Research on the impact of COVID-19 on the spatiotemporal distribution of carbon dioxide emissions in China

Abstract

Since the outbreak of COVID-19 at the end of 2019, the Chinese government has imposed strict control measures on affected cities, which may have impacted the spatial and temporal pattern of carbon dioxide emissions. This paper follows the quantitative analysis method, experimental method, mathematical method, etc., and quantitatively studies the impact of the epidemic on China's carbon emissions. The combination model of ARIMA and BP neural network is used to predict the actual impact of epidemic situation on China's carbon emissions in 2020, and the spatial autocorrelation analysis method is used to analyze the spatial characteristics of China's provincial carbon emissions, which indicate that China's carbon emissions have consistently maintained a growth trend, from 2.05 billion tons in 2005 to 3.89 billion tons in 2019. Furthermore, the growth rate of carbon emissions and the changing trend of the emission intensity are the same, dropping from 12% in 2005 to 3% in 2019. The emission intensity also dropped from 1.1 in 2005 to 0.6 in 2019, indicating that the trend of increasing carbon emissions in northern provinces and Xinjiang changed significantly from 2005 to 2019. The overall carbon emissions of the 30 provinces in 2020 are predicted to be 4.068 billion tons, while the actual energy carbon emissions will be 3.921 billion tons, suggesting that the pandemic significantly reduced carbon emissions. Among affected provinces, carbon emissions from Hubei, Jiangsu, Shandong, Shanghai, and other places changed significantly, from 0.99, 0.25, 0.43, and 76 million tons in 2019 to 0.88, 0.24, 0.42, and 72 million tons in 2020, respectively. The results show a positive spatial correlation between China's provincial carbon emissions; the high-high and bottom-high agglomeration are mainly among the provinces, mainly distributed in North China and East China. Although the pandemic seriously impacts China's carbon emissions, each province's spatial relationship has not changed significantly.

1. Introduction

At the end of 2019, COVID-19 began spreading worldwide, significantly impacting the life and economy of various countries and regions [1]. China is the first country to be affected by COVID-19, and it is also the first to take measures to prevent and control the pandemic. From the earliest “closed city” in Wuhan to the suspension of work, production, and classes in various industries and regions, pandemic prevention measures have greatly influenced China's development [2]. Implementing various pandemic prevention measures has significantly reduced CO₂ emissions from fossil energy consumption and combustion [3,4]. Relevant studies have shown that due to the impact of COVID-19, global emissions of greenhouse gases and air pollutants have dropped significantly [[5], [6], [7], [8], [9]]. According to the International Energy Agency (IEA), global fossil fuel combustion reduced CO₂ emissions by 2.6 billion tons in 2020 due to the impact of the COVID-19 pandemic. This reduction represents a decrease of about 8% from 2019, mainly due to declining oil and coal consumption. The IEA also emphasized that the economic recovery plan should prioritize clean energy to avoid a sharp rebound in carbon emissions [10].

According to the current research, the primary research focuses on the influence of COVID-19 on carbon emissions, and the analysis mainly measures energy transition and economic recovery after the pandemic [[11], [12], [13]]. Gabbatiss estimated the impact of COVID-19 on the global CO₂ concentration per month according to the monthly forecast of the IEA for oil demand, finding that the annual average CO₂ concentration in 2020 was 0.32 ml/m3 lower than in 2019. This decrease is equivalent to the CO₂ concentration decreasing by 11%, and while the CO₂ accumulation rate will be slower than expected due to COVID-19, this will be insufficient to significantly reduce global warming [14]. For example, the impact of the global financial crisis of 2008–2009 on carbon emissions was short-lived; global CO₂ emissions rebounded rapidly in 2010, mainly due to rapidly falling energy prices, significant government investment in many countries to promote swift economic recovery, and the rapid economic growth of developing countries [15]. Since the outbreak of COVID-19, government departments and industry associations in some countries have called for delaying the implementation of “green policies,” relaxing automobile emission standards, and suspending research on clean energy deployment and supply issues. How countries make decisions will affect the annual CO₂ emission path over the coming decades [5].

Combined model of ARIMA and BP is simple and flexible, and can deal with nonlinear relationship well, so it is widely used in carbon emission prediction research. Xiao uses the combination model of ARIMA and BP to forecast the carbon emissions in China. The results show that the model has a small deviation, and the carbon emissions from 2014 to 2020 are well predicted [16]. Wang developed ARIMA1-BPNN and BPNN-ARIMA using the combination model of ARIMA and BP to simulate the carbon emissions of China, India, the United States and the European Union without epidemic, with a relative error of 1%. This study will fully understand the impact of epidemic on carbon emissions [17]. In view of the advantages of this model in carbon emission prediction, this study uses it to predict the carbon emissions of different provinces in China in 2020.COVID-19 significantly affected China's greenhouse gas emissions and the reduction of atmospheric pollutants from January to April 2020 [9,18]; however, the spatiotemporal pattern of the impact of the COVID-19 pandemic on the national provincial-level carbon emissions is not yet apparent.

