Do the urban polycentricity and dispersion affect multisectoral carbon dioxide emissions? A case study of 95 cities in southeast China based on nighttime light data

ABSTRACT Effectively exploring the impacts of urban spatial structures on carbon dioxide emissions is important for achieving low-carbon goals. However, most previous studies have examined the impact of urban spatial structure on total carbon emissions based only on polycentricity. Fine-grained studies on subsectoral carbon emissions and other dimensions of urban spatial structure are lacking. Therefore, our study comprehensively explores the impact of urban dispersion and polycentricity on total carbon emissions and carbon emissions of four subsectors (industry, power, civilian, and transportation) from 2012 to 2017 while considering the effects of city size. Results reveal that the nighttime light data is useful for measuring urban spatial structure, and a polycentric, decentralized urban spatial structure correlates with the reduced total carbon emissions and transportation carbon emissions. Meanwhile, a decentralized urban spatial structure gives rise to lower industrial carbon emissions and civilian carbon emissions, whereas a multicenter urban spatial structure contributes to minimizing carbon emissions from power systems. However, in small and medium-sized cities, urban spatial structure differently affects the total carbon and transportation carbon emissions.


Introduction
Since the industrial revolution, the increasing greenhouse effect has accelerated global warming (Wang et al. 2014).According to Zhang et al. (2021), from 1990 to 2014, global greenhouse gas emissions increased by about 15 billion tons, with carbon dioxide emissions considerably contributing to such an increase.China has inevitably experienced rapid industrialization and urbanization, which has caused high energy consumption and carbon dioxide emissions (Shi et al. 2018).Meanwhile, China has pledged to meet its peak carbon target by around 2030 (Yang et al. 2020).Therefore, the reduction of carbon emissions has become a primary task in China.
Similar to traditional technologies and strategies of energy efficiency enhancement, an appropriate urban spatial structure can effectively reduce carbon dioxide emissions (Lee and Lee 2014;Shi et al. 2023).Previous studies have focused on two aspects.First, the impact of polycentricity on carbon emissions is primarily explored.For example, Makido, Dhakal, and Yamagata (2012) found that polycentric cities in Japan tend to reduce carbon emissions.Liu et al. (2020) showed limited effects of polycentric structure on carbon emissions.Second, landscape indicators are often used to characterize urban spatial structure and determine its impact on carbon emissions.For instance, Shi et al. (2020) demonstrated that complex and decentralized urban structures are positively correlated with carbon emissions, while cohesive and rule-based urban structures exhibit the opposite trend.She et al. (2015) analyzed the geographic changes in 30 cities in the Yangtze River Delta from 1990-2015, revealing positive correlations between carbon emissions and factors, such as patch size, number of urban patches, maximum patch index, road network density, and average perimeter-area ratio.Bereitschaft and Debbage (2013) examined 86 metropolitan areas in the United States in 2000 as the study area, concluding that urban continuity, urban form complexity, residential density, centrality, street accessibility, and Smart Growth America sprawl index were correlated with carbon emissions.
However, these studies only focused on the influence of urban spatial structures on total carbon emissions and the potential relation between urban spatial structures and carbon emissions of different sectors is still unknown.
In previous studies, the urban spatial structure was always measured from a single dimension of polycentricity (Garreau 1991).However, the urban spatial structure should contain two dimensions: polycentricity and dispersion (Li 2020).Polycentricity reflects the concentration of the population and employment in a small number of locations (i.e.main centers and subcenters), whereas the decentralized dimension represents the dispersion of employment and population outside the main center of a city (Burger and Meijers 2010).According to these two dimensions, the transformation of an urban spatial structure from a monocentric structure can be grouped into four potential patterns (Figure 1).For example, if the transformation has diverse destinations that are scattered and not well developed, the urban spatial structure has high dispersion and low polycentricity (Figure 1a). Figure 1b represents an urban spatial structure with high polycentricity and dispersion, whereas Figure 1c  specific destination that has developed well, the urban spatial structure becomes polycentric with low dispersion (Figure 1d).Therefore, using the two dimensions simultaneously can help us better quantitatively understand the urban spatial structures (Li and Liu 2018;Zhu, Tu, and Li 2022).
