Carbon-Emission Density of Crop Production in China: Spatiotemporal Characteristics, Regional Disparities, and Convergence

Abstract

An analysis of carbon emissions of crop production provides paths for global warming mitigation. Existing studies have focused on the magnitude of the carbon emissions from crop production, which is unreasonable for inter-location comparison due to neglecting regional variations in cultivation technologies and planting scale. Different from the conventional idea, this paper estimated the carbon-emission density of crop production (CEDCP) based on carbon emissions per hectare of crop production. With the 30 Chinese provinces between 2000 and 2020 as the study area, temporal dynamics and spatial patterns of the CEDCP were explored, regional disparities of the CEDCP were discussed based on the Theil index, and the possibility of regional coordinated optimization for the CEDCP was explored by relying on the convergence tests. The results show that the average annual CEDCP in China was 1.462 t/hm², reaching a peak of 1.576 t/hm² in 2015. The national carbon-emission densities of agricultural materials, rice fields, soil management, and straw burning were 0.492 t/hm², 0.390 t/hm², 0.189 t/hm², and 0.391 t/hm², respectively. In many provinces, the CEDCP increased first and then decreased, presenting a spatial pattern of high in the eastern region and low in the western region. Regional disparities of CEDCP shrank early but expanded later, and the disparities within the western region had always contributed considerably to the overall disparities. The CEDCP had shown σ- and β- convergence in both national and regional scales, and the convergence process had positive spillover effects. These findings suggest that inter-provincial cooperation may facilitate the CEDCP to converge.

1. Introduction

Countries worldwide have taken effective carbon-reducing measures to mitigate global warming. Ranking first in carbon emissions among all countries [1], China committed to peak carbon dioxide (CO₂) emissions by 2030 and strive for carbon neutrality by 2060. It has imposed low-carbon constraints on various emission sources, and relevant sectors are accelerating the pace of emission reduction. Since the 21st century, crop production in China has been mechanized and fertilized, becoming a substantial source of carbon emissions [2]. To identify potential strategies for emission mitigation, researchers have attempted to calculate the total carbon emissions from crop production. However, crop production, the backbone of the country’s economy, is expected to supply crop products and guarantee food security. As different planting scales result from various natural conditions of provinces, it is difficult for agricultural provinces to avoid having high emissions from crop production, whereas non-agricultural ones naturally achieve the target of releasing low emissions. Comparisons based on total emissions obscure the variations in planting areas and output contribution among the provinces, making it unreasonable to assess how emissions have been reduced. The uniqueness of crop production, therefore, lies in the fact that we must discuss the carbon emissions while considering the planting areas in various provinces.

Crop production has more complicated carbon effects than most production activities, and the main approaches for estimating its emissions include field trials [3,4], life cycle assessment [5,6], and model simulation [7,8,9]. Based on life cycle assessment, many researchers applied some widely recognized emission coefficients to calculate the carbon emissions from crop production. This idea, which requires simple operation and eases regional comparison, is commonly used in national and provincial scale studies [10,11]. With different research aims, accounting based on the emission coefficients can be summarized into two perspectives: one is to focus on a certain emission source, such as rice fields [12], soil management [13], and agricultural waste [14,15,16], which is conducive to investigating the greenhouse gas emissions in specific links; the second is to measure the total emissions generated by multiple emission sources [17,18], grasping the carbon effect of the whole process of crop production. The latter perspective is more common. Several factors, including fertilizers, pesticides, mulch, electricity, and machinery, comprised early carbon accounting inventory of crop production [19,20]. With research deepening, rice fields and soil management have been gradually added to the list [21,22]. Based on survey data, Zhang et al. [23] took agricultural materials, energy consumption, soil management, rice field, and straw burning into the carbon accounting of grain production, mainly involving three kinds of crops: maize, wheat, and rice.

Carbon accounting reveals the magnitude of emissions but fails to eliminate the impact of production scale. Instead, a relative index, which has drawn attention in recent research, can be used to compare carbon emissions more effectively than an absolute index [24]. Carbon-emission intensity is a relative index of carbon emissions, reflecting the emissions per unit of output value [25,26,27]. Some researchers [28,29] analyzed the carbon-emission intensity of crop production, and others defined the carbon emissions per unit of yield as the product carbon footprint [4,23,30] and per unit of area as the farm carbon footprint [31,32] based on the idea of the functional unit. These researchers also referred to these carbon emissions as density or intensity. For example, Li et al. [30] used the carbon emissions per unit grain output to express the carbon-emission intensity, then further explored the driving factors and spatial effects of the carbon-emission intensity in China’s grain-producing provinces. Yan et al. [31] calculated the agricultural carbon emissions per unit of planting area in China between 1991 and 2012, analyzing their inflection points and spatiotemporal characteristics. Chen et al. [24] used Xianyang, a city in China, to carry out agricultural carbon accounting, finding that the emissions per hectare of cropland were 3.68 t C. Xu et al. [33] found that the farm carbon footprints of rice, wheat, and maize corresponded to 7285 ± 78, 2800 ± 222, and 2707 ± 151 kg CO₂ eq/hm², respectively. Liu et al. [34] estimated the farm carbon footprint throughout all of China between 2000 and 2015, revealing that the value was 1.34 t CE/hm² in 2000 and 1.21 t CE/hm² in 2015.

