Evaluation, Driving Mechanism and Spatial Correlation Analysis of Atmospheric Environmental Efficiency in the “2+26” Cities Based on the Nonradial MEA Model

Abstract

The “2+26” cities are 26 cities in Beijing, Tianjin and the surrounding cities, constituting a region characterized by economic prosperity and diverse industries but plagued by severe atmospheric pollution. As a focal area for atmospheric pollution control, a scientific assessment of atmospheric environmental efficiency in the “2+26” cities that measures the degree of coordination between the economy and air pollution is very important for winning the battle of blue sky defense. Based on this, this study comprehensively used the nonradial multi-directional efficiency analysis (MEA) model, Global Reference Malmquist Model and spatial correlation analysis to evaluate the atmospheric environmental efficiency, calculate the driving factors and explore the spatiotemporal distribution characteristics of the “2+26” cities from 2009 to 2018. The research findings indicate the following: (1) Atmospheric environmental efficiency showed a trend of first decreasing and then increasing, with a significant improvement potential of 26.7% in the future. (2) There was a significant discrepancy between the best- and worst-performing cities, with the best being 0.910 and the worst being 0.573, demonstrating imbalanced development between cities. The relatively low-efficiency cities were mainly located in Hebei, Shanxi and Henan provinces. (3) A value of technological efficiency change (EC) less than 1 was the main restrictive factor for improving atmospheric environmental efficiency, whereas a value of technological change (TC) greater than 1 enhanced it. (4) The atmospheric environmental efficiency presented a distinct spatial distribution pattern of high–high and low–low aggregation, forming high-value areas centered in the Beijing–Tianjin region and along the Zibo–Zhengzhou line. The western and central regions were relatively low, whereas the northern and eastern regions were relatively high, with significant regional differences in spatial distribution. The conclusions from this article’s empirical analysis can help concerned developing countries determine key factors to improve their atmospheric environmental efficiency and then formulate policies for sustainable economic and environmental development.

1. Introduction

The development of China’s economy has entered a critical stage from speed-oriented growth to quality-oriented development, which has also brought about many challenges, such as environmental pollution, ecological damage and resource scarcity. China is not alone in facing these challenges; other countries also face the problem of environmental degradation caused by economic growth [1,2,3]. Thus, in recent years, many countries have been exploring how to balance economic development and environmental protection, adopting various green development policies. For example, ASEAN-9 presented energy transition to curb environmental degradation. At the 75th session of the UN General Assembly, China formulated the “Carbon Neutrality Action Plan” to achieve the goals of peaking carbon emissions by 2030 and achieving carbon neutrality by 2060. Chinese local governments attach great importance to green development and actively achieve the goals of carbon peak and carbon neutrality. Compared to developed countries, the lack of sufficient technological investment in developing countries may lead to more severe environmental pollution [4]. Therefore, the Chinese Ministry of Science and Technology has taken the lead in formulating the “Implementation Plan for Science and Technology to Support Carbon Peaking and Carbon Neutrality (2022–2030)”. All regions in China are actively developing new energy, researching low-carbon technologies and establishing a carbon neutrality management system.

At the same time, with the rapid urbanization of China, China has experienced frequent regional air pollution, attracting high attention from the Chinese government. As controlling regional air pollution is the top priority for improving the quality of the atmospheric environment, the Ministry of Ecology and Environment of the People’s Republic of China issued the “2017 Air Pollution Prevention and Control Work Plan for Beijing- Tianjin-Hebei and Surrounding Areas” and defined Beijing–Tianjin–Hebei and the surrounding areas (“2+26” cities) as air pollution transmission channel cities. The economy of the “2+26” cities is prosperous. In 2018, the regional GDP was approximately CNY 13.9 trillion, accounting for approximately 15.54% of the gross domestic product. However, according to the 2018 China Environmental Status Bulletin, the air quality of 13 cities in the region ranks 20th from the bottom among the 169 key monitored cities nationwide [5], and air quality was identified as needing to be improved (see Figure 1).

Against this backdrop, the scientific assessment of atmospheric environmental efficiency, which measures the coordination between the value generated by economic activities and the atmospheric environment, is an essential prerequisite for controlling regional air pollution. Therefore, the objective and innovation of this study lie in presenting results that no previous studies have investigated, i.e., the current situation, driving factors and spatial patterns of atmospheric environmental efficiency in the “2+26” cities. This study creatively devised an indicator system for regional air pollution that includes CO₂ emissions. It also sets up a nonradial MEA model as an extension and supplement to the DEA method. Based on the above discussion, this study attempted to address the following key inquiries: (1) What objective methods and indicator system can be used to evaluate the atmospheric environmental efficiency in the “2+26” cities? (2) What was the impact of the driving factors on atmospheric environmental efficiency? (3) How can the atmospheric environmental efficiency be effectively improved to promote sustainable development and achieve the emission peak and carbon neutrality goals?

2. Literature Review

In terms of research methods, domestic and foreign scholars mainly use data envelopment analysis (DEA), stochastic frontier analysis (SFA) and the slack-based measure (SBM) [6,7,8,9,10,11].

The scope of environmental efficiency research is very broad, including national, regional and industry research. Halkos and Tzeremes [12] considered greenhouse gases undesirable outputs and calculated the regional environmental efficiency of the United Kingdom based on a conditional directional distance function. Henry et al. [13] analyzed the environmental efficiency of dairy farms in Ireland using the DEA model. Combining industry characteristics and regional environmental impact factors, the number of farm livestock, capital and variable costs were taken as the inputs, the sales revenue of milk and other products was taken as the expected output, and the surplus of nitrogen and phosphorus was taken as the undesired output. Dirik et al. [14] evaluated the environmental efficiency of 51 cement factories in Turkey using the DEA method, with carbon dioxide (CO₂) emissions as the undesired output. Bibi et al. [10] used the SFA method to analyze the technical and environmental efficiency of the agricultural sector in South Asia from 2002 to 2016. Galeana-Pizaña [15] defined the Food Environmental Efficiency Index (FEE) as an arithmetical calculation of the significant correlation between food security and land use change.

Furthermore, with further deepening of the research, as the negative effects of environmental pollution affect economic benefits to a certain extent, domestic and foreign scholars have incorporated environmental factors into efficiency evaluations, combining different research fields to form a variety of efficiency terms, such as ecological efficiency, resource environmental efficiency and industrial environmental efficiency. These efficiency indicators have similar meanings to environmental efficiency. In essence, environmental indicators are considered undesirable outputs or harmful inputs, and the relationship between research and economic output reflects the connotation of environmental efficiency. Bian and Yang [16] combined DEA and Shannon’s entropy (Shannon–DEA) to calculate the comprehensive efficiency of provincial resources and environments by taking sulfur dioxide (SO₂) discharge, chemical oxygen demand and nitrogen discharge as three adverse environmental factors. Coluccia et al. [17] used a DEA model to evaluate the agricultural ecological efficiency in Italy, taking land, fertilizer and irrigation area as environmental variables. Ding et al. [18] aimed to assess the industrial water efficiency in the Yangtze River Economic Belt using the undesirable SE-SBM model, which included wastewater as an undesirable output. In addition to the mentioned studies, atmospheric environmental efficiency is an extension of environmental efficiency research in the atmospheric field. From the perspective of air pollution emissions, research is mainly conducted by domestic scholars from the perspective of provinces, economic belts and industries. Cai and Ye [19] carried out an industrial atmosphere environmental efficiency analysis at the interprovincial level through the virtual frontier undesirable output SBM model. Lv and Deng [20] measured the atmospheric environmental efficiency of 17 cities in Shandong from 2007 to 2017 based on the Super-SBM model. Wang et al. [21] employed the SBM-Undesirable model to evaluate the atmospheric environmental efficiency of 13 cities in the Beijing–Tianjin–Hebei region.

