The Influencing Mechanism and Spatial Effect of the Digital Economy on Agricultural Carbon Emissions

Abstract

As a progressive and systematic initiative that necessitates the collective participation of society, achieving the goals of carbon peaking and carbon neutrality has had a significant and positive impact on the transformation of the energy structure, the development of the new energy industry, the enhancement of economic efficiency and environmental quality, and the deepening of international cooperation across multiple dimensions. This study examines how the digital economy affects carbon reductions in the context of China’s pursuit of carbon peak and carbon neutrality targets. To thoroughly examine how regional digital economy development influences agricultural carbon emissions and uncover its underlying mechanism, this study uses regression analysis models using panel data from 31 Chinese provinces (not including Hong Kong, Macau, and Taiwan) from 2013 to 2022. In the meantime, the study investigates the spatial effects of the digital economy on agricultural carbon emissions. The results show that the rapid development of the digital economy plays a significant role in reducing agricultural carbon emissions. In particular, every 1 unit increase in the level of digital economy development is associated with a 0.125-unit reduction in agricultural carbon emissions. Second, the expansion of the digital economy allows regional labor transfer, which indirectly influences its suppressive effect on agricultural carbon emissions through this channel. Third, the expansion of the digital economy in one area has significant spatial spillover effects, leading to agricultural carbon emissions in other provinces and cities. Fourth, these spatial spillover effects vary depending on the topography and economic production. In particular, flat regions and high-yield agricultural areas see greater carbon reduction spillover effects from the digital economy compared to steep regions and low-yield agricultural areas. Therefore, research on the impact of the digital economy on agricultural carbon emissions can help to reveal the path of the digital-technology-driven green transformation of agriculture and provide a scientific basis for optimizing agricultural carbon-emission-reduction policies and achieving sustainable agricultural development.

1. Introduction

Food production is an important foundation for national security and stability and for socio-economic development. Throughout China’s thousands of years of agricultural development, a food security strategy centered on high yields has long been implemented to ensure food security and a stable supply of agricultural products. As a result, this focus has fostered a production model dependent on agrochemicals, resource over-exploitation, and the expansion of animal husbandry, leading to diminishing marginal benefits of agricultural production and increased carbon emissions. According to the Food and Agriculture Organization of the United Nations (FAO), agroforestry activities and the food industry chain account for 31% of global greenhouse gas (GHG) emissions [1]. Systematically reducing carbon emissions has therefore become a common problem for the international community [2,3]. As the world’s largest developing country, 14% of China’s total carbon dioxide emissions come from the agri-food system, making it imperative to solve the problem of agricultural carbon emissions [4].

With the rollout of national strategies such as “Broadband China”, “Digital Countryside”, and “Digital Agriculture”, the impact of the digital economy on the environment and other aspects is becoming the focus of academic research [5,6]. Some scholars have shown that digital technology can support carbon-emission-reduction goals by accelerating the diffusion of green innovation, driving the environmental management processes, and then optimizing the energy-allocation efficiency [7,8]. As the digital economy continues to penetrate the agricultural sector, it facilitates the upgrade of the information infrastructure of agricultural production and management, improves the efficiency of resource use and the application of agricultural machinery and technologies, and enables the transition toward greener production practices. In view of this, by integrating the technological potential of the digital economy, we can synergize the promotion of agricultural carbon emission reductions and industrial upgrading, which has the dual value of theoretical innovation and practical breakthroughs.

Most domestic and international studies on agricultural carbon emissions have focused on the practical level [9]. Regarding the measurement methods, current research mainly focuses on carbon emissions from animal husbandry and crop cultivation [10]. Foreign scholars were among the first to identify the sources of agricultural carbon emissions [11,12], including fertilizers, pesticides, rice cultivation, animal husbandry, and agricultural waste. Domestic scholars initially focused on the six key factors, pesticides, agricultural films, fertilizers, irrigation, agricultural diesel, and cropland [13], and used carbon emission factors to estimate agricultural carbon emissions. Later, some scholars focused on measuring carbon emissions from rice cultivation [14], livestock manure [15] energy use and agricultural inputs [16], and calculating the intensity of agricultural carbon emissions based on the proportion of the crop area [17] or the agricultural output value [18]. They usually use methods such as the discounting method [19] or carbon emission factor method [20] to determine the CO₂ emissions. Regarding the relationship between the digital economy and agricultural carbon emissions, scholars at home and abroad have found that the digital economy significantly inhibits agricultural carbon emissions [21,22]. This is primarily attributed to the construction of digital villages, which improves economies of scale and optimizes resource allocation through advanced digital technologies [23]. These technologies enable a precise agricultural analysis, refine the management of agricultural production and processing, and help to reduce resource and energy waste in agricultural production [24]. By solving the problems caused by operational fragmentation, the digital economy also serves as a catalyst for improving total factor productivity and promoting sustainable agricultural development [25]. At the same time, the digital economy promotes the innovation and application of green technologies, such as renewable energy technologies and bio-environmental technologies [24], which can effectively integrate resources at all stages of agricultural production and enhance the transparency within the agricultural supply chain [26]. These advancements have reduced traditional energy consumption and lowered agricultural carbon emissions. Numerous scholars have delved into how the digital economy empowers high-quality agriculture and scrutinized its driving mechanisms [27,28]. They also empirically tested the digital economy’s positive inhibitory effect on agricultural carbon emissions by taking the regional economic development level, agricultural machinery technology progress, and other factors as mechanism variables. Other scholars have approached the topic from a spatial correlation perspective, finding that the data elements are able to break the spatial constraints and show strong mobility [29,30]. Digital economy development in a region not only suppresses its own agricultural carbon emission intensity but also produces spatial spillover effects on neighboring regions [31]. These spillovers show obvious spatial heterogeneity, following the pattern of “east high and west low” [32]. In addition, some scholars have found that the relationship between the digital economy and carbon emissions is not simply linear but rather complex and nonlinear in nature [33,34]. So far, scholars at home and abroad have performed in-depth and multi-dimensional evaluations of the role of technological innovation in reducing agricultural carbon emissions [35,36,37].

Current research on the impact of the digital economy on agricultural carbon emissions still has several limitations. First, the theoretical framework remains underdeveloped, with an insufficient exploration of farmers’ individual behavior from a micro perspective. Second, the analysis of how the digital economy promotes the reduction of agricultural carbon emissions tends to be superficial, indicating a need for more thorough research. Third, although some scholars have incorporated the two into a unified framework from a spatial relevance perspective, there is a relative lack of comparisons regarding the spatial spillover effect heterogeneity of the digital economy’s role in reducing agricultural carbon emissions. Filling these research gaps is of great significance for constructing a more systematic theoretical framework, guiding the practice of agricultural carbon emission reductions, and formulating global policies on agricultural carbon emission reductions.

