and “optimization”. With the introduction of a dynamic assessment system, real-time tracking of pollution sources, and responsibility tracing through 5G and blockchain technologies, the efficiency of PM_2.5 emission reduction has been improved significantly. By 2021, China has selected a total of 290 pilot smart cities in three batches, covering municipalities, provincial capitals, and small and medium-sized cities, as shown in Figure 1.

From the perspective of relevant policies, the objectives of SC are mainly to enhance the efficiency of urban management, improve the quality of life of residents, promote sustainable economic development, and promote environmental protection and green and low-carbon development. The main elements of SC include the following: I. Strengthening digital infrastructure. II. Building 5G networks, cloud computing centers, and IoT platforms to support intelligent urban operations. III. Provide intelligent public services. Promote smart transportation, smart healthcare, and smart education to enhance the convenience of residents’ lives. IV. Promote the adjustment of the industrial system in the direction of both the digital economy and the green economy, improve economic efficiency and quality, and reduce resource consumption and pollution emissions. V. Greening environmental governance. Real-time tracking of pollution sources through intelligent detection systems (e.g., PM_2.5 sensors), optimization of energy structure, and promotion of clean technologies. VI. Multi-departmental governmental governance. Break the data silo, realize the environmental protection, transportation, and industrial sector data sharing, and come to joint decision-making. Through the SC stage-by-stage policy evolution, it provides a systematic solution for urban environmental governance, which is conducive to the precise control and fine management of PM_2.5 emissions in cities, and promotes China’s PM_2.5 urban governance capacity steady improvement.

Overall, as an important part of the national new urbanization strategy, smart city construction, through cutting-edge digital innovations, is reconfiguring the urban governance model and driving coordinated advancement across economic, social, and environmental spheres.

Figure 2 presents the dynamics of PM_2.5 concentrations in the treatment and control groups before and after the implementation of SC. It is evident that the gap between the treatment and control groups progressively shrank following the implementation of the policy, strongly reaffirming the effectiveness of the intervention in managing PM_2.5 levels. Specifically, the PM_2.5 concentration in the treatment group dropped significantly following the policy’s implementation, in stark contrast to the control group. This clearly illustrates the SC policy’s usefulness in lowering pollutant emissions and enhancing air quality.

2.2. Research Hypotheses

2.2.1. Smart Cities Directly Affect PM_2.5 Concentrations

Smart cities play a key role in reducing PM25C as an important transformation of modern urban governance. These policies include, among others, smart air pollution detection, smart transportation, smart energy management, smart manufacturing, and infrastructure development.

Specifically, smart air pollution detection effectively lowers the levels of micron-scale pollutants, such as PM_2.5, by establishing an intelligent atmospheric monitoring system that enables environmental protection authorities to remotely target pollution sources and quickly make interventions [37,38]. Intelligent transportation focuses on establishing an intelligent transportation system to optimize traffic efficiency by shortening the residence time of vehicles on the road and guiding the diversion of vehicles, thereby curbing vehicular PM_2.5 particulate discharge [39]. Effectively reduce PM_2.5 emissions during urban transportation operations. Smart energy management promotes the establishment of the Internet of Energy, regional energy cooperation, energy market reform, and clean energy quota system, diminishes energy consumption, improves the overall efficiency of the urban energy system, and promotes the transformation of the urban energy structure towards low carbonization [40]. Smart manufacturing is the real-time equipment tracking and operational refinement of production processes through digital technologies (e.g., IoT sensor networks, big data analytics, and automated equipment), precise deployment of resources, and reduction of energy consumption and waste of raw materials in the production process, which directly reduces the emission of the pollutant PM_2.5 [41]. In addition, the construction of intelligent infrastructure reduces financing costs through information transparency, making it easy for clean energy projects to gain more attention from external capital, which directly attracts green investment and promotes the development of clean energy [42], reducing the use of fossil fuels, and thus reducing PM_2.5C.

In this regard, this paper proposes:

Hypothesis 1. 

SC can effectively reduce the concentration of PM_2.5 in the pilot area.

2.2.2. Smart Cities Influence PM_2.5 Concentrations Through Technological Innovation

Technological innovation refers to the process of improving and upgrading products, production processes, and business models through the application of digital technologies (e.g., information, computing, and communication technologies) and the assimilation of external knowledge, aiming at transforming technological potentials into real value [43,44]. In SC, technological innovations are mainly seen in the integration of digital technologies with the environment, transportation, manufacturing, and power grids.

SC promotes the intelligent transformation of cities and technological innovation in environmental governance by integrating digital technologies (technologies such as the Internet of Things, cloud computing, spatial geographic information integration, and artificial intelligence) with the optimization of physicochemical processes. Specific paths include: on the ecological side, combining AI, big data, and IoT with meteorological detection technologies to promote the development of meteorological detection technologies and the construction of an integrated meteorological detection system. Specifically, artificial intelligence technology integrates satellite remote sensing and ground sensor networks to realize excellent data processing and collection efficiency, which greatly increases the accuracy of meteorological detection technology in tracking pollution sources, and makes the prediction of pollution sources of meteorological integrated detection system more accurate [45]; second, the establishment of meteorological detection system requires the composition of a large amount of data and the synergy of multi-sectoral data. The emergence and application of big data and blockchain technology promote the resource sharing of meteorological data and provide technical support for the comprehensive meteorological detection system to provide information collection and information storage [46]; the use of IoT technology is reflected in the arrangement of sensors within the city, which allows the meteorological monitoring system to carry out real-time urban monitoring and provide early warning when the pollution concentration exceeds the standard for the PM_2.5 management to provide a technical path.