From the perspective of the literature analysis, the existing research still has the following shortcomings: (1) In terms of research objects, it mainly focuses on the national or provincial level, and research on the prefectural city level and atmospheric pollution areas composed of prefecture-level cities is still lacking. Moreover, research mostly at the national and provincial levels has overlooked regional atmospheric pollution, so there is insufficient research examining regional atmospheric pollution. (2) At present, most of the studies are based on the DEA and SFA models. The DEA series of models suffers from issues such as overestimation of efficiency values, infeasible solutions and impractical improvement methods. SFA is a parametric method that needs to assume the production function in advance. (3) The existing literature mostly focuses on atmospheric environmental efficiency and its influencing factors, with insufficient research on the evolution of spatial patterns and few analyses on regional differences and evolution trends. (4) The evaluation system for atmospheric environmental efficiency often relies on indicators such as sulfur dioxide (SO₂) emissions and industrial smoke (dust) emissions as undesired outputs, whereas greenhouse gas emissions, specifically CO₂ emissions, have not been taken into consideration.

Accordingly, to compensate for the shortcomings of existing research, this study attempted to take atmospheric pollution areas (the “2+26” cities) as the study object and calculated the atmospheric environmental efficiency of a total of 28 cities in the “2+26” region from 2009 to 2018 based on an improved nonradial multi-directional efficiency analysis (MEA) method. Combined with spatial autocorrelation analysis, the analysis in this study adopted multiple perspectives, such as spatiotemporal analysis, regional differences, evolution trends and spatial patterns. In order to further explore the driving factors, this study decomposed the Malmquist index of atmospheric environmental efficiency over time. In terms of constructing the indicator system, this article introduced CO₂ emissions into the construction of the indicator system as a negative output, making the evaluation of atmospheric environmental efficiency more comprehensive.

3. Materials and Methods

3.1. Study Area

The “2+26” cities are located in the middle latitude region of eastern and central China, with longitudes ranging from 112°30′ to 118°30′ and latitudes ranging from 34°60′ to 40°18′. The region encompasses two municipalities directly under the central government of Beijing and Tianjin, as well as some prefecture-level cities in Hebei, Shandong, Shanxi and Henan (see Figure 2). The “2+26” cities exhibit a vibrant landscape of industrial production, with prominent contradictions between economic development and environmental protection. The output of steel, coke, electrolytic aluminum, flat glass, cement, raw material and pesticides in this region respectively accounts for 43%, 47%, 38%, 33%, 19%, 60% and 40% of the national output. The “2+26” cities region is adjacent to the Yanshan Mountains in the north and the Taihang Mountains in the west. Most of the region is located in the North China Plain, high in the northwest and low in the southeast. Given the natural, social and economic conditions prevalent in the “2+26” cities, the air pollution is concentrated and does not easily spread, leading to the frequent occurrence of regional air pollution. Consequently, the “2+26” cities are known as major transmission channels of air pollution.

3.2. Research Method

3.2.1. Nonradial Multi-Directional Efficiency Analysis (MEA) Model

At present, DEA [22,23,24,25] and SFA [26,27,28] are the most popular research methods for efficiency measurement. Both are based on the directional distance function to construct the production frontier and calculate the relative efficiency of each evaluation unit [29]. The main difference between the two is that DEA is a non-parametric method that does not need to assume the production function in advance [30,31], but SFA can eliminate the influence of random errors [32,33]. As a supplement to and extension of data envelopment analysis, the multi-directional efficiency analysis (MEA) method was first proposed by Bogetoft and Hougaard in 1999 [34] and then gradually became widely used in energy [35], agriculture [36,37], transportation [38], banking [39] and other fields. Bogetoft and Hougaard [34] proposed a new concept of production potential in theory. The MEA method seeks to improve methods that are influenced by production potential, reflecting the reduction potential of inputs and undesirable outputs and the expansion potential of desirable outputs based on the potential improvement ratio rather than the past. That is, different variables in each period are constructed by the decision-making unit with different direction vectors, which change according to the changes in technical boundaries in different periods. The evaluation unit considers the optimal direction vector a potential improvement direction, reflecting different behavioral decisions for efficiency improvement [40].

This study incorporated undesirable outputs into the MEA model [41] and calculated the atmospheric environmental efficiency of the “2+26” cities. Nonradial MEA has the following advantages over the traditional DEA model: (1) It measures the individual efficiency values of each evaluation unit at different periods based on the improvement potential. Therefore, not only can the efficiency status of different evaluation units be detected, but different efficiency modes can also be detected. Therefore, it helps with the analysis of the future improvement direction of each decision-making unit. (2) It incorporates undesirable output variables into the MEA model to obtain more comprehensive calculation results. (3) It avoids equal proportional improvements in the radial model to make the results more accurate.

First, the ideal reference points are determined. The ideal reference point for the input variable is determined using Equation (1), that for the desirable output variable is determined using Equation (2), and that for the undesirable output variable is determined using Equation (3).

• $\mathrm{m}\mathrm{i}\mathrm{n}{d}_{i0}$

$$\mathrm{s}.\mathrm{t}.\left\{\begin{array}{l}{\sum }_{j=1}^n{\lambda }_j{x}_{ij}\le d_{i0},\\ {\sum }_{j=1}^n{\lambda }_j{x}_{-ij}\le x_{-i0},-i=1,\cdots ,i-1,i+1,\cdots m\\ {\sum }_{j=1}^n{\lambda }_j{y}_{rj}\ge y_{r0},r=1,\cdots ,s_1\\ {\sum }_{j=1}^n{\lambda }_j{c}_{kj}=c_{k0},k=1,\cdots ,s_2\\ {\lambda }_j\ge 0,j=1,\cdots ,n\end{array}\right.$$

(1)

• $\mathrm{m}\mathrm{a}\mathrm{x}{\delta }_{r0}$

$$\mathrm{s}.\mathrm{t}.\left\{\begin{array}{l}{\sum }_{j=1}^n{\lambda }_j{y}_{rj}\ge {\delta }_{r0},\\ {\sum }_{j=1}^n{\lambda }_j{y}_{-rj}\ge y_{-r0},-r=1,\cdots ,r-1,r+1,\cdots s_1\\ {\sum }_{j=1}^n{\lambda }_j{x}_{ij}\le x_{i0},i=1,\cdots ,m\\ {\sum }_{j=1}^n{\lambda }_j{c}_{kj}=c_{k0},k=1,\cdots ,s_2\\ {\lambda }_j\ge 0,j=1,\cdots ,n\end{array}\right.$$

(2)

• $\mathrm{m}\mathrm{i}\mathrm{n}{\phi }_{k0}$

$$\mathrm{s}.\mathrm{t}.\left\{\begin{array}{l}{\sum }_{j=1}^n{\lambda }_j{c}_{kj}={\phi }_{k0}\\ {\sum }_{j=1}^n{\lambda }_j{c}_{-kj}=c_{-k0},-k=1,\cdots ,k-1,k+1,\cdots ,s_2\\ {\sum }_{j=1}^n{\lambda }_j{x}_{ij}\le x_{i0},i=1,\cdots ,m\\ {\sum }_{j=1}^n{\lambda }_j{y}_{rj}\ge y_{r0},r=1,\cdots ,s_1\\ {\lambda }_j\ge 0,j=1,\cdots ,n\end{array}\right.$$

(3)

In Equations (1)–(3), $n$ represents the decision-making units. In this study, $n$ represents the “2+26” cities, $m$ represents the number of input variables $x$, $s_1$ represents the number of desirable outputs $y$, and $s_2$ represents the number of undesirable outputs $c$. By calculating Equations (1)–(3), an ideal reference point $(d_{i0}^{*},{\delta }_{r0}^{*},{\phi }_{k0}^{*})$ can be determined, which is also the optimal solution. The ideal reference point is then substituted into Equation (4). Using the linear programming Equation (4), the MEA efficiency of each variable of the decision-making unit can be determined.