In view of this, the unique contributions of this study are as follows: first, this paper adopts a multidimensional index system to measure the development indices of the digital economy and agricultural carbon emissions, incorporating both into a comprehensive framework. Second, it chooses labor force transfer as a mechanism variable to explore its effect on agricultural carbon emissions, thereby offering a powerful complement to existing research. Finally, this paper establishes a spatial panel model to systematically investigate the presence and heterogeneity of the spillover effect arising from the digital economy’s influence on agricultural carbon emissions, providing a useful reference for policy formulation.

2. Theoretical Analysis and Research Hypotheses

The Cobb–Douglas production function identifies labor and technology as the two main determinants of input–output connections. As a new production factor, the digital economy will unavoidably reshape the allocation of traditional production factors when integrated into agricultural production, promoting the further optimization and upgrading of production factors. The digital economy is characterized by two key components: first, it emphasizes the gathering, storing, processing, analyzing, and use of digital data; second, it integrates digital technologies like cloud computing and precision identification. Its entry into the agricultural sector represents not only the emergence of new, modern agricultural industries with distinctive characteristics but also the slow modification of conventional agricultural techniques [38]. By integrating and analyzing big data resources in agricultural production, this model provides a scientific foundation for agricultural decision-making, thereby propelling the sector toward greater informatization and intelligence. Technologies such as climate monitoring and satellite imagery have enabled precision farmland management, including targeted fertilization and disaster early warning. Meanwhile, advances in satellite positioning systems, remote sensing, geographic information systems, and high-precision sensors have facilitated the widespread adoption of smart agricultural machinery. Completely automated farming technology advancements have decreased needless soil disturbance, reducing possible effects on soil carbon reserves and offering a practical means of lowering agricultural carbon emissions. Second, the digital economy has sped up the development of the rural digital infrastructure, narrowing the digital divide between urban and rural areas and encouraging the adoption of eco-friendly and green ideas. As agricultural production and processing become increasingly digitized, the integration of data factors with conventional production elements raises the marginal return, creating amplification, synergy, and multiplier effects that drive agricultural economic growth [39,40]. These effects, in turn, contribute directly to the reduction of agricultural carbon emissions. In addition to creating platforms for environmental services, the expansion of digital inclusive finance has given farmers financial support, which helps to improve carbon sequestration and changes in agricultural production practices, thus reducing agricultural carbon emissions. Based on the above discussion, Hypothesis 1 is proposed.

H1: 

Carbon emissions from agriculture are inhibited by the expansion of the digital economy.

In 1954, Lewis’s dual economy model revealed a trend: labor tends to shift from traditional agricultural production to more productive modern industries [41]. The development of the digital economy has brought new vitality to the labor market, especially in emerging areas, such as the Internet, e-commerce, software development, and telework. As urbanization accelerates and the tertiary industry thrives, an increasing number of rural laborers have been drawn to cities and non-agricultural industries by the promise of more employment opportunities and higher income levels, thereby accelerating the transformation of the rural labor force. Meanwhile, the deep penetration of the digital economy has profoundly reshaped the agricultural production model. The widespread application of intelligent and automated agricultural tools has contributed to the transformation of traditional extensive agriculture to intensive agriculture, promoting the more refined agricultural management of agriculture, significantly improving agricultural production efficiency, and reducing the labor demand. Similarly, the reduction in agricultural labor has objectively accelerated agricultural mechanization, replacing human and animal power to a certain extent. This transformation not only significantly improves production efficiency but also further promotes the optimal allocation and transformation of rural labor.

At the same time, in light of China’s “red line” policy on food security and arable land, the State will encourage highly skilled agricultural personnel to return to rural areas through digital agricultural training programs and other means. This approach aims to support agricultural and rural development and achieve talent revitalization in the countryside. Specifically, the transfer of high-quality labor into the low-carbon emission agricultural industry improves production efficiency and optimizes resource allocation. These improvements not only drive green technological innovation in agriculture but also help reduce carbon emissions. Mechanization can reduce the carbon emissions caused by improper or inefficient human operation. In addition, the transfer of high-skilled labor to rural areas will lead to changes in cropping structures and facilitate the extension and upgrading of the industrial chain. These changes are conducive to reducing carbon emissions and promoting sustainable agricultural development. Therefore, through the rational allocation of labor resources, it can effectively promote the adjustment and optimization of the regional economic structure and has a positive impact on agricultural carbon emission reductions [42]. However, it is worth noting that Huang et al. point out that the labor force size, to a certain extent, constrains agricultural carbon emission reductions [13]. In this context, investigating how the digital economy affects agricultural carbon emissions by promoting rural labor force transfer is of far-reaching significance and practical value for promoting the rural digital economy, thus reaching the goal of reducing agricultural carbon emissions. We put out Hypothesis 2 in light of the previously described analyses.

H2: 

The growth of the digital economy can promote labor mobility, which indirectly influences agricultural carbon emission reductions.

According to the First Law of Geography, everything is intrinsically interconnected, and the tightness of linkages and the efficiency of factor mobility are greatly increased by increasing the distance. There is a certain degree of synergy in the economic development and manufacturing activities of neighboring regions. Given the trans boundary nature of the digital economy and the external characteristics associated with carbon emissions, interregional factor transfers are accelerated, increasing the dynamic and changing nature of cooperative and competitive partnerships. Moreover, the digital economy’s phased development causes adjacent regions’ digital economies to influence local agricultural carbon emissions, which shows up in three primary ways: 1. significant peer effects, agricultural carbon emissions in one region are often influenced by those in surrounding, especially neighboring, regions, serving as a critical reference for adjustment and optimization [42]. This is primarily due to the shared environmental conditions and regulatory frameworks among neighboring regions, which intensify competition among regions with similar levels of economic development. To achieve good results in both economic and environmental aspects, governments often draw on the successful experiences of nearby regions in leveraging the digital economy to enhance agricultural carbon emission management. They deepen the local application of digital technologies and flexibly adjust emission-reduction strategies, striving to advance toward a new phase of “economy-environment” harmonious development alongside neighboring regions. 2. Prominent learning effects: the progress of the digital economy has facilitated the widespread dissemination of knowledge, technologies, and experiences related to low-carbon agricultural production [43]. Geographical proximity fosters the flow of information between neighboring regions and farmers, enabling the adoption of advanced technologies, thereby effectively decreasing carbon emissions. Meanwhile, digital platforms provide convenient channels for technology sharing, helping to address challenges in development. However, it is worth noting that competitive barriers between neighboring regions may also hinder technological spillovers. 3. Significant diffusion effects: due to the substantial initial investment of the digital economy, it often emerges first in developed regions, creating a demonstration effect. According to the “core-periphery” theory, surrounding regions will actively imitate and learn from these developments, promoting the widespread spillover of digital technologies and knowledge. The acceleration of this process, enabled by digital media [44], improves efficiency and extends the benefits to a broader range of regions, further enhancing agricultural carbon reduction. Moreover, the progress of the digital economy in neighboring regions expands agricultural markets. While this may increase input in production factors, it also raises consumer awareness of agricultural product quality, driving the popularization of green production methods and indirectly improving carbon emission efficiency. In light of this, Hypothesis 3 is put out in this paper:

H3: 

Spatial spillover effects of the digital economy affect agricultural carbon emissions.