Existing research on carbon emissions of crop production has been improved and extended. There are still certain limitations, though: (1) Most country-scale studies have focused on total emissions, which scarcely consider the variations in production technologies and planting scales between provinces, which, consequently, lessens the fairness of comparison. A few research studies addressed this issue by examining the carbon-emission intensity based on yield or area. However, some researchers did not conduct comprehensive carbon accounting lists, ignoring the emissions from straw burning [31] or soil management [34], hardly measuring the carbon emissions during the entire production. Moreover, some researchers calculated the carbon-emission intensity by dividing the emissions by cropland area rather than the actual planting area. Using the area of cropland as the denominator will obscure the effects of multiple cropping and abandoned cropland, as western China has many mountainous and abandoned croplands while southern China has a common practice of multiple cropping. There are also studies [4,23] focusing on a particular crop, which only comprehends the carbon-emission intensity of a specific crop and cannot accurately reflect the overall situation of crop production. (2) While previous research has made efforts to analyze the carbon emissions from crop production at the national level, extensive characterization of provincial emission composition is lacking. The outcomes have not yet met the demands for targeted emission reduction. (3) Convergence tests and disparities analyses are quite prevalent in the field of carbon emissions from crop production, but they are less seen in the case of derived indices like farm carbon footprint, product carbon footprint, and carbon-emission intensity. Most research discovered that absolute convergence does not occur in overall emissions, that is, regional disparities can hardly ever disappear naturally. However, the potential effects of planting scale and production conditions have been diminished in the case of the relative index of carbon emissions, facilitating the convergence formation. If convergence is proven to exist, regions may come up with a coordinated strategy to control the emissions.

Instead of using the output value, crop yield, or cropland area as the denominator to measure carbon-emission intensity as is done in most studies, this study uses the actual crop planting area as the basis and defines the outcome as the carbon-emission density of crop production (CEDCP). The CEDCP reflects the carbon emissions caused by crop production in a unit of area, eliminating the variation brought on by planting scale and ensuring the validity of inter-location comparisons. Based on this, we calculated the CEDCP of the 30 provinces in China between 2000 and 2020, taking four emission sources into account: agricultural materials, rice fields, soil management, and straw burning. Then, we examined the CEDCP’s spatial pattern and temporal evolution at the national and provincial levels, as well as its specific carbon-emission composition. The Theil index-based regional disparities of CEDCP were explored, and convergence tests were used to investigate the feasibility of regionally coordinated optimization. The study expands the research of relative indices of carbon emissions and contributes to mitigating carbon emissions from the standpoint of crop production.

2. Approaches and Data

2.1. Calculations of the CEDCP

This paper defines the CEDCP as the carbon emissions generated during crop production per unit area, involving three main gases: CO₂, methane (CH₄), and nitrous oxide (N₂O). The emissions are typically called carbon emissions since they can be converted to carbon equivalents based on global warming potential (GWP). Equation (1) states that the CEDCP is calculated by dividing the total carbon emission of crop production by the area planted with crops:

$$CEDCP=\frac{E}{L}$$

(1)

where E denotes the carbon emissions from crop production, L is the planting area of crops. It is necessary to identify individual emission sources and activity data to estimate the overall carbon emissions from crop production, and the specific equations and carbon coefficients can be referred to the research [20,35,36,37,38] and also in the Supplementary Materials. Four emission sources—agricultural materials, rice fields, soil management, and straw burning—were taken into account. It is possible to determine the carbon-emission density (CED) of each emission source, and the sum of them is CEDCP, as expressed by Equation (2):

$$CEDCP=CE{D}_{\mathrm{materials}}+CE{D}_{\mathrm{rice}}+CE{D}_{\mathrm{soil}}+CE{D}_{\mathrm{straw}}$$

(2)

where CED_materials refers to the CED of agricultural materials, CED_rice refers to the CED of rice fields, CED_soil refers to the CED of soil management, and CED_straw refers to the CED of straw burning. During the calculation, CH₄ and N₂O were converted into carbon equivalents according to GWP [38], as expressed by Equations (3) and (4).

$${GWP}_{{\mathrm{CH}}_4\mathrm{to} \mathrm{CE}}=9.2727$$

(3)

$${GWP}_{{\mathrm{N}}_2\mathrm{O} \mathrm{to} \mathrm{CE}}=81.2727$$

(4)

2.2. Theil Index

Theil index assists in clarifying the contributions of inter- and intra-regional disparity in regional disparity. In this paper, the Theil index is employed to analyze the spatial disparity in the CEDCP, which is expressed as Equations (5)–(8):

$$T=T_W+T_B$$

(5)

$$T_p={\displaystyle \sum _{i=1}^{n_p}\frac{1}{n_p}}\cdot \left(\frac{e_i}{{\overline{e}}_p}\right)\cdot \ln \left(\frac{e_i}{\overline{e_p}}\right)$$

(6)

$$T_W={\displaystyle \sum _{p=1}^m\left(\frac{n_p}{n}\cdot \frac{\overline{e_p}}{\overline{e}}\right)}\cdot T_p$$

(7)

$$T_B={\displaystyle \sum _{P=1}^m\frac{n_p}{n}}\cdot \left(\frac{\overline{e_p}}{\overline{e}}\right)\cdot \ln \left(\frac{\overline{e_p}}{\overline{e}}\right)$$

(8)

where T_W, T_B, and T denote the intra-regional, inter-regional, and total Theil index of the CEDCP, respectively. T_p denotes the Theil index of region p. m is the number of regional clusters. n_p is the number of provinces belonging to region p. n is the number of provinces. e_i, e_p and $\overline{e}$ are the CEDCP in province i, region p, and the entire country, respectively. T ranges in value from 0 to 1, and the higher T, the greater the regional disparity.