Traditionally, two types of method are widely employed to quantify urban dispersion.The first is directly using statistical data to refer to the degree to which the population is clustered or concentrated in urban centers (Lee 2007), such as the number of counties (Hill 1974), room and dwelling density (Bramley and Power 2009;Glaeser and Kahn 2010), and job opportunity (Glaeser et al. 2001).Generally, these studies treated all urban centers equally, which is not consistent with reality as urban centers could have their own sizes of influence on surrounding areas.Therefore, the second type is calculating indices with the weights of each unit, including the Herfindahl-Hirschman Index (Hirschman 1964), the space Gini coefficient (Audretsch and Feldman 1996), the Ellison-Glaeser Index (Ellison and Glaeser 1994), and the DO Index (Duranton and Overman 2005).However, these indices could exaggerate the contribution of larger urban centers as they are calculated using the percentage sum of squares of contributions of each urban center, such as economy, population, or employment.Thus, the metropolitan fragmentation index (MFI) was proposed based on the percentage sum of the square root of a contribution instead of the percentage sum of the square of a contribution (Mitchell-Weaver, Miller, and Deal 2000).Because MFI has a betterbalanced value range, we decided to use this index to quantify the urban dispersion herein.
The methods for measuring polycentricity focused on two main perspectives (Liu, Derudder, and Wang 2018).The first is the importance of individual center, which is traditionally measured from their population, employment, and gross domestic product (Sha et al. 2020) as well as the interactions among centers (Camagni and Capello 2004).Furthermore, the measurement of the balance of importance among the centers can be grouped into three types: the rank size approach (Burger and Meijers 2010), the comparison between the observed actual distribution and some ideal-typical benchmark scenarios (Hanssens et al. 2014), and a social network-based approach (Green 2007).However, previous methods only focused on one of the two perspectives.Thus, we adopted a novel polycentric indicator based on the number of centers and the above two aspects to comprehensively understand urban polycentricity (Amindarbari and Sevtsuk 2012).
In addition, although the MFI from Mitchell-Weaver, Miller, and Deal (2000) and the polycentric metric from Amindarbari and Sevtsuk (2012) performed well in representing urban dispersion and polycentricity, they highly relied on statistical data.These statistics often have a time lag and are incomplete in some backward areas, seriously limiting their usefulness.Therefore, an alternative data source is required to enhance the measurement of urban spatial structures.
Nighttime light (NTL) remote sensing uses sensors to record artificial light emitted by cities and towns on the Earth's surface (Elvidge et al. 1997).NTL datasets have been utilized to estimate gross the domestic product (Chen and Nordhaus 2011;Shi et al. 2014), carbon dioxide emissions (Shi et al. 2016), electricity consumption (Elvidge et al. 1997), postwar economic recovery assessment (Li et al. 2018;Li et al. 2018), housing vacancy rates (Chen et al. 2015), poverty (Yu et al. 2015), and economic development quality (Chen et al. 2023;Yang et al. 2019).NTL remote sensing data not only has wide coverage, high efficiency and objectivity but also can characterize the attributes of human economic activities; therefore, it can be applied to the extraction and research of urban spatial information and a lot of research results have been obtained from the analysis of urban spatial structures (Chen et al. 2017).
Herein, we aimed to measure the polycentricity and dispersion of urban spatial structures from NTL data and use a pooled regression model to reveal how urban spatial structures affect multisectoral carbon emissions within 96 cities in southeast China from 2012 to 2017.The objectives of this study are as follows: 1) to accurately quantify urban spatial structures from the two dimensions of polycentricity and aggregation using NTL data and 2) to study what urban spatial structures can reduce carbon emissions.