Figure 1 illustrates the mechanisms of the impact of digital economy on agricultural carbon emissions.

3. Research Design

3.1. Variable Selection and Measurement

3.1.1. Dependent Variable: Agricultural Carbon Emission Intensity (Aci)

This study focuses on crop farming as the research subject for agricultural carbon emissions. It selects six carbon sources (agricultural diesel, fertilizers, pesticides, agricultural films, irrigation, and plowing) to calculate carbon emissions. The total agricultural carbon emissions (TAs) are represented by summing these sources. The formula for calculation is as follows:

$$TA={\displaystyle \sum A_{\mathrm{i}}}={{\displaystyle \sum Q}}_{\mathrm{i}}\times {\sigma }_{\mathrm{i}}$$

(1)

where $TA$ represents the total agricultural carbon emissions, $A_{\mathrm{i}}$ denotes the absolute statistics of the i-th carbon source, and $Q_{\mathrm{i}}$ is the coefficient for the i-th carbon source (details in Table 1).

The logarithmic expression of agricultural carbon emission intensity is calculated by dividing the total carbon emissions from agricultural activities by the regional GDP [45]. The calculation formula is as follows:

$$A\mathrm{ci}=L\mathrm{n}\left({\displaystyle \frac{TA}{GDP}}\right)$$

(2)

where $A\mathrm{ci}$ is the agricultural carbon emission intensity, and $GDP$ represents the gross domestic product.

Table 1. Sources of agricultural carbon emissions and their coefficients.

Carbon Sources	Carbon Emission Coefficients	Reference Sources
Agricultural diesel	0.59 kg/kg	IPCC2013
Agricultural fertilizers	0.89 kg/kg	Oak Ridge National Laboratory, Oak Ridge, TN, USA
Pesticides	4.93 kg/kg	Oak Ridge National Laboratory, Oak Ridge, TN, USA
Agricultural film	5.18 kg/kg	Institute of Resources, Ecology and Environment of Agriculture, Nanjing Agricultural University, Nanjing, China
Agricultural irrigation	266.48 kg/hm2	Duan Huaping et al. [46]
Plowing	312.60 kg/km2	Li Bo et al. [44]

Table 1. Sources of agricultural carbon emissions and their coefficients.

Carbon Sources	Carbon Emission Coefficients	Reference Sources
Agricultural diesel	0.59 kg/kg	IPCC2013
Agricultural fertilizers	0.89 kg/kg	Oak Ridge National Laboratory, Oak Ridge, TN, USA
Pesticides	4.93 kg/kg	Oak Ridge National Laboratory, Oak Ridge, TN, USA
Agricultural film	5.18 kg/kg	Institute of Resources, Ecology and Environment of Agriculture, Nanjing Agricultural University, Nanjing, China
Agricultural irrigation	266.48 kg/hm2	Duan Huaping et al. [46]
Plowing	312.60 kg/km2	Li Bo et al. [44]

Given that the differences in various sources of agricultural carbon emissions may significantly impact the empirical analysis results, this paper boosts the robustness and reliability of the research conclusions by selecting alternative dependent variables for robustness testing. These include agricultural carbon emissions (Ac), livestock and livestock farming, and agricultural rice cultivation carbon emissions (Act).

In calculating carbon emissions from agricultural rice cultivation, estimates are made according to the actual rice-planting area for the year. Considering regional differences in rice cultivation seasons and growth cycles, the corresponding single-season methane emission coefficients for rice are selected for each province [47]. This study uses the stock numbers at the end of each year as baseline data and chooses significant livestock and poultry species, such as cattle, mules, goats, sheep, donkeys, camels, horses, and pigs, based on the real development of China’s livestock business [48,49]. Given that the average lifecycle of pigs is about 220 days, the calculation method for their average annual feeding amount is adjusted accordingly [50], to more accurately reflect the actual situation.

$$M_{\mathrm{i}}=D_{\mathrm{i}}\times {\displaystyle \frac{N_{\mathrm{i}}}{365}}$$

(3)

where $N_{\mathrm{i}}$ is the amount of livestock i that is slaughtered annually, $D_{\mathrm{i}}$ is the average lifetime of livestock i, and $M_{\mathrm{i}}$ is the average amount of livestock i that is fed annually. To facilitate the calculation of carbon emissions, methane and nitrous oxide are both substituted with standard carbon based on the IPCC Fourth Assessment Report.

3.1.2. Core Explanatory Variable: Digital Economy Development Level (Digl)

There is a general trend of uniformity in the methods for measuring the level of digitalization, with the evaluation systems increasingly refined over time. Nonetheless, there are still some variations in the choice of particular assessment metrics. This research uses the measuring techniques used in previous studies [31,50,51,52] and chooses indicators that reflect the degree of digitalization in “issues relating to agriculture, rural areas, and rural people”, while taking into account the actual conditions of digital economic advancement across various Chinese provinces and the viability of data collection. A thorough assessment indicator criterion is built based on three dimensions: rural digital infrastructure, digital development capabilities, and digital industrial services [53,54,55]. The entropy weight method is used for measurement. This targeted, multi-dimensional evaluation index system not only improves the reliability of research on the rural digital economy but also enhances its general applicability and policy guidance. Details are presented in

Table 2. Measurement indicator system for assessing the digital economy’s development level.

Target Layer	Criterion Layer	Indicator Layer	Nature
Digital Economy Development Indicator System	Digital Infrastructure	Network broadband coverage rate in administrative villages (%)	Positive
		Proportion of administrative villages with postal services (%)	Positive
		Number of broadband access ports (unit: 10,000)	Positive
		Length of optical cable lines (unit: 10,000 km)	Positive
		Level of digitalization in inclusive finance	Positive
	Digital Development Capabilities	Number of agricultural meteorological experiment service stations	Positive
		Taobao Villages	Positive
		E-commerce sales (unit: 100 million yuan)	Positive
		Rural electricity consumption (unit: 100 million kWh)	Positive
	Digital Industrial Services	Number of broadband users in rural areas (unit: 10,000 households)	Positive
		Number of computer users per 100 households nationwide	Positive
		Number of mobile phone users per 100 households nationwide	Positive
		Per capita transportation and communication expenses in rural areas	Positive

.

