*4.1. Results for Growing and Shrinking Cities*

Based on the value of UDD and the categorization rules, 145 cities shrank during 2009–2018, accounting for 51.79%, mostly distributed in northeastern, central, and western China. A total of 135 cities grew, accounting for 48.21%, mostly distributed in the eastern coastal areas (Figure 2). Among the growing cities, 126 had slight growth, and 9 had rapid growth, accounting for 45% and 3.21%, respectively. Among the shrinking cities, 33 were rapidly shrinking, and 112 were slightly shrinking, accounting for 11.79% and 40%, respectively. Cities with slight growth account for the highest proportion, while those with rapid growth account for the lowest proportion. Most cities in China are SGCs or SSCs, while only a few are RGCs or RSCs.

**Figure 2.** Spatial distribution of shrinking and growing cities in China, 2009–2018.

4.1.1. Characteristics of Growing Cities

The UDD of RGCs exceeds 0.05; these include Nanjing, Hangzhou, Zhengzhou, Wuhan, Xi'an, Chongqing, Chengdu, Kunming, and Qingdao. Most of these cities are

provincial capitals with convenient transportation, good medical conditions, and high levels of education. Their economic development is at the highest level, with per capita GDPs of 50,000–110,000 yuan, and their economies are rapidly expanding. All of these RGCs have populations close to 10 million and are growing rapidly. Moreover, these cities have completed their industrial transformation and upgrading. In 2018, tertiary industry accounted for more than 55% of all industry in these cities (Figure 3a).

**Figure 3.** Industrial structure of growing and shrinking cities.

The UDD of SGCs ranges from 0 to 0.05; examples include Jinan, Suzhou, Ningbo, Jiaxing, Xiamen, and Fuzhou, mostly distributed in Shandong, Jiangsu, Zhejiang, and Fujian Provinces. Most of these cities are economically developed provincial capitals, with per capita GDPs of 40,000–80,000 yuan. Their economic development levels are lower than those of RGCs, but their economic growth is steady. Furthermore, the SGCs have established a leading position in tertiary industry by optimizing their industrial structures, with the proportion of tertiary industry accounting for more than 45% of the total (Figure 3b). This means that the service industry is constantly improving in these cities. Finally, their total populations show a trend of steady growth.

#### 4.1.2. Characteristics of Shrinking Cities

The UDD of RSCs is below −0.02. Examples include Anshan, Fushun, Zhangjiakou, Tangshan, Baotou, and Daqing, which are mostly distributed in Liaoning, Hebei, and Inner Mongolia. Most of these cities are resource based, mainly relying on mineral resources during their early stage of development, with secondary industry accounting for more than 50% of the total (Figure 3c). Adjustments to the industrial structure started late in these cities. Their economic development is below the middle level, with per capita GDPs mostly between 30,000 and 60,000 yuan. RSCs are experiencing economic decline and population loss.

The UDD of SSCs is between −0.02 and 0. This includes Yichun, Liaoyuan, Datong, Kaifeng, Zigong, and other cities, mostly distributed in Heilongjiang, Jilin, Shanxi, Henan, Gansu, and Sichuan Provinces. Compared with RSCs, SSCs are mostly resource-based cities, taking secondary industry as the leading industry. After industrial restructuring, however, secondary industry accounted for less than 50%, while the service industry accounted for nearly 45% (Figure 3d). In addition, these cities are far away from economically developed cities, have poor geographical location conditions and transportation accessibility, and have relatively low levels of economic development. Their per capita GDP is 20,000–50,000 yuan. Their economies are growing but their populations are diminishing.

#### *4.2. Emission Characteristics of Growing and Shrinking Cities*

The carbon emissions of RGCs show a trend of fluctuating growth, RSCs present an inverted U-shaped trend, and SGCs and SSCs rise first and then develop steadily (Figure 4). The carbon-emission trends of the four groups are different, and their driving factors may be different as well. Here, we analyze the possible effects of economies, industrial structure, and technology on the carbon-emission characteristics of the four city groups.