2.3. Convergence Test

2.3.1. σ-Convergence Test

Can the CEDCP of provinces eventually achieve the same steady-state level? σ-convergence test can explore the development of regional disparities in the CEDCP to provide an answer to this question. The σ-convergence indicates a tendency for the regional disparities to decline over time, reflecting the deviation of the provincial level from the overall average. The equation is shown in Equation (9).

$${\sigma }_t=\frac{\sqrt{\left[{\displaystyle \sum _{i=1}^n{\left(CEDC{P}_{it}-\overline{CEDC{P}_t}\right)}^2}\right]/n}}{\overline{CEDC{P}_t}}$$

(9)

where CEDCP_it is the CEDCP in province i in period t. $\overline{CEDC{P}_t}$ is the average CEDCP of all provinces in period t. n is the number of provinces. It is possible to determine σ_t for each year, and if the value falls with time, there is σ-convergence in CEDCP.

2.3.2. Conditional β-Convergence Test with Spatial Effects Considered

β-convergence indicates that the growth rate of CEDCP is negatively correlated with the initial level, including two types: absolute and conditional. Absolute β-convergence assumes that all provinces possess the same production base, such as climate conditions and planting structure, and the CEDCP will eventually converge to the same steady state. In comparison, conditional β-convergence takes the provincial heterogeneity into account, and the CEDCP of provinces will reach the local steady state instead of the same level, which is more realistic than absolute convergence. Therefore, the conditional β-convergence test is adopted in this study, as shown in Equation (10).

$$\ln \left(\frac{CEDC{P}_{it}}{CEDC{P}_{i,t-1}}\right)=\beta \ln \left(CEDC{P}_{i,t-1}\right)+\gamma X+c_i+{\eta }_t+{\epsilon }_{it}$$

(10)

where CEDCP_it and CEDCP_i,t−1 represent the CEDCP in province i in year t and year t−1, respectively; X denotes control variables that may affect the convergence, and γ is the coefficient of the control variables; c_i and η_t are the individual effect and time effect, respectively; ε_it is the error term. If β is negative significantly, the existence of conditional β-convergence is verified.

The possible spatial correlation of CEDCP does not meet the assumption of conventional econometrics. To ensure the reliability of the model, the Lagrange multiplier (LM) test is applied to determine the spatial correlation of the residuals based on the ordinary least squares (OLS). If a spatial correlation exists, spatial econometric models, such as spatial lag model (SAR), spatial error model (SEM), and spatial Durbin model (SDM), are appropriate to use in the conditional β-convergence test. SDM is utilized as the benchmark since it is the general form of SAR and SEM, and the relevant equation of the convergence test is represented as Equation (11).

$$\ln \left(\frac{CEDC{P}_{it}}{CEDC{P}_{i,t-1}}\right)=\rho W\ln \left(\frac{CEDC{P}_{it}}{CEDC{P}_{i,t-1}}\right)+\beta \ln \left(CEDC{P}_{i,t-1}\right)+\lambda W\ln \left(CEDC{P}_{i,t-1}\right)+\gamma X+\phi WX+c_i+{\eta }_t+{\epsilon }_{it}$$

(11)

where, ρ is the spatial autoregressive coefficient, showing the interaction of growth rates of CEDCP among provinces. W is the spatial weight matrix, which is formed by the inverse of the squared distance among the geographical center of provinces in this study. λ reflects the spillover effect of the CEDCP, and φ denotes the spillover effect of the control variable. The meanings of other variables are consistent with Equation (10). The SDM is suitable to be estimated by the method of maximum likelihood. In addition, Wald and likelihood ratio (LR) tests can determine whether the SDM should be simplified to SAR or SEM, and Hausman test can provide information for selecting fixed effect model or random effect model.

The convergence speed, v, describes how fast the CEDCP tends to a steady state in different regions throughout period T, while the semi-convergence period, τ, represents the time elapsed when the initial gap is reduced by 50%. The two indicators can be calculated by Equations (12) and (13), respectively:

$$v=-\ln (1-\left|\beta \right|)/T$$

(12)

$$\tau =\ln 2/v$$

(13)

According to the model setting of the conditional β-convergence test, the logarithm of the growth rate of CEDCP is the dependent variable, and the logarithm of CEDCP is the core independent variable. Several control variables concerning the environment and characteristics of crop production were introduced.

Table 1. Descriptive statistical analysis of explanatory variables in the conditional β-convergence test.