In the manufacturing industry, first of all, the sharing platform established based on big data can integrate industrial emission data, which is beneficial to the relevant departments for the monitoring of corporate emissions, thus promoting the development and application of low-consumption technologies (such as renewable energy, biomanufacturing, end-of-pipe treatment equipment, etc.) such as clean-technology energy production processes, and will also bring about the benefits of shortening the cycle of research and development of green technologies [41]. In addition, information sharing reduces information costs, facilitates the rational allocation of R&D investment, reduces innovation risks, and accelerates the commercialization of scientific and technological achievements, thus stimulating the continuous growth of R&D investment and promoting green technology innovation [47]. Secondly, the application of artificial intelligence in manufacturing energy monitoring is of great significance, which can effectively improve the efficiency of energy utilization by accurately predicting energy demand, optimizing the production and consumption process of energy, and realizing intelligent regulation. This not only helps to reduce the cost of energy consumption, but also significantly reduces environmental pollution and air pollutant PM_2.5 [48]. In addition, SC also emphasized that local governments should use financial guarantees as a policy tool to increase government investment in STI activities, the allocation of funds to technological innovation. The development and introduction of emission reduction technological innovations, such as clean energy and pollution control equipment, have improved pollution control capabilities and significantly reduced PM_2.5 emissions [49]. It is found that there is a typical U-curve relationship between technological innovation and air pollution, and the development and utilization of clean technology and renewable energy can effectively reduce air pollution and improve air quality during the effective stage of technological research and development [50]. Thirdly, energy efficiency and pollution control efficiency during industrial production can be improved through energy-saving technologies and improved production processes, thus reducing PM_2.5 emissions [51].

In the field of transportation, when big data and IoT are applied to transportation, relying on the interplay between urban big data platforms and urban sensors, urban traffic operation signs can be monitored by all elements, improving the city’s perception and prevention of scenarios such as water on the road surface, icy roads, defective bridges, etc., and avoiding the paralysis of the transportation system in these emergencies [46]; and secondly, after the sensors have monitored the abnormal situation, AI will dynamically adjust the signal light cycle to reduce unnecessary tailpipe emissions in traffic operation, further reducing PM_2.5C and forming truly intelligent transportation. In terms of urban energy management, when big data technology is applied to urban power grids, the monitoring of the energy sector is also included in the urban safety operation monitoring system. Secondly, with advanced control, performance prediction, and optimization supported by AI technologies and computational intelligence, smart cities are able to achieve unprecedented levels of energy efficiency and sustainability [52]. In summary, SC is effective in reducing PM_2.5 emissions through technological pathways in these four major areas.

In this regard, this paper proposes the following research:

Hypothesis 2. 

SC can contribute to the reduction of PM_2.5 concentration through innovation effects.

2.2.3. Moderating Effect of Economic Agglomeration on the Relationship Between Smart Cities and the Concentration of PM_2.5

During China’s rapid economic growth phase, the rapid advancement of industrialization and urbanization significantly exacerbated the PM_2.5 pollution problem [18]. Under the early rough development mode, the industrial structure dominated by heavy industry, the energy consumption structure centered on coal, and the surge in motor vehicle ownership directly pushed up PM_2.5 concentrations [53,54], leading to the frequent occurrence of regional haze. As the economy steps into the stage of high-quality development—characterized by innovation-driven growth, optimized resource allocation, and green and low-carbon transitions [55]—the energy consumption structure turns cleaner [56]; while the promotion of clean energy has been incorporated into local development planning, cleaner production methods have been widely adopted, and the new energy proportion is increased and the use of coal is reduced, which, in turn, pushes the city to realize the energy transition, improve the efficiency of urban resource utilization, and reduce the emission of pollutants such as PM_2.5.

In this regard, this paper proposes research hypothesis 3: High-quality economic development may reduce urban PM_2.5 emission levels.

Based on the above analysis, this paper draws a theoretical framework diagram, which can be seen in Figure 3.

3. Model Setup and Variable Description

3.1. Modeling

3.1.1. Multi-Temporal Double Difference Models

Difference-in-Differences (DID) modeling is commonly used to compare the difference between treatment and control groups before and after the implementation of a policy in order to estimate the net effect of the policy, and is a widely used method of causal inference in econometrics and social sciences [57,58,59]. The Multiple Time Period Difference-in-Differences (MDID) model, as an extension of the traditional DID model, is able to analyze the effects of policy interventions at multiple time points. Given that the SC pilots were distributed across 2012, 2013, and 2014, this research uses the MDID model to assess the effects of SC policies. Drawing on Callaway et al. [60], the model setup of this paper is as follows:

$$Y_{it}={\beta }_0+{\beta }_{1}Treate{d}_i\times Pos{t}_{it}+\lambda {\displaystyle \sum Control{s}_{it}}+{\nu }_i+{\tau }_t+{\epsilon }_{it}$$

(1)

In model (1), Y_it is the explanatory variable, denoting the PM_2.5 concentration of city i in year t. Treated_i is the grouping variable, city i takes 1 if it belongs to the treatment group, and 0 if it belongs to the control group; Post_it is the treatment-period dummy variable, and city i takes 1 if it is a smart city in year t, otherwise it takes 0. Controls is the set of control variables in this paper. ν, τ, and ε are city-fixed effects, time-fixed effects, and random error terms, respectively.