Therefore, this research based on ARIMA and BP model analyzes the carbon emissions of 30 provinces in China to more deeply understand the impact of COVID-19 on various provinces and study the pandemic's long-term impact on China's carbon peaking and carbon neutrality goals. The second part of this paper introduces the specific model method, the third part analyzes and settles the spatial and temporal changes of carbon emissions in 2005-2-01 and the specific impact of COVID-19 on carbon emissions, the fourth part discusses the impact of COVID-19 on carbon emissions, and the fifth part summarizes the research results of this paper.

2. Research methods

At present, the research on carbon emission caused by epidemic situation is mainly based on satellite remote sensing images and air quality measurement. Based on energy carbon emissions, this study adopts the calculation method proposed by the Intergovernmental Panel on Climate Change (IPCC), which is more conducive to quantitative research. Therefore, based on energy consumption data, this study uses ARIMA and BP combined model and spatial autocorrelation analysis to predict carbon emissions without epidemic more accurately, which will make the research on the impact of epidemic on carbon emissions from emissions to spatial changes more intuitive and prominent.

2.1. Calculating carbon emissions

This paper uses the method proposed by the Intergovernmental Panel on Climate Change to calculate CO₂ emissions. We denote the time series as {X_t}, and the formula [19]is as follows：

$${\mathrm{C}\mathrm{O}}_2=\sum _{i=1}^n{A}_i\times B_i\times C_i$$ (1)

CO₂ represents the carbon dioxide emissions measured in 10,000 tons per unit. i is the energy type, which includes eight energy types; A is energy consumption; B is the converted coal coefficient of various energy; C is the emission factor. Table 1 presents the specific coefficients.

Table 1.

Carbon emission coefficients of various energy sources and conversion coefficients of standard coal.

i	coal	coke	crude	gasoline	kerosene	diesel fuel	fuel oil	natural gas
Bi (kgce/kg)	0.7143	0.9714	1.4286	1.4714	1.4714	1.4571	1.4286	1.33①
Ci (kg/kgce)	0.7559	0.855	0.5857	0.5538	0.5714	0.5921	0.6185	0.4483

Note: The carbon emission numbers refer to the research of [20,21]. The conversion coefficient of standard coal is referred to in the China Energy Statistical Yearbook 2020; ① Unit = kilogram of coal equivalent (kgce)/m³.

2.2. Autoregressive integrated moving average model

The autoregressive integrated moving average (ARIMA) model is built based on the moving average model MA (q), the autoregressive model AR (p), and the autoregressive moving average model ARMA (p, q). The stochastic process comprises autoregressive and moving averages, called the autoregressive moving average process. Its expression [17] is:

$${\mathrm{X}}_{\mathrm{t}}={\mathrm{\phi }}_1{\mathrm{X}}_{\mathrm{t}–1}+{\mathrm{\phi }}_2{\mathrm{X}}_{\mathrm{t}–2}+\dots \dots {\mathrm{\phi }}_{\mathrm{p}}{\mathrm{X}}_{\mathrm{t}-\mathrm{p}}+{\mathrm{\epsilon }}_{\mathrm{t}}+{\mathrm{\theta }}_1{\mathrm{\epsilon }}_{\mathrm{t}–1}+{\mathrm{\theta }}_2{\mathrm{\epsilon }}_{\mathrm{t}–2}+\dots \dots +{\mathrm{\theta }}_{\mathrm{q}}{\mathrm{\epsilon }}_{\mathrm{t}-\mathrm{q}}$$ (2)

In the ARIMA model (p, d, q), p represents the order of the autoregressive model, determined by the number of partial autocorrelation coefficients (PACF). Additionally, q represents the order of the moving average model, determined by the autocorrelation coefficient (ACF) number, d represents the number of differences made when the time series becomes stationary, φ is the autoregressive coefficient, and θ is the moving average coefficient.

2.3. Backpropagation model

The backpropagation (BP) neural network is the most widely used and mature artificial neural network method. It is a multilayer feedforward network trained by an error backpropagation algorithm that is most suitable for simulating the relationship between input and output [9]. It includes three layers to complement existing theoretical analysis and aid decision-making processes: input, hidden, and output. The layers are not interchangeable and cannot be transferred; only the upper and lower layers can be entirely connected through the weighted network. Fig. 1 shows the basic principle.

Fig. 1.

Three-layer BP neural network.