Variable	Symbol	Calculation	Mean	Std. Dev.	Min	Max
Cropland per person (hm2·person−1)	area	Cropland area/employees of planting sector	0.936	0.661	0.161	3.410
Multiple cropping	multi	Crops sown area/cropland area	1.428	0.511	0.421	2.848
Cropping structure	structure	Grain area/crops sown area	0.659	0.130	0.354	0.971
Agricultural structure	agriculture	Output value of planting/output value of agriculture	0.524	0.086	0.339	0.740
Disaster	disaster	Disaster area/crops sown area	0.238	0.162	0.002	0.936
Fiscal support to agriculture	fiscal	Fiscal expenditure for agriculture/total fiscal expenditure	0.089	0.041	0.012	0.190

presents the descriptive statistical analysis of the explanatory variables.

2.4. Study Area and Data

The study area covered 30 provinces in mainland China, while Tibet, Hong Kong, and Macao were not included in the study due to data lacking. According to the conventional zoning criteria, the study area was divided into four regions (Figure 1).

There are notable spatial differences in the distribution of cropland across China. For instance, five provinces, namely Heilongjiang, Inner Mongolia, Henan, Jilin, and Xinjiang, account for the largest share of cropland, with a combined coverage of 40%. Conversely, Tianjin, Beijing, Shanghai, Qinghai, and Ningxia have relatively smaller areas of cropland. Northern regions of the country face harsh climatic conditions, such as lower winter temperatures, decreased rainfall, and drier weather in Inner Mongolia, the Loess Plateau region, Gansu, and Xinjiang. Conversely, southern parts of China enjoy more abundant rainfall, particularly in the middle and lower reaches of the Yangtze River and southwest China. These distinct climate zones influence crop production patterns, with cereals representing 40–70% of provincial crop structures, while vegetables make up 0–30% of crops. Since different climatic environments and cropland conditions result in diverse crop production patterns, it is crucial to discuss the CEDCP rather than total carbon emissions.

The study required data involved in calculating CEDCP and control variables in the conditional β-convergence test. They were obtained from the China Statistical Yearbook, China Rural Statistical Yearbook, and provincial statistical yearbooks.

3. Results and Analysis

3.1. Spatiotemporal Characteristics of CEDCP in China

3.1.1. Temporal Evolution of CEDCP in China

CEDCP in China from 2000 to 2020 was calculated, and a graph was created to depict its development and composition (Figure 2).

The four emission sources showed both parallels and variances in the evolution of CED. The CED of agricultural materials had a yearly average value of 0.492 t/hm², first rising and then falling. It was 0.384 t/hm² in 2000 and continued to grow steadily until it peaked at 0.548 t/hm² in 2014 before falling to 0.470 t/hm² in 2020. The material-driven development of agriculture in China is reflected in this trend. Rice fields had an annual average CED of 0.390 t/hm². In the early stages of development, there were noticeable fluctuations. The CED of rice fields decreased from 0.409 t/hm² to 0.370 t/hm² between 2000 and 2003. It subsequently bounced back, increasing to 0.402 t/hm² in 2006 and eventually falling to 0.367 t/hm² in 2020. The annual average CED for soil management was only 0.189 t/hm². Its inverted V-shaped development saw a steady rise from 0.171 t/hm² in 2000 to 0.203 t/hm² in 2012 before gradually declining to 0.169 t/hm² in 2020. Straw burning resulted in an annual CED of 0.391 t/hm². Its early development resembled the CED of rice fields. It displayed a negative trend between 2000 and 2003, falling from 0.317 t/hm² to 0.298 t/hm². In 2004, it recovered and grew rapidly to 0.448 t/hm² in 2017, where it later stabilized. In conclusion, the CED of the other three emission sources had all demonstrated a consistent downward trend except straw burning, which remained stable at its plateau.

The CEDCP in China increased by 0.628% annually between 2000 and 2020, averaging 1.462 t/hm². The CEDCP went through three stages of development: from 2000 to 2003, it ranged between 1.27 t/hm² and 1.28 t/hm². Peasants had insufficient motivation to produce because of the poor income from crop output during the time, and, consequently, the carbon emissions throughout the entire process from sowing to waste treatment were reduced. The CEDCP started to rise in 2004 and peaked in 2015 at 1.576 t/hm². A general rising tendency was seen in both the crops sown area and carbon emissions at this point. The former climbed from 153.321 million hm² to 166.596 million hm², while the latter soared by 207.956 Mt to a peak of 262.648 Mt. It is speculated that since 2004, the Chinese government had begun to exempt agricultural taxes, increased financial investment in agricultural production, and gradually introduced a number of peasant-friendly measures. These actions successfully boosted peasants’ ardor for production. The resources going into agriculture had gradually increased, as had the amount of energy consumed, and the carbon emissions from activities like burning straw and cultivating rice had also increased. The CEDCP gradually decreased after 2015, reaching 1.452 t/hm² in 2020. The total emissions decreased throughout this period as well, reaching 242.861 Mt in 2020, although the area of crops cultivated remained at almost 166 million hm².