The entropy weight approach provides more objective weights for gauging the development degree of the digital economy. As a result, this study uses this measurement technique, following the following precise steps:

First, the data undergo normalization processing because the indicators’ dimensions differ. Since all the indicators in this study are positive, the following standardization formula is applied:

$${\mathrm{Y}}_{\mathrm{i}\mathrm{j}}={\displaystyle \frac{{\mathrm{X}}_{\mathrm{i}\mathrm{j}}-\min \left({\mathrm{X}}_{1\mathrm{j}},{\mathrm{X}}_{2\mathrm{j}},\mathrm{}\dots ,{\mathrm{X}}_{\mathrm{n}\mathrm{j}}\right)}{\max \left({\mathrm{X}}_{1\mathrm{j}},{\mathrm{X}}_{2\mathrm{j}},\mathrm{}\mathrm{}\dots ,{\mathrm{}\mathrm{X}}_{\mathrm{n}\mathrm{j}}\right)-\min \left({\mathrm{X}}_{1\mathrm{j}},{\mathrm{X}}_{2\mathrm{j}},\dots ,{\mathrm{X}}_{\mathrm{n}\mathrm{j}}\right)}}\mathrm{}\left(\mathrm{i}\mathrm{}=\mathrm{}1,\mathrm{}2,\mathrm{}\dots \mathrm{n}; \mathrm{}\mathrm{j}\mathrm{}=\mathrm{}\mathrm{1,2},\mathrm{}\dots \mathrm{}\mathrm{m}\right)$$

(4)

where $Y_{\mathrm{ij}}$ is the standardized value of indicator j in province i, and $X_{\mathrm{ij}}$ is the original value of indicator j in province i.

By computing the ratio of the indicator for year I under the j-th indication to the indicators at that level, the entropy weight method is used to ascertain the objective weight of the digital economy development level:

$${\mathrm{f}}_{\mathrm{i}\mathrm{j}}={\displaystyle \frac{{\mathrm{Y}}_{\mathrm{i}\mathrm{j}}}{\sum _{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{Y}}_{\mathrm{i}\mathrm{j}}}}$$

(5)

Calculating the entropy value of the j-th indicator,

where $\mathrm{k}>0$, $e_j>0$; $\mathrm{k}={\displaystyle \frac{1}{\ln 31}}$, $0\le \mathrm{e}\le 1$:

$$e_j=-{\displaystyle \frac{1}{ln\left(n\right)}}\ast \sum _{i=1}^{31}{f}_{ij} \ast ln{f}_{ij}$$

(6)

Calculating the weights of each indicator:

$$W_j={\displaystyle \frac{1-e_j}{\sum _{j=1}^{13}1-e_j}}$$

(7)

Calculating the comprehensive score:

$$S={\displaystyle \sum _{\mathrm{j}=1}^{13}{\mathrm{f}}_{ij}\times W_{\mathrm{ij}}}$$

(8)

3.1.3. Mediating Variable: Labor Transfer (It)

Labor transfer (It) is used in this study as a proxy variable to measure the level of digital economic advancement [56,57]. It is determined by the ratio of primary sector workers to rural residents. A higher value indicates a larger scale of rural labor transfer, and in the actual analysis, the logarithm of this value is used.

3.1.4. Control Variables

To effectively mitigate the estimation bias caused by omitted variables, this study draws on existing literature and opts for the following control variables to comprehensively capture the influencing mechanisms:

Agricultural Internal Industrial Structure (Ais): Measured by dividing the overall output value of agriculture, forestry, animal husbandry, and fisheries by the combined output value of crop farming and animal husbandry. It reflects the relative scale and structural characteristics among different agricultural production sectors [58].

Level of Agricultural Economic Development (Agdp): Measured based on the proportion of the added value of the primary sector to the number of workers in the primary industry. This indicator reflects both the output efficiency and growth potential of the agricultural economy [20].

Level of Agricultural Financial Support (Asup): Measured based on the proportion of government fiscal expenditures on agriculture, forestry, and water affairs to the combined production value of agriculture, forestry, animal husbandry, and fishery. This reflects the extent of government financial support for the agricultural sector and its direct impact on agricultural development [59].

Labor Productivity (Wp): Measured based on the proportion of the added value of the primary industry to the rural population. It reflects agricultural production efficiency and the effective utilization of labor resources [60].

Urbanization Rate (Ic): Represented in logarithmic form of the proportion of the urban population to the total regional population. It helps analyze the potential impact of urbanization on agriculture and related fields [60].

Level of Environmental Regulation (Er): Determined based on the proportion of environmental pollution control investments to the regional GDP. This indicator reveals the intensity of regional investment in environmental protection and its impact on sustainable agricultural development [61].

3.1.5. Other Variables

To further enhance the scientific validity and rationality of the study’s conclusions, other variables are included during the subsequent robustness checks and heterogeneity analysis as tools for the supplementary analysis. The development of the digital economy is closely tied to high-quality talents, while the improvement in human capital is closely related to the level of higher education. Higher education can provide high-quality talents for the digital economy, especially those capable of supporting agricultural carbon emission efforts especially by improving the innovation abilities and spirit of rural workers [62,63]. Considering that regional education levels may influence the acceptance of the digital economy in rural areas [14,64], the education level (Edu) is selected as another variable, assessed based on the mean years of education per individual in rural areas.

3.2. Model Construction

3.2.1. Baseline Model

In order to examine the impact of the digital economy on agricultural carbon emission reductions, the subsequent model has been developed:

$$AC{E}_{\mathrm{it}}={\alpha }_0+{\alpha }_{1}D{\mathrm{igl}}_{\mathrm{it}}+{\alpha }_{2}C{\mathrm{ontrol}}_{\mathrm{it}}+{\beta }_{\mathrm{i}}+{\omega }_{\mathrm{t}}+{\epsilon }_{\mathrm{it}}$$

(9)

where i is the province; t is time; $AC{E}_{\mathrm{it}}$ denotes agricultural carbon emission indicators, including the agricultural carbon emission intensity or total agricultural carbon emissions; $D{\mathrm{igl}}_{\mathrm{it}}$ is the level of digital economic development; $C{\mathrm{ontrol}}_{\mathrm{it}}$ is a series of control variables; ${\beta }_{\mathrm{i}}$ is the province fixed effects; ${\omega }_{\mathrm{t}}$ is the time fixed effects; and ${\epsilon }_{\mathrm{it}}$ is the random error.

