High-resolution mapping of carbon dioxide emissions in Guizhou Province and its scale effects

Abstract

Accurate spatial distribution of carbon dioxide (CO₂) emissions is essential information needed to peaking emissions and achieving carbon neutral in China. The aim of this study was to map CO₂ emissions with high spatial resolution at provincial scale and then explore the scale effect on mapping results. As an example, the spatiotemporal pattern and factors influencing CO₂ emissions were examined in Guizhou Province in Western China. With the proposed method, a reasonable spatial distribution of CO₂ emissions with high spatial resolution was obtained, which had relatively accurate information on spatial details. The optimal resolution of CO₂ emissions at the provincial scale under high spatial resolution was approximately 90 m and 1260 m. More detailed grid data can better reflect the spatial variability of CO₂ emissions. Emissions of CO₂ were spatially heterogeneous in Guizhou, with high emissions in centers of big cities that gradually spread and decreased from city centers. From 2009 to 2019, the spatial distribution of CO₂ emissions developed from agglomeration to dispersion. Areas of high carbon emissions decreased, those of medium carbon emissions increased, and many areas changed from no emissions to carbon emissions. Industrial land had the highest emissions, followed by commercial and transportation lands. Over 10 years, changes occurred in the relation between interregional economic level of Guizhou and CO₂ emissions, with the relation changing from linear into an inverted U-shaped relation. The effect of industrial structure on CO₂ emissions decreased, and the linear increase between CO₂ emissions and the urban scale became more evident. The results of this study will contribute to accurate monitoring and management of carbon emissions in Guizhou, as well as provide support to formulate policies related to controls on carbon emissions in different regions.

Introduction

The accounting of carbon dioxide (CO₂) emissions serves as the foundation for monitoring regional emissions and also for government departments to review emission sources and development patterns^1,2. It offers a quantitative perspective on CO₂ emissions from fossil fuels, facilitating the evaluation of viable reduction strategies and informing projections related to socioeconomic trends, thereby guiding the formulation of policies aimed at mitigating the impacts of climate change^3,4. The emission of CO₂ should be reflected not only in quantity but also in space^5,6,7,8. The spatialization of CO₂ emissions is of great significance to regional emission control policies, and low carbon spatial planning^5,9,10. Information on the fine structure of spatial emissions can assist to find effective points of intervention to promote emissions reduction by showing how emissions change between regions, between urban centres of different types and between urban and rural populations¹¹. What’s more, the produced emission maps can be used to see which parts of the region have highest emissions, and from what sources, which provide data support for carbon emission control policies.

Scale effects are one of the main concerns in the spatialization of CO₂ emission. Mapping CO₂ emissions has been explored at global, national, provincial, and even county-level scales^5,7. Spatial resolution of global carbon emissions has been examined at 5°, 1°, 0.5°, 0.25°, and 0.1°^12,13,14,15, whereas spatial resolution of CO₂ emissions at national, regional and provincial scales ranges from 1.0 to 10 km, which is relatively coarse^{11,16,17,18,19}. With the development of emission inventories towards higher resolution, the study of scale effects of carbon emission spatialization is becoming increasingly important. Hogue et al.²⁰ found the larger the grid sizes, the smaller is the relative uncertainty in emissions per grid space at four different levels of resolution (1 × 1, 2 × 2, 3 × 3, and 4 × 4 degrees). Long et al.⁵ investigated the scale effect of carbon emissions at different resolutions and determined the optimal spatial resolution for county-level emissions. Zheng et al.¹⁷ analyzed the influence of spatial proxies on the precision of gridded carbon emission data at varying spatial scales (36, 12 and 4 km resolutions). However, the spatial resolution settings for existing studies on carbon emission scale effects are mostly relatively coarse. There is still a relative lack of research on scale effects with high spatial resolution.

Currently, the mapping of CO₂ emissions mainly include two methods, which are bottom-up estimation method and downscaling method^{17,21,22,23,24}. Bottom-up estimation relies on mass emission sources with geographical location information. Carbon dioxide emissions are first accounted at the facility-level data source based on their energy consumptions. The emissions of each spatial grid cell are the sum of the emissions of various facility in the grid cell^6,25. Sources of emissions include large power plants, polluting enterprises, and traffic, among others^17,21. The bottom-up method is relatively accurate for point source emissions, but the locations and their energy consumption information of these mass facility data at a provincial scale are difficult to obtain. In particular, the location and relevant data of some facilities are easy to be omitted. In addition, the bottom-up method is difficult to extend through all sectors and regions. The mapping of carbon emissions from nonpoint source data with this method also usually requires use of auxiliary data, which have the limitations associated with downscaling method⁶.

Due to the difficulty of obtaining mass facility data for bottom-up method, downscaling method becomes an alternative method for the mapping of CO₂ emissions. In this method, total carbon emissions of a region are first accounted at the regional level based on regional energy consumption and emission factors. Then the total CO₂ emissions are distributed to each spatial grid cell according to the relationship between emissions and spatial auxiliary data, so as to obtain the spatial distribution of emissions^11,26. Population density, nighttime light data, and other remote sensing data are widely used as spatial auxiliary data because they reflect human activities^2,27. Downscaling method is feasible easily and requires relatively less data with low cost, it has become widely used for developing gridded emission inventories, especially in large-scale regions¹⁷. Downscaling method has great potential in mapping of CO₂ emissions with high spatial resolution at relative large-scale regions, but it still has some limitations. First, the correlations between CO₂ emissions and auxiliary data are incomparable under different sectors, which is easy to lead to a poor correlation on the fine-scale spatial resolutions¹¹. For example, for the catering service sector and residential energy use, similar nighttime light values do not indicate similar CO₂ emissions, which is also the case for transportation and industrial sectors. Among different emission sectors, the relationship between CO₂ emissions and auxiliary data is different. Using the same relationship between CO₂ emissions and auxiliary data in a region will lead to greater uncertainty. Second, auxiliary data do not accurately indicate the variability in intensity of emissions from point source s without additional information, such as power plants².