The following formula [17] calculates the output of the hidden layer neurons:

$$x_i^1=f\left({\sum }_{j=1}^n{w}_{ij}^0{x}_j+w_{i0}^0\right)，i=\mathrm{1,2}，\cdots ，p$$ (3)

where $x_i^1$ represents the neuron output of the hidden layer; $\sum _{j=1}^n{w}_{ij}^0{x}_j+w_{i0}^0$ represents the weighted summation of n input layer neurons; $w_{ij}^0$ represents is the weight coefficient of the influence of the input layer neuron j on the hidden layer neuron i; $w_{i0}^0$ is the threshold of the hidden layer neuron i; f is a nonmemory nonlinear excitation function to change the neuron. The following formula [17] calculates the output layer neuron's output:

$$y^k＝f\left({\sum }_{j=1}^p{w}_{jk}^1{x}_j^1＋{\mathrm{w}}_{\mathrm{k}0}^0\right),\mathrm{k}＝\mathrm{1,2}，\cdots ,\mathrm{m}$$ (4)

The output layer's error function [17] is:

$$\mathrm{E}=\frac{1}{2}\sum _k\left(d^k-y^k\right)$$ (5)

2.4. ARIMA and BP composite pattern

According to existing research [16,22], China's historical carbon emissions are linear and nonlinear. Therefore, combining ARIMA and BP models can make up for both deficiencies and make the prediction results more accurate. The composite model is constructed as follows:

① Use model (3) to establish an ARIMA model for the time series X t and obtain the fitted value X^t and residual error e_t = X_t − X^_t.

② BP neural network model fitting residual sequence [16]: Assuming n network inputs exist, the residual relationship can be expressed as

$$e_t=f(e{{}_t}_{-}{}_1{}^{},e{{}_t}_{-}{}_2{}^{},\dots ,e{{}_t}_{-}n)+n{}_t$$ (6)

Among them, f is the nonlinear function of the neural network model, and n t is the random error.

③ Models (3) and (6) establish the combined model as follows:

$$\varnothing \left(B\right){\nabla }^d{X}_t=\Theta \left(B\right){\epsilon }_t$$

$$e_t=X_t-\widehat{X_t}$$

$$e_t=f\left(e_{t-1}，e_{t-2}，\cdots e_{t-n}\right)+n_t$$

2.5. Spatial autocorrelation analysis

Spatial autocorrelation includes global and local spatial autocorrelation.

2.5.1. Global autocorrelation

The global spatial autocorrelation can reflect the overall characteristics of the spatial relativity of carbon emissions among provinces to judge the overall relationship between spatial dimensions and the difference in provincial energy carbon emissions; its index is mainly the global Moran's I index [23]. The following is a specific expression:

$$I＝\frac{n{\sum }_{\mathrm{i}＝1}^n\sum _{\mathrm{j}=1}^n{w}_{\mathrm{i}\mathrm{j}}\left(x_{\mathrm{i}}-\bar{x}\right)\left(x_j-\bar{x}\right)}{\sum _{\mathrm{i}=1}^n\sum _{\mathrm{j}＝1}^n{w}_{ij}\sum _{\mathrm{i}＝1}^n{\left(x_i－\bar{x}\right)}^2}$$ (7)

In the formula [23], I is the global Moran's I value; n is the number of provinces and cities under study. x_i and x_j are the carbon emissions of provinces and cities i and j, and W_ij is the spatial weight matrix of each province and city under study, based on Rook adjacency. $\bar{x}$ is the average value of the carbon emission attribute value of the research object.

Moran's I ranges from −1 to 1; values closer to 1 indicate that the observed value is positively correlated in space, while values less than 0 indicate that the observed value is negatively correlated in space. Values equaling 0 mean that the observations are randomly distributed in space, and there is no spatial autocorrelation. Moran's I can be divided into four types: HH (high–high), HL (high–low), LH (low–high), and LL (low–low). HH (LL) agglomeration means positive spatial autocorrelation between adjacent provinces, and the spatial unit has homogeneity. HL (LH) indicates a negative spatial autocorrelation between adjacent provinces; high (low) carbon emission intensity provinces are surrounded by low (high) carbon emission intensity provinces, and there is heterogeneity in spatial units [24]. After the calculation, the Z-value test must be conducted on the results. The specific formula [19] is:

$$\mathrm{Z}\left(I\right)＝\frac{I-E\left(I\right)}{\sqrt{\mathrm{V}\mathrm{A}\mathrm{R}\left(\mathrm{I}\right)}}$$ (8)

where E(I) is the expected value of autocorrelation of carbon emissions, and VAR(I) is the variance. Z(I) > 1.96 indicates a significance level of 5%, while Z(I) > 2.58 indicates a significance level of 1.

2.5.2. Local autocorrelation

To further study the spatial correlation and difference between a province and surrounding provinces and cities, it is essential to use the local spatial autocorrelation method for further analysis. This paper uses the ArcGIS platform's clustering and outlier analysis (LISA) to judge each province and city. The degree of similarity (positive correlation) or difference (negative correlation) with surrounding provinces and cities. The specific formula is as follows:

$${I_i=x}_i^{\prime }\sum _{j=1}^n{w}_{ij}{x}_j^{\prime }$$ (9)

In the formula [23], x′_i and x′_j are the standardized unit observations, w_ij is the weight, and a LISA diagram is drawn based on the Z(I) test.