3.2.2. Mediating Effect Model

Referring to the method adopted by Jiang Ting [65], labor transfer is tested with the following model:

$$AC{E}_{\mathrm{it}}={\alpha }_0+{\alpha }_{1}D{\mathrm{igl}}_{\mathrm{it}}+{\alpha }_{2}C{\mathrm{ontrol}}_{\mathrm{it}}+{\beta }_{\mathrm{i}}+{\omega }_{\mathrm{t}}+{\epsilon }_{\mathrm{it}}$$

(10)

$$L{\mathrm{t}}_{\mathrm{it}}={\chi }_0+{\chi }_{1}D{\mathrm{igl}}_{\mathrm{it}}+{\chi }_{2}C{\mathrm{ontrol}}_{\mathrm{it}}+{\beta }_{\mathrm{i}}+{\omega }_{\mathrm{t}}+{\mu }_{\mathrm{it}}$$

(11)

where $L{\mathrm{t}}_{\mathrm{it}}$ is the mediating variable; ${\alpha }_1$ is the regression coefficient of the digital economy on the agricultural carbon emission intensity; ${\chi }_1$ is the regression coefficient of the digital economy development level on the mediating variable; ${\epsilon }_{\mathrm{it}}$ and ${\mu }_{\mathrm{it}}$ are random errors. The causal relationship between the mediating and dependent variables is further addressed in light of current theories.

3.2.3. Lagged Model

To thoroughly analyze the temporal dynamics of its impact on the agricultural carbon emission intensity, a lagged treatment of the core explanatory variable, the digital economy, is applied in Equation (9). This approach seeks to track the possible long-term impacts of the current condition of digital economy development on the regional agricultural carbon emission intensity during the upcoming periods. The following model is created:

$$AC{E}_{\mathrm{it}}={\alpha }_0+{\alpha }_{1}D{\mathrm{igl}}_{\mathrm{i}\left(\mathrm{t}\text{-}\mathrm{m}\right)}+{\alpha }_{2}C{\mathrm{ontrol}}_{\mathrm{it}}+{\beta }_{\mathrm{i}}+{\omega }_{\mathrm{t}}+{\epsilon }_{\mathrm{it}}$$

(12)

where m represents the lag period of the variable.

3.2.4. Spatial Econometric Model

To further investigate the spatial distribution characteristics of the digital economy and agricultural carbon emission intensity, as well as the specific mechanisms through which the digital economy affects the agricultural carbon emission intensity, this study chooses to use a spatial econometric model. This is because the carbon emission intensity of regional agriculture is not only relevant to the local growth of the digital economy but is also indirectly influenced by digital economic activities in neighboring regions. Consequently, the following spatial Durbin model (SDM) is established in accordance with model (9).

$$AC{E}_{\mathrm{it}}={\alpha }_0+\gamma WAC{E}_{\mathrm{it}}+{\eta }_{1}WD{\mathrm{igl}}_{\mathrm{it}}+{\alpha }_{1}D{\mathrm{igl}}_{\mathrm{it}}+{\eta }_{2}WC{\mathrm{ontrol}}_{\mathrm{it}}+{\alpha }_{2}C{\mathrm{ontrol}}_{\mathrm{it}}+{\beta }_{\mathrm{i}}+{\omega }_{\mathrm{t}}+{\epsilon }_{\mathrm{it}}$$

(13)

where $\gamma$ represents the spatial autocorrelation coefficient; W denotes the spatial weight matrix; ${\eta }_1$ and ${\eta }_2$ represent the elasticity coefficients of the spatial interaction terms of the core independent variables and control variables; ${\beta }_{\mathrm{i}}$ is the province fixed effects; ${\omega }_{\mathrm{t}}$ is the time fixed effects; and ${\epsilon }_{\mathrm{it}}$ is the random error.

3.3. Data Sources and Statistical Description

This study selects 31 Chinese provinces (not including Hong Kong, Macao, and Taiwan) as the research sample for the years 2013–2022 based on the validity and completeness of the data. The National Bureau of Statistics, China Statistical Yearbook, China Rural Statistical Yearbook, Peking University’s Digital Finance Research Center, and the Ali Research Institute are the primary sources of the fundamental data for the different variables. Interpolation, approximation to the average, or average value approaches were used to fill in the missing data.

Table 3. Descriptive statistics.

Variable Type	Variable Name	Sample Size	Average	Standard Deviation	Minimum	Maximum
Dependent Variable	Agricultural Carbon Emission Intensity	310	−4.506	1.010	−8.002	−2.815
	Agricultural Carbon Emissions	310	5.354	1.141	2.543	6.093
Core Independent Variable	Level of Digital Economic Development	310	−1.986	0.697	−4.011	−0.232
Mechanism Variable	Labor Transfer	310	0.383	0.125	0.0795	0.642
Independent Variables	Internal Industrial Structure of Agriculture	310	0.141	0.0826	0.004	0.437
	Level of Agricultural Economic Development	310	3.503	1.811	0.821	10.760
	Level of Fiscal Support for Agriculture	310	0.315	0.403	0.071	2.091
	Labor Productivity	310	1.246	0.613	0.337	3.581
	Urbanization Rate	310	−0.525	0.211	−1.428	−0.110
	Level of Environmental Regulation	310	0.013	0.009	0.003	0.094
Other Variables	Education Level	310	7.722	0.828	3.804	9.915

displays the findings of the statistical description.

4. Research Results and Analysis

4.1. Analysis of Trends in Agricultural Carbon Emissions in Time and Space

To visually capture the spatial and temporal evolution characteristics of agricultural carbon emission intensity, Figure 2 demonstrates the spatial pattern of agricultural carbon emissions in 31 provinces in China. The data are based on the agricultural carbon emission intensity collected and calculated in the previous period. In this paper, the natural breakpoint method is adopted, with four time benchmarks, namely 2013, 2016, 2019, and 2022, selected to provide an intuitive analysis of the intensity of agricultural carbon emissions in each region.

As shown in Figure 2, there are significant regional differences in the spatial distribution of agricultural carbon emissions in China. These emissions show a clear pattern: higher in the west, lower in the east, and higher in the north than in the south. Moreover, China’s agricultural carbon intensity has seen a significant decline over the past decade. By 2025, it is expected to fall below the 0.05 threshold in most regions.

4.2. Analysis of Baseline Regression

The variance inflation factor (VIF), as determined through a multicollinearity study, is 2.46, which is far less than the crucial threshold of 5. This suggests that the variables do not significantly suffer from multicollinearity. Additionally, it was found that using a two-way fixed effects model is the best approach using LM tests, F-tests, and Hausman tests.

This study examines the direct relationship between digital economic development and agricultural carbon emission intensity using a two-way fixed effects model. Column (1) of

Table 4. Outcomes for baseline regression.