Study areas
Herein, 96 cities in the southeastern region of China were selected to analyze the correlation between urban spatial structures and carbon emissions based on the following two main reasons (Figure 2).First, coastal towns develop considerably faster than inland cities because of their natural advantage of being open to the outside world.This rapid development causes rapid economic growth and thus a sharp increase in carbon dioxide emissions.Therefore, the task of reducing carbon dioxide emissions is more challenging in coastal areas than in other regions.Second, China's southeast region includes cities with diverse functions, such as resource-intensive cities (e.g.Shanghai, Shenzhen, and Guangzhou) and tourism service cities (e.g.Jingdezhen, Jiujiang, and Huangshan).Notably, because the carbon emission data in Hainan Province were incomplete, we excluded the Hainan Province from this study.
Urban extent data from harmonized NTLs (Zhao et al. 2022) were used to integrate with the NPP-VIIRS-like NTL data for the measurement of urban dispersion.This dataset fits well with the traditional global urban boundaries derived from impervious surface area.This dataset was obtained from https://doi.org/10.6084/m9.figshare.16602224.
The LandScan high-resolution global population dataset was adopted to identify urban centers for further polycentricity measurements.The LandScan population dataset captures the average population distribution over 24 h, with a grid resolution of approximately 1 km (30 ′′ × 30 ′′ ).The dataset can be freely accessed from the US Department of Energy's Oak Ridge National Laboratory (https://landscan.ornl.gov/).
Carbon emission data were derived from the Multi-resolution Emission Inventory for China (MEIC) model (Li et al. 2017;Zheng et al. 2018).The MEIC model contains data on the total carbon emissions and carbon emissions in four subsectors: industry, power, civilian, and transportation.Meanwhile, the dataset covers the whole of mainland China and has a spatial resolution of 0.25°× 0.25°.Because the dataset was available from 2008 to 2017 and the growth rate of China's emissions first decelerated in 2012, we set the study period to 2012-2017 (Zheng et al. 2019) (Figure 3).

Control variables
Carbon emissions may be influenced by factors other than urban spatial structure; therefore, four control variables were set up herein to better explore the impact of urban spatial structure on carbon emissions.Based on previous studies (Shi et al. 2019;Shi et al. 2020), this study included several control variables such as gross domestic product per capita (GDP), normalized difference vegetation index (NDVI), number of patents (TEC), and total nighttime light intensity (TNL) in the regression model.( 1) GDP represents the level of economic development of the city, and rapid economic growth may lead to a sharp increase in carbon emissions.( 2 of the Chinese Academy of Sciences (https://www.resdc.cn/).Annual NDVI data were used in this study, which mainly used the maximum value in the monthly NDVI from January to December of each year.The data covered the period from 1998 to 2019.The spatial resolution of this data was 1 km, which met the accuracy requirement of this study.

Methodology
As shown in Figure 4, we quantified the urban polycentricity and dispersion of urban spatial structures using NPP-VIIRS-like NTL data and investigated their impacts on subsectoral carbon emissions using a pooled regression model.This comprised three components.(1) Urban dispersion of each city was calculated using the MFI based on the total night light intensity within the urban extents.(2) The urban polycentricity was calculated according to the results of the urban center identification from LandScan gridded population data and the exploratory spatial data analysis (ESDA) method at the city level.(3) The pooled model was used to explore the effects of urban polycentricity and dispersion on subsectoral carbon emissions.

Measurement of urban dispersion
Urban dispersion was calculated using the MFI method from Mitchell-Weaver, Miller, and Deal (2000).The detailed expressions are as follows: where x i (i = 1, 2, … , n, where n is the number of urban extents in the urban area) represents the luminance value in the urban extent, y i represents the proportion of the luminance value in the ith urban extent block to the total urban extent blocks, and I is the fragmentation index, where a larger value of I means that the city has a higher level of urban dispersion.The range of I varies in different cities as its maximum value is controlled by the number of urban extents of each city, which makes the degree of dispersion based on I not comparable between any two cities.Therefore, the Min-Max normalization method was applied to make the degree of dispersion comparable across cities (Panda and Jana 2015), using equation 3.
where I nor is the normalized value of I.I Max and I Min are the maximum and minimum values of I, respectively.