2.6. Data sources

We use energy consumption data to calculate carbon emissions. The energy consumption data from the 30 provinces and autonomous regions are from China Energy Yearbook (2006–2021), and the gross domestic product (GDP) data are from China Statistical Yearbook (2006–2021).

3. Results

3.1. Spatiotemporal analysis of carbon emissions

3.1.1. Overall analysis of China's carbon emissions

3.1.1.1. Total carbon emissions and growth rate

The process of carbon emissions in China in the past 15 years (Fig. 2) includes two stages. The first stage was from 2005 to 2011, when carbon emissions rapidly increased from 2.05 billion tons in 2005 to 3.34 billion tons in 2011; the emission in 2011 was 1.6 times that of 2005. The second stage was from 2012 to 2019, when the carbon emissions’ growth rate slowed significantly compared with the previous period. Emissions increased by 460 million tons—from 3.44 billion tons in 2012 to 3.89 billion tons in 2019—with an average annual growth rate of 1.8%, indicating that carbon emission reduction policies positively promote carbon emission reduction.

Fig. 2.

Change trend of China's carbon emissions from 2005 to 2019.

The growth rate trend changes more and can be divided into four stages. Phase one covers 2005–2008, during which the carbon emission rate rapidly declined. The largest decline was in 2005–2006, dropping from 12% in 2005 to 8% in 2006, reaching its lowest point in 2008. Phase two covers 2008–2011. During this period, the growth rate was rapidly recovering because China's economy was in the recovery and development stage, rebounding the growth rate of carbon emissions compared to previous years. The third stage covers 2011–2013, where two main reasons account for the fast decline in the rate of carbon emissions increase. One is because China's economy and technology rapidly developed, and increasing attention was paid to development quality, reducing carbon emissions to a certain extent. Implementing emission reduction measures has gradually banned some backward industrial and mining enterprises and industries that consume significant energy and resources, thereby reducing carbon emissions. The fourth stage covers 2013–2019, during which time the growth rate was in a stage of steady and slow growth. This slowdown in growth rate shows that China significantly affects carbon emission control, decreasing the annual growth in carbon emissions.

3.1.1.2. Energy intensity of carbon emissions

The GDP of 2005–2019 was calculated at the constant price of 2005 to avoid the influence of economic development on carbon emissions intensity. The national energy intensity of the carbon emissions change map from 2005 to 2019 was calculated (Fig. 3) based on each province's emission data. The figure indicates that the emission intensity change can be subdivided into three stages. First, 2005–2008 was a period of rapid decline, where emission intensity fell from 1.1 tons/10,000 Chinese yuan (CNY) in 2005 to 1.0 tons/10,000 CNY in 2008, representing a decline rate of 10%. The rapid decline of carbon emission intensity at this stage has a significant relationship with the proposal of China's “Eleventh Five-Year Plan,” which puts pollution prevention and control as the top priority, strengthens pollution control, and simultaneously emphasizes speeding up structural adjustment. Implementing a series of stringent measures has ensured a reduction in carbon intensity. Second, 2009–2011 was a stable stage, where changes in carbon emission intensity were minimal, with annual measures of 0.94, 0.93, and 0.94, respectively. The reason is related to the global financial crisis in 2008. Although a series of effective measures from the Chinese government prevented a severe financial crisis, the country was inevitably affected to a certain extent. China's development was affected to some degree, and the growth trend of carbon emissions has slowed, so there was no obvious downward trend in carbon emission intensity from 2009 to 2011.

Fig. 3.

China's carbon emission intensity from 2005 to 2019.

The third trend, from 2012 to 2019, maintained a steady decline. During these 7 years, the carbon emission reduction rate was about 29%, and the lowest point was 0.64 tons/10,000 CNY in 2019. Fig. 3 shows a downward trend, inseparable from China's active implementation of national strategies to deal with climate change, adjustment of industrial structure and energy structure, improvement of the carbon market, and increase of forest carbon sinks. The peak carbon neutrality target was proposed, and China has been implementing stricter carbon control measures to reduce carbon emissions and intensity.

3.1.2. Spatial evolution characteristics of carbon emissions

We use natural breakpoints in 2005, 2010, 2015, and 2019 to divide Chinese provinces of carbon emissions into 7 levels: no data (0), emissions (0–0.25), slight emissions (0.25–0.5), mild emissions (0.5–0.75), moderate emission (0.75–1.0), high emission (1.0–1.25), and heavy emission (>1.25). The spatial analysis function of ArcGIS 10.7 was used for analysis, resulting in Fig. 4. Data are unavailable for Tibet, Taiwan, Macau, and Hong Kong, which were excluded from this study.

Fig. 4.

Temporal and spatial analysis of carbon emissions from 2005 to 2019.