Therefore, when adopting downscaling methods for creating high-resolution carbon emission maps, restricting emissions from each sector to their corresponding land use types will better establish the relationship between emissions and auxiliary variables. Existing studies have confirmed a clear correspondence between carbon emissions from socioeconomic sectors and land use types^28,29. Zhao and Huang³⁰ matched land use types with carbon emission items from various sectors to investigate emissions resulting from energy consumption in different land use types in Jiangsu Province. Wu et al.³¹ established the relationship between carbon emissions of sectors and different land use types in Zhejiang Province and indicated that carbon emissions were primarily propelled by the inflexible demands for energy-intensive land. Zhao et al.²⁸ analyzed the characteristics of spatial and temporal carbon emissions of Shanxi based on the land use type data and fossil energy consumption data from different sectors. The corresponding relationship between CO₂ emissions from different sectors and land use has received widespread attention. However, due to the lack of information on land use within urban areas, few studies have focused on utilizing the correspondence between national economic sectors and land use to enhance the refinement and accuracy of CO₂ emission mapping.

To address the incomparable of the relationships between CO₂ emissions and auxiliary data under different sectors for the downscaling methods and the lack of research on high spatial resolution scale effects in carbon emission mapping, this study obtained detailed urban land use data and established a correspondence between land use and sectors. The emissions of different sectors are restricted within their corresponding land use areas, and then high-resolution CO₂ emission mapping is conducted using different spatial auxiliary data in different land uses. The scale effect and optimal spatial resolution of CO₂ emissions at the provincial level was analyzed and determined. Lastly, the spatiotemporal patterns and factors influencing CO₂ emissions in the study area were analyzed. This study will provide data to help meet goals of peak carbon emissions and carbon neutrality in China.

Materials and methods

Study area

Most studies in China focus on mapping CO₂ emissions in the east or some developed provinces, and less attention is paid to the economically less developed provinces in western China^5,7. Most provinces in western China are in the initial stage of rapid economic development, and their CO₂ emissions are also in a stage of rapid increase. To balance economic growth and control of carbon emissions in these regions in the future, it is important to understand temporal and spatial distribution patterns of carbon emissions. Thus, in this study, the western province of Guizhou was used as a case study.

Guizhou Province is in the hinterland of Southwest China, and it covers approximately 176,167 km² and includes nine cities (prefectures) and 88 county-level administrative divisions (Fig. 1). Guizhou Province is a transportation hub in Southwest China and an important part of the Yangtze River economic belt. With implementation of a western development strategy, the economy of Guizhou entered a stage of rapid development. By 2019, the regional gross domestic product (GDP) in the province reached 1,676.934 billion yuan, with an annual growth rate of 8.3%³². The growth rate ranked first in China for nine consecutive years and was first in China for three consecutive years. Acceleration of industrialization and urbanization has led to rapid growth in energy demand and a significant increase in carbon emissions. The contradiction between demand for rapid economic growth and control of CO₂ emissions is increasingly prominent in Guizhou Province. Guizhou Province is a typical area of “rich coal, lack of oil and less gas,” and thus, coal has long dominated energy consumption.

Fig. 1

Location of study area and its nighttime light data with 30 m resolution.

Carbon accounting

In this study, the emissions of CO₂ are sourced from China Emission Accounts and dataset (CEADs), which were proposed by Shan et al.³³. This study mainly focuses on the years 2009 and 2019. Carbon emissions were considered primarily from two aspects: energy-related emissions and industrial process-related emissions. Energy-related carbon emissions were those emitted during the combustion of fossil fuels, which included emissions from 17 fossil fuels burned in 47 socioeconomic sectors in this study. Emissions were calculated according to the following equation:

$${{\varvec{C}}{\varvec{E}}}_{{\varvec{i}}{\varvec{j}}}={{\varvec{F}}{\varvec{C}}}_{{\varvec{i}}{\varvec{j}}}\times {{\varvec{N}}{\varvec{C}}{\varvec{V}}}_{{\varvec{i}}}\times {{\varvec{C}}{\varvec{C}}}_{{\varvec{i}}}\times {{\varvec{O}}{\varvec{R}}}_{{\varvec{i}}{\varvec{j}}}$$

(1)

where ${CE}_{ij}$ and ${FC}_{ij}$ are the CO₂ emissions and the consumption of fossil fuel type i in socioeconomic sector j, respectively; ${NCV}_{i}$ is the net calorific value of fossil fuel type i; ${CC}_{i}$ is the CO₂ emissions per unit of net heat generated by fossil fuel i; and ${OR}_{ij}$ is the oxidation rate during fuel combustion³³. Fossil fuel consumption (${FC}_{ij}$) was obtained based on China’s energy statistical yearbooks and Guizhou’s statistical yearbooks in 2009 and 2019.

Forty-seven socioeconomic sectors includes farming, forestry, animal husbandry, fishery and water conservancy, coal mining and dressing, petroleum and natural gas extraction, ferrous metals mining and dressing, production and supply of electric power, steam and hot water, transportation, storage, post and telecommunication services, wholesale, retail trade, and catering services, other service sectors, urban resident energy usage, and rural resident energy usage, among others. The 17 fossil fuels included raw coal, cleaned coal, other washed coal, briquette, coke, coke over gas, other gas, other coking products, crude oil, gasoline, kerosene, diesel oil, fuel oil, other petroleum products, liquefied petroleum gas, refinery gas, and natural gas.