3.1.2. Mediated Effects Model

The two-stage mediation framework developed by Jiang Boat is applied to assess operational mechanisms. The model is divided into two main steps: first, clarifying the influence of the mediating variables on the explanatory variables through theoretical analysis; second, assessing the influence of the explanatory variables on the mediating variables through empirical analysis. Compared with the traditional three-stage model, the two-stage model simplifies the analysis process and is more direct and efficient. This paper constructs model (2) based on the above method:

$$M_{it}={\beta }_0+{\beta }_{1}Treate{d}_i\times Pos{t}_{it}+\lambda {{\displaystyle \sum Control{s}_{it}+{\nu }_i+{\tau }_t+\epsilon }}_{it}$$

(2)

M is the mediating variable in the model (2), which mainly has an innovation effect.

3.1.3. Moderating Effect Model

Given that the speed of economic development may affect the role of smart city pilot policies on PM_2.5, reference is made to Wang and Li’s approach to empirically test the mechanism of action through the lens of economic agglomeration [61].

Building upon Model (1), we develop a regression analysis to assess SC’s impact on economic agglomeration. Introducing the cross-multiplier term between the level of urban economic development and the policy implementation dummy variable, the moderating effect model shown in model (3) is constructed.

$$\begin{array}{c}{Y}_{it}={\beta }_0+{\beta }_{1}Treate{d}_i\times Pos{t}_{it}+{\beta }_{2}N+{\beta }_{3}Treate{d}_i\times Pos{t}_{it}\times N+\\ \lambda {{\displaystyle \sum Control{s}_{it}+{\nu }_i+{\tau }_t+\epsilon }}_{it}\end{array}$$

(3)

Y_it in model (3) is the mechanism variable, measured using the logarithm of urban economic density. Urban economic density is the ratio of gross regional product to the land area of the administrative region.

3.2. Variable Description and Source of the Data

3.2.1. Explanatory Variables

State Smart City (SC). This variable takes the value of 1 in the year a city is approved as an SC and in subsequent years, and 0 otherwise. Given that county-level cities are also eligible to apply to become pilots, this paper collects and analyzes data at the district and county levels accordingly to ensure that the study covers all pilot regions.

3.2.2. Explained Variables

PM_2.5 concentration data were sourced from the Atmospheric Composition Analysis Group (ACAG), integrating satellite-derived aerosol optical depth (AOD) measurements (MODIS, VIIRS, MISR, SeaWiFS) with multi-algorithm retrievals (Dark Target, Deep Blue, MAIAC) and GEOS-Chem model simulations [62,63]. Estimates of global and regional PM_2.5 concentrations are generated by geophysical estimation using relative uncertainties determined from ground-based solar photometer (AERONET) observations to account for most of the variance in ground-based PM_2.5 measurements. Building upon this foundation, the study utilizes ArcGIS tools to transform geospatial raster datasets into annual average PM_2.5 concentration estimates across Chinese counties.

3.2.3. Control Variables

Referring to the existing literature [64,65], the following control variables are selected in this study to control for potential confounders:

• Economic development level (lnpgdp): quantifying using the logarithmic transformation of per capita gross regional product (GRP), serving as an indicator of regional economic scale;
• Fiscal self-sufficiency (fisdes): assesses the financial autonomy of local governments through the ratio of local general budget revenues to expenditures;
• Industrial structure upgrading (is): the ratio of tertiary to secondary value added and is an indicator of a country’s industrial modernization process;
• Financial sector depth (fd): the year-end loan balance of financial institutions as a percentage of regional GDP is used to reflect the level of regional financial service provision levels;
• Education resource allocation (student): the coverage density of basic education is reflected by the ratio of students enrolled in general secondary schools in the household population;
• Healthcare supply (hos): quantifying the intensity of medical infrastructure investment by the number of hospital beds per 1000 population.

Table 1. Descriptive Statistics.

Variable	N	Mean	SD	Min	Max
PM2.5	32546	0.419	0.199	0.010	1.393
SC	32546	0.178	0.382	0	1
lnpgdp	32546	10.060	0.836	7.394	13.430
fisdes	32546	0.316	0.242	0.006	9.463
is	32546	1.237	1.354	0.043	54.980
fd	32546	0.631	0.437	0	7.635
student	32546	0.496	0.172	0.007	2.069
hos	32546	0.035	0.020	0	0.332

demonstrates the data characteristics of the main variables in this paper, where the mean value of SC is 0.178, indicating that 17.8% of the 32,546 samples belong to the SC pilot region. This suggests that the pilot area of SC is relatively large and covers a sufficient number of samples, and this proportion reflects the fact that the implementation of SC policy nationwide is in the stage of deepening and expanding.

3.3. Sample Selection and Data Sources

Given substantial county-level data gaps, interpolation-based estimations might distort true environmental patterns and therefore decide to exclude all the missing data samples. Eventually, this paper forms a data set covering 32546 observation samples for the period of 2006 to 2021. The data on PM_2.5 concentrations were obtained from the Atmospheric Composition Analysis Group (ACAG), which generates estimates of global and regional PM_2.5 concentrations using satellite observations and ground-based measurements. Through geospatial processing with ArcGIS, the gridded environmental datasets were transformed into county-level annual PM_2.5 concentration measurements across China. The data for SC were obtained from the Chinese government website, and control variables and mediator data were sourced from the China County Statistical Yearbook and State Intellectual Property Office (SIPO).