It demonstrates how the increasing speed of the CEDCP has been successfully regulated since the establishment of ecological protection, energy conservation, and emission reduction regulations in recent years. When comparing the growth of the total emissions and the crop areas between 2000 and 2015, they both declined first and subsequently increased, with the exception of the final five years showing variances (2016–2020). CEDCP, as the ratio of the two, had practically the same developments with the total emissions throughout the study period, indicating that the carbon emissions expanded faster than the scale of crop planting in the first 16 years. With 2015 being an obvious watershed, the CEDCP began to decline after that year due to the increasing attention to green crop production and low-carbon targets.

3.1.2. Spatial Pattern of CEDCP in China

To explore the situation of CEDCP at the provincial level, we mapped the spatial pattern of the total amount, density, and composition of carbon emissions of crop production in China (Figure 3).

According to the average CEDCP, the provinces with the highest CED for agricultural materials, rice fields, soil management, and burning straw were Zhejiang (1.041 t/hm²), Jiangxi (1.278 t/hm²), Fujian (0.497 t/hm²), and Hunan (0.760 t/hm²), respectively. In contrast, those with the lowest CED for the four sources of emissions were Guizhou (0.218 t/hm²), Qinghai (0 t/hm²), Qinghai (0.058 t/hm²), Guizhou (0.027 t/hm²), in turn.

The provinces with over 50% emissions from agricultural materials included Beijing, Tianjin, Hebei, Inner Mongolia, Shandong, Shaanxi, Gansu, Qinghai, and Xinjiang, primarily located in northern coastal and northwestern China. Agricultural materials in other provinces contributed 10–50% to their own total emissions. Rice fields contributed close to 50% of the total emissions in Hunan and Jiangxi due to their significant role in the agricultural structure. The proportion of soil management was not high in all provinces, with the highest value in Guizhou (21%) and the lowest one in Jiangxi (6%). In conclusion, the four emission sources contributed differently to each province, demanding consideration of provincial peculiarities in order to execute distinct strategies for emission mitigation.

As for the average annual emissions, Hunan was the only province with emissions above 20 Mt (20.550 Mt). As a rice-producing province in China, Hunan generated substantial methane carbon emissions from rice fields and relied heavily on high-carbon agricultural materials like fertilizers and pesticides. Moreover, over 40% of straw was burned in open fields, leading to a CEDCP that surpassed that of other grain-producing provinces. The provinces with emissions between 16 and 20 Mt were Henan (19.239 Mt), Anhui (17.087 Mt), and Jiangsu. Average yearly emissions from five provinces, including Shandong and Hubei, were dispersed over 12–16 Mt. Eleven provinces, including Hebei, Jilin, and Yunnan, have average annual emissions at the level of 4–12 Mt. The other ten locations all have yearly emissions of less than 4 Mt, with Shanghai, Tianjin, Beijing, and Qinghai having emissions even lower than 1 Mt.

Observing the average CEDCP, Guangdong was far ahead among provinces with a CEDCP of 2.726 t/hm², followed by Hainan, Hunan, and Fujian, all of which had CEDCP over 2.5 t/hm². In Zhejiang, Jiangxi, Jiangsu, and Shanghai, the CEDCP were greater than 2 t/hm², but they were all less than 1 t/hm² in Qinghai, Guizhou, Inner Mongolia, Gansu, Ningxia, Xinjiang, and Shanxi, and varied between 1 and 2 t/hm² in other provinces. In many provinces, the CEDCP throughout time climbed initially and decreased later. Additionally, areas with high CEDCP were consistently found in eastern China, whereas those with low CEDCP persistently agglomerated in western China.

3.2. Regional Disparities in CEDCP in China

To discover the regional disparities of CEDCP, we calculated the total Theil index, the Theil index within regions, and the Theil index between regions. The results are shown in Figure 4.

The total Theil index was 0.132 in 2000. It then displayed an erratic declining pattern, hitting a low of 0.092 in 2013, before gradually rising after that. According to this trend, the regional variations in CEDCP initially diminished and then grew. The total Theil index in 2020 was much lower than it was in 2000, indicating that overall provincial disparities in CEDCP had narrowed. Comparing the Theil index inside and between areas in 2000, intraregional disparities contributed 56.0%, and interregional disparities contributed 44.0%. The former’s contribution declined with time, while the latter’s performed oppositely. Interregional disparities had increased in proportion to 53.8% by 2020, and the share of intraregional disparities had decreased to 46.2% by that time.

Observing the average share of intraregional disparities, the rate reached 22.4% in the western region, ranking second (12.8%) in the eastern region, third (12.0%) in the central region, and last (3.5%) in the western. The provincial disparities in the western region contributed 24.0% to the overall disparities, decreasing to 22.6% in 2020. Similarly, the central region’s contribution fell from 15.5% in 2000 to 9.1% in 2020. Even while the two remaining regions’ contributions periodically changed, they generally showed a downward tendency, particularly in the northeastern region, where the share of intraregional disparities that contributed to total disparities fell to 1.7% in 2020. Overall, the variances within the western area had always been a significant contributor to the national disparities in CEDCP.

3.3. Convergence Tests for CEDCP in China

3.3.1. σ-Convergence Tests

The σ-convergence test results are displayed in Figure 5.