Carbon dioxide emissions from industrial processes are primarily due to physical and chemical reactions in production processes. Cement production is the main source of carbon emissions, accounting for approximately 75% of total CO₂ emissions from industrial production in China³³. Therefore, in this study, emissions from the cement production process were determined according to the following equation:

$${{\varvec{C}}{\varvec{P}}}_{{\varvec{k}}}={{\varvec{F}}{\varvec{C}}}_{{\varvec{k}}}\times {{\varvec{C}}{\varvec{F}}}_{{\varvec{k}}}$$

(2)

where ${CP}_{k}$ is the CO₂ emissions during cement production and ${FC}_{k}$ is the activity factor in the carbon emissions accounting for cement production in year k, that is also the cement output. The activity factor was obtained from the official data set of the National Bureau of Statistics. The emission factor of cement production is ${CF}_{k}$ , which was equal to 0.2906³³. Carbon dioxide emissions from cement production were classified as manufacturing in this study.

Mapping CO ₂ emissions

The spatialization of CO₂ emissions in this study included three steps (Fig. 2). Firstly, obtain detailed land use data with a resolution of 30 m and establish the corresponding relationship between land use and national economic sectors. The emissions of different sectors were restricted within their corresponding land use areas. Second, within each land use type, the total CO₂ emissions will be allocated to square grids with size of 30 m based on nighttime light data, population, and other auxiliary data.

Fig. 2

Spatial mapping scheme of CO₂ emission.

CO ₂ emissions from sectors to land use types

This study established the linkage of different sectors with land use types^29,30,34,35. First, the 47 economic sectors were first merged into eight sectors, which including production and supply of electric power sector, the farming, forestry, animal husbandry, fishery and water conservancy sector, mining sector, transportation service sector, manufacturing industries, wholesale, retail trade, catering, and other service sectors, urban residents’ energy usage, and rural residents’ energy usage.

The land use data in this study were obtained by combining urban and other areas using different methods (Fig. 3). The urban land use data were generated based on the 30 m Landsat 8 OLI/TIRS, OpenStreetMap, nighttime lights, and points of interest (POI) using the method developed by Gong et al.³⁶. The types of urban land use include land for catering and other service land, industrial land, land for transportation site, urban resident land, transportation road. Land use data in non-urban areas was obtained by using supervised classification method based on the 30 m Landsat 8 OLI/TIRS, which including forest land, garden land, cultivated land, grass land, unutilized land, water body, land for agricultural and water facilities, rural resident land. The POI data was acquired utilizing Python through the Application Programming Interface (API) provided by Baidu Maps (http://map.baidu.com/), a prominent web mapping service in China.

Fig. 3

The spatial distribution of land use types.

Carbon dioxide emissions from different industries were allocated to their corresponding land use types based on the land use data and emission data. As point-source data, CO₂ emissions generated by power plants were allocated to industrial land parcels where the power plant is located. Data on power plants were obtained from the world power plant database (https://www.wri.org/research/global-database-power-plants). Carbon dioxide emissions from manufacturing were allocated to industrial lands excluding those where power plants were located. Emissions from combustion of oil fuels in transportation, storage, and post and telecommunication services sectors were allocated to roads, and emissions from combustion of nonoil fuels were allocated to transportation sites. Carbon dioxide emissions from mining sectors were allocated to mining lands. Emissions from energy use by urban and rural residents were allocated to urban and rural residential lands, respectively. Emissions generated by wholesale, retail trade, catering services, and other service sectors were allocated to facility lands of the commercial service industry. Emissions from agriculture, forestry, animal husbandry, and sideline and fishery industries were allocated to lands for agricultural facilities, hydraulic construction, and other land types. Due to the lack of OpenStreetMap and some other data, the land use classification in 2009 does not separately list lands for the service industry. Emissions from urban residents and the service industry were uniformly allocated to urban land.

High spatial resolution mapping of CO₂ emissions in different land use types

After emissions of different industries were allocated to different land use types, emissions on each grid pixel cell were allocated adjusting for the weight of nighttime light or population in each land type^2,27,37. The grid size was set to be consistent with the land use data, which is 30 m. Carbon dioxide emissions in urban and rural residential lands were spatially distributed weighted by population data, whereas those from other land use types were spatially distributed weighted by nighttime light data. Emissions allocated increased as light or population values increased. Emissions were estimated according to the following equation:

$${{\varvec{C}}}_{{\varvec{t}}{\varvec{j}}}={{\varvec{C}}{\varvec{E}}}_{{\varvec{j}}}\times \frac{{{\varvec{N}}{\varvec{L}}}_{{\varvec{t}}}}{\sum_{{\varvec{t}}=1}^{{\varvec{n}}}{{\varvec{N}}{\varvec{L}}}_{{\varvec{t}},{\varvec{j}}}}$$

(3)

where ${C}_{tj}$ is the CO₂ emissions of grid t in ground class j; ${CE}_{j}$ is the total emissions allocated to land class j; ${NL}_{t}$ is the nighttime light brightness value or population value of the grid; n is the number of grid t in land type j; and $\sum_{t=1}^{n}{NL}_{t,j}$ is the sum of light values or populations of all grids in ground class j. Ultimately, spatial distributions of CO₂ emissions in 2009 and 2019 were obtained.