**Figure 4.** CO2 emission characteristics of growing and shrinking cities.

#### 4.2.1. Rapidly Growing Cities (RGCs)

The carbon emissions of RGCs are the highest among the four groups, with a trend of fluctuating growth (Figure 4a). From 2009 to 2012, carbon emissions grew at an annual rate of 3.3%. They fluctuated from 2013 to 2016 and continued to grow at an average annual rate of 2.4% in 2017 and 2018. In terms of economic growth, the GDP of RGCs shows an increase of 298.9%, with primary, secondary, and tertiary industries increasing by 202.9%, 247.6%, and 361.7%, respectively. This indicates rapid economic expansion, which contributes to increases in emissions. Between 2013 and 2016, however, the growth rate of GDP slowed down (annual GDP growth during this period was around 10%, lower than in other periods) (Figure 5a), which slowed the growth of carbon emissions over the period. In terms of

industrial structure, the proportion of primary and secondary industry continues to decline, while the proportion of tertiary industry continues to rise. The service industry in RGCs is relatively mature and has occupied a dominant position for a long time, making increasing contributions to the economy. Therefore, the development of services may contribute to the increase in carbon emissions. In terms of energy structure, the proportion of coal in RGCs generally shows a downward trend, especially during 2012–2015. The proportion of coal shows a rapid decline (Figure 6), which may help to reduce carbon emissions. In terms of technology, energy intensity drops from 0.74 to less than 0.26 (tec/10<sup>4</sup> yuan), an annual decrease of about 11%, showing a rapid downward trend (Figure 7). This means that technology was greatly improved during the study period. We can see that the energy structure and technology of RGCs have been optimized and improved, which helps reduce carbon emissions. However, the carbon-increasing effect of rapid economic expansion and expanded tertiary industry is greater than the carbon-reducing effect of energy structure optimization and technological progress. Therefore, overall urban carbon emissions show an increasing trend.

**Figure 5.** GDP and GDP growth rate of the four city groups.

#### 4.2.2. Slightly Growing Cities (SGCs)

The carbon emissions of SGCs are low among the four city groups, showing a trend of first rising and then changing only slightly (Figure 4b). Specifically, from 2009 to 2012, the carbon emissions of SGCs increased from 24.83 million tons to 30.18 million tons, with an annual growth rate of 5%. Between 2013 and 2018, emissions were around 30 million tons, a small change. In economic terms, the economies of SGCs show an increase of 258.2%. However, GDP growth generally shows a downward trend after 2012 (Figure 5b). Although economic expansion may increase carbon emissions, a decline in economic growth might also slow down the growth of emissions, accounting for the slowed growth of emissions after 2012. In terms of industrial structure, the proportion of primary and secondary industry continues to decline, while that of tertiary industry keeps rising. Especially after 2012, the proportion of secondary industry dropped rapidly (annual decline of 2%). This

means that the cities reduced their dependence on energy-intensive industries, which helps reduce carbon emissions. In terms of energy structure, the proportion of coal in SGCs is 62%–66%, and there is a small decline (Figure 6). The urban energy structure has been optimized, and the decline in the proportion of coal is conducive to reducing carbon emissions. In terms of technology, the energy intensity of SGCs decreases year by year, from 0.7 in 2009 to 0.34 (tec/104 yuan) in 2018, an annual decrease of 7.6% (Figure 7). This could be another reason for the slowed growth of carbon emissions. In general, economic expansion in the SGCs increased carbon emissions. After 2012, however, with slowed economic growth, accelerated changes in industrial structure, energy structure optimization, and technological progress, the growth of carbon emissions was restrained, stabilizing changes in emissions.

**Figure 6.** Coal share of growing and shrinking cities.

**Figure 7.** Energy intensity of growing and shrinking cities.

#### 4.2.3. Rapidly Shrinking Cities (RSCs)

Among the four city groups, RSCs' carbon emissions are high, presenting an inverted U-shaped trend; 2013 is a turning-point year (Figure 4c). Regarding the economy, despite RSC economies showing growth, their GDP growth rate first rises and then declines rapidly, indicating that their economies declined after a period of growth (Figure 5c). This is consistent with the change characteristics of carbon emissions. We can infer, therefore, that

economic changes triggered the changes in emissions. In terms of industrial structure, the proportion of secondary industry increases first and then decreases, while tertiary industry shows the opposite trend. This means that the cities gradually reduced their dependence on secondary industry and developed tertiary industry, which could curb carbon emission growth. In terms of energy structure, the overall change in coal consumption is relatively small (Figure 6), which differs greatly from the change in carbon emissions. Therefore, energy structure might not have an effect on carbon emissions. Meanwhile, the energy intensity of RSCs decreases from 0.88 to 0.42 (tec/104 yuan) (Figure 7); technology was improved, which is conducive to reducing carbon emissions. This analysis reveals that economic and industrial structure (secondary industry) and carbon emissions show the same change characteristics; thus, they may be the main reasons for changes in emissions. Although technological progress can restrain carbon emissions, emissions still show an increasing trend in the early stage. Therefore, technological progress had a weak effect on reducing emissions in the early part of the study period but might account for the reduced emissions in the latter part.