• $\mathrm{m}\mathrm{a}\mathrm{x}{\beta }_{i0}+{\beta }_{r0}+{\beta }_{k0}$

$$\mathrm{s}.\mathrm{t}.\left\{\begin{array}{l}{\sum }_{j=1}^n{\lambda }_j{x}_{ij}\le x_{i0}-{\beta }_{i0}\left(x_{i0}-d_{i0}^{\mathrm{*}}\right),i=1,\cdots ,m\\ {\sum }_{j=1}^n{\lambda }_j{y}_{rj}\ge y_{r0}+{\beta }_{r0}\left({\delta }_{r0}^{\mathrm{*}}-y_{r0}\right),r=1,\cdots ,s_1\\ {\sum }_{j=1}^n{\lambda }_j{c}_{kj}=c_{k0}-{\beta }_{k0}\left(c_{k0}-{\phi }_{k0}^{\mathrm{*}}\right),k=1,\cdots ,s_2\\ {\lambda }_j\ge 0,j=1,\cdots ,n\end{array}\right.$$

(4)

$({\lambda }_j^{*},{\beta }_{i0}^{*},{\beta }_{r0}^{*},{\beta }_{k0}^{*})$ is the optimal solution for Equation (4). Finally, by combining the above calculated parameter values, the MEA efficiency values of each decision-making unit $(x_{i0},y_{r0},c_{k0})$ and all its variables can be calculated.

The calculation formula for the MEA efficiency value of each input variable is outlined in Equation (5):

$${\theta }_i=\frac{x_{i0}-{\beta }_{i0}^{\mathrm{*}}\left(x_{i0}-d_{i0}^{\mathrm{*}}\right)}{x_{i0}}$$

(5)

The calculation formula for the MEA efficiency value of each desirable output variable is demonstrated by Equation (6):

$${\theta }_r=\frac{y_{r0}}{y_{r0}+{\beta }_{r0}^{\mathrm{*}}\left({\delta }_{r0}^{\mathrm{*}}-y_{r0}\right)}$$

(6)

The calculation formula for the MEA efficiency value of each undesirable output variable is expressed in Equation (7):

$${\theta }_k=\frac{c_{k0}-{\beta }_{k0}^{\mathrm{*}}\left(c_{k0}-{\phi }_{k0}^{\mathrm{*}}\right)}{c_{k0}}$$

(7)

By combing the MEA efficiency of the overall variables, based on the SBM model of Tone [42], a comprehensive MEA efficiency can be established. The calculation formula is described in Equation (8):

$${\theta }_0=\frac{1-\frac{1}{m}{\sum }_{i=1}^m\frac{{\beta }_{i0}^{\mathrm{*}}\left(x_{i0}-d_{i0}^{\mathrm{*}}\right.)}{x_{}{x}_{i0}}}{1+\frac{1}{s_1+s_2}\left[{\sum }_{r=1}^{s_1}\frac{{\beta }_{r0}^{\mathrm{*}}\left({\delta }_{r0}^{\mathrm{*}}-y_{r0}\right)}{y_{r0}}+{\sum }_{k=1}^{s_2}\frac{{\beta }_{k0}^{\mathrm{*}}\left(c_{k0}-{\phi }_{k0}^{\mathrm{*}}\right)}{c_{k0}}\right]}$$

(8)

Obviously, it is more comprehensive and effective than the initial MEA efficiency measure, including every input and output variable efficiency.

3.2.2. Global Reference Malmquist Model

The Malmquist index is a commonly used indicator for measuring productivity changes, also known as the Malmquist Total Factor Productivity Index (MI), which can be used to evaluate changes in productivity during two different periods. The Malmquist index based on DEA can also be decomposed into two components: technological efficiency change (EC) and technological change (TC) [43]. This article uses the nonradial MEA model as an alternative method for measuring the efficiency in DEA [44].

The global reference Malmquist model is a Malmquist index calculation method proposed by Pastor and Lovell [45]. Since all periods are compared to the same global frontier respectively, only one Malmquist index can be obtained. The global reference Malmquist value ($M_g$) is described in Equation (9):

$$M_g\left(x^{t+1},y^{t+1},x^t,y^t\right)=\frac{E^g\left(x^{t+1},y^{t+1}\right)}{E^g\left(x^t,y^t\right)}$$

(9)

In Equation (9), $E^g\left(x^{t+1},y^{t+1}\right)$ indicates the efficiency value of $\left(x^{t+1},y^{t+1}\right)$, obtained by referring to the global frontier; $E^g\left(x^t,y^t\right)$ indicates the efficiency value of $\left(x^t,y^t\right)$, obtained by referring to the global frontier.

EC can be defined as demonstrated in Equation (10):

$$\mathrm{E}\mathrm{C}=\frac{E^{t+1}\left(x^{t+1},y^{t+1}\right)}{E^t\left(x^t,y^t\right)}$$

(10)

In Equation (10), $E^{t+1}\left(x^{t+1},y^{t+1}\right)$ indicates the efficiency value of $\left(x^{t+1},y^{t+1}\right),$ obtained by referring to the period frontier; $E^t\left(x^t,y^t\right)$ indicates the efficiency value of $\left(x^t,y^t\right)$, obtained by referring to the period frontier.

Thus, TC can be defined as demonstrated in Equation (11):

$$\mathrm{T}{\mathrm{C}}_g=\frac{M_g\left(x^{t+1},y^{t+1},x^t,y^t\right)}{\mathrm{E}\mathrm{C}}=\frac{E^g\left(x^{t+1},y^{t+1}\right)/E^{t+1}\left(x^{t+1},y^{t+1}\right)}{E^g\left(x^t,y^t\right)/E^t\left(x^t,y^t\right)}=\frac{E^g\left(x^{t+1},y^{t+1}\right)}{E^{t+1}\left(x^{t+1},y^{t+1}\right)}\frac{E^t\left(x^t,y^t\right)}{E^g\left(x^t,y^t\right)}$$

(11)

3.2.3. Spatial Correlation Analysis

(1)

Global Moran Index

The most commonly used spatial autocorrelation analysis is the Global Moran Index (Moran’s I), which can be used to represent spatial distribution characteristics. The formula for calculating the global Moran’s I is as follows:

$$\mathrm{G}\mathrm{l}\mathrm{o}\mathrm{b}\mathrm{a}\mathrm{l}\mathrm{M}\mathrm{o}\mathrm{r}\mathrm{a}\mathrm{n}’\mathrm{s}\mathrm{I}=\frac{1}{{\sum }_{i=1}^n{\sum }_{j=1}^n{W}_{ij}}\times \frac{{\sum }_{i=1}^n{\sum }_{j=1}^n{W}_{ij}\left(x_i-\bar{x}\right)\left(x_j-\bar{x}\right)}{{\sum }_{j=1}^n{\left(x_i-\bar{x}\right)}^2/n}$$

(12)

In Equation (12), $n$ is the number of cities;$W_{ij}$ is the spatial weight matrix; and $x_i\mathrm{a}\mathrm{n}\mathrm{d}{x}_j$ represent observations from different cities. At significance levels of 1%, 5% or 10%, when Moran’s I > 0, there is a spatial positive correlation, exhibiting an aggregation phenomenon; when Moran’s I < 0, there is a spatial negative correlation and a dispersion phenomenon; and when Moran’s I = 0, there is no spatial autocorrelation.

(2)

Local Moran’s Index

The local Moran’s I can denote the spatial agglomeration (dispersion) state of a region at a specific time, usually represented by a Moran scatter plot. If Moran’s I > 0, it indicates that similar units in the local area tend to gather spatially, and the surrounding areas with high (low) values in the local area also have high (low) values. If Moran’s I < 0, it indicates that similar units in the local area tend to diverge, and the surrounding areas with high (low) values in the local area also have low (high) values. If Moran’s I = 0, it indicates that adjacent units in the local area are randomly distributed in space. The formula for calculating the local Moran’s I is as follows:

$$\mathrm{L}\mathrm{o}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{M}\mathrm{o}\mathrm{r}\mathrm{a}\mathrm{n}’\mathrm{s}\mathrm{I}=\frac{\left(x_i-\bar{x}\right){\sum }_{j=1}^n{W}_{ij}\left(x_j-\bar{x}\right)}{{\sum }_{j=1}^n{\left(x_i-\bar{x}\right)}^2/n}$$

(13)

The meanings of the variables in Equation (12) are the same as those in Equation (13).