3.1.2.1. The tred of the emission and slight emission

The changing trend of the number of provinces with emissions and slight emissions is basically the same, showing a decreasing trend, with a higher decline degree for slight-emission provinces. Emission provinces primarily comprise Qinghai, Hainan, Ningxia, Guangxi, and Yunnan. At the same time, the carbon emissions of Qinghai, Yunnan, and Hainan increased from 7.61, 14.26, and 4.56 million tons, respectively, in 2005 to 16.71, 30.07, and 21.16 million tons in 2019. In 2005, China's carbon emissions were mainly light emissions, with 11 provinces in this range; however, by 2019, only 5 provinces were included. The changing trends suggest a relatively apparent upward emission trend in each province from 2005 to 2019.

3.1.2.2. The tred of the light and high emission

Provincial variation characteristics are stable, with high emissions comprising seven provinces and light emissions maintaining three; the changes are mainly reflected in different provinces. Overall, the light-emission provinces in 2005 were mainly high emission in 2019, such as Heilongjiang, Sichuan, and Hubei. Provinces in high-emission areas changed from Jiangsu, Liaoning, and Henan in 2005 to Zhejiang, Anhui, and Henan in 2019. In 2005, Jiangsu, Liaoning, and Henan relied on their advanced manufacturing industries, strong industrial bases, and superior location conditions, so carbon emissions were in the high range in 2005. Furthermore, Zhejiang and Anhui relied on technological innovation, balanced development, and geographical advantages, and their rapid economic development and carbon emissions in 2019 also entered the high-emission range.

3.1.2.3. The tred of the moderate and heavy emissions

The number of provinces with moderate and heavy emissions has steadily increased, with heavy emitters growing more than moderate ones. Provinces in the heavy emission range in 2005 included Shanxi, Hebei, and Shandong, while in 2019, this number expanded to include Shaanxi, Xinjiang, Guangdong, Liaoning, Jiangsu, Hebei, Inner Mongolia, Shanxi, and Shandong.

Fig. 4 shows that the trend of provincial carbon emissions in the northern region and Xinjiang from 2005 to 2019 has changed significantly, from light emissions at the beginning to heavy emissions. The development of Xinjiang is inseparable from its advantages in natural resources and the inclination of national policies. The development of the entire northern region relies on its superior location and strong economic foundation; however, carbon emissions are proliferating.

3.2. Impact of COVID-19 on carbon emissions

3.2.1. Analysis of annual change rate of carbon emissions

We used the energy consumption data in 2020 and calculated each province's and city's carbon emissions in 2020 to see more clearly the difference in carbon emissions between 2019 and 2020. Fig. 5 presents the results. In terms of overall carbon emissions, those in 2020 were 3.921 billion tons, a significant decrease compared with 2019. The growth rate of carbon emissions in 2019 was 3.90%, and in 2020, it was 0.67%. Since the end of 2019, the COVID-19 pandemic has proliferated worldwide, affecting everyone and bringing significant disadvantages. This impact is evident in industry, transportation, and service, making it difficult to maintain normal order in production and life. Many factories and institutions are facing shutdowns and even bankruptcy; however, this reduces energy consumption to a certain extent, leading to a slower increase in carbon emissions. Changes in the increase in carbon emissions can also reflect the degree of impact of the pandemic on provinces and cities. The most obvious changes occurred in Hubei, Jiangsu, Shandong, and Shanghai, judging from the carbon emission increases in various provinces and cities in 2019 and 2020. Since the outbreak, all parts of China have attached great importance to the pandemic and responded actively. Hubei and other provinces have entered wartime states, assuming the strictest antipandemic measures, such as internal nonproliferation, external antiexport, shutdowns, and closed management. Implementing these measures has led to a large-scale reduction in human activities, a decline in energy consumption and industrial production activities, a short-term reduction in carbon emissions, and a slowdown in growth. Hubei's carbon emissions have dropped from 99 million tons in 2019 to 88 million tons in 2020, indicating that these measures significantly inhibit carbon emissions. Jiangsu, Shandong, and Shanghai have also been affected to a certain extent, and their respective carbon emissions decreased from 0.25, 0.43, and 76 million tons in 2019 to 0.24, 0.42, and 72 million tons in 2020. After the outbreak in Wuhan, all provinces and cities in China adopted strict prevention and control measures. Therefore, each province's carbon emissions should not increase much in 2020, and many provinces should decline.

Fig. 5.

Carbon emissions in 2019 and 2020.

3.2.2. Carbon emission analysis under ideal scenarios

Carbon emissions in the imaginary scenario refer to the normal development of social production and life, without the pandemic, based on the development law of carbon emission over the years. This study used the ARIMA and BP combined model to simulate the carbon emission in 2020.

3.2.2.1. Data stationarity test

Fig. 2 shows that the {X_t} time series has an apparent upward trend, indicating that this series is a nonstationary series. For further verification, the carbon emission data from 2005 to 2019 was recorded as the {X_t} sequence for the unit root test, and p = 0.93 > 0.05 was obtained by calculation; thus, the data were significantly nonstationary. Difference processing was then performed to obtain stationary time series data. Using R software to perform differential processing on the {X_t} sequence, the results show that the data has stabilized when the second-order difference is used. Fig. 6 presents the results.