	Aci	Aci	Aci	Lt
	(1)	(2)	(3)	(4)
Digl	−0.183 ***	−0.159 ***	−0.125 ***	−0.046 ***
	(0.050)	(0.045)	(0.041)	(0.014)
Ais		0.580	0.538	0.264 **
		(0.592)	(0.564)	(0.125)
Agdp		−0.037 ***	−0.042 ***	−0.031 ***
		(0.009)	(0.010)	(0.004)
Asup		−0.193 ***	−0.229 ***	−0.057 ***
		(0.053)	(0.057)	(0.016)
Wp		0.155 ***	0.163 ***	0.054 ***
		(0.043)	(0.043)	(0.012)
Ic			−0.376 **	−0.268 ***
			(0.186)	(0.060)
Er			−0.697	0.567 **
			(0.581)	(0.237)
_cons	−4.868 ***	−4.905 ***	−5.000 ***	0.165 ***
	(0.099)	(0.137)	(0.154)	(0.037)
N	310	310	310	310
R2	0.074	0.326	0.344	0.403
YearFE	YES	YES	YES	YES
IDFE	YES	YES	YES	YES

Note: the values in parentheses represent standard errors, ** p < 0.05, *** p < 0.01.

displays the pertinent regression results. The outcomes of the univariate regression are shown in Table 4. After controlling for both time and individual fixed variables, the regression coefficient for the digital economy’s effect on the agricultural carbon emission intensity is significantly negative at the 1% significance level. Control variables are then progressively added to the regression model, with the outcomes displayed in columns (2) and (3) of Table 4. It is found that the model’s goodness of fit increases with the addition of each control variable, and the digital economy coefficient stays significant with only little variations in its value. It shows a significant inhibitory effect of the digital economy on agricultural carbon emissions, confirming hypothesis H1.

The findings in Table 4′s column (3) make it clear that the degree of growth in the digital economy has a particular effect on the intensity of carbon emissions from agriculture. Specifically, for every 1 unit increase in the digital economy’s development level, agricultural carbon emissions decrease by 0.125 units. The development of the digital economy facilitates the infusion of information and technology into the agricultural sector. On the one hand, online environments make it easier to obtain and share information quickly, optimize resource allocation in the agricultural production sector, and assist farmers in developing green development concepts, thereby establishing an unofficial cloud-based environmental control. On the other hand, farmers can more effectively monitor field production conditions, increase the efficiency of agricultural resource use, and lower greenhouse gas emissions from resource waste thanks to the deep integration of digital technology with agriculture. This helps farmers reach the goal of lowering agricultural carbon emissions.

The degree of agricultural economic development, agricultural fiscal support, and the pace of urbanization all significantly reduce the intensity of agricultural carbon emissions in the control variables. As the agricultural economy grows, it maximizes production efficiency and other factors, thereby lowering carbon emissions. Fiscal support for agriculture efficiently reduces the intensity of agricultural carbon emissions, boosts farmers’ incomes, and increases agricultural production efficiency [59]. It is thought that increasing rate of urbanization will alter the consumption and lifestyle habits of rural workers, optimize the structure of energy consumption, and lower carbon emissions from agriculture. The government can effectively reduce polluting agricultural activities by enforcing environmental regulations and carbon-reduction strategies, as indicated by the coefficient for the effect of the environmental regulation level on the agricultural carbon emission intensity, which is negative but not significant. At the same time, tailored environmental governance and restoration programs can effectively protect agricultural ecosystems, improve their carbon sequestration, and lessen carbon emissions. The intensity of agricultural carbon emissions is positively impacted by worker productivity and the internal agricultural industrial structure. This is mostly because the livestock and crop farming sectors account for a large share of agricultural carbon emissions. Furthermore, increases in labor productivity may further boost the share of these two sectors, potentially worsening carbon emissions. Although the use of agricultural technology or equipment may raise the intensity of agricultural carbon emissions, technological developments can increase worker productivity. The issue of carbon emissions from agriculture could get worse if production does not incorporate environmental protection principles.

Zaozhuang Xuecheng Zero-Carbon Digital Industrial Park, as the first modern agricultural park themed on zero-carbon digital agriculture in China, achieves an annual power generation of 8.61 million kWh by integrating photovoltaic power generation, an intelligent water–fertilizer integration system, and group cultivation technology, relying on a 3.5 km photovoltaic corridor to save 2535 tons of standard coal and 4780 tons of carbon dioxide emissions, and forms a whole industry chain mode of “zero-carbon production, seedling cultivation and agro-tourism integration”. It has formed a comprehensive industry chain model that integrates ”zero-carbon production-seedling and cultivation-agro-tourism”. Recognized as the provincial digital intelligent agriculture application base, the park provides a replicable technical path and industrial model for carbon emission reductions and digital transformation in agriculture.

4.3. Robustness Test

4.3.1. Replacement of Dependent Variables

The robustness test uses agricultural carbon emissions (Ac) and carbon emissions from livestock farming and rice cultivation (Act) as alternative dependent variables because China is in the monsoon climate zone, where rice cultivation is the primary agricultural activity and livestock farming contributes significantly to the overall agricultural output value. Consistent with the earlier findings, the results, which are shown in

Table 5. Analysis based on robustness and endogeneity tests.

	Ac (1)	Act (2)	Aci (3)	Aci (4)	Digl (5)	Aci (6)
Digl	−0.063 ** (0.030)	−0.141 *** (0.060)	−0.147 *** (0.040)			−0.270 ** (−2.315)
Digl(t−1)				−0.072 ** (0.031)
Edu			−0.621 *** (0.190)
Instrumental Variable					0.000 *** (4.153)
Kleibergen-Paaprk LM Statistic					4.344 **
Cragg–Donald Wald F Statistic					140.206 (17.246)
Control	Control	Control	Control	Control	Control	Control
YearFE	Control	Control	Control	Control	Control	Control
IDFE	Control	Control	Control	Control	Control	Control
N	310	310	310	279	310	310
R2	0.803	0.803	0.372	0.783		0.876

Note: ** p < 0.05, *** p < 0.01.

′s columns (1) and (2), show that the suppressive effect of the degree of digital economic advancement on agricultural carbon emission intensity is still negative at the 5% and 1% significant levels.

4.3.2. Add Control Variables

After adding the education level (Edu) as a variable to the regression model for robustness verification, the negative effect of the digital economy on the agricultural carbon emission intensity is still present, as shown in column (4) of Table 5. Specifically, for every 1 unit increase in the development level of the digital economy, total agricultural carbon emissions decrease by 0.147 units.

4.3.3. Endogeneity Test

To reduce the potential impact of unobserved factors on the empirical analysis results, the previous baseline regression and robustness tests used a two-way fixed effects model of time and individuals and included multiple control variables to mitigate the endogeneity issues caused by omitted variables. Nevertheless, endogeneity may still persist due to two-way causality. This research will use instrumental variable techniques and expand the temporal window method as part of the methodology to further examine and validate these possible endogeneity issues.

4.3.4. Extension of the Time Window

This study introduces a one-period lag and further expands the time observation window for the primary explanatory variable, the agricultural carbon emission intensity, which is impacted by the growth of the digital economy. As seen in column (4) of Table 5, the regression coefficient is negative at the 5% significance level, further supporting the validity of the main finding and successfully resolving the endogeneity problem caused by reverse causality.