3.3. Indicator System

According to most existing studies, the establishment of an atmospheric environmental efficiency evaluation system should fully consider the resources, economy and environment, following the principles of science, practicality and data availability. Most studies use total energy consumption, labor force and capital stock as input indicators, gross domestic product (GDP) as a desirable output indicator and pollution emissions as undesirable indicators [46,47,48,49,50].

3.3.1. Input Variables

The three input variables were total energy consumption, capital stock and labor force. Furthermore, capital stock was not directly available, so the capital stock of each prefecture-level city from 2009 to 2018 was calculated using the perpetual inventory method and the total fixed assets investment of the whole society with a constant price in 2009 [51]. The total energy consumption was calculated using the product of energy consumption per CNY 10,000 of GDP and GDP. The labor force was measured using the number of employees at the end of the year. The above data were all from the China City Statistical Yearbook (2010–2019) [52,53,54,55,56,57,58,59,60,61].

3.3.2. Desirable Output Variable

This study selected the regional GDP as the desirable output variable. In order to unify the data caliber, the annual regional GDP was converted into 2009 constant prices with the data selected from the China City Statistical Yearbook (2010–2019) [52,53,54,55,56,57,58,59,60,61].

3.3.3. Undesirable Output Variables

The reference studies selected SO₂ emissions and industrial smoke (dust) emissions as the undesirable output variables, but the difference is that this study added CO₂ emissions as another undesirable output variable. Although CO₂ is not an atmospheric pollutant, it is the main cause of global warming. Therefore, incorporating CO₂ emissions into the atmospheric environmental efficiency indicator system is more comprehensive. Among them, carbon dioxide emissions were sourced from the CEADs database [62]. SO₂ and industrial smoke (dust) emissions were obtained from the China City Statistical Yearbook (2010–2019) [52,53,54,55,56,57,58,59,60,61].

The descriptive statistics of all the above variables are shown in

Table 1. Descriptive statistics of all variables, 2009–2018.

Variables	Unit	Average	Std. Dev.	Maximum	Minimum
Total energy consumption	105 tons of standard coal	2798.40	5270.61	11,097.00	555.77
Capital stock	108 CNY	10,763.98	11,232.88	70,457.25	995.66
Labor force	104 persons	417.69	401.60	819.30	16.08
GDP	108 CNY	3282.93	3834.00	22,875.13	348.71
SO2 emissions	104 tons	7.61	5.99	33.19	0.09
Smoke (dust) emissions	104 tons	5.34	7.76	53.61	0.08
CO2 emissions	106 tons	51.61	28.78	159.81	9.36

.

4. Results

4.1. Evaluation of Atmospheric Environmental Efficiency in “2+26” Cities

This paper used the nonradial MEA model to evaluate the atmospheric environmental efficiency in the “2+26” cities. The establishment of an atmospheric environmental efficiency evaluation system and the data sources are presented in Section 3.3. All results are shown in

Table 2. Urban atmospheric environmental efficiency and ranking in the “2+26” cities during 2009–2018.

City	2009	2010	2011	2012	2013	2014	2015	2016	2017	2018	Mean Value	Rank
Beijing	0.940	1.000	1.000	0.964	0.687	0.671	0.728	0.974	1.000	1.000	0.896	3
Tianjin	0.785	0.803	0.787	0.804	0.788	0.724	0.841	1.000	1.000	1.000	0.853	5
Shijiazhuang	0.463	0.637	0.659	0.625	0.634	0.634	0.584	0.593	0.619	0.603	0.605	23
Tangshan	0.739	0.835	1.000	0.858	0.909	0.921	1.000	0.839	1.000	1.000	0.910	1
Handan	0.554	0.556	0.924	0.711	0.627	0.663	0.515	0.532	0.469	0.447	0.600	24
Xingtai	0.713	0.547	0.737	0.527	0.555	0.671	0.629	0.589	0.617	0.604	0.619	22
Baoding	0.696	0.691	0.684	0.670	0.677	0.730	0.692	0.746	0.759	0.774	0.712	17
Cangzhou	0.763	0.780	0.699	0.763	0.781	0.649	0.754	0.877	1.000	1.000	0.807	9
Langfang	0.730	0.726	0.596	0.571	0.622	0.774	0.731	0.744	0.684	1.000	0.718	16
Hengshui	0.829	0.809	0.735	0.579	0.612	0.782	0.734	0.747	1.000	0.769	0.759	13
Taiyuan	0.841	0.619	0.560	0.613	0.604	0.605	0.602	0.679	0.719	0.740	0.658	19
Yangquan	0.830	0.826	0.839	0.829	0.683	0.662	0.600	0.583	0.387	0.340	0.658	20
Changzhi	1.000	0.887	1.000	0.987	0.704	0.852	0.823	0.640	0.649	0.679	0.822	7
Jincheng	1.000	0.586	0.649	0.652	0.576	0.604	0.429	0.543	0.338	0.351	0.573	28
Jinan	0.633	0.688	0.685	0.662	0.674	0.722	0.717	0.760	0.846	1.000	0.739	14
Zibo	1.000	1.000	1.000	0.890	0.660	0.612	0.579	0.691	0.769	1.000	0.820	8
Jining	0.982	0.888	0.829	0.773	0.738	0.648	0.636	0.729	0.838	0.954	0.802	10
Dezhou	0.646	0.771	0.828	0.827	0.759	0.636	0.687	0.683	0.739	0.775	0.735	15
Liaocheng	1.000	1.000	0.899	0.837	0.906	0.831	0.747	0.800	0.816	1.000	0.884	4
Binzhou	0.939	0.907	0.832	0.783	0.760	0.673	0.654	0.816	0.662	0.710	0.773	11
Heze	0.616	0.692	0.936	1.000	1.000	1.000	0.952	0.983	0.819	1.000	0.900	2
Zhengzhou	0.711	0.743	0.730	0.723	0.721	0.672	0.684	0.787	0.858	1.000	0.763	12
Kaifeng	1.000	0.948	0.908	0.887	0.761	0.709	0.733	0.703	0.794	1.000	0.844	6
Anyang	0.708	0.638	0.646	0.611	0.613	0.636	0.635	0.405	0.420	0.460	0.577	27
Hebi	0.610	0.676	0.623	0.631	0.624	0.617	0.618	0.422	0.445	0.661	0.593	25
Xinxiang	0.616	0.631	0.652	0.670	0.641	0.630	0.648	0.681	0.730	0.536	0.643	21
Jiaozuo	0.605	0.641	0.615	0.620	0.597	0.482	0.459	0.471	0.564	0.754	0.581	26
Puyang	0.535	0.544	0.657	0.676	0.628	0.651	0.667	0.759	0.808	1.000	0.693	18
Mean value	0.767	0.752	0.775	0.741	0.698	0.695	0.681	0.706	0.727	0.791	0.733	—

.

4.1.1. Discussion from the Perspective of All “2+26” Cities

Looking at the “2+26” cities as a whole, the average atmospheric environmental efficiency was 0.733, indicating that the overall level of the “2+26” cities was not high, with a remaining improvement potential of 26.7% (see Table 2). The average growth rate of the atmospheric environmental efficiency in the “2+26” cities from 2009 to 2018 was 0.3%, and the average growth rate of real GDP was 8%. This showed that the atmospheric environmental efficiency, a metric gauging the balance between atmospheric environmental and economic development, has experienced some growth. However, the rate of atmospheric environmental improvement was much slower than the rate of economic growth.