Both the nighttime light data and the population data were downscaled to a resolution of 30 m. The nighttime lighting data utilized were sourced from the corrected global DMSP NTL (Defense Meteorological Satellite Program Nighttime Lights) dataset with a 1 km resolution³⁸. The population data were obtained from the kilometer-grid data from the Resource and Environmental Science and Data Center (https://www.resdc.cn/). For the downscaling of nighttime light data, the HPANI-OK method developed by Guo et al. was employed³⁹. Firstly, human activity indicators, including road density and POI density were generated at a 30-m resolution. The road network data for 2019 and 2009 are obtained from the Resource and Environmental Science Data Platform (https://www.resdc.cn/data.aspx?DATAID=237) and the National Earth System Science Data Center (https://www.geodata.cn/data/index.html?publisherGuid=126744287495931&categoryId=20), respectively. POI density was generated using kernel density analysis. The POI data was acquired utilizing Python through the Application Programming Interface (API) provided by Baidu Maps (http://map.baidu.com/). Secondly, a power function regression was used to characterize the relationship between DMSP NTL and human activity factors at a coarse resolution (1 km). Lastly, both power function regression and Ordinary Kriging were applied to predict the nighttime light trends and residuals at a resolution of 30 m (Fig. 1)^39,40. The detailed steps of this process are outlined in the research conducted by Guo et al.³⁹.

For downscaling population data, random forests (RF) based on remote sensing and POI data were utilized.^41,42. Initially, seven proxies with a 30-m resolution were generated, including elevation, slope, nighttime light data, NDVI, POI density, road density, and distance to the nearest POI. Secondly, all proxies were aggregated to a 1000-m resolution and then correlated with the natural logarithm of the population data at a 1-km resolution to fit the RF model. The RF model's objective is to calculate the distribution weight for each gridded area. Finally, the weights derived from the RF model were employed to disaggregate the population from a 1-km resolution to a 30-m resolution. The specific steps of this process are detailed in the research conducted by Ye et al.⁴¹.

Scale effect analysis and optimal resolution selection

Scale dependence of spatial heterogeneity has always been a concern in geography and ecology, which includes grain and extent. Extent is the geographic scale, which was the provincial scale in this study. Grain represents the spatial resolution of a map. If the spatial resolution is too large, the location information of carbon emissions will not be expressed accurately. If the spatial resolution is too small, the mapping process will be time-consuming and the amount of data will increase, especially in a large area. Therefore, it is very important to determine the appropriate spatial resolution for the mapping of CO₂ emissions. In this study, the spatial distribution maps of CO₂ emissions with different resolutions were produced by changing the grid size. The scale effect and the optimal resolution of the spatial distribution of emissions under different resolutions were analyzed. At the highest spatial resolution, the minimum grid size is set to 30 m. The spatial distributions of CO₂ emissions under different grid sizes are obtained by increasing grid size at 30-m increments up to 2010 m.

Referring to previous studies, this study used the landscape metrics to analyze the scale effect of the spatial distribution of carbon emissions and determine optimal resolution^5,43. The landscape metrics was calculated by FRAGSTATS 4.2 software⁴⁴. The changes and scale effects of the index at different scales were analyzed. Scale effect indicators were Shannon’s diversity index (SHDI) and evenness index (SHEI), the aggregation index (AI), and the proportion of like adjacencies (PLADJ), which are sensitive to scale changes^43,45. The two Shannon indices indicate diversity (SHDI) and evenness (SHEI) of patch types and therefore are measures of landscape heterogeneity. Both indices are especially sensitive to unbalanced distributions of each patch type in the landscape. The AI indicates the spatial aggregation of patch types, and PLADJ indicates the agglomeration of a landscape. All indicators were calculated at the landscape level. When a landscape index changes with scale, there may be an obvious “scale turning point (inflection point)”, and the two adjacent inflection points are called the scale domain. In a scale domain, the landscape pattern index is relatively stable, and therefore, its pattern characteristics are relatively stable, which can better reflect characteristics of the regional landscape pattern and also indicate optimal resolution.

Analysis on influencing factors of regional differences in CO ₂ emissions

Based on the spatial distribution of CO₂ emissions, this study further explored the influencing factors of regional differences in CO₂ emissions in the study area. The influencing factors mainly considered the economic development conditions, industrial structure, and urbanization level. With counties (districts) as the unit, CO₂ emissions, economic development conditions, industrial structure, and urbanization level were determined for 88 counties (districts) in Guizhou Province in 2009 and 2019. Economic development condition was indicated by GDP and per capita GDP. Industrial structure was indicated by proportions of primary, secondary, and tertiary industries. Area of built-up area, per capita construction land area, and proportion of construction land area were used to indicate urbanization level. The scatter diagram between each impactor factors and CO₂ emissions were constructed and their relationship were analyzed based on the statistical models, such as linear function, univariate quadratic polynomial, exponential function, logarithmic function, etc.

Results

Carbon dioxide emissions accounting results

Carbon dioxide emissions in Guizhou Province increased from 184.7 million tonnes in 2009 to 258 million tonnes in 2019. Thus, rapid economic development of Guizhou over the 10 years led to an approximate increase of 73 million tonnes in CO₂ emissions, an increase of approximately 7.3 million tonnes per year. Carbon dioxide emissions under different economic sectors and land use types in Guizhou Province are shown in Table 1. The highest emissions were in production and supply of electric power sector, which reached 114.48 million tonnes in 2019 and accounted for 58.63% of total emissions. The second largest emitters of CO₂ were manufacturing sector, which emitted 49.74 million tonnes of CO₂ emissions. Therefore, the land use type with the highest emission was the industrial lands, which reached 164.22 million tonnes in 2019 and accounted for 63.65% of total emissions. Compared with 2009, emissions increased by 26.88 million tonnes. Rapid development of industrialization in Guizhou over the 10 years was the main reason for the increase. Therefore, how to reduce carbon emissions from industrial land is the primary problem that Guizhou currently needs to address. Owing to rapid development of tertiary industry in Guizhou Province, CO₂ emissions from wholesale, retail trade, catering, and other service sectors also increased significantly, which reached 43.92 million tonnes and accounted for 17.03% of total emissions. Those service sectors were becoming the third largest emitters of CO₂ and their CO₂ emissions increased by 26.52 million tonnes over the 10 years of the study area. Carbon dioxide emissions from transportation land reached 16.19 million tonnes, which accounted for 6.28% of total emissions and was an increase of approximately 7.89 million tonnes compared with 2009 emissions.