#### 4.2.4. Slightly Shrinking Cities (SSCs)

SSCs have the lowest total carbon emissions among the four city groups. Their carbonemission trend is similar to that of SGCs, showing a trend of first rising and then changing only a little (Figure 4d). Specifically, carbon emissions rose from 15.14 million tons to 18.09 million tons from 2009 to 2012 and then remained around 18 million tons. In terms of economic development, the GDP of SSCs expanded overall, but the GDP growth rate declined year by year, especially after 2012 (Figure 5d). In terms of industrial structure, the change characteristics for SSCs are similar to those of RSCs, with the proportion of secondary industry first rising and then falling and that of tertiary industry first falling and then rising. Specifically, the proportion of secondary industry rose from 49.3% in 2009 to 51.16% in 2013, and then dropped to 42.18% in 2018. The proportion of tertiary industry fell from 34.5% to 32.8% and then rose to 44.6%. The expansion of economic scale increased carbon emissions, but the slowdown of economic growth and the adjustment of industrial structure after 2012 helped to curb emissions, making them relatively stable. In terms of energy structure, the proportion of coal in SSCs increased from 62.9% in 2009 to 65.8% in 2018 (Figure 6); this means that the energy structure did not improve. Therefore, the energy structure might not be the reason for reduced carbon-emission growth. The energy intensity of SSCs declined annually by 5.4% (Figure 7). The development of low-carbon technologies might be the reason for the reduced growth of emissions.

#### *4.3. Regression Results*

According to the F-test and Hausman test results, for SGCs, RSCs and SSCs, the fixedeffects model should be used for regression, while the mixed-effects model should be used for RGCs, as shown in Table 2. However, by comparing the mixed-effects regression results with the fixed-effects regression results, the significance of the coefficients of influencing factors has not changed a lot. Therefore, in order to be comparable with the regression results for the other three groups of cities, a fixed-effects model was used for RGCs.

**Table 2.** F-test and Hausman test for four city groups.


Table 3 shows the estimation results of the fixed-effects model for four city groups. It can be found that the regression results of the models in SGCs, RSCs and SSCs perform better, showing that the R-squared values are more significant. The regression result for RGCs is relatively insignificant, mainly due to the small samples, which makes its explanatory power limited. However, the regression result for RGCs is similar to SGCs, mainly because they belong to growing cities, so the results are reasonable to a certain extent.


**Table 3.** Regression results.

Note: \*\*\* *p* < 0.01, \* *p* < 0.1.

Based on the regression results, population and economic growth have a positive effect on carbon emissions, while technological progress has a negative effect on carbon emissions. The proportion of tertiary industry has a positive impact on growing cities, but has a negative impact on shrinking cities. For RGCs, the proportion of tertiary industry has the largest and most significant impact on emissions, which means that the proportion of tertiary industry is the primary influencing factor for carbon emissions in RGCs. An increase of 1% in the proportion of tertiary industry increases carbon emissions by 1.337%. For SGCs and RSCs, population is the most important factor affecting carbon emissions, showing that carbon emissions increase by 0.943% and 0.722% for every 1% increase in population. The technological progress has the greatest negative impact on SGCs, RSCs and SSCs. For every 1% increase in technological level, carbon emissions decrease by 0.709%, 0.518% and 0.547%. For SSCs, the economic growth has the largest positive impact, as every 1% increase in the economy increases carbon emissions by 0.74%.

In order to further verify the robustness of the regression results, we run a Stochastic Frontier Analysis (SFA) model on the extended STIRPAT, and the SFA results are in high agreement with the regression results, indicating our result is robust. Moreover, referring to Wang et al. [21], we compare the fitting data of CO2 emission (lnCO2) calculated by the regressions coefficients in Table 3 with the actual CO2 emissions (lnCO2). The margin of error is within 10%, and most are less than 5%, as shown in Table 4, indicating that the regression model is available.