Measurement of urban polycentricity
The measurement of urban polycentricity developed by Amindarbari and Sevtsuk (2012) requires the identification of urban centers.Herein, we used ESDA to detect the urban centers from Land-Scan gridded population data (Li and Liu 2018).First, we calculated the local Moran's I index based on the LandScan population dataset.We considered the population grid identified as a high-high cluster (HH) as the urban center.Then, the extracted HH grids were aggregated under the criterion of rook contiguity to obtain potential urban centers.Finally, potential urban centers with an area exceeding 2 km 2 and a population higher than 50,000 were selected as the final urban centers (Li, Xiong, and Wang 2019;Sha et al. 2020).
The degree of intracity polycentricity (PC) from Amindarbari and Sevtsuk (2012) contained three dimensions: the number of urban centers (N), the degree of agglomeration (R), and the degree of balance of urban centers (HI), as shown in equation ( 4).The three dimensions are specified as follows: (1) cities with more urban centers have a more substantial level of polycentricity than cities with fewer urban centers; (2) the more urban elements are clustered in urban centers, the higher the level of polycentricity, where the number of urban elements is measured by the intensity of luminosity in this study; and (3) the more balanced the power among urban centers within a city, the higher the level of urban polycentricity, where the degree of center balance is calculated based on the entropy index by Limtanakool, Schwanen, and Dijst (2009).The calculation of the degree of polycentricity can be expressed as where p i represents the total luminous value in the ith urban center; N represents the number of urban centers.As this formula is valid only for polycentric cities, we arbitrarily set the PC value for monocentric cities as 0.

Developing the pooled regression model
The pooled regression model was adopted to analyze the effects of urban spatial structures on total and subsectoral carbon emissions because of its two advantages.First, it has a particular time dimension, so it can solve the problem of insufficient sample size to a certain extent (Hsiao 2014).Second, it can reduce the mutual influence between independent variables and weaken the influence of multicollinearity (Fang, Wang, and Li 2015).The specific formula is as follows: In addition to two urban spatial structure indicators, four indicators (GDP, TEC, NDVI, and NTL) were selected as control variables.CE it represents the carbon dioxide emissions of the ith city in the tth year, α represents the intercepts of all individuals, β1-β6 represents regression parameters for each independent variable, GDP represents the GDP per capita, TEC represents the level of technological innovation, and ε is the random error.

Trends in urban dispersion and urban polycentricity
In the southeast region, the urban spatial structure is moving toward dispersion and polycentricity.Over the years, the median line for urban dispersion and urban polycentricity has been on the rise (Figure 5).This implies that cities in the southeast region, in general, have experienced a dispersion of population and urban factors from the urban centers, along with improved equitable distribution of these factors between the main centers and subcenters.Our empirical results are similar to those of previous studies (Angel and Blei 2016).As depicted in Figure 6 (a and b), the degree of urban dispersion in southeast China increased from 2012 to 2017 in most cities, which means more and more cities in this region moved toward a decentralized spatial structure, especially in the east of Guangxi Province and west of Fujian Province.These changes corresponded to the urban master plan introduced in the area.For example, Chizhou, a city in Anhui Province, experienced a rapid increase in urban dispersion as it has a particular focus on renowned tourist destinations in the 'Chizhou City Master Plan .' Chizhou has a natural landscape and human resources, especially in suburban areas, such as Jiuhua Mountain and Guliujiang.The development of these regions made them tourist centers and substantially increased urban dispersion.Furthermore, efficient communication technology facilitates interactions between central business districts (CBDs) and suburbs located farther from CBDs, which also accelerates urban dispersion (Arribas-Bel and Sanz-Gracia 2014; Tao, Sheng, and Wen 2023).However, a minority of cities are moving toward an aggregated urban spatial structure.These cities exhibit lower economic development levels compared to their southeast counterparts, which is consistent with the results from Wang (2022).These cities usually struggle with limited land resources and insufficient infrastructure.In these cities, urban planners and developers tend to build more compact urban spatial structures to maximize land use efficiency (Ding 2013;Yan et al. 2021), and citizens prefer to live in urban centers for high accessibility to public services (Ojo 2020).
Meanwhile, we found that polycentricity had an increasing trend similar to that of urban dispersion Figure 6 (c and d).First, more cities had a polycentricity value higher than 0, which meant the number of polycentric cities increased significantly from 2012 to 2017.Second, the urban centers in each city experienced a balanced development, according to the increment of the polycentricity value in most cities.
The formation of urban polycentricity is influenced by physical geography and human behavior (Krugman 1993).First, the topography of southeast China is complex and dominated by hills and mountains, making it impossible to concentrate the population in a single center and thus increasing the number of urban centers in the cities in this topography (e.g.Heyuan and Jieyang).Second, most cities in southeast China have high economic levels and population concentrations, making them experience agglomeration diseconomies.Therefore, these cities (e.g.Xiamen and Suzhou) tend to be polycentric to alleviate the pressure on the main center and weaken the agglomeration diseconomies (Sun and Lv 2020).