The σ-coefficients of the eastern, northeastern, central, and western regions in 2000 were 0.271, 0.053, 0.166, and 0.349, respectively. It demonstrates that the internal disparities were greatest in the western region, followed by the eastern and central regions, and were minimal in the northeastern region. The σ-coefficient in the eastern region initially reduced and then increased over time, the northeastern region saw oscillations in its reduction, and the central region displayed a gradual declining trend. Similar to the σ-coefficient of the eastern region, that of the western region decreased first and then increased. The σ-coefficients of the four regions in 2020 were 0.248, 0.034, 0.110, and 0.301 in turn, and the corresponding declines were 0.023, 0.019, 0.056, and 0.048, indicating that σ-convergence existed in all regions. A rebound of the σ-coefficient in the western region in recent years revealed a slight dispersion in CEDCP.

The national σ-coefficient began with a value of 0.544 in 2000 and fluctuated at first, dropping to its lowest point of 0.464 in 2008. Following that, it marginally recovered, reverting to the range of 0.47 to 0.48 and fluctuating nearby, reaching a value of 0.476 in 2020. According to the transition process, the national CEDCP converged initially, but eventually reached a standstill and the gap between the provinces remained balanced. The fact that there was σ-convergence across the entire nation was confirmed by the difference in σ-coefficients between the beginning and end. The CEDCP in the western region needs to be particularly optimized to preserve the national σ-convergence, though, as indicated by the minor recovery of σ-coefficients.

3.3.2. β-Convergence Tests

As σ-Convergence of CEDCP has been verified, how did the convergence speed perform? To address this question, we performed β-convergence testing for CEDCP across the nation and in four different regions. We also took into account potential spatial effects that may influence the convergence process, given that there is a spatial correlation in CEDCP according to the locations of provinces due to similarities in climate conditions and planting practices [34,39]. Using Stata 15.1 software, we performed original least square regression on the convergence equations and then conducted LM tests and robust LM tests on the residuals to determine the specific model. The results are presented in

Table 2. Statistical tests for model selection.

Test	Eastern		Northeastern		Central		Western		National
	Statistics	p	Statistics	p	Statistics	p	Statistics	p	Statistics	p
Moran’s I	1.746	0.081	4.016	0.000	3.637	0.000	3.966	0.000	8.203	0.000
LM Error (Burridge)	2.295	0.130	13.466	0.000	10.989	0.001	13.693	0.000	62.55	0.000
LM Error (Robust)	0.042	0.838	1.234	0.267	3.586	0.058	0.388	0.534	0.238	0.626
LM Lag (Anselin)	2.71	0.100	12.234	0.000	18.831	0.000	16.8	0.000	72.851	0.000
LM Lag (Robust)	0.457	0.499	0.002	0.965	11.428	0.001	3.494	0.062	10.539	0.001
Spatial econometric model	No		Yes		Yes		Yes		Yes
Wald-SAR			32.44 ***		10.21		20.31 ***		50.99 ***
Wald-SEM			13.93 *		12.66 *		16.23 **		40.91 ***
LR-SAR			24.80 ***		9.98		19.69 ***		49.07 ***
LR-SEM			14.32 **		13.40 *		17.11 **		42.54 ***
Hausman test	23.64 ***		24.37 ***		25.07 ***		30.46 ***		57.91 ***
Model	FE		SDM-FE		SDM-FE		SDM-FE		SDM-FE

Note: * means that the coefficients passed the z-test at a 10% significance level, ** means that the coefficients passed the z-test at a 5% significance level, and *** means that the coefficients passed the z-test at a 1% significance level.

.

The convergence equation of the eastern region cannot pass the spatial correlation test, requiring the application of an ordinary panel regression model. The statistics of other equations, in contrast, passed the LM test or robust LM test (p < 0.05), confirming the existence of both the spatial lag effect and the spatial error effect and necessitating the use of the SDM. According to Wald and LR tests, the SDM could not be converted into either a SAR or a SEM. All equations also significantly passed the Hausman test and were appropriate for a fixed-effect model. The results of the conditional β-convergence test are shown in

Table 3. Conditional β-convergence test results of the CEDCP.