Fig. 6.

Second-order differential timing sequence diagram.

The T-test indicates that P = 0.01 < 0.05, and the null hypothesis of the existence of a unit root is significantly rejected at the significance level of 0.05, so the sequence after the second-order difference is stationary.

3.2.2.2. Determine the model

Since the data is stable after the second-order difference, the ARIMA model can be judged as a (P, 2, q) model. The ACF (Fig. 7) and PACF (Fig. 8) functions of the second-order difference data were plotted with R software. Fig. 7 shows that the ACF values of orders 1 to 11 all fall within the confidence interval (within the dotted line in the figure); thus, it is determined that the sequence is the end, and the model is MA (0), so P = 0. Fig. 8 indicates that in the third order, the PACF value tends to 0, and the model is AR (3), q = 3, that is, ARMA (3, 0).

Fig. 7.

Autocorrelation function diagram.

Fig. 8.

Partial autocorrelation function diagram.

Still, this order determination method is subjective, arbitrary, and insufficiently detailed and scientific. Therefore, the minimum AIC principle of R software's auto_arima is used to verify this model, and the best model of ARIMA is (0, 2, 0). Therefore, the ARIMA (0, 2, 0) model is finally adopted. The expression of this model is as follows:

Y = X_t = 2X_{t – 1} − X_{t – 2} + ε_t,{ε_t}∼N(0, 0.8377)

The Ljung–Box test was performed on the model residuals, and the P value was 0.482 > 0.05, indicating that the Ljung–Box test was passed. This result shows that the model information is good and can be used for prediction.

3.2.2.3. Model fitting and prediction

The ARIAM(0, 2, 0) model is used to fit the {X_t} time series, and the value X^t and residual et are obtained, as shown in Table 2. Fig. 9 shows the fitted line graph of the true and predicted values. We applied this model to forecast the carbon emissions in 2020, and the forecast result is 4 billion tons, denoted as Xt*. Table 2 and Fig. 9 show that the prediction results have high precision, and only a few years, such as 2012 and 2013, have high deviation.

Table 2.

Fitting values and residual sequences from 2005 to 2019.

Year	2005	2006	2007	2008	2009	2010	2011	2012	2013
fitted value X^t	20.526	23.054	24.025	25.300	27.712	31.360	33.956	31.977	33.106
residual value et	0.0091753	−0.021804	−0.80478	−0.64211	0.3290259	1.2677073	0.5773248	−2.343731	−1.078372
Year	2014	2015	2016	2017	2018	2019
fitted value X^t	35.349	34.374	35.580	36.743	37.565	39.122
residual value et	0.6503675	−0.41962	0.3456612	0.5341245	0.1906035	0.291413

Fig. 9.

Fitting line of real and predicted carbon emissions from 2005 to 2019.

3.2.2.4. Constructing the BP neural network

The BP model is established using the residuals from 2005 to 2019, as shown in Fig. 10. According to the prediction, the et* in 2020 is 0.68. Based on the forecast of the combined model, the final carbon emission forecast for 2020 is 4.068 billion tons. This forecast aligns with the carbon emission trend from 2005 to 2019, indicating a stable increase. Furthermore, the carbon emission forecast of the combined model in 2020 increased by 147 million tons compared to the actual carbon emission. This increase also shows, to a certain extent, the impact of COVID-19 on emissions.

Fig. 10.

BP model.

3.3. Spatial autocorrelation analysis

We used GeoDa and ArcGIS to analyze the spatial distribution of carbon emissions. Hainan Province is not geographically adjacent to other provinces; therefore, when setting spatial weights, nearby Guangdong Province and Guangxi Province were set as adjacent to Hainan Province. Due to a lack of data, Taiwan was not used.

3.3.1. Global autocorrelation

Table 3 is obtained through the global spatial autocorrelation analysis of China's provincial carbon emissions from 2005 to 2019. Table 3 shows that Moran's I from 2005 to 2019 are all positive values, and all pass the 95% significance level test, indicating that China's provincial energy carbon emissions have a positive spatial correlation. Statistics on Moran's Ι values in each year show that overall, Moran's Ι values from 2005 to 2019 trended gently downward, indicating that the spatial agglomeration of provinces with similar energy carbon emissions has declined. This paper takes 2005, 2010, 2015, and 2019 as examples, adopting the natural breakpoint method to obtain Moran's I scattergram of China's energy carbon emissions from 2005 to 2019, as shown in Fig. 11. Fig. 11 shows that the proportions of the H–H and L–L quadrants in the four years are 60%, 57%, 57%, and 57%, indicating a strong positive spatial correlation of carbon emissions. This correlation means that provinces with high (low) emissions are next to those with high (low) emissions, indicating homogeneity between regions; the proportions of the second and fourth quadrants are 40%, 43%, and 43%. These results suggest obvious spatial heterogeneity: high (low) carbon provinces are surrounded by provinces with low (high) carbon emissions. In general, the carbon emission space has an obvious positive correlation. The quantity distributed in the first quadrant is generally higher than in the third, showing that the spatial distribution of carbon emission is highly agglomerated.