Effects of polycentricity and urban dispersion on total and subsectoral carbon dioxide emissions
We entered the indicators of 96 cities in southeast China between 2012 and 2017 as samples in a pooled regression model.First, we ensured that there was no colinearity between any two variables using the variance inflation factor (VIF) method and tolerance values (Table 1).Then, all variables were inputted into the pooled regression model, and the results showed that the total carbon emissions (OCEs) had the highest R 2 value of 0.861 when fitting with urban spatial structures.For the carbon emissions of four subsectors (industry, power, civilian, and transportation), industrial carbon emissions (ICEs) and transportation carbon emissions (TCEs) had more important relationships with urban spatial structures (with R 2 values of 0.800 and 0.768, respectively) than those of power carbon emissions (PCEs) and civilian carbon emissions (CCEs), whose R 2 values were 0.418 and 0.467, respectively.
As shown in Table 1, the degrees of both dispersion and polycentricity had an important negative effect on OCEs and TCEs.The degree of dispersion negatively impacted ICEs and CCEs, whereas the polycentricity level could considerably reduce PCEs.Specifically, a 1% increase in urban dispersion reduced OCEs by 0.085%, ICEs by 0.134%, CCEs by 0.126%, and TCEs by 0.049%.Simultaneously, a 1% increase in polycentricity can reduce OCEs by 0.064%, PCEs by 0.071%, and TCEs by 0.099%.

Impacts of city size on the correlation between urban spatial structure and carbon emissions
Owing to the differences in the development and size of cities in southeast China, the following question arises: do small and medium-sized cities require distinct spatial planning strategies to reduce carbon emissions compared to the larger cities? To answer this, a pooled regression model was adopted again excluding 11 large cities (such as Shanghai, Shenzhen, Xiamen, Ningbo, and six provincial capitals).
In contrast to the findings in Table 1, the urban spatial structure of the small and medium-sized cities affects the OCEs and TCEs differently.First, the polycentricity of small and medium-sized cities no longer exhibited a significant reduction in OCEs.Second, the negative effects of the dispersion and polycentricity on the TCEs within these cities were diminished.

Validation and robustness analysis
Verifying the capacity of NTL data in measuring urban spatial structure is imperative.We used traditional measurements (Mitchell-Weaver, Miller, and Deal 2000) as the reference and found that the urban dispersion and polycentricity degrees, derived from NTL data in this study, exhibited high accuracy.The R 2 value of each year exceeded 0.8 and reached up to 0.976, as shown in Figure 7.This validation affirms the viability of the NTL data as an alternative proxy for measuring urban spatial structure in terms of urban dispersion and polycentricity (Table 2).
In addition, considering potential time lags in the influence of urban spatial structure on carbon emissions and to mitigate potential endogeneity concerns (Cooper et al. 2020), urban dispersion and urban polycentricity degrees were lagged by one period.The results presented in Table 3 reveal that the effects of urban dispersion and polycentricity on total and subsectoral carbon emissions align with those in Table 1, indicating the robustness of the results of this study.