4.4. Instrumental Variable Method

The interaction term between the number of phones per capital in each province and the number of internet users from the prior year is used as the instrumental variable for the digital economy [66]. The following are the rationales behind this choice: first, the relevance condition is met by the number of phones and internet users from the prior year, which reflect the state of the regional communication infrastructure and have a significant impact on the internet- and informatization-centered digital economy. Secondly, the exogeneity condition is satisfied as, after adjusting for other variables, the interaction term between the number of phones per capital and the number of internet users from the prior year does not directly correlate with agricultural carbon emissions. The regression findings for the two stages are displayed in Table 5 in columns (5) and (6), respectively.

In the first stage, the expansion of the digital economy was significantly boosted by the instrumental variable, and the F-statistic was significantly higher than the crucial value of 10, indicating that there is no problem with a weak instrumental variable. The predicted coefficient for the degree of digital economic development in the second stage was substantially negative. The regression coefficient’s absolute value rose in comparison to the baseline regression findings when the instrumental variable was included, suggesting that the instrumental variable had an amplifying effect. Additionally, this raises the possibility that the baseline regression estimates were understated. Overall, even after accounting for the possible endogeneity problem, the digital economy continues to have a noticeable impact on reducing carbon emissions from agriculture.

4.5. Mediating Effect Test

To further analyze the transmission mechanism through which the digital economy impacts the agricultural carbon emission intensity, this study introduces the variable of labor transfer (Lt) and uses Jiang Ting’s two-step method [67] for stepwise regression verification. Table 4′s column (4) displays the findings of the regression analysis: the regression coefficient between labor transfer and the digital economy is strongly negative at the 1% confidence level, suggesting a strong correlation between the two. Previous research has shown that the outflow of rural labor can successfully reduce agricultural carbon emissions. In particular, labor transfer often leads to a decline in the proportion of land dedicated to grain crops, resulting in lower carbon emissions due to changes in cropping patterns [68]. Additionally, the movement of rural labor shows a “U-shaped” trend in agricultural carbon emissions, first declining and then rising [69]. The rapid development of the digital economy has not only transformed production practices and lifestyles but has also sped up the migration of workers from conventional agriculture to other sectors of the economy, which has an effect on the carbon emission intensity of agricultural output. This transfer indirectly contributes to a reduction in agricultural carbon emissions by shrinking the scale of agricultural production and reducing the land area under cultivation. The volume of agricultural production has shrunk, and the number of farming households has decreased as a result of the digital economy’s explosive growth, which has encouraged rural workers to migrate to cities or other non-agricultural industries. As a result, there is now less demand for farmland irrigation and less area used for crop cultivation, which lowers associated energy consumption and carbon emissions. Furthermore, the development of the digital economy has sped up the technological advancement and modernization of agriculture. The use of digital technology is causing agricultural production to shift more and more toward intelligent and sophisticated management, which drastically lowers energy consumption and resource waste in conventional operational procedures. In conclusion, labor transfer serves as a mediating mechanism through which the digital economy promotes agricultural carbon emission reductions, thus validating hypothesis H2.

4.6. Spatial Spillover Effect Analysis

The spatial measurement model provides a scientific basis and a powerful analytical tool for studying regional economy and environment issues. By quantifying the spatial dependence and heterogeneity among geographic units, it reveals the cross-regional interaction effects that traditional models often neglected and explores the real relationship among variables. With the rapid development of the digital economy, economic ties between areas have gotten closer, making the interdependent impacts between regions increasingly noticeable. In light of this, this study analyzes the spatial implications of the expansion of the digital economy on the intensity of carbon emissions from agriculture using a spatial distance matrix. To investigate whether there is spatial auto correlation between the intensity of agricultural carbon emissions and the degree of the digital economy, the Global Moran’s Index was utilized.

Table 6. Global Moran’s I test results for the relationship between the digital economy and agricultural carbon emissions.

Year	Morans’I	p-Value	Z-Statistic	Year	Morans’I	p-Value	Z-Statistic
2013	0.056	0.007	2.688	2018	0.063	0.004	2.892
2014	0.055	0.008	2.661	2019	0.061	0.004	2.855
2015	0.057	0.006	2.727	2020	0.062	0.004	2.861
2016	0.057	0.006	2.726	2021	0.061	0.004	2.842
2017	0.059	0.005	2.778	2022	0.060	0.005	2.803

displays the empirical findings.

The findings of the Global Moran’s I test shows that the Moran’s I values for both digital economy development and agricultural carbon emission intensity were positive and exhibited a varying rising trend between 2013 and 2022. The null hypothesis of no spatial connection was rejected at a 1% significance level based on the p-value, which indicated a positive spatial relationship between agricultural carbon emissions in nearby provinces and the degree of digital development. The local Moran’s I scatter plot helps to visualize the spatial distribution of agricultural carbon intensity. As shown in Figure 3, the values of the Moran’s I for agricultural carbon intensity are mainly concentrated in quadrants 1 and 3, which indicates that China’s agricultural carbon emissions show obvious local spatial clustering characteristics in the economic field.

Based on these results, we used a spatial econometric model for empirical testing to examine the spatial link between the agricultural carbon emission intensity and digital economic development. The results of the study’s successive LM, LR, Wald, and Hausman tests are shown in

Table 7. Results for spatial model test.

Test	Statistic	Test	Statistic
LM (error) test	20.128 ***	LR(sdm sar) test	63.880 ***
Robust LM (error) test	19.807 ***	Wald (sdm sar) test	68.070 ***
LM (lag) test	0.322	LR(sdm sem) test	73.050 ***
Robust LM (lag) test	0.001	Wald (sdm sem) test	80.560 ***
Joint significance test	Ind (73.850 ***)	Time (992.380 ***)

Note: *** p < 0.01.

.

The fixed effects model is supported by the Hausman test results, which decisively refute the superiority of the random effects model. The LM test shows that the model has significant spatial bias effects at a 1% significance level. Furthermore, the findings of the LR test support the spatial Durbin model (SDM) as the best model. Additionally, the findings of the Wald test exclude the idea of reducing the SDM model to a spatial error model (SEM) or a spatial autoregressive model (SAR). The dual-fixed SDM model with individual and time was finally selected for testing after the joint significance test revealed that the dual fixed effects for both were the best. The findings are shown in

Table 8. Spatial spillover effect test results.