4. Empirical Analysis

4.1. Benchmark Regression

Model (1) assesses the effect of smart cities (SC) on PM_2.5 levels, with fixed-effects regressions using robust standard errors reported in columns 1–4 of

Table 2. Benchmark regression results.

	(1)	(2)	(3)	(4)
	PM2.5	PM2.5	PM2.5	PM2.5
SC	−0.015	−0.015	−0.015	−0.015
	(−12.807)	(−12.802)	(−12.804)	(−13.422)
lnpgdp		0.031	0.032	0.036
		(15.509)	(15.387)	(17.454)
fisdes		−0.041	−0.041	−0.040
		(−4.978)	(−4.978)	(−5.011)
is			0.003	0.003
			(5.538)	(5.608)
fd			−0.002	−0.001
			(−1.412)	(−0.422)
student				−0.031
				(−9.246)
hos				−0.598
				(−15.535)
_cons	0.422	0.125	0.111	0.105
	(1203.184)	(6.674)	(5.482)	(5.277)
County FE	√	√	√	√
Year FE	√	√	√	√
N	32,341	32,341	32,341	32,341
R2	0.940	0.941	0.941	0.942

z statistics in parentheses. p < 0.1, p < 0.05, p < 0.01. Clustering to the county.

, covering the year of inclusion and district/county-level data. Column (1) presents the baseline estimates excluding control variables, where SC demonstrates statistically negative coefficients (β = −0.015, t = −13.422, p < 0.01), confirming smart city initiatives’ PM_2.5 mitigation effects. This corresponds to a reduction of approximately 3.58% relative to the sample mean (0.419 units). The 95% confidence interval for the true effect ranges from 3.05% to 4.11%, further validating the statistical significance. Subsequent columns incrementally add control variables, showing minimal changes in coefficient magnitudes and consistent statistical significance, indicating robust conclusions across model specifications. In this regard, Hypothesis 1 is verified.

4.2. Parallel Trend Test

The validation of parallel trend assumptions constitutes a fundamental prerequisite for difference-in-differences (DID) methodology, essential for establishing causal inference validity in policy evaluation studies [66,67]. This diagnostic procedure examines whether treatment and control groups exhibited comparable temporal trajectories in outcome variables prior to policy implementation, a critical condition ensuring that post-intervention divergences reflect genuine treatment effects rather than pre-existing differential trends [68]. Parallel trend tests are usually conducted using the Event Study Method (ESM), which looks at whether the trends in the two groups are parallel by comparing different time points before and after the treatment. Specifically, if the coefficients of the pre-treatment time points do not significantly deviate from zero, it indicates that the parallel trend hypothesis is valid [69,70]. Based on this idea, we used event analysis to test whether the data changes before and after the policy were smooth (parallel trend test), and then built an analytical model according to this law.

$$Y_{it}=\mathsf{\partial }+{\displaystyle {\sum }_{k=-7,k\ne -1}^9{\beta }_{k}Treated(k)}+\lambda {\displaystyle \sum Control{s}_{it}+{\nu }_i+{\tau }_t+{\epsilon }_{it}}$$

(4)

In Model (4), we take the year before the start of the policy (k = −1) as the baseline for comparison, and exclude this point in time from the calculations, with the other settings as in the base model. If none of the coefficients are significant before the policy is implemented (k < 0), the data are consistent with the parallel trend assumption and the results are reliable.

Figure 4 displays the parallel trend analysis outcomes. Pre-policy coefficients oscillate near zero without statistical significance, while post-implementation estimates exhibit pronounced negative magnitudes, revealing sustained PM_2.5 reductions following policy enactment.

4.3. Robustness Tests

4.3.1. Replacement of Variables

To confirm the stability of key results, alternative explanatory variables are tested through substitution analysis. Specifically, this paper replaces the PM_2.5 concentration with the logarithmic value of the sum of PM_2.5 (PM₁), while the standard deviation of PM_2.5 is replaced with the maximum (PM₂) and minimum (PM₃) values of PM_2.5. Through regression analysis, the empirical estimates are presented in

Table 3. Robustness test I.

	(1)	(2)	(3)	(4)	(5)	(6)
	PM1	PM2	PM3	PM2.5	PM2.5	PM2.5
SC	−0.013 ***	−0.014 ***	−0.016 ***		−0.009 ***	−0.015 ***
	(−6.583)	(−13.229)	(−13.089)		(−8.703)	(−13.198)
L.SC				−0.014 ***
				(−10.106)
lnpgdp	0.013 ***	0.037 ***	0.039 ***	0.041 ***	0.019 ***	0.036 ***
	(3.888)	(21.140)	(18.648)	(17.438)	(10.071)	(18.827)
fisdes	−0.033 ***	−0.060 ***	−0.067 ***	−0.065 ***	−0.031 ***	−0.062 ***
	(−5.230)	(−17.310)	(−15.856)	(−12.199)	(−8.808)	(−16.407)
is	0.003 ***	0.002 ***	0.003 ***	0.002 ***	−0.000	0.003 ***
	(3.535)	(5.430)	(5.996)	(4.260)	(−1.178)	(5.210)
fd	−0.014***	−0.000	−0.001	−0.005 **	0.012 ***	−0.002
	(−4.722)	(−0.016)	(−0.313)	(−2.206)	(9.159)	(−1.102)
student	−0.043 ***	−0.024 ***	−0.036 ***	−0.035 ***	0.007 *	−0.031 ***
	(−7.379)	(−7.913)	(−9.712)	(−8.527)	(2.428)	(−9.180)
hos	−1.275 ***	−0.524 ***	−0.648 ***	−0.668 ***	−0.380 ***	−0.584 ***
	(−16.549)	(−16.011)	(−14.782)	(−12.145)	(−8.901)	(−15.663)
_cons	−2.597 ***	0.029 ***	0.159 ***	0.060 **	0.271 ***	0.111 ***
	(−80.674)	(1.672)	(7.642)	(2.685)	(14.857)	(5.809)
County FE	√	√	√	√	√	√
Year FE	√	√	√	√	√	√
N	32,341	32,341	32,341	20,138	26,325	31,324
R2	0.990	0.944	0.938	0.950	0.963	0.946