What type of urban spatial structure is conducive to reducing OCEs?
According to the results shown in Table 1, the higher the degree of dispersion and polycentricity, the lower the OCE let off.Specifically, a polycentric urban structure can accelerate the population migration from the main urban center to subcenters, which can considerably reduce the OCEs in the main center but slightly increase the carbon emissions in the subcenters (Sun, Han, and Li 2020).Because of decentralized urban spatial structures, more urban ecological corridors are created to speed up the diffusion of carbon emissions (Guo et al. 2007) and cause positive spatial spillovers on carbon emissions of its surrounding regions (Wu et al. 2023).
We also found that the impact of urban dispersion on total urban carbon emissions is much stronger than that of urban polycentricity, according to the coefficients of urban dispersion and polycentricity in model (1) (Table 1).A higher urban polycentricity means urban centers have a balanced development, which implies that each urban center has similar spillover effects on carbon emissions to surrounding urban centers.In other words, spillover effects are repulsive; subsequently, the reduction of carbon emissions of the polycentric structures seems weakened.

What type of urban spatial structure is conducive to reducing subsectoral carbon emissions?
Herein, we discussed the effects of urban spatial structures on four subsectoral carbon emissions.For instance, we revealed that decentralized urban spatial structures can considerably reduce ICEs and CCEs.There are two reasons for this finding.First, dispersed factories and settlements lead to dispersed sources of carbon emissions, facilitating the diffusion of carbon dioxide to other urban areas and reducing the local carbon dioxide concentration.Second, decentralized urban spatial structures are good for forming urban ecological corridors, which can help absorb  considerable amounts of carbon dioxide (Guo et al. 2007) and introduce fresh air into urban interiors to reduce the concentration of carbon dioxide (Fan et al. 2022).
The results of model ( 3) in Table 1 show that urban spatial structures cannot distinctly affect PCEs as its R 2 of 0.418 is the lowest among those of all the models.However, we found that polycentric cities can have lower carbon emissions from power systems.Polycentric urban structure involves the emergence of new urban centers within the existing urban areas.The influx of talent often accompanies technological advancements, accelerating transformations in energy structures, such as the adoption of natural gas and solar energy (Chen et al. 2022;Junfeng et al. 2021).As usual, these new energy structures let off less carbon emissions than those of the traditional coal energy in old urban areas (Vazquez Hernandez, Gonzalez, and Fernandez-Blanco 2019).For example, the Lishui Urban Master Plan (2013-2030) clearly proposes the setting of Bihu and Dagangtou towns as subcenters, where the gasification rate must reach 90% in their gas engineering plan.
A decentralized and polycentric urban structure is beneficial for reducing TCEs.There are two main reasons for this.On the one hand, a decentralized urban structure can considerably alleviate traffic jams and consequently reduce exhaust emissions (Jung, Kang, and Kim 2022).Moreover, a polycentric urban structure implies that these subcenters are comparatively advanced, boasting well-established infrastructures.Consequently, such a spatial layout can lower the frequency of intercenter mobility.Moreover, because of the shorter commuting distances, residents tend to switch from private vehicles to public transportation (Ismiyati and Hermawan 2018).Therefore, polycentric urban spatial structures can considerably reduce TCEs.An empirical study conducted by Yang et al. (2012) focused on the 50 largest metropolitan areas in the United States and found that the spatial distribution of high-density areas, similar to our urban centers, played a critical role in reducing commuting times while reducing the carbon emissions from transportation.