Variables	Eastern		Northeastern		Central		Western		National
	Coefficients	z	Coefficients	z	Coefficients	z	Coefficients	z	Coefficients	z
β	−0.094 ***	−4.13	−0.660 ***	−5.45	−0.116 **	−1.97	−0.133 ***	−2.69	−0.154 ***	−6.69
area	0.005 ***	5.86	0.080 **	2.18	0.236 ***	3.41	−0.026	−0.84	−0.016	−1.01
multi	0.004	0.48	0.235 ***	2.87	0.128 ***	3.87	0.016	1.05	0.004	0.61
structure	−0.051	−0.63	1.322 ***	3.51	0.023	0.16	−0.066	−1.06	−0.123 **	−2.07
agriculture	−0.224	−1.03	0.226	1.01	−0.031	−0.27	0.116	0.84	−0.073	−1.37
disaster	0.060 ***	3.05	−0.030	−0.68	0.050 *	1.83	0.011	0.60	0.047 ***	3.55
fiscal	−0.131	−1.12	0.308	1.14	0.096	0.84	−0.065	−0.53	0.053	0.47
W × β			0.616 ***	4.20	0.047	0.51	0.155 ***	2.58	0.170 ***	4.58
W × area			−0.059	−1.25	−0.007	−0.06	−0.022	−0.43	0.038	0.94
W × multi			−0.196 *	−1.81	0.039	0.62	0.006	0.36	0.046 **	2.46
W × structure			−1.578 ***	−3.18	0.192	0.65	−0.213	−1.37	0.120	1.17
W × agriculture			−0.472 **	−1.99	−0.140	−0.78	−0.231	−1.41	−0.168	−1.29
W × disaster			0.068	1.33	0.035	0.80	0.089	1.50	0.019	0.55
W × fiscal			−0.398	−1.55	−0.253 *	−1.73	0.009	0.04	−0.183	−1.20
ρ			0.491 ***	6.13	0.322 ***	3.31	0.325 ***	4.25	0.331 ***	4.29
v	0.005		0.051		0.006		0.007		0.008
τ	153.145		14.021		122.335		106.285		90.236
R2	0.2094		0.4303		0.4757		0.3029		0.2881
Log-pseudolikelihood			119.7114		276.7035		421.6345		1146.8701
Observations	200		60		120		220		600

Note: * means that the coefficients passed the z-test at a 10% significance level, ** means that the coefficients passed the z-test at a 5% significance level, and *** means that the coefficients passed the z-test at a 1% significance level. The significance level for coefficients is expressed consistently with that in Table 2.

.

The β-convergence coefficients were all significantly negative (p < 0.01), indicating that conditional convergence existed and that the growth rate of CEDCP was negatively related to the initial CEDCP on both the national and regional levels. The spatial autoregressive coefficients in all test outcomes were significantly positive (p < 0.01), demonstrating that the CEDCP’s convergence had a positive spatial spillover impact. Each province’s CEDCP development was influenced by neighboring provinces’ policies and production as well as local production practices.

The CEDCP in each region has its steady state, forming β-convergence clubs. Among the four regions, the eastern region ranked last in the convergence speed with a value of 0.005, while the convergence speed in the northeastern region was far higher than that of the other regions with a value of 0.051. It is speculated that the highly homogeneous production conditions and cropping structure in the three northeastern provinces eased the convergence of CEDCP. The convergence speed was 0.006 and 0.007 in the central and western regions, respectively, and the overall convergence rate of the country (0.008) was at the average level of the four regions. According to the semi-convergence cycle, to naturally reduce the disparities of CEDCP between provinces within the four regions to half of the current level, it would take 153.145, 14.021, 122.335, and 106.285 years, respectively, while for the country as a whole, it would take 90.236 years.

4. Discussion

4.1. The Results of CEDCP

Because of the disparity in the carbon-accounting inventories and study periods, the results could vary slightly among the literature [39]. However, having different definitions and estimations with CEDCP, the idea of the studies of Yan et al. [31] and Liu et al. [34] was similar to ours. Therefore, we compared their results with our results, as Figure 6 presents.

With the study period overlapping from 2000 to 2012 in the research of Yan et al. [31] and this paper, the general trends of the two curves are consistent, but a certain gap exists. The carbon-emission intensity of crop production measured in their research increased from 0.996 t/hm² to 1.160 t/hm², and the CEDCP measured in this paper increased from 1.282 t/hm² to 1.561 t/hm², higher than their results. The reason is that Yan et al. [31] only included three emission sources—agricultural products, rice fields, and soil management—when estimating overall carbon emissions; however, this paper further included emissions from straw burning. During the corresponding period, the CED of straw burning rose from 0.317 t/hm² to 0.424 t/hm², and when it was removed from CEDCP, the CEDCP would increase from 0.965 t/hm² to 1.137 t/hm², which was more consistent with the results measured by Yan et al. [31].

There is also some discrepancy between the two curves when comparing this work to the study of Liu et al. [34] Between 2000 and 2015, the farm carbon footprint as determined by Liu et al. [34] declined with variations. Two factors account for the disparity. In their study, various emission sources, such as pesticides, mulch, energy used for irrigation, and soil management, were not included in the carbon-accounting list, which may cause the carbon emissions to be underestimated. Second, they defined the farm’s carbon footprint as the carbon emissions per hectare of cropland, which used the area of the cropland as the denominator. Early data on farmland acreage in China contain a few small inaccuracies and discrepancies since survey standards and technical procedures are limited. Therefore, this paper used the actual sown area as the denominator when measuring the CEDCP, which not only ensured data accuracy but also achieved fairness of regional comparison caused by multiple cropping and land abandonment.

According to Zhang et al. [23], the emissions per hectare for maize, wheat, and rice were 4.052 t CO₂-eq (1.105 t C-eq), 5.455 t CO₂-eq (1.488 t C-eq), and 11.881 t CO₂-eq (3.240 t C-eq), respectively. Although the carbon sources addressed in their study were comparable to those in ours, Zhang et al. [23] measured carbon emissions by crops, and, therefore, it was not appropriate to directly compare them with our results. Nevertheless, the range of the carbon-emission intensity of the three crops was close to the result in this study.