z statistics in parentheses. The following tables follow the same format. * p < 0.1, ** p < 0.05, *** p < 0.01.

(1)–(3). SC coefficients in this process are −0.013, −0.014, and −0.016, All estimates remain statistically significant. The empirical results corroborate the stability of our principal conclusions, demonstrating analytical consistency across methodological variations.

4.3.2. A Lag Phase

It may take some time for smart city construction to have an impact on the PM_2.5 emission behavior of urban economic agents. Consequently, this study employs a one-period policy lag in econometric modeling, with results documented in Table 3(4). The negative regression coefficient for SC (−0.014) with 1% significance demonstrates the stability of core analytical outcomes.

4.3.3. Replacement of Sample Intervals

Given that COVID-19 may have caused some interference with the SC policy effect, this paper excludes the relevant data from 2019 to 2021 and reruns the regression analysis. Subsequent regression outputs in Table 3(5) reveal SC parameters of −0.009 with 1% significance, demonstrating analytical consistency.

4.3.4. PSM-DID

In this study, caliper matching was used for propensity score matching. It ensured the similarity of features between the treatment and control groups after matching. Figure 5 shows the results of caliper nearest neighbor matching, from which it can be observed that the % bias (%bias) after matching is close to the 0-axis, indicating the success of matching by this method. The paper is re-run employing model (1) and the regression results are presented in Table 3(6). The SC coefficient remains significantly negative, confirming the conclusion’s robustness against confounding variable exclusion.

4.3.5. Exclusion of Contemporaneous Policies

To address potential confounding effects from concurrent urban policy initiatives, our analysis systematically filters out counties participating in four major national programs: the intellectual property court (ipc), the pilot national innovative city (nic), the pilot low carbon city (lccp) and the broadband china demonstration city (bbc). Post-sample refinement, we re-estimate the core specification through multiple robustness checks. In addition, considering that the exclusion of a single policy may not be able to exclude all the impacts, this paper excludes all the pilot counties of these four policies and returns them again. From the data in each column of

Table 4. Robustness test II.

	(1)	(2)	(3)	(4)	(5)
	PM2.5 ipc	PM2.5 nic	PM2.5 lccp	PM2.5 bbc	PM2.5
SC	−0.015 ***	−0.015 ***	−0.015 ***	−0.015 ***	−0.015 ***
	(−13.855)	(−13.926)	(−13.855)	(−13.855)	(−13.926)
lnpgdp	0.036 ***	0.036 ***	0.036 ***	0.036 ***	0.036 ***
	(20.029)	(20.027)	(20.029)	(20.029)	(20.027)
fisdes	−0.061 ***	−0.061 ***	−0.061 ***	−0.061 ***	−0.061 ***
	(−16.626)	(−16.694)	(−16.626)	(−16.626)	(−16.694)
is	0.003 ***	0.003 ***	0.003 ***	0.003 ***	0.003 ***
	(5.578)	(5.582)	(5.578)	(5.578)	(5.582)
fd	−0.000	−0.000	−0.001	−0.000	−0.000
	(−0.155)	(−0.141)	(−0.155)	(−0.155)	(−0.141)
hos	−0.059 ***	−0.584 ***	−0.590 ***	−0.590 ***	−0.584 ***
	(−15.630)	(−15.516)	(−15.630)	(−15.630)	(−15.516)
student	−0.028 ***	−0.029 ***	−0.028 ***	−0.028 ***	−0.029 ***
	(−9.102)	(−9.169)	(−9.102)	(−9.102)	(−9.169)
_cons	0.107 ***	0.107 ***	0.107 ***	0.107 **	0.107 ***
	(5.925)	(5.923)	(5.925)	(5.925)	(5.923)
County FE	√	√	√	√	√
Year FE	√	√	√	√	√
N	32,341	32,327	32,341	32,341	32,327
R2	0.947	0.947	0.947	0.947	0.947

** p < 0.05, *** p < 0.01.

, it can be seen that the SC variable has a significant effect on PM_2.5 reduction, all coefficients are negative and pass the test regardless of whether other pilot policy samples are excluded, validating the analytical robustness.

4.4. Placebo Test

The placebo test is a statistical method used to verify the robustness of double difference (DID) models and to rule out pseudo-causality. The method tests the reliability of model results by randomly generating an experimental or control group and repeating the experiment multiple times [71] (usually no less than 500 times). It centers on verifying whether the observed effects are due to random factors or omitted variables, thus enhancing the credibility and precision of causal identification.

The specific steps of the placebo test are as follows: first, individuals in the sample are randomly assigned to experimental and control groups; second, DID regression analyses are performed on the sample after each random assignment, and the policy effect coefficients are recorded; The placebo test procedure involves three steps: (1) compiling all simulated coefficients into a kernel density plot to graphically contrast empirical versus pseudo-treatment effects; (2) statistically assessing if the observed effect exceeds randomized samples through distributional comparison; (3) validation requires the empirical coefficient surpassing 95%/99% confidence thresholds of placebo estimates, where significant divergence confirms non-random causation.