Table 1 Carbon dioxide emissions in different land use types in Guizhou (mt).

Carbon dioxide emissions from rural residential energy usage reached 15.15 million tonnes in Guizhou Province, which was much higher than that of urban residential energy usage with emissions of 5.18 million tonnes. Compared with 2009, the growth of emissions from rural residential areas was much greater than that of urban residential land. The continuous increase in rural residential area in Guizhou was the primary reason for the difference, which was much greater than that in urban residential area. The result also indicates that Guizhou still needs to further accelerate the pace of urbanization and promote intensive economic development. In the future, identifying the best solution to balance accelerating urbanization and control of carbon emissions will be the major challenge for development of Guizhou. Emissions from agriculture, forestry, animal husbandry, and sideline fisheries did not increase significantly over the 10 years, whereas the only land type with reduced carbon emissions among all land types was mining land.

Spatial patterns of CO ₂ emission

Spatial distribution of CO₂ emissions in the 30-m grid is shown in Fig. 4. Carbon dioxide emissions in Guizhou Province in 2009 were spatially aggregated, and most emissions were concentrated in urban areas of various cities (prefectures). High carbon emission areas were primarily concentrated in Guiyang City and Zunyi City. However, in 2019, CO₂ emissions were more spatially diffuse and dispersed. Because of urban expansion, areas with CO₂ emissions increased significantly over the 10 years. Many areas changed from noncarbon emission areas to those with carbon emissions. Although distribution of medium carbon emission areas increased, that of high carbon emission areas decreased significantly. In addition, carbon emissions gradually spread from west to east. Thus, the spatial distribution of carbon emissions developed from agglomeration to decentralization in Guizhou, which reflected the trend toward decentralized and balanced development of industrial layout and expansion of urbanization.

Fig. 4

Spatial distribution of CO₂ emissions. (a) 2009, (b) 2019.

The hot spot maps of CO₂ emissions from various sources at the county level in 2019 are presented in Fig. 5. Regarding the total emissions of each county, Guiyang City, the southern part of Zunyi City, Panzhou County, and Xingyi County, located in the southwest of the study area, were hotspots for emissions, with an average emission of approximately 8 million tonnes. Conversely, a few counties in the east and south of the study area were cold areas. As for the carbon emissions originating from the mining sector, the hotspots were primarily situated in the southern counties of Bijie City and Zunyi City, as well as the southwestern counties of Guizhou Province. The CO₂ emissions from mining sector totaled approximately 0.5 million tonnes. The hotspots for carbon emissions resulting from urban residents' energy usage were primarily located in the southern regions of Zunyi City and Guiyang City, as well as in several adjacent counties, with each of these counties emitting around 0.3 million tonnes of CO₂.

Fig. 5

Hot spots maps of CO₂ emissions from different sources at county scale in 2019.

The hotspots for carbon emissions arising from rural residents' energy usage were notably distinct from those of cities, being primarily situated in the central and southern regions of Bijie City and the northern part of Anshun City. Furthermore, the three counties of Weining, Panzhou, and Xingyi, located in the western part of the study area, are also considered emission hotspots. The emissions from these hotspots were all approximately 0.5 million tonnes. The carbon emissions stemming from rural residents' energy usage are very low in the eastern counties of the study area (most of which are in Qiandongnan Miao and Dong Autonomous Prefecture). The hot spot maps based on CO₂ emissions from transportation, wholesale, retail trade, catering, and other service sectors resembled the emission hotspot map for urban residents' energy usage. The hot spot areas were primarily distributed in regions with higher economic levels. For instance, the southern regions of Zunyi and Guiyang City, among others, had carbon emissions that were roughly in the range of 2 million tonnes to 0.5 million tonnes. The hot spot map of CO₂ emissions from manufacturing, power production, and other industrial sectors is essentially the same as the hot spot map for total emissions. However, there are few emission hotspots from the farming, forestry, animal husbandry, fishery, and water conservancy sectors, which are mainly concentrated in a few counties in Bijie and Qianxinan.

Scale effect and optimal resolution

To compare mapping results of carbon emissions under different grid sizes, spatial distributions of CO₂ under 54 grid sizes at 30-m increments were created (later, the increment was changed to 60 m and then to 90 m). The minimum grid size was 30 m, and the maximum was 2,010 m. Some of the results are shown in Fig. 6. In the 90-m grid, there was spatial heterogeneity in the spatial distribution of CO₂ emissions. The spatial heterogeneity of CO₂ emissions became more evident with increasing grid size to a certain point, but with further increases in grid size, spatial heterogeneity and details were reduced. Overall, CO₂ emissions from big city centers, such as Guiyang and Zunyi, were high but gradually diffused and decreased around those cities. Emissions from traffic and roads and rural areas were intermediate.

Fig. 6

Spatial maps of CO₂ emissions at different resolutions in 2019. (a) 90 m, (b) 600m, (c) 1260m, (d) 2010m.