Table 3.

Moran's I global energy carbon emissions in China.

Year	2005	2006	2007	2008	2009	2010	2011	2012	2013
Moran's I	0.264361	0.267983	0.254484	0.26474	0.240529	0.235412	0.242217	0.234283	0.22725
Z	2.7185	2.6375	2.6563	2.6157	2.4646	2.3274	2.4140	2.3424	2.2821
P	0.009	0.012	0.008	0.009	0.012	0.018	0.016	0.015	0.015
Year	2014	2015	2016	2017	2018	2019
Moran's I	0.217493	0.221132	0.203897	0.19293	0.200259	0.150992
Z	2.1876	2.2461	2.1278	2.0625	2.0656	1.7507
P	0.025	0.022	0.028	0.033	0.037	0.049

Fig. 11.

Moran's I scatter diagram of China's energy carbon emissions from 2005 to 2019.

3.3.2. Local autocorrelation

To further study the spatial relationship of the emissions among provinces, a LISA map of the spatial distribution of Chinese carbon emissions was created through the ArcGIS platform, as shown in Fig. 12. According to the analysis, in 2005, the high agglomeration areas were mainly Inner Mongolia, Liaoning, Shanxi, Shandong, Henan, and Jiangsu provinces. This agglomeration is due to the proximity to Beijing and Tianjin, the high level of urbanization, and the developed industry and agriculture, which increased carbon emissions to form a high concentration zone. The high–low correlation area was mainly Guangdong, which relies on unique geographical location, capital investment, scientific and technological progress, and industrial progress. The province's economy is far more developed than the surrounding provinces, forming a high-value isolated area. The bottom-high correlation areas are mainly Beijing, Tianjin, Jilin, and Anhui. Although Beijing and Tianjin have small carbon emissions, they are surrounded by Hebei, which has high carbon emissions; thus, forming low-value isolated areas. Jilin and Anhui are in areas with significant carbon emissions, forming bottom-high correlation areas. There is only one province in Guangxi in the low–low agglomeration area, which has a low level of urbanization and a relatively backward economy based on intraprovincial cooperation; thus, forming a low–low agglomeration area. Compared with 2005, the changes in 2010 were small, and the differences were mainly reflected in Shaanxi, Anhui, and Shanghai. Shaanxi and Anhui have become new high–high agglomeration areas, mainly because they are located in the periphery of high–high agglomeration areas. Under the influence of the diffusion effect of the surrounding provinces, carbon emissions are increasing, and Shaanxi, Anhui, and Shanghai have become high–high agglomeration areas. In 2010, Shanghai belonged to the bottom-high correlation area. It is located in the Yangtze River Delta with convenient sea and land transportation, concentrated talents, and developed technology. The pursuit of high-quality, high-level green development has made Shanghai a low-value isolated area. Compared with 2005 and 2010, the changes in the local spatial pattern of carbon emissions in 2015 and 2019 were mainly reflected in Xinjiang. Since 2015, Xinjiang has become a high-low correlation area, inseparable from the Belt and Road strategy, the western development policy, and unique natural resources. This change shows the impact of these policies and measures on the local economy in Xinjiang, indicating that development has a clear positive role in promoting.

Fig. 12.

Carbon emission concentration of China's provinces.

Fig. 12 shows that under the influence of the pandemic in 2020, the spatial relationship of carbon emissions in Chinese provinces and cities has not changed significantly. Only Jilin has changed from a high correlation area at the bottom in 2019 to insignificant; other provinces and cities have not changed.

4. Discussion

This paper calculates the carbon emissions of each province from 2005 to 2019. The calculation results are consistent with the calculation results and forecast trends of Wang Ying et al. [25] and Tong Xin et al. [26], who use the carbon emission coefficient method to calculate emissions. Due to the different choices of energy types and carbon emission coefficients, there are certain differences in the specific values of carbon emissions; however, they are within a reasonable range. The research results of spatial autocorrelation and local spatial autocorrelation of provincial carbon emissions are similar to those of Wang Ying et al. [25] and Sun He et al. [27]. The Moran I values are all greater than zero, and all pass the 95% significance level test, indicating that the provincial carbon emissions are mainly characterized by high–high and bottom-high agglomeration; the high–high agglomeration is mainly distributed in the central and northern regions.

Impacted by the pandemic, most countries’ emissions have changed to different degrees compared with the same period. European countries, such as Spain, Italy, and France, reduced carbon emissions by 10.66 million tons due to the pandemic [28,29]. This study shows that Hubei has the most considerable contribution to carbon emission reduction, mainly due to the decline in the share of secondary industry; however, due to the lack of energy data, few studies on carbon emissions exist at the provincial level. This paper compared the true emissions in 2020 with the carbon emissions predicted by the model in 2020 to quantify the influence of the outbreak on carbon emissions. The study showed that under normal economic and social development conditions, the actual carbon emissions in 2020 were 147 million tons less than predicted. Studies have shown that different blockade measures affect air quality differently [[30], [31], [32]]. When the pandemic broke out, strict lockdown measures were taken in Wuhan, China, which significantly improved air quality in the Yangtze River Delta region.