Figure 6 displays PM_2.5 placebo test results. The red density curve clusters near zero, showing that randomized treatment groups lacked significant policy impacts in most trials. The dashed line represents the true regression coefficient, which does not intersect the red curve, indicating that the actual policy effect is significantly different from the placebo test results, i.e., the policy effect is not caused by random factors, and the placebo test passes.

4.5. Mechanism Analysis

4.5.1. Innovation Effects

Referring to the existing literature [70,71], Two innovation proxies, the total number of granted invention patents (patent_grant) and the number of patents per 10,000 inhabitants (patent_density), are used in this study for regression analysis. As shown in columns 1–2 of

Table 5. Mechanism Analysis.

	(1)	(2)	(3)
	patent_grant	patent_density	Economic Aggregation
SC_econ			−0.019 ***
			(−27.176)
SC	0.178 ***	0.185 ***	0.109 ***
	(0.025)	(0.028)	(25.113)
ln_econ_agg			−0.015 ***
			(−3.765)
lnpgdp	0.082 **	0.188 ***	0.050 ***
	(0.032)	(0.039)	(11.43)
fisdes	−0.235 ***	−0.318 ***	−0.058 ***
	(0.054)	(0.066)	(−16.180)
is	−0.012	−0.007	0.003 ***
	(0.009)	(0.016)	(5.254)
fd	0.066 **	0.110 **	−0.003 *
	(0.033)	(0.055)	(−1.615)
hos	2.929 ***	3.630 ***	−0.520 ***
	(0.587)	(0.718)	(−14.143)
student	0.139 ***	0.228 ***	−0.027 ***
	(0.043)	(0.054)	(−8.855)
_cons	−0.711 **	−1.700 ***	0.055 *
	(−2.153)	(0.419)	(2.296)
County FE	√	√	√
Year FE	√	√	√
N	20,291	20,291	32,116
R2	0.728	0.726	0.944

* p < 0.1, ** p < 0.05, *** p < 0.01.

, the SC coefficients (0.178 and 0.185) indicate that the innovation-driven decline in PM_2.5 is statistically significant. This demonstrates that SC can reduce the concentration of PM_2.5 through the innovation effect. In this regard, this paper proves Hypothesis 2.

4.5.2. Moderating Effects of Economic Agglomeration

As regional economies advance, environmental governance ascends policy agendas, while agglomeration-driven scale economies and knowledge spillovers enhance emission mitigation efficacy through technological upgrading. Table 5(3) analyzes the moderating effect of the increase in economic growth rate on the relationship between SC and PM_2.5. The Empirical analysis reveals that the coefficient of the main effect of economic density (ln_econ_agg) is statistically negative (−0.015), which represents that for every 1% increase in economic density, the PM_2.5 concentration decreases by 0.015 units, indicating that the higher the level of economic agglomeration, the lower the PM_2.5 pollution. Concurrently, the coefficient of the interaction term (SC_econ) between economic agglomeration and SC is significantly negative (−0.019), which further indicates that with the gradual shift of the economy to high-quality development, the stronger the effect of SC policy, the better the effect of reduction of PM_2.5, and that economic agglomeration is an important condition for SC to exert its environmental benefits. In this regard, Hypothesis 3: High-quality economic development may reduce urban PM_2.5 emission levels is verified.

5. Heterogeneity Analysis

5.1. Digital Infrastructure Heterogeneity

Sound and improved new digital infrastructure is an important engine for promoting the development of smart cities. Digital infrastructure construction refers to the establishment and improvement of digital technology and network systems in modern society to support the development and application of information technology. According to the annual statistics of the communications industry in the mobile communications capacity, 5G base stations by province, Guangdong, Jiangsu, Zhejiang, Shandong, and Henan have long been in the leading position in the main indicators of the communications industry. These provinces are significantly higher than other regions in terms of the number of 5G base stations, Internet broadband access ports, fixed-line telephone subscribers, and mobile telephone subscribers. In this paper, the above five provinces are set as high digital infrastructure zones. The remaining provinces are classified as low digital infrastructure zones, and regression analyses are conducted separately.

Table 6. Digital infrastructure heterogeneity.

	(1)	(2)	(3)
	PM2.5 High dinf	PM2.5 Low dinf	PM2.5
SC	−0.002	−0.018 ***	−0.010 ***
	(−0.732)	(−14.972)	(−8.363)
SC_dinf			−0.029 ***
			(−12.551)
lnpgdp	0.049 ***	0.024 ***	0.035 ***
	(11.125)	(12.628)	(19.273)
fisdes	−0.073 ***	−0.056 ***	−0.061 ***
	(−10.370)	(−13.740)	(−16.814)
is	0.007 ***	0.002 ***	0.003 ***
	(3.487)	(4.757)	(5.533)
fd	0.036 ***	−0.007 ***	−0.001
	(7.979)	(−3.901)	(−0.456)
student	−0.058 ***	−0.026 ***	−0.027 ***
	(−10.507)	(−7.342)	(−8.587)
hos	−0.716 ***	−0.571 ***	−0.595 ***
	(−8.836)	(−14.768)	(−15.760)
_cons	0.075	0.205 ***	0.119 ***
	(1.593)	(10.818)	(6.612)
County FE	√	√	√
Year FE	√	√	√
N	5984	26,356	32,340
R2	0.963	0.944	0.947

*** p < 0.01.

reveals significant disparities in SC policy effectiveness on PM_2.5 mitigation across digital infrastructure capacities. In the low digital infrastructure zone (column 2), the implementation of SC significantly reduces PM_2.5 concentration by 1.8% (coefficient = −0.018, p < 0.01), while in the high digital infrastructure zone (column 1), the direct effect of SC is not significant (coefficient = −0.002, p > 0.1).