The atmospheric environmental efficiency of the “2+26” cities exhibited a general trend of an initial decrease and a subsequent increase from 2009 to 2018. The analysis of the whole atmospheric environmental efficiency can be divided into two stages. In the first stage, from 2009 to 2015, the atmospheric environmental efficiency gradually decreased. During this stage, the maximum efficiency was 0.767 (in 2009), and the minimum was 0.681 (in 2015). This was mainly due to the resumption of production in many heavy industrial enterprises in the region after the Olympic Games, resulting in increased industrial waste gas pollution emissions and CO₂ emissions. Among them, there was a slight increase in efficiency in 2011, possibly due to policies such as energy conservation, emission reduction and the promotion of economic transformation, which played a certain role in the opening year of the “Twelfth Five-Year Plan” in the same year. It is worth noting that the atmospheric environmental efficiency dropped below 0.7 in 2013 and only recovered to above 0.7 after 2016. Furthermore, the three years from 2013 to 2015, with the lowest atmospheric environmental efficiency in the “2+26” cities, coincided with the occurrence of haze in the Beijing–Tianjin–Hebei region. Although the implementation of the “Air Pollution Prevention and Control Action Plan” [63] and the “12th Five-Year Plan on Air Pollution Prevention and Control in Key Regions” [64] in 2013 played a promoting role in the control of air pollution, improvement in atmospheric environmental efficiency cannot be achieved in the short term. The implementation of environmental policies, the optimization and upgrading of industrial structures, and the energy saving and emission reduction of green technologies were all phased processes. In addition, heavy industries such as cement, steel and coking were developed in the region. At this stage, SO₂, smoke and CO₂ emissions in the “2+26” cities were still in a high-value stage, and economic development was heavily based on fossil energy consumption. In the second stage, after 2015, the atmospheric environmental efficiency value gradually increased year by year until 2018 to a maximum of 0.791. It can be seen from

Table 3. Ratio of reducible energy input and air pollution emissions during 2009–2018.

Year	Energy	SO2 Emissions	Smoke (Dust) Emissions	CO2 Emissions
2009	18.41%	18.44%	35.60%	31.45%
2010	15.38%	19.73%	44.19%	35.74%
2011	8.50%	26.68%	38.21%	31.69%
2012	3.64%	37.32%	51.73%	34.03%
2013	4.56%	48.44%	55.78%	35.86%
2014	7.17%	57.06%	50.81%	35.42%
2015	7.20%	56.48%	57.33%	33.60%
2016	9.46%	45.42%	42.31%	34.32%
2017	13.97%	36.47%	32.30%	31.88%
2018	15.20%	24.89%	19.89%	24.71%

that, after 2015, although the proportion of energy input increased, the proportions of SO₂, smoke (dust) and CO₂ emissions that could be reduced decreased significantly, and the growth efficiency of the atmospheric environmental efficiency also increased. In other words, the reasons for this phenomenon were the implementation of environmental policies and strengthened supervision, as well as significant reductions in industrial waste gas emissions through energy-saving and emission-reduction measures. The implementation of the “2017 Plan for Prevention and Control of Atmospheric Pollution in Beijing-Tianjin-Hebei and Surrounding Areas” in early 2017, which included capacity reduction, the establishment of “no coal zones” and the restriction of waste gas emissions, had good implementation effects in accordance with the time nodes. Therefore, there was a significant improvement in the atmospheric environmental efficiency from 2017 to 2018.

4.1.2. Discussion from the Perspective of Provinces in the “2+26” Cities

From the perspective of provinces, as shown in Figure 3, the division was based on the provinces to which the “2+26” cities belong. The atmospheric environmental efficiency in the Beijing–Tianjin region was the highest. However, the efficiency decreased from 0.88 to 0.69 between 2012 and 2014 but gradually improved after 2014, with efficiency values of 1 in both 2017 and 2018, positioning the region at the forefront of production with a balanced development of economy and atmospheric environment. Following the Beijing–Tianjin region, the ranking of atmospheric environmental efficiency was Shandong > Hebei > Shanxi > Henan. During the research period, the atmospheric environmental efficiency of the Beijing–Tianjin region and Shandong province were always higher than the average, showing good performance. The atmospheric environmental efficiency of Hebei province showed a trend of fluctuating up, but it was still below the average level. Except for 2009, the atmospheric environmental efficiency of Shanxi was consistently below the average level, showing a downward trend with a significant decrease from 0.91 to 0.52. The atmospheric environmental efficiency of Henan was always lower than average. Overall, the gap in atmospheric environmental efficiency between different provinces demonstrated an expanding trend. From a specific analysis perspective, the Beijing–Tianjin region had a developed economy, a high proportion of the tertiary industry, a developed high-tech industry and industrial upgrading, resulting in low pollution emissions. In contrast, the lower rankings of Henan and Shanxi may be attributed to their urban economic size, which stood at the bottom among the “2+26” cities, excluding provincial capital cities. These cities faced challenges in industrial structure adjustment and the implementation of green technologies. In particular, Shanxi was in the middle stage of industrial development, where the industrial economy showed growth but lacked sufficient investment in industrial technology, leading to fluctuating and declining efficiency. Another contributing factor could be the rapid development of cities like Beijing and Tianjin, which were located at the forefront. These cities demonstrated significant disparities in terms of economic development, industrial structure rationality and the implementation of environmental protection policies when compared to Henan and Shanxi. In recent years, Shandong has achieved good results in urban transformation, especially in the development of high-end industries. As a strong industrial province, it had a high degree of industrial agglomeration and many advantageous industries. However, the atmospheric environmental efficiency of Hebei ranked in the middle among the five regions. The main reason is that Hebei’s major industries were traditional heavy industries such as steel, metallurgy and chemical manufacturing. Even though the Beijing–Tianjin–Hebei development strategy has promoted Hebei’s development, its own basic conditions were poor, and it was difficult to change its development model in the short term. Despite these cities actively adjusting their industrial structure and developing their tertiary industries, a lack of funding, technology and talent have hindered their ability to achieve economies of scale, resulting in insufficient economic output and low atmospheric environmental efficiency.

4.1.3. Discussion from the Perspective of Prefecture-Level Cities in the “2+26” Cities

From the perspective of prefecture-level cities, the values of atmospheric environmental efficiency were between 0.573 and 0.910 (see Table 2). There was a significant discrepancy between the best- and worst-performing cities, demonstrating imbalanced development between cities. The structure of the atmospheric environmental efficiency of the “2+26” cities presented an olive shape, with few cities at either end and more in the middle. The top five cities with the highest atmospheric environmental efficiency were Tangshan, Heze, Beijing, Liaocheng and Tianjin, all with values above 0.85. Fifteen cities fell between 0.65 and 0.85, including Xingtai, Baoding and Langfang, where economic and environmental development was not well coordinated. In addition, the eight cities (Shijiazhuang, Handan, Xingtai, Jincheng, Anyang, Hebi, Xinxiang and Jiaozuo) below 0.65 were key construction cities for improving the atmospheric environmental efficiency. The primary reason for the low atmospheric environmental efficiency in these eight cities was the inefficiency in undesirable output MEA efficiency. Therefore, the future improvement direction is to reduce the emissions of SO₂, smoke (dust) and CO₂.

4.2. Analysis of the Dynamic Evolution and Driving Mechanisms of Atmospheric Environmental Efficiency in the “2+26” Cities

From 2009 to 2018, the mean value of the global Malmquist (GM) index was 1.0007, indicating an average annual increase of 0.07%. Although the increase was small, it also indicated that the contradiction between atmospheric environmental protection and economic development in the “2+26” cities was alleviated, and there was no trend toward further deterioration. The GM index of the atmospheric environment exhibited an initial increase, followed by a decrease, and then a fluctuating rise. It rose to 1.032 in 2010–2011, declined to 0.947 in 2012–2013 and subsequently experienced a fluctuating increase to 1.081 in 2017–2018 (see Figure 4). From the index decomposition results (see

Table 4. GM index in the “2+26” cities (decomposing the EC and TC) during 2009–2018.