Changes in spatial grid size significantly affected CO₂ mapping results. The optimal spatial resolution for high spatial-resolution mapping of CO₂ emissions at the provincial scale was a grid size of 90 m and 1,260 m (Fig. 7). According to SHDI and SHEI of different grid sizes at the landscape level, the finest grain was from 30 to 90 m and from 1,140 m to 1,260 m, because curves of the two indices increased steadily within those ranges (Fig. 7a,b). The curve of AI values indicated the finest grain was from 30 to 90 m, from 1,160 m to 1,260 m, and from 1,740 m to 1,830 m, according to the same criteria (Fig. 7c). The curve of PLADJ values indicated the most suitable grid size was from 30 to 90 m and from 1,200 m to 1,290 m (Fig. 7d). The turning point and scale domain reflected the scale sensitivity and stability of landscape change. A small grid usually leads to a power exponential increase in the amount of data, resulting in a reduction in operational efficiency. Therefore, in a relatively stable scale domain, a relatively large grid should be selected as the optimal resolution. Thus, in Guizhou Province, based on the turning point and scale domain of the above landscape indices, the optimal resolution at high spatial resolution was 90 m, followed by 1,260 m. The appropriate resolution can be determined according to the actual application requirements.

Fig. 7

Scale effect of landscape indices. (a) for the Shannon’s diversity index, (b) for the Shannon’s evenness index, (c) for the aggregation index, (d) for the proportion of like adjacencies. (Grid size refers to the length and width of the grid, which are equal.)

Comparison with other gridded data

Spatial allocation results of this study were compared with spatial data of CO₂ emissions obtained by other methods, including the Regional Emission Inventory in Asia (REAS) with grid size of 0.25°¹³ and grid data in Peking University’s global pollutant emission inventory (PKUEI) with grid size of 0.1^14,16. The REAS data covers up to 2014, whereas Peking University's global pollutant emission inventory covers up to 2015. A comparison is undertaken between the grid data from Peking University's global pollutant emission inventory and this study for the years 2009 and 2014. Similarly, a comparison is made between the REAS data and this study for the years 2009 and 2015.

Table 2 shows the statistical values of different grid data. In terms of total emissions, the REAS total emissions in 2009 were relatively low, while the PKUEI total emissions were relatively high, and the emissions results in this paper fall between the two. However, by around 2015, the PKUEI emissions were not significantly different from the results in this paper, while the REAS emissions remained relatively low. From the perspective of maximum and minimum values, since the research in this paper has refined to a spatial resolution of 30 m, the minimum carbon emission value is 0, which is lower than that of other grid data, while the maximum value is much higher than that of other grid data. For the median values, in 2009, due to the relatively concentrated emissions, most regions in Guizhou had no carbon emissions, so the median value was lower than that of other grid data. However, with the expansion of urbanization, most regions shifted from having no carbon emissions to having carbon emissions. In 2015, the median value of our study was close to the PKUEI data, which had similar total emissions. In terms of average values, the results in 2009 were between the two sets of data, while in 2015, they were greater than the other two sets of data, mainly due to the relatively large total emissions. Overall, more detailed grid data can better reflect the spatial variability of emissions.

Table 2 The statistical values of different grid data (kg/m²).

When comparing the spatial distributions of the REAS data and the global pollutant emission inventory with the data presented in this paper, they exhibit similar patterns for different grid data (Fig. 8). In 2009, CO₂ emissions were relatively concentrated, and high carbon emission regions were primarily concentrated in the western region of Guizhou Province. By 2015, the distribution of CO₂ emissions was more balanced, with a significant decrease in high carbon emission areas, an increase in medium carbon emission areas, and carbon emissions gradually spreading from west to east. However, because of limitations in spatial resolution, more refined emission distributions, such as those associated with roads, could not be determined. Therefore, the REAS and global pollutant emission inventory cannot provide the more refined grid data needed to manage CO₂ emissions at the provincial level. The method proposed in this paper can obtain more precise spatial distribution of CO₂ emissions.

Fig. 8

Spatial data of global CO₂ emissions from Peking University in (a) 2009 and (b) 2014 and from regional emission inventory in Asia in (c) 2009 and (d) 2015 for Guizhou Province.

Analysis on influencing factors of regional differences in CO ₂ emissions

The scatter diagram of economic level (GDP and GDP per capita) and carbon emissions in different counties of Guizhou Province for 2009 and 2019 were shown in Fig. 9. Previous studies showed two main views on the relationship between economic level and carbon emissions: growth leads to more emissions, or an inverted U-shaped relationship^46,47,48. For example, Yang et al. pointed out that there exists a turning point in the relationship between carbon emissions and GDP economic⁴⁸. Zhang et al. found that economic growth significantly promoted the increase of carbon emissions in China before 2012, but the positive effect was significantly weakened after 2012⁴⁹. The results of this study are consistent with these two mainstream views. The relationship between regional economic level and CO₂ emissions developed from a linear relation to an inverted U-shaped one in the 10 years from 2009 to 2019 (Fig. 9). In the period prior to 2009, Guizhou's economic development was still largely extensive, primarily driven by the consumption of fossil fuels to promote industrial growth and subsequently boost the economy. Consequently, the economy and CO₂ emissions have a significant positive correlation (Fig. 9a,b). Around 2009, the 11th Five-Year Plan for national development, emphasizing the adjustment of industrial structure and the transformation of economic growth patterns, was proposed, and implemented. After that, Guizhou initially reversed its extensive development path and gradually shifted towards an intensive development model. Thus, by 2019, relations between economic level and CO₂ emissions of counties in Guizhou changed to an inverted U-shaped relation because of agglomeration effect and economic transformation (Fig. 9c,d).