To further examine the different effects of SC on different digital infrastructure cities, this study constructs the dummy variable “dinf”. High_dinf is set to 1, low_dinf is set to 0. It forms the interaction term between dinf and SC, which is included in the regression model, and the results are shown in column (3) of Table 6. The additional 2.9% reduction in PM_2.5 concentration when the SC policy is implemented in the high digital infrastructure area suggests that the level of digital infrastructure has a reinforcing effect on the abatement effect of SC. This result may stem from the technological synergies of high digital infrastructure zones: 5G networks and IoT platforms support real-time pollution monitoring and promote the application of cleaner technologies, thus amplifying the environmental benefits of SC policies. The insignificant regression results for high infrastructure zones alone may be related to sample size limitation or policy implementation lag.

5.2. Regional Heterogeneity

In general, different geographic locations of cities directly determine the differences in economic development levels between cities, which may impact smart city construction. To test whether there is geographic location heterogeneity in the impact of smart city construction on PM_2.5C. The study categorizes the sample into eastern, central, and western geographical clusters for regional subgroup regressions. Regression analyses in

Table 7. Regional heterogeneity.

	(1)	(2)	(3)
	PM2.5 Landlord	PM2.5 Center	PM2.5 West
SC	−0.007	0.000	−0.018
	(−3.480)	(0.073)	(−10.956)
lnpgdp	0.046	0.024	−0.011
	(12.090)	(6.674)	(−4.772)
fisdes	−0.097	−0.047	−0.027
	(−13.877)	(−7.165)	(−5.171)
is	−0.000	0.005	0.002
	(−0.246)	(4.033)	(3.257)
fd	−0.014	0.019	−0.004
	(−3.484)	(7.550)	(−2.545)
student	−0.101	−0.014	0.006
	(−15.509)	(−2.639)	(1.478)
hos	−0.416	−0.936	−0.503
	(−5.584)	(−11.780)	(−10.570)
_cons	0.140	0.264	0.491
	(3.474)	(7.601)	(21.897)
County FE	√	√	√
Year FE	√	√	√
N	8215	9013	11,348
R2	0.959	0.934	0.946

reveal site-specific PM_2.5 reduction effects, the coefficients follow the pattern of “west > east > central”. The reason for this result is as follows: the western region relies on the national new infrastructure strategy, and the smart city technology is directly applied to clean energy substitution and real-time pollution monitoring, which fills the traditional environmental governance gap; while the eastern region is subject to the complexity of the digitalization level, the new challenges of cross-region pollution, mobile population, and other new challenges are beyond the scope of the management of a single city’s smart system; and the central region performs weakly because it is in the transition of industrialization, and the smart transformation of heavy industry is lagging, which makes it difficult for the central region to manage the pollution of the city.

5.3. Heterogeneity of Urban Hierarchies

To investigate the heterogeneous effects of smart city measures on air quality, this study categorizes cities into those with high administrative levels (provincial capitals and sub-provincial cities) and ordinary cities to examine the policy effects. The regression analyses in

Table 8. Urban hierarchical heterogeneity.

	(1)	(2)
	PM2.5 High Administrative Level Cities	PM2.5 Ordinary Cities
SC	−0.018	−0.015
	(−10.709)	(−13.419)
lnpgdp	−0.012	0.036
	(−5.505)	(17.452)
fisdes	−0.010	−0.041
	(−2.214)	(−5.011)
is	0.002	0.003
	(2.938)	(5.609)
fd	−0.004	−0.001
	(−2.685)	(−0.422)
hos	−0.514	−0.058
	(−10.769)	(−15.536)
student	0.006	−0.031
	(1.425)	(−9.243)
SC_level
_cons	0.506	0.105
	(22.284)	(5.278)
County FE	√	√
Year FE	√	√
N	11,346	32,340
R2	0.945	0.942

show that SC significantly reduces PM_2.5 concentrations in both city types, but the effect is stronger in cities with high administrative levels (coefficient = −0.018, p < 0.01) than in ordinary cities (coefficient = −0.015, p < 0.01).

This heterogeneity may stem from different governance capabilities: (1) High administrative level cities benefit from centralized resource allocation and mature digital ecosystems. Integrated smart platforms (e.g., cross-sectoral pollution databases and 5G-enabled emissions monitoring) allow precise targeting of industry- and transportation-related PM_2.5 sources. (2) While deploying IoT sensors and ITS, ordinary cities face bottlenecks in policy implementation due to weak institutional coordination and dependence on external technical support, limiting marginal gains. Notably, economic development (lnpgdp) exhibits the opposite effect: it is associated with high administrative level cities with PM_2.5 reduction (−0.012), but increases pollution in average cities (0.036). This contrast highlights the need to tailor sustainable development strategies to city-specific industrial structures and governance frameworks.