Number	Region	GM	EC	TC	Trend
1	Beijing	1.007	1.000	1.007
2	Tianjin	1.027	1.000	1.027
3	Shijiazhuang	1.030	0.986	1.044
4	Tangshan	1.034	1.000	1.034
5	Handan	0.976	0.952	1.025
6	Xingtai	0.982	1.000	0.982
7	Baoding	1.012	0.995	1.017
8	Cangzhou	1.031	1.000	1.031
9	Langfang	1.036	1.000	1.036
10	Hengshui	0.992	0.984	1.007
11	Taiyuan	0.986	0.988	0.998
12	Yangquan	0.905	1.000	0.905
13	Changzhi	0.958	0.982	0.976
14	Jincheng	0.890	0.915	0.973
15	Jinan	1.052	1.000	1.052
16	Zibo	0.999	1.000	0.999
17	Jining	0.997	0.997	1.000
18	Dezhou	1.021	0.982	1.040
19	Liaocheng	1.005	1.000	1.005
20	Binzhou	0.969	0.980	0.989
21	Heze	1.055	1.000	1.055
22	Zhengzhou	1.039	1.000	1.039
23	Kaifeng	1.006	1.000	1.006
24	Anyang	0.953	0.993	0.960
25	Hebi	1.009	0.963	1.048
26	Xinxiang	0.985	1.006	0.979
27	Jiaozuo	1.025	0.981	1.044
28	Puyang	1.072	1.000	1.072
29	Mean Value	1.001	0.989	1.012

), the EC was less than 1, but the TC was greater than 1, overall, which comprehensively promoted the improvement of the GM index. However, the deterioration of the EC served as the primary reason affecting the increase in the GM index. Technological progress drove the rise in atmospheric environmental efficiency. Specifically, although technological advancements in the industrial processes, research and development of new industrial products and improvements in production methods promoted enhancement of the GM index, they failed to fully offset the decline in technical efficiency resulting from outdated industrial production and management practices. Despite active efforts to adjust the urban industrial structure in the “2+26” cities, the reduction in industrial scale and inadequate funding and technology in the emerging tertiary sector led to insufficient output, thereby contributing to a decline in technical efficiency.

Based on the decomposition results of the GM index from 2009 to 2018, the 28 cities were divided into two groups for analysis based on whether their GM index was greater or less than 1. There were 16 cities with a GM index greater than 1, accounting for 57% of the total sample. Among them, the cities of Jiaozuo, Dezhou, Baoding, Hebi and Shijiazhuang had EC values of less than 1, but their TC values increased by 4.4%, 4%, 1.7%, 4.8% and 4.4%, respectively. Therefore, these cities achieved an improvement in the atmospheric environmental MI solely through TC. However, there were 12 cities with a GM index of less than 1. Among them, the cities of Xinxiang, Xingtai, Yangquan and Zibo had EC values greater than or equal to 1, but their TC values decreased by 2.09%, 1.83%, 9.4% and 0.1%, respectively. TC was the main direction for improving the atmospheric environmental MI in these cities. In the cities of Hengshui, Handan and Jining, their TC values were greater than or equal to 1, and their EC values decreased by 1.6%, 4.8% and 0.3%, respectively. Therefore, improving the EC is the main direction for enhancing the atmospheric environmental MI in these cities. The remaining cities with a GM index of less than 1 were influenced by both the EC and TC. These cities should be the key focus for atmospheric environmental efficiency improvement, with the simultaneous promotion of the EC and TC as dual drivers to enhance the atmospheric environmental MI.

4.3. Spatial Correlation Analysis and Spatial Evolution

4.3.1. Spatial Correlation Analysis of the Atmospheric Environment in the “2+26” Cities

Considering the “2+26” cities as a whole, the global spatial correlation of the atmospheric environmental efficiency of the “2+26” cities was explored based on Moran’s I.

Table 5. Global Moran’s I of atmospheric environmental efficiency in the “2+26” cities.

Year	Global Moran’s I	p-Value	Year	Global Moran’s I	p-Value
2009	−0.178	0.155	2014	−0.005	0.404
2010	−0.072	0.400	2015	0.240	0.019
2011	−0.041	0.489	2016	0.443	0.000
2012	−0.033	0.489	2017	0.357	0.002
2013	0.145	0.085	2018	0.399	0.001

presents the results of Moran’s I used to measure the “2+26” cities from 2009 to 2018. Over the entire study period, the distribution of atmospheric environmental efficiency was initially irregular and gradually became more concentrated, demonstrating significant agglomeration features. From 2009 to 2014, except for 2013, Moran’s I did not pass the significance test, indicating that the atmospheric environmental efficiency did not exhibit spatial autocorrelation and had no obvious spatial distribution features. However, in 2015, Moran’s I passed the 5% significance level test, and from 2016 to 2018, Moran’s I passed the 1% significance level test, demonstrating a significant positive spatial correlation. That is, cities with a high atmospheric environmental efficiency value were adjacent to other cities with a high atmospheric environmental efficiency value, whereas cities with a low atmospheric environmental efficiency value were adjacent to other cities with a low atmospheric environmental efficiency value. $\mathrm{M}\mathrm{o}\mathrm{r}\mathrm{a}\mathrm{n}’\mathrm{s}\mathrm{I}$ increased from 0.24 in 2015 to 0.399 in 2018, with an annual growth rate of 18.5%, indicating that the urban atmospheric environmental efficiency was rapidly agglomerating, and the spatial agglomeration degree was quickly becoming stronger.

To further depict the local spatial characteristics of the atmospheric environmental efficiency in the “2+26” cities, this paper employed Stata 16 software to draw Moran scatter plots of atmospheric quality efficiency in 2009, 2013 and 2018, which facilitated the exploration of the spatial distribution characteristics (Figure 5). The reasons for selecting these three years for specific analysis were as follows: 2009 and 2018 were the start and end years, and 2015 was the middle year between 2009 and 2018. The p-value accurately reflects the trend of Moran’s index of changes.

The four quadrants of the Moran scatter plot represent different types of spatial autocorrelation: The first quadrant (H–H) represents high-value areas around high-value areas, whereas the third quadrant (L–L) represents low-value areas around low-value areas, reflecting positive spatial autocorrelation in high- or low-value areas. The second quadrant (L–H) represents high-value areas around low-value areas, whereas the fourth quadrant (H–L) represents low-value areas around high-value areas, both reflecting negative spatial autocorrelation. Although the significance test was not passed in 2009, the majority of cities showed an H–L or L–H distribution, with more than half of the cities located in the second (13 cities) and fourth (7 cities) quadrants, only 5 cities in the first quadrant (H–H) and 3 cities in the third quadrant (L–L). This indicates that more cities had negative spatial correlation than positive spatial correlation in 2009. In 2013, the overall dispersion level of the “2+26” cities increased significantly, with the distribution pattern shifting from L–H and H–L distribution clustering to H–H and L–L distributions. In 2018, the clustering level of the “2+26” cities further increased, mainly reflected in the increase in H–H city clustering, with a total of 23 cities in the first (H–H) and third (L–L) quadrants. The local Moran scatter plots from 2009 to 2018 show that the “2+26” cities were significantly clustered, showing spatial positive correlation and demonstrating a distribution pattern of clustering toward H–H and L–L.

Table 6. The same locations in the Moran scatter plots of atmospheric environmental efficiency in the “2+26” cities.