Fig. 9

Scatter diagram of economic level (GDP and GDP per capita) and carbon emissions in different counties of Guizhou Province. (a, b) 2009 and (c, d) 2019.

The relationship between industrial structure and CO₂ emissions is shown in Fig. 10. The effect of regional industrial structure on CO₂ emissions weakened over the 10 years. Regional CO₂ emissions showed a significant exponential decline with the increase in proportion of primary industry, with an R² value of 0.7 and a decline rate of 2.433 in 2009. By 2019, the R² value decreases to about 0.2 and the exponential decline rate was 1.064 (Fig. 10a,d). Regions with a high proportion of primary industry had low degree of industrialization and urbanization, and those regions had less CO₂ emissions. However, with the upgrading of the industrial structure and the acceleration of the urbanization process, the impact of the primary industry on emissions has significantly weakened. The relations between proportion of secondary industry and carbon emissions showed exponential increases in both years, with R²values of 0.33 in 2009 and 0.29 in 2019 (Fig. 10b,e). There was no relation between proportion of tertiary industry and carbon emissions (Fig. 10c,f). The absence of a relation might be because of the significant differences in emissions from different sectors within the tertiary industry, such as transportation and the catering industry had high carbon emissions, whereas tourism and the high-tech industry had low carbon emissions, such as. Therefore, development of low-carbon emission industries should be considered in development of tertiary industries. Adjustment of industrial structure will be an important factor in the control of CO₂ emissions in Guizhou.

Fig. 10

Scatter diagram of industrial structure and carbon emissions. (a–c) 2009 and (d–f) 2019.

The relationship between the primary and secondary industries and carbon emissions discussed in this study is consistent with most of the previous researches^49,50. However, it is worth noting that previous studies have indicated that industrial structure, specifically the development of the tertiary industry, exerts a restraining influence on the increase in carbon emissions^1,51. In this study, the development of the tertiary industry did not exhibit such a restraining effect. The study indicates that the potential of Guizhou's industrial structure has not been fully unleashed and exploited. This finding is supported by Zhang et al.⁴⁹. Further optimize the industrial structure will be an important direction for emission reduction in Guizhou.

Over the 10 years of the study, the influence of urbanization level on CO₂ emissions in Guizhou Province increased. Carbon dioxide emissions increased with increasing area of regional built-up areas in both 2009 (R² = 0.48) and 2019 (R² = 0.73) (Fig. 11a,d). There was no significant relation between area of construction land per capita and carbon emissions (Fig. 11b,e). However, in large per capita construction areas (more than 2.7 km²/10,000 people in this study), that is, in areas with low population concentration, CO₂ emissions were low. Urbanization level and economic development of those areas were relatively low, and therefore, CO₂ emissions were not high. There was an inverted U-shaped relation between proportion of construction land in the total area and CO₂ emissions (Fig. 11c,f). Previous studies have shown that an increase in built-up area will increase CO₂ emissions. But with the acceleration of urban expansion, the proportion of built-up area will decrease carbon emissions to a certain extent^52,53. This is consistent with the conclusion of this study. The main reason is that urbanization brings about the agglomeration of population and industries, which is conducive to reducing energy consumption in transportation and other sectors, improving the utilization of infrastructure, making industrial division of labor and cooperation more reasonable and efficient, and thus reducing carbon emissions.

Fig. 11

Scatter diagram of urbanization level and CO₂ emissions. (A–C) 2009 and (D–F) 2019.

Discussion

Remote sensing data such as nighttime light have great potential for the spatial distribution of CO₂ and other pollutant emissions^2,38,54. Such data are easily accessible at low cost and are suitable for large-scale mapping of carbon emission⁵⁵. In this study, a method was presented to spatialize CO₂ emissions at a provincial scale with high spatial resolution that combined data on nighttime light, populations, point sources data, and land use types. The approach is more reasonable than that of the previous method of space allocating that used auxiliary data in an administrative region⁵⁵. Human activities represented by the intensity of nighttime light values in the same land use type were comparable. Emissions of power plants were allocated to the industrial parcel, which can avoid underestimation caused by mapping of emissions using nighttime lighting data. Nighttime lighting data provide good indication of the flow of vehicles on roads, emissions of crude oil, gasoline, diesel oil, fuel oil, and other petroleum products in the transportation service industry were allocated to road land with the help of nighttime light data in this study⁵⁶. In this study, population data was used as a proxy for mapping carbon emissions in urban residential and rural areas. However, considering only the population may cause some bias in the mapping results, as people's consumption patterns vary. A smaller number of people with high incomes and high consumption patterns would release more CO₂ into the atmosphere than a larger population with low incomes but also high consumption patterns. To enhance the accuracy of emission mapping in residential areas, an additional proxy, such as median income or consumption level, reflecting the diversity of consumption, should be considered in the future.

The downscaling method used in the mapping of CO₂ emissions assumed linear relations between CO₂ emissions and spatial proxy variables^11,16,54. Previous studies suggest that underlying assumption is potentially valid at a coarse scale but is highly suspect at a fine scale¹⁷. The main reason for the difference may be that a linear relation is assumed in a given district rather than in the same land use type or the same sector. When the resolution is high, proxy variables such as nighttime light easily reflect characteristics of different land uses. However, proxy data values are not comparable, and there are poor correlations between proxy variables and emissions. Zheng et al. (2017) also found that emissions were sensitive to proxy variables at finer scales in various sectors, especially for residential and transportation emissions. Therefore, it is reasonable to associate land uses with industry sectors and then allocate CO₂ emissions in different land uses with the help of selected proxies. In high spatial-resolution mapping of carbon emissions, auxiliary variables must be used with caution and cannot be used directly regardless of land type characteristics.