Optimal urban spatial structures for subsectoral carbon emissions reduction across different city sizes
First, the limited capacity of polycentricity in reducing OCE in small and medium-sized cities can be attributed to their insufficient population density.Polycentric urban spatial structures are effective in reducing carbon emissions within a certain population density range (Han, Sun, and Zhang 2020).Large cities generally exhibit high population density, particularly their main centers, which gathers a large number of people, ultimately causing polycentric urban structures to alleviate the concentration of pollution in the main center (Wu et al. 2022).In contrast, small and mediumsized cities feature low population densities, and their main centers lack strong agglomeration effects, diminishing the effects of polycentric urban structures on carbon emission reduction (Shi et al. 2023).Second, the reasons for the attenuation of the negative effects of decentralization and polycentricity on TCE in small and medium-sized cities differ.Traditionally, traffic congestion is less prevalent in small and medium-sized cities compared to large cities, limiting the potential for improved traffic conditions through urban decentralization.Therefore, urban decentralization exhibits a limited effect on TCEs.Moreover, the urban centers in small and medium-sized cities usually exhibit imbalances between job opportunities and housing, resulting in most routine commuting being intercenter (Burgalassi and Luzzati 2015), making it challenging for polycentricity to directly reduce TCEs in such regions.

Study limitations and future research directions
Although we have demonstrated that urban spatial structures can have important impacts on total urban carbon emissions and four subsectoral carbon emissions, there are still some limitations in this study.First, because the current carbon emission data in China did not cover recent years, we could not empirically illustrate how urban spatial structures currently affect carbon emissions.However, once we get sufficient carbon emission data, we can quickly make an empirical analysis as we have proposed an efficient framework.Second, this study only analyzed the impact of urban spatial structure on carbon emissions within a six-year period, without accounting for inter-annual variations, a topic that warrants future investigation.In addition, the effects of urban spatial structures on OCEs are diverse at different spatial scales (Shi et al. 2020) but the effects on subsectoral carbon emissions at county or region scales remain unknown.

Conclusion
This study quantified urban dispersion and urban polycentricity using the NTL data.Subsequently, a pooled regression analysis was performed on the samples from 2012 to 2017 in 96 cities in southeastern China to analyze how urban spatial structures impact carbon emissions, especially subsectoral carbon emissions, while considering the effects of city scale.From 2012 to 2017, we concluded that decentralized and polycentric structures benefit the reduction of OCEs and TCEs.A decentralized urban structure is conducive to reducing ICEs and CCEs, whereas a polycentric urban structure gives rise to lower PCEs.Moreover, we found that polycentricity in small and medium-sized cities exhibited no correlation with OCEs.Additionally, neither polycentricity nor decentralization in these cities influenced TCEs.The relations between subsectoral carbon emissions and urban spatial structures can provide valuable insights for each city.First, for industrial cities, industries should be dispersed and the size of industrial parks should be limited.This strategy can better reduce ICEs.Second, for cities with good transportation, the infrastructure within each urban center should be well-designed and constructed.This can help reduce intercenter commuting and consequently cut TCEs.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Figure 2 .
Figure 2. Cartographic representation of southeastern China.Blue regions represent the study area of this work.
Figure 3. Spatial distribution of nighttime light intensity (a-f) in the southeastern region from 2012 to 2017.

Figure 4 .
Figure 4. Framework of the methodology.GDP: gross domestic product per capita, NDVI: normalized difference vegetation index; TEC: number of patents, TNL: total nighttime light intensity, ICE: industrial carbon emissions, PCE: power carbon emissions, CCE: civilian carbon emissions, and TCE: transportation carbon emissions.

Figure 5 .
Figure 5. Box plots of urban dispersion and urban polycentricity from 2012 to 2017.

Figure 7 .
Figure 7. Scatter plots of urban (a-f) dispersion and (g-l) polycentricity based on NTL data and traditional built-up area and urban center area from 2012 to 2017.

Table 1 .
Relations between urban spatial structures and carbon emissions.

Table 2 .
Relations between urban spatial structures and carbon emissions in small and medium-sized cities.

Table 3 .
Relations between lag-phase urban spatial structure and carbon emissions.