Agglomeration Feature	2009	2013	2018
H–H (first quadrant)	Tianjin, Taiyuan, Binzhou, Zibo, Jincheng	Jining, Kaifeng, Tianjin, Tangshan, Binzhou, Dezhou, Cangzhou	Jining, Beijing, Langfang, Cangzhou, Tianjin, Tangshan, Jinan, Zibo, Heze, Zhengzhou, Kaifeng
L–H (second quadrant)	Shijiazhuang, Puyang, Handan, Jinan, Dezhou, Heze, Jiaozuo, Xinxiang, Baoding, Cangzhou, Langfang, Tangshan, Zhengzhou	Jinan, Zibo, Puyang, Langfang, Xingtai	Binzhou, Baoding, Dezhou
L–L (third quadrant)	Hebi, Xingtai, Anyang	Beijing, Xinxiang, Taiyuan, Handan, Baoding, Yangquan, Jincheng, Jiaozuo, Anyang, Hebi, Shijiazhuang, Hengshui	Yangquan, Jincheng, Handan, Anyang, Xinxiang, Xingtai, Hengshui, Jiaozuo, Shijiazhuang, Hebi, Changzhi, Taiyuan
H–L (fourth quadrant)	Heng shui, Yangquan, Beijing, Kaifeng, Liaocheng, Jining, Changzhi	Changzhi, Zhengzhou, Liaocheng, Heze	Puyang, Liaocheng

shows the distribution of cities in each quadrant. Overall, in 2009, 2013 and 2018, the number of cities in the first and third quadrants increased, accounting for 29%, 68% and 82%, respectively. That is, most cities had similar efficiency values compared to the surrounding cities, with the overall spatial difference narrowing. In 2018, the cities located in the first quadrant included Beijing, Tianjin (both municipalities directly under the central government), Jinan and Zhengzhou (both provincial capital cities), with abundant funds and technology, radiating and driving the development of surrounding cities and demonstrating a concentration of atmospheric environmental efficiency (H–H). Shijiazhuang and Taiyuan were located in the third quadrant, although they are provincial capital cities, and their radiation-driven development effect was not strong, showing a concentration of atmospheric environmental efficiency (L–L). It was necessary to explore a path suitable for one’s own economic and environmental development to improve the atmospheric environmental efficiency.

4.3.2. Spatial Evolution in the “2+26” Cities

Based on previous research and incorporating the calculation results of this study, the research units were categorized as follows: 0–0.65, representing a relatively low-efficiency city in terms of MEA; 0.65–0.85, representing a relatively medium-efficiency city; 0.85–1, representing a relatively high-efficiency city; and 1, representing a high-efficiency city in terms of MEA. The mapped areas of the spatial evolution of the atmospheric environmental efficiency (city MEA efficiency) scores of the “2+26” cities are depicted in Figure 6. This provides a more intuitive representation of the uneven spatial distribution and demonstrates the process of spatial evolution.

In terms of the city MEA efficiency (Figure 6), the proportions of cities categorized as having high MEA efficiency, relatively high MEA efficiency, relatively medium MEA efficiency and relatively low MEA efficiency in 2009 were 5:3:11:9. However, in 2018, 12 cities, including Beijing, Tianjin and Jinan, ascended to become cities with high MEA efficiency, adding 7 cities compared to 2009, which represents significant improvement. The proportions of cities with high, relatively high, relatively medium and relatively low MEA efficiency were 12:1:8:7. The proportion of cities categorized as having high MEA efficiency increased by 25%, whereas the proportions of those with relatively high, relatively medium and relatively low MEA efficiency decreased by 7%, 11% and 7% respectively. From 2009 to 2015, the urban MEA efficiency generally declined, but from 2015 to 2018, there was a noticeable improvement in the urban MEA efficiency, primarily concentrated in the northern region of the “2+26” cities, including the Beijing–Tianjin area, as well as the Zhengzhou–Zibo line in the southern and southeastern parts. Throughout the study period, there were significant variations in the MEA efficiency values of the “2+26” cities, and in terms of spatial distribution, a cluster of cities with high MEA efficiency gradually formed, centered around the Beijing–Tianjin region, including cities like Beijing, Tianjin, Tangshan, Langfang, Cangzhou and the Zhengzhou–Zibo line, which included Liaocheng, Zhengzhou and Kaifeng. In the central region, there was a cluster of cities with medium efficiency, including Baoding, Hengshui, Dezhou and Binzhou. In the western region, there was a cluster of cities with low efficiency along the Shijiazhuang–Xinxiang line and in western areas, including Jincheng, Handan and Xinxiang, comprising a total of seven cities, primarily located in Hebei, Henan and Shanxi provinces.

In terms of city smoke (dust) MEA efficiency (Figure 7), in 2009, the number of cities with high smoke (dust) MEA efficiency accounted for half of the “2+26” cities. However, by 2012, it had decreased to only six cities (Heze, Yangquan, Zibo, Dezhou, Liaocheng and Kaifeng). Among them, Baoding, Cangzhou, Langfang, Hengshui, Taiyuan, Handan and Jincheng transitioned from high smoke (dust) MEA efficiency to relatively low smoke (dust) MEA efficiency. Especially by 2015, only Tianjin and Tangshan remained cities with high smoke (dust) MEA efficiency. This indicates an overall decline in smoke (dust) MEA efficiency from 2009 to 2015, with a significant reduction in cities with high smoke (dust) MEA efficiency. It was not until 2015–2018 that there was some improvement in smoke (dust) MEA efficiency. In 2018, the number of cities with high smoke (dust) MEA efficiency significantly increased to 15. Smoke (dust) MEA-efficient cities were mainly distributed in the northern part of the “2+26” city region and along the Tangshan–Heze line, including the southernmost cities of Zhengzhou and Kaifeng. On the contrary, the area along the Shijiazhuang–Xinxiang line and its western region consisted of relatively low-efficiency cities. This highlights a significant exacerbation in regional disparities in smoke (dust) MEA efficiency. Notably, the cities of Shijiazhuang and Anyang consistently remained in a state of relatively low smoke (dust) MEA efficiency. The provincial capital cities of Jinan and Zhengzhou experienced a substantial improvement in smoke (dust) MEA efficiency, transitioning from a state of relatively low smoke (dust) MEA efficiency to high efficiency, indicating further optimization of the input and output.

In terms of SO₂ MEA efficiency (Figure 8), the high-efficiency cities in 2009 included 12 cities, such as Baoding and Langfang, whereas the remaining cities, except for Beijing, which showed relatively high efficiency, were considered to have relatively low SO₂ MEA efficiency. By 2012, the number of cities with high SO₂ MEA efficiency had decreased by three-fourths compared to 2009, with only Kaifeng and Heze maintaining high efficiency in both smoke (dust) MEA and SO₂ MEA. Significant declines in efficiency were observed in Baoding, Langfang, Hengshui, Taiyuan, Jining, Liaocheng, Jincheng and Binzhou, highlighting noticeable disparities among the cities, with over half of the “2+26” cities having relatively low SO₂ MEA efficiency. In 2015, the spatial distribution pattern of SO₂ MEA efficiency was similar to that of 2012, with the exception of specific cities such as Beijing, Tangshan and Baoding. However, in 2018, there was a significant improvement in SO₂ MEA efficiency, indicating a shift from deteriorating efficiency to a positive trend. This resulted in the formation of high SO₂ MEA efficiency clusters centered around the Beijing–Tianjin region in the north and along the southeastern borderlines. The cities with relatively medium efficiency were primarily located in Henan and Shanxi provinces.

In 2009, there were nine cities that demonstrated high CO₂ MEA efficiency (Figure 9), which was lower compared to the cities with high smoke (dust) MEA and SO₂ MEA efficiency. These cities were sparsely distributed in the eastern and western regions of the “2+26” cities, without any apparent distribution pattern. In 2012, Dezhou exhibited remarkable performance and ascended to the status of a city with high CO₂ MEA efficiency. However, the CO₂ MEA efficiency of this city displayed significant fluctuations. In 2015, it declined to a relatively medium-efficiency city, only to rise again in 2018 as a city with high CO₂ MEA efficiency. Yangquan, Changzhi, Liaocheng and Zibo consistently maintained high CO₂ MEA efficiency in both 2009 and 2012. In 2015, the number of high-efficiency cities further decreased, but the distribution pattern of relatively low-efficiency cities remained similar to that of 2012, with a few exceptions, such as Anyang and Zhengzhou. In 2018, the distribution of cities with high CO₂ MEA efficiency exhibited a trend of concentration from the surrounding areas toward the east and north, whereas the relatively low-efficiency cities were concentrated along the Baoding–Xinxiang line and its western side.