Due to the divergence in natural conditions and planting preference, crop production differs in scale and structure across provinces, resulting in a wide gap in total carbon emission. Contrarily, the CEDCP, which is less impacted by the magnitude, is better suited for provincial comparison. Comparing the total carbon emissions and CEDCP in the empirical analysis, although the overall performance is consistent, the total emissions are slightly scattered in the spatial pattern, while CEDCP showed spatial agglomeration.

4.2. The Results of Convergence Tests

Researchers have conducted convergence tests on carbon emissions from crop production. For example, Wu et al. [40] found that neither σ convergence nor conditional β-convergence appeared in agricultural carbon emissions in China between 2000 and 2014. This conclusion was different from our findings on the convergence of CEDCP. Carbon emissions are affected by many factors, such as production scale and technologies. The visible disparity in agricultural developments between provinces makes it more onerous for total emissions to converge, but the CEDCP largely removes the natural circumstances and production scale, easing the conditions of convergence.

As there was a spatial correlation of CEDCP among provinces, if the spatial interaction was ignored, the regression of the convergence test would be biased. Therefore, instead of the ordinary panel regressions applied in most studies, we employed spatial econometric models. Accordingly, the geographic location of the provinces was incorporated into the analysis, facilitating examination of how nearby provinces interacted during convergence. The outcome confirmed that the spatial interaction positively affected the convergence of the CEDCP. The addition of the spatial effect to the convergence study broadened the scope of the previous research.

4.3. Limitations and Future Directions

Limited by data availability, there are still some limitations in this study: (1) We focused on the past situation of the CEDCP in China and did not discuss the potential in the future. Exploring the possibility of reducing emissions from crop production is of great significance for the dual carbon goal. Under this assumption, the potential contribution of carbon mitigation of crop production requires forecasting. (2) We calculated the carbon emissions of crop production based on the coefficients released by the Chinese government, IPCC, and some widely referenced literature, but there was still uncertainty in the results. Through the integration of field experiments, model simulation, farm surveys, and statistical data, follow-up studies may improve the data source. The accuracy of emission coefficients is also expected to be improved for further optimizing carbon accounting.

5. Conclusions and Policy Recommendations

5.1. Conclusions

In this study, we defined the carbon-emission density of crop production (CEDCP) and calculated it for the 30 provinces in China between 2000 and 2020. Then, we examined the CEDCP’s spatial pattern and temporal evolution at the national and provincial levels. Finally, we explored the regional disparities of CEDCP and investigated the possible convergence. The conclusions are as follows:

(1)

The CEDCP in China averaged 1.462 t/hm² annually, possessing a growth rate of 0.628%. It fluctuated in 2000–2003, rose steadily in 2004–2015, reaching the highest point of 1.576 t/hm², then decreased slowly in 2016–2020. The annual CED of agricultural materials, rice paddies, soil management, and straw burning were 0.492 t/hm², 0.390 t/hm², 0.189 t/hm², and 0.391 t/hm² in order. Provinces with high CEDCP were agglomerated in the eastern region, while those with low CEDCP were mainly located in the western region.

(2)

The regional disparities in CEDCP initially declined but later expanded, as shown by the total Theil index, which first decreased and then increased. The share of inter-regional disparities in total disparities declined with time, while intra-regional disparities performed oppositely. From the intra-regional side, the western region had the highest contribution rates (22.4%) to total disparities, followed by the eastern region (12.8%), the central region (12.0%), and the western region (3.5%). The disparities within the western region have always been a major component of the national disparities of CEDCP.

(3)

σ-Convergence was seen in the CEDCP of the whole country and all regions. The β-convergence speed decreased in the order of northeastern region, western region, central region, and eastern region, and that of the whole country was at the average level of the four regions. The β-convergence process had positive spillover effects. According to the semi-convergence cycle, to naturally reduce the disparities of CEDCP between provinces within the four regions to half of the current level, it would take 153.145 years for the eastern region, 14.021 years for the northeastern region, 122.335 years for the central region, and 106.285 years for the western region, while for the country as a whole, it would take 90.236 years.

5.2. Policy Recommendations

Attention should be paid to key emission sources, especially agricultural materials and straw burning. Precise application of fertilizers and pesticides is encouraged, and high-efficiency fertilizer and nitrification inhibitors are recommended. A tougher ban on the open burning of straw should be implemented, while straw returning and transiting to energy or feed are promoted. It is also suggested to continue developing new varieties of low-methane high-yield rice with effective farming measures and field management applied in rice-producing areas like Hunan and Jiangxi. Despite the recent decreasing trend in both total carbon emissions and CEDCP, the overall level is still high. It results from the expanding use of inputs and machinery in exchange for crop output, and the input-driven model should be gradually replaced by a technology-driven model.

CEDCP convergence clubs had nearly developed in each of the four regions and the entire country, but in recent years, a slight divergence was observed in the western region, necessitating caution. Without government assistance, the CEDCP would take a long period to reach overall convergence. Inter-regional differentiated and regional synergistic approaches for emission reduction should be considered because natural conditions and farming patterns vary between areas but are similar within regions. The gap could be closed by utilizing the positive spatial spillover effects occurring throughout the convergence process. It is anticipated that measures like sharing low-carbon technology and inter-provincial cooperation would hasten CEDCP’s general convergence.