Studies have shown that the pandemic has improved air quality and reduced many cities' carbon emissions and water pollution [[33], [34], [35]]. Conversely, it has also set back economic development, increased medical waste, and energy industry challenges [36]. The pandemic's impact is twofold. According to Refs. [37,38], the specific impact of COVID-19 on China can be seen from multiple angles and industries. In the short term, it will lead to setbacks in economic development and increased industrial challenges; however, the adoption of targeted low-carbon economic stimulus policies can help to restore the economy and adjust the energy structure and industrial institutions. Moreover, in the long run, implementing “green” policies and measures after the COVID-19 pandemic is the key for China to accelerate the development of low-carbon technologies and take the lead in future competition [39,40].In general, the impact of the pandemic on different countries, regions, and provinces is not the same; however, mitigating its impact requires actively changing the mode of economic development, improving the utilization rate of resources, and achieving green and healthy economic development [41].

To sum up, most studies focus on different countries or the same country, mainly studying the impact of the epidemic on carbon emissions, or the different measures taken by different countries to deal with the epidemic. This paper focuses on the impact of epidemic situation on carbon emissions in different provinces of China in the same time period. This study will provide different emission reduction directions and measures for carbon emission research in different provinces. In the provinces where the epidemic has a great impact on carbon emissions, the main industries affected include cement, transportation, industry, power generation, etc., which is a challenge even if it is an opportunity to reduce environmental pollution. In the meantime, we should formulate appropriate policies to reduce greenhouse gas emissions. For example: continue to deepen the adjustment of industrial structure and improve energy-saving and emission-reduction technologies; In the next carbon emission reduction work, the provinces with little epidemic impact should strengthen the supervision and inspection of carbon emission reduction, pay attention to the change of green production mode, reduce the use of fossil fuels, and pay attention to the emission of greenhouse gases from daily life to production activities, which will greatly promote the carbon emission reduction work.

5. Conclusion

During the epidemic in COVID-19, it affected the carbon emissions of all countries in the world to varying degrees. Taking China as an example, this paper quantitatively studied the impact of the epidemic on the carbon emissions of various provinces. Based on the energy consumption, it predicted the carbon emissions in 2020 by using the combination model of ARIMA and BP, and obtained the actual impact of the epidemic on carbon emissions by comparing the predicted value with the actual carbon emissions. In order to further study the carbon emission space of provinces in 2020, a spatial autocorrelation analysis was conducted, and the specific conclusions are as follows.

This study made several findings. First, from 2005 to 2020, China's carbon emissions increased in two stages: a rapid increase from 2005 to 2011 and a slow increase from 20012 to 2020. From 2005 to 2020, the change rate of carbon emissions, growth rate, and emission intensity indicate that China's carbon emission reduction policies and measures have achieved good results, and more stringent measures must be taken to reduce future carbon emissions.

Second, by applying the combined model of ARIMA and BP to more scientifically and accurately predict carbon emissions in 2020. The model predicts that carbon emissions will be 4.068 billion tons in 2020, and the actual carbon emissions will be 3.921 billion tons. Comparing the actual carbon emission increase in each province and city in 2020 with the increase in 2019 indicates that the carbon emission increase in most provinces and cities in 2020 is significantly smaller than that in 2019. In particular, the carbon emissions in Hubei, Jiangsu, Shandong, Shanghai, and other areas changed significantly, decreasing from 0.99, 0.25, 0.43, and 76 million tons, respectively, in 2019, to 0.88, 0.24, 0.42, and 72 million tons in 2020. These results show that COVID-19 will somewhat impact various provinces and cities in China in 2020.

Finally, the research on the global spatial autocorrelation of China's provincial emissions from 2005 to 2019 shows that Moran's Ι is positive, and all passed the 95% significance level test, indicating that China's provincial energy carbon emissions have a significant positive spatial correlation. Local autocorrelation studies show that the 2020 COVID-19 pandemic has had little impact on China's overall provincial spatial carbon emission pattern.

Author contribution statement

Li Guo: Conceived and designed the experiments; Performed the experiments; Wrote the paper.

Lifang Bai: Performed the experiments.

Yixuan Liu: Analyzed and interpreted the data.

Yuzheng Yang: Contributed reagents, materials, analysis tools or data.

Funding statement

This work was supported by the Foundation of Intelligent Ecotourism Subject Group of Chongqing Three Gorges University (zhlv20221005) and the China Scholarship Council (CSC202108505061).

Data availability statement

Data included in article/supp. material/referenced in article.

Declaration of interest’s statement

The authors declare no conflict of interest.paper.