Based on the aforementioned analysis, it can be observed that, in 2018, the spatial agglomeration characteristics of the urban MEA efficiency and the undesirable output MEA efficiency (smoke (dust) MEA efficiency, SO₂ MEA efficiency and CO₂ MEA efficiency) were essentially consistent. These high-efficiency cities were primarily concentrated in the northern region centered around Beijing and Tianjin, as well as along the southeastern border. Over the course of the study period, these cities gradually witnessed improvements in their MEA efficiency values. The reasons for variations in the undesirable output MEA efficiency values in other cities can be attributed to the differing levels of technological prowess, which ultimately determined the potential direction of improvement based on the optimal direction vector determined by the individual indicators. This direction vector changed with the shifting technological frontier, resulting in variations in efficiency values among different cities and time periods, leading to distinct spatial distribution characteristics. In actual production, factors such as the enhancement of production technology, the application of new processes and equipment and the form of production organization affected the level of technology. Cities with relatively low urban MEA efficiency were located along the Shijiazhuang–Xinxiang line and its western region. Although Shijiazhuang is the provincial capital city, its urban MEA efficiency values remained relatively low, possibly due to a lack of robust economic growth stimuli. Despite active industrial restructuring, the tertiary industry’s growth was not sufficiently strong, and the input–output ratio was not reasonable, indicating a need for improvement in production technology levels. Cities with relatively low MEA efficiency were primarily distributed in Hebei, Henan and Shanxi provinces. This could be attributed to the underdeveloped nature of these cities’ economies, with a lower proportion of high-tech and emerging industries. These cities faced challenges in terms of economic development and technological progress, with industrial technological development failing to keep pace with economic demands, relying more on the consumption of resources and energy. These cities should prioritize improving their undesirable output MEA efficiency by focusing on technological innovation and management levels. This includes the introduction of industrial equipment for pollution recovery, the improvement of production process techniques, the enhancement of industrial technological levels, gradual reductions in the proportion of heavy industries, the active development of emerging technology industries and the gradual establishment of a virtuous cycle between economic development and environmental protection.

5. Discussion

This section compares the research findings of this paper with existing studies based on the DEA model. Wang et al. [65] pointed out that, at the provincial level, there was a significant imbalance in the performance of atmospheric environmental efficiency, with Beijing and Tianjin outperforming Shandong, and Henan, Hebei and Shanxi consistently lagged behind. Chen et al. [66] stated that there has been uneven development of atmospheric environmental efficiency in the “2+26” cities, with significant differences in efficiency between cities. Wang et al. [21] found that the deterioration of EC was the main constraint factor affecting atmospheric environmental efficiency in the Beijing–Tianjin–Hebei region, and TC promoted the improvement of the atmospheric environmental efficiency. From 2013 to 2015, the atmospheric environmental efficiency was significantly low. Furthermore, it is noteworthy that Beijing and Tangshan exhibited superior atmospheric environmental efficiency compared to the other cities. These findings are consistent with the research conclusions of this paper. However, Wang et al. [21] also found that the atmospheric environmental efficiency in the “2+26” cities slightly increased in 2009-2010 and 2012–2013. However, this study showed a significant decrease during these two stages. Apart from the different scopes of the studies, the reason for the inconsistent conclusions may be due to the different indicators selected. This study added CO₂ emissions as an undesirable output index, which can more comprehensively and accurately measure the atmospheric environmental efficiency.

6. Conclusions

6.1. Results and Policy Implications

This study focused on the “2+26” cities and used the nonradial MEA model to measure the atmospheric environmental efficiency of these cities. It investigated the temporal and spatial evolution patterns and driving factors. The following are the main conclusions. First, from a temporal perspective, the atmospheric environmental efficiency values of the “2+26” cities showed a downward trend followed by an upward trend during the study period, with still 26.7% room for improvement. Second, from provincial and municipal perspectives, there was a significant difference in the atmospheric environmental efficiency among the “2+26” cities, indicating uneven development. The atmospheric environmental efficiency in the regions of Beijing–Tianjin and Shandong consistently remained higher than the average, whereas in the regions of Shanxi and Henan, it was consistently lower than the average, thereby dragging down the overall atmospheric environmental efficiency of the “2+26” cities. Third, in terms of driving factors, EC was the main limiting factor for improving the atmospheric environmental efficiency, whereas TC contributed to improving the atmospheric environmental efficiency. The 12 cities, including Jining, Xinxiang and Handan, with a GM index of less than 1, were the key cities for future atmospheric environmental efficiency improvement. Fourth, from the perspective of spatial distribution, the atmospheric environmental efficiency of the “2+26” cities showed significant positive spatial correlation, gradually showing a distribution pattern of H–H and L–L aggregation from a random distribution, demonstrating obvious local clustering characteristics. Fifth, in 2018, the spatial agglomeration characteristics of high-efficiency cities in terms of urban MEA efficiency and undesirable output MEA efficiency were essentially consistent. These cities were primarily concentrated along the northern region centered around Beijing and Tianjin, as well as the southeastern border, where significant improvements in efficiency values were exhibited.

To promote further improvement of the atmospheric environmental efficiency in the “2+26” cities, this study proposes the following policy recommendations based on the aforementioned conclusions: (1) Adopting targeted measures is crucial to eliminate the disparity in atmospheric environmental efficiency among regions with varying economic levels and degrees of air pollution. Considering that cities with low atmospheric environmental efficiency were predominantly located in Hebei, Shanxi and Henan provinces, this seriously hinders the overall atmospheric environmental efficiency in the “2+26” cities. This is imperative to strengthen regional collaborative governance, formulate common guidelines for industrial development and pollution control and establish comprehensive supervisory mechanisms. Technologically advanced cities like Beijing and Tianjin should actively provide technical assistance and foster alignment between cities. As provinces affected by pollution transfer, they should avoid burdening resource providers with the full cost of pollution transfer and instead establish comprehensive mechanisms for ecological compensation, environmental taxes and fees. (2) Capital investment should be strategically redirected toward high-value-added industries, and regional location advantages should be taken into account. By fully exploring local resource endowments, administrators can promote capital allocation that emphasizes both high quality and high coordination, thereby shifting away from a solely resource-driven investment path that relies on economies of scale. (3) Enterprises should strive to enhance energy efficiency and reduce per unit GDP energy consumption. The “2+26” cities should gradually phase out outdated production capacity and embrace the development of high-end assembly and modern manufacturing industries, thereby facilitating industrial ecological transformation. It is also necessary to extend product chains and increase the value added. (4) From the decomposition of the GM index, it can be seen that both the EC and TC have played a certain role in the “2+26” cities. Hence, the government should actively encourage businesses to adopt low-carbon, clean production and recycling technologies. Companies should embrace new processes and equipment, elevate management practices and promote responsible corporate social obligations.

6.2. Limitations and Future Research

This study has certain limitations, mainly arising from data collection and factors affecting the atmospheric environment. Scholars have generally focused on environmental indicators such as PM_2.5, PM₁₀ and nitrogen oxides when studying atmospheric pollution. Due to the lack of relevant data at the city level, this study did not adopt such indicators. Further research could be conducted in the future if relevant data become available. In addition, the atmosphere is fluid, and the selected indicators may not fully cover the influencing factors of atmospheric environmental efficiency. This study had multiple high-efficiency cities with a value of 1. If scholars want to further distinguish between these decision-making units, where the calculated efficiency value is equal to 1, they can improve the nonradial MEA model in future studies. Future research needs to further deepen and improve these aspects.