6. Conclusions and Policy Recommendations

6.1. Conclusions

Utilizing the staggered implementation of China’s smart city (SC) pilot policy as a quasi-natural experiment, this study assesses the abatement effect of digital governance on PM_2.5 and the mechanism of its action through the MDID system, based on panel data of 2,163 counties from 2006 to 2021. The study found the following: significant emission reduction effect of SC policy: SC construction reduced PM_2.5 concentration by 3.58% on average in the pilot region, a finding verified by robustness tests such as parallel trend test, PSM-DID, and placebo test. Emission reduction pathway driven by technological innovations: mediating effect analysis showed that SC directly reduced PM_2.5 emissions by promoting the research and development and application of green technologies. The moderating role of agglomeration: SC policy emission reduction efficiency in high economic density regions is increased by 15%, indicating that the high-quality development stage can strengthen the effectiveness of environmental governance. Spatial heterogeneity characteristics: PM_2.5 reduction is more significant in the western region, mature counties with digital infrastructure, and high administrative level cities, whereas the emission reduction effect of the eastern complex urban agglomeration is relatively limited due to the insufficient trans-regional pollution synergistic management.

Compared with the existing literature, this study achieves breakthroughs in the following three aspects: first, the refined analysis of policy pathways. While existing studies focus on the overall environmental effects of SC [28,29], this paper breaks down the policy mechanism into four major areas: ecological monitoring, intelligent transportation, industrial upgrading, and digital infrastructure for the first time. Second, dynamic assessment of policy iteration. By including the smart city deepening phase (ISC, 2016–2021), the synergistic effect of policy upgrading is revealed, which compensates for the limitation of missing phase effects in municipal data studies [34,35]. Third, the micro breakthrough in governance scale. Unlike studies relying on prefecture-level city data [36], this paper captures the governance heterogeneity of sub-administrative units based on county-level data, which provides empirical evidence for “precise pollution control”. The findings of this study provide key insights for differentiated smart city policies: western counties should prioritize the layout of digital infrastructure and ecological monitoring networks, while eastern urban agglomerations should build a cross-regional platform for collaborative pollution management.

6.2. Limitation

Control variable data limitations: This paper suffers from deficiencies in the selection of control variables and data coverage, and fails to adequately incorporate key environmental and economic factors. Coal consumption dynamics: changes in coal consumption (e.g., industrial coal, residential heating coal) are not systematically tracked across counties, while differences in coal dependence may significantly affect PM_2.5 emissions; and strength of environmental regulations: lack of quantitative controls on the strength of enforcement of local environmental regulations (e.g., frequency of emissions fines, proportion of polluting firms shut down), leading to potential omitted variable bias in the assessment of policy effects.

Potential impact of endogeneity issues: smart city pilot selection may be driven by unobserved factors (e.g., local government governance capacity, financial resource endowment), and although PSM-DID is used to mitigate selectivity bias, the failure to introduce instrumental variables (e.g., geographic location, historical digitization level) or breakpoint regression designs to further validate the causal chain may weaken the causal explanatory power of the findings.

6.3. Policy Recommendations

Based on the above analysis, the policy recommendations of this paper are as follows:

(1)

Based on the effectiveness of PM_2.5 emission reduction in the SC pilot counties, it is recommended to expand the coverage of the policy in a phased manner, prioritizing the focus on regions with serious PM_2.5 pollution, high potential for economic agglomeration, and a better level of digital infrastructure.

(2)

The mechanism of SC to reduce PM_2.5 through technological innovation has been verified, and technological investment needs to be further strengthened to promote the intelligent transformation of highly polluting industries and encourage enterprises to apply cleaner production processes.

(3)

In high economic density areas (e.g., central cities), the effect of SC in reducing PM_2.5C is more significant, and it is recommended to prioritize the layout of intelligent environmental protection monitoring and sharing of pollution control facilities. In regions with lower economic density, supporting industrial cultivation policies are needed.

(4)

Setting up a special fund for “digital infrastructure transfer payments” to subsidize the purchase of smart equipment and provide technical operation and maintenance training to economically backward counties, to break the financial bottleneck of their digital transformation.

(5)

In response to digital infrastructure heterogeneity, priority is given to promoting SC in regions with mature digital infrastructure, while strengthening digitalization in low-infrastructure regions to fill in the gaps. Given regional heterogeneity, it is necessary to formulate differentiated promotion strategies. In the western region, relying on the “East Counts, West Counts” project, the layout of the intelligent monitoring network; in the eastern region to build provincial and municipal “air pollution collaborative governance platform”, the integration of the Yangtze River Delta, Beijing Tianjin Hebei region pollution source data, the use of blockchain technology to realize the responsibility for pollution control and ecological compensation mechanism; in the central region to set up heavy pollution control platform, the use of blockchain technology to realize the responsibility for pollution control and ecological compensation mechanism. In the central region, a transition fund for the intelligent transformation of heavy industry has been set up to help build digital infrastructure.

(6)

To ensure the long-term effectiveness of smart city policies, there is a need to improve the policy hiring and dynamic adjustment mechanism. In addition, public participation and social cooperation should be strengthened. Referring to the experience of universal action in the “dual-carbon” goal, we should promote the “Citizen Environmental Protection Data Platform”, encourage the public to query the local PM_2.5 data in real-time through the government app, and open the channel for reporting pollution sources to form a closed loop of “monitoring-feedback-governance”. Further, we should be carrying out “intelligent pollution control” science popularization actions, popularizing the application scenarios of the Internet of Things and big data in environmental governance in communities and schools, and enhancing the public’s recognition of and participation in smart city policies.