Land as the carrier of economic activities, CO₂ emissions from different sectors are reflected in different land use types⁵⁷. In the method in this paper, a relationship was established between emissions and land use types. Because of the linkage with land use type, CO₂ emissions can be updated while detecting land use change. The details of a land use classification will affect the accuracy of spatial distributions of CO₂ emissions. For example, most land use data do not distinguish between urban residential land and commercial land, although the nighttime light value of residential land is inconsistent with that of commercial land in indicating spatial differences in carbon emissions. Only nighttime light values between similar industries are comparable. Therefore, the more detailed the land use classification is, the more accurate the spatial mapping of CO₂ emissions, or the coarser the land use classification is, the greater the uncertainty of spatial distribution of CO₂ emissions.

The research results in this paper can provide policy support for government. Guizhou’s CO₂ emissions are at an intermediate level in China, but its annual growth rate in recent years is relatively high. Emissions in 2009 and 2019 in Guizhou indicated the primary CO₂ emission source was the industry sector, especially production and supply of electric power. Thus, policy makers should require upgrades to emission control technologies of existing industries and consider increasing standards for industrial emissions²⁵. At the same time, Guizhou should give full play to its advantages of rich water resources, increase the utilization of clean energy, and reduce the carbon emission of the production and supply of electric power. Carbon dioxide emissions from rural residential lands in Guizhou Province were much higher than those from urban residential lands, and this gap is still increasing in recent 10 years. According to the land use map in 2019, the area of rural residential lands in Guizhou is about 4.5 times that of urban residential lands. Reducing the energy consumption of rural residential areas and improving their utilization efficiency are the focus of Guizhou in the future. In urban planning, land for various facilities should be rationally distributed, and the balance between work and housing should be advocated. For new houses, we should promote the use of new energy-saving materials and improve the level of central heating to improve energy efficiency. Moreover, the coverages of subways in various cities of Guizhou are limited due to the limitation of topography. At present, only four subways have been opened in Guiyang, the provincial capital of Guizhou province. The government needs to further accelerate the construction of subway and tram, to improve the level of public travel and reduce carbon emissions.

Some studies indicated that there was a linear positive relationship between regional economic and CO₂^58,59, while some studies suggested that there was an inverted U-shaped relationship between them^60,61. The main reason may be that the economic level of some regions is still in the initial stage, which is far from reaching the inflection point of inverted U-shape. The empirical evidence of Guizhou indicated the relationship between urbanization and CO₂ emission inevitably changes from linear relationship to inverted U-shaped relationship. However, China’s plan to reach the peak of carbon emissions in 2030 inevitably needs to accelerate the transformation from linear relationship to U-shaped relationship under human intervention. How to balance the relationship between the growth of economic and the control of carbon CO₂ emission is particularly important, especially for the western region of China, where is still in the initial development stage.

Conclusion

Accurate spatial distribution of CO₂ emissions is essential information of refined grid management for controlling CO₂ emissions. This study established a correspondence between land use and sectors. The emissions of different sectors are restricted within their corresponding land use areas, and then high-resolution CO₂ emission mapping is conducted using different spatial auxiliary data in different land uses. Furthermore, the scale effect and the optimal spatial resolution of CO₂ emissions at the provincial level were analyzed and determined. With Guizhou Province as an example, this study also explored the influence factors of regional differences in CO₂ emissions. Results showed that compared to other grid data, the high spatial resolution grid data proposed in this study can better reflect the spatial variability in emissions at the microscale. Changes in spatial grid size significantly affected CO₂ emissions mapping results. The optimal resolution for high spatial-resolution mapping of CO₂ emissions at the provincial scale was 90 m and 1,260 m. Over the past 10 years, the distribution of CO₂ emissions developed from agglomerated to dispersed, areas of high carbon emissions decreased, areas of medium carbon emissions increased, many areas changed from no carbon emissions to carbon emissions, and carbon emission areas gradually spread from west to east.

Carbon dioxide emissions in Guizhou were approximately 258 million tonnes in 2019, which was an increase of approximately 7.3 million tonnes per year in the 10 years from 2009 to 2019. Industrial land was the land use type with the highest emissions, accounting for 63.65% of total emissions. Therefore, how to reduce carbon emissions from industrial land is the primary task in reaching the goal of peak carbon emissions in Guizhou in 2030. For industrial land, the southern regions of Guiyang, Zunyi, and Liupanshui are the key areas for emission reduction. The second largest emissions were from lands for the service industry, which accounted for 23.3% of total emissions. The southern regions of Guiyang and Zunyi, the northern part of Anshun, and the eastern part of Liupanshui are key areas for emission reduction in service industry. Carbon dioxide emissions from rural residential lands in Guizhou Province were much higher than those from urban residential lands, indicating that Guizhou still needs to accelerate the future pace of urbanization. The southern regions of Guiyang and Zunyi, the northern part of Anshun, and the eastern part of Liupanshui are key areas for emission reduction in service industry land use. How to balance accelerating urbanization with carbon emission control is a major challenge in the development of Guizhou. In the 10 years from 2009 to 2019, the relation between regional economic level and CO₂ emissions in Guizhou developed from a linear one to an inverted U-shaped one. In addition, there was a decrease in the effect of industrial structure on CO₂ emissions but an increase in that of urbanization. The results of this study will provide basic support for overall national planning of policies addressing peak carbon emissions and also be a reference for studies on carbon emissions and scale effects in other regions.