	SDM (1)		SEM (2)		SAR (3)
Digl	−0.071 **	(0.031)	−0.099 ***	(0.035)	−0.105 ***	(0.032)
Ais	−0.124	(0.392)	0.514	(0.378)	0.452	(0.377)
Agdp	−0.024 ***	(0.008)	−0.039 ***	(0.007)	−0.037 ***	(0.007)
Asup	−0.240 ***	(0.039)	−0.220 ***	(0.036)	−0.228 ***	(0.036)
Wp	0.180 ***	(0.029)	0.159 ***	(0.028)	0.151 ***	(0.028)
Ic	−0.264 *	(0.147)	−0.304 **	(0.143)	−0.371 ***	(0.134)
Er	−0.830	(0.523)	−0.696	(0.548)	−0.759	(0.544)
W × Digl	−0.963 ***	(0.194)
W × Ais	−4.481	(3.007)
W × Agdp	−0.057	(0.053)
W × Asup	−0.361	(0.269)
W × Wp	0.561 **	(0.234)
W × Ic	−2.879 ***	(0.889)
W × Er	−1.095	(3.553)
Spatial rho	0.368 **	(0.163)			0.584 ***	(0.118)
Spatial lambda			0.440 ***	(0.160)
Variance sigma2_e	0.003 ***	(0.001)	0.004 ***	(0.001)	0.004 ***	(0.001)
N	310		310		310
R2	0.126		0.749		0.717
Province Fixed Effects	Control		Control		Control
Time Effects	Control		Control		Control

Note: * p < 0.1, ** p < 0.05, *** p < 0.01; values in parentheses represent t-values. In the table, W × Digl, W × Ais, W × Agdp, W × Asup, W × Wp, W × Lc, W × Er, and W × Er represent the spatial lag terms of the digital economy, agricultural internal industrial structure, agricultural economic development level, level of fiscal support for agriculture, labor productivity, urbanization rate, and environmental regulation level, respectively (same below).

.

As shown in column (1) of Table 8, the spatial lag term of digital economy development is significantly negative at the 1% significance level, while the regression coefficient for the level of digital economy development is negative at the 5% significance level. These results indicate that the development of the digital economy not only promotes agricultural carbon emission reductions but also significantly affects the intensity of agricultural carbon emissions in neighboring regions. This result is consistent with the previous findings and supports earlier conclusions. In addition, the regression results show that the spatial autoregressive coefficient (rho) is significantly positive, indicating a positive spatial spillover effect of agricultural carbon emissions in China. To put it another way, due to geographic proximity, agricultural carbon emissions in one area are spatially associated with those in nearby regions. A spatial rivalry impact on the agricultural carbon intensity may be the result of competition for agricultural production inputs, such as land, labor, capital, natural resources, and technology, which is fueled by the development of digital technology. Therefore, while t agricultural carbon emissions exhibit spatial spillover characteristics, the digital economy also demonstrates a spatial spillover effect in promoting carbon emission reductions, thus confirming H3.

4.6.1. Spatial Effect Decomposition

The coefficients in spatial econometric models cannot accurately represent the nonlinear spatial effects of the digital economy on carbon reductions. For an additional study, the regression findings are therefore broken down into direct and indirect effects.

Table 9. Decomposition of effects in the spatial Durbin model.

Variable	Direct Effect		Indirect Effect		Gross Effect
	Coefficient	T Value	Coefficient	T Value	Coefficient	T Value
Digl	−0.105 **	(0.043)	−1.809 *	(1.041)	−1.914 *	(1.072)
Ais	−0.339	(0.437)	−8.860	(7.711)	−9.199	(8.025)
Agdp	−0.025 ***	(0.008)	−0.119	(0.113)	−0.144	(0.115)
Asup	−0.248 ***	(0.046)	−0.702	(0.598)	−0.950	(0.627)
Wp	0.198 ***	(0.040)	1.151	(0.730)	1.349 *	(0.752)
Ic	−0.348 **	(0.148)	−4.880 ***	(1.856)	−5.227 ***	(1.909)
Er	−0.843	(0.590)	−1.523	(6.657)	−2.366	(6.974)

Note: * p < 0.1, ** p < 0.05, *** p < 0.01.

displays the findings of the decomposition of direct effects, indirect effects, and total impacts. The direct effect of the digital economy is −0.105, which is considerably negative at the 5% confidence level, based on the figures in Table 9. Both pass significance tests, and the indirect effect’s regression result is −1.809, which is negative at the 10% significance level. The main way in which the digital economy stimulates carbon emission reductions in agriculture is through indirect consequence. For example, while the digital economy can increase the local agricultural carbon emission intensity, it tends to reduce the intensity in neighboring regions. This may be due to the polarization effect of factors of production, such as the labor force and information technology, where higher levels of the digital economy lead to the concentration of production in developed regions. When the spillover effect of digital technology and management ideas from these regions cannot offset this polarization, the phenomenon will occur. This indicates that the intensity of agricultural carbon emissions is significantly impacted by spatial agglomeration and spillover effects of the digital economy.

4.6.2. Robustness Tests

This study employed two robustness testing techniques to validate the findings on spatial spillover effects: replacing the spatial weight matrix and excluding certain samples. On the one hand, the development level of the digital economy varies significantly across different provinces. Remote areas, such as Tibet and Xinjiang, have underdeveloped digital economies due to weak economic foundations and outdated technology, which may lead to insignificant or heterogeneous results. Therefore, this paper excludes Tibet and Xinjiang from the robustness test, focusing on 29 provinces and municipalities directly under the central government. This reduces sample heterogeneity, making the research results more robust and reliable. Column (1) of

Table 10. Robustness tests for spatial spillover effects.

	SDM (1)		SDM (2)
Digl	−0.0717 **	(0.0341)	−0.0518 *	(0.0293)
Ais	−0.106	(0.397)	−0.1830	(0.329)
Agdp	−0.0283 ***	(0.00771)	−0.0124 *	(0.00744)
Asup	−0.291 ***	(0.0481)	−0.182 ***	(0.0346)
Wp	0.148 ***	(0.0300)	0.100 ***	(0.0288)
Ic	−0.299	(0.190)	−0.487 ***	(0.144)
Er	−0.424	(0.549)	−1.054 **	(0.470)
W × Digl	−0.725 ***	(0.197)	−0.347 ***	(0.0738)
W × Ais	−4.458	(2.894)	−3.892 ***	(0.909)
W × Agdp	−0.0838	(0.0523)	−0.0655 ***	(0.0204)
W × Asup	−0.575 **	(0.272)	−0.0831	(0.0882)
W × Wp	0.523 **	(0.228)	0.268 ***	(0.0778)
W × Ic	−3.984 ***	(0.971)	−0.443	(0.370)
W × Er	0.609	(3.404)	−0.444	(1.088)
Spatial rho	0.350 **	(0.166)	0.482 ***	(0.0782)
Spatial lambda
Variance sigma2_e	0.00311 ***	(0.000261)	0.00275 ***	(0.000226)
N	290		310
R2	0.328		0.237
Province Fixed Effects	Control		Control
Time Effects	Control		Control

Note: * p < 0.1, ** p < 0.05, *** p < 0.01.

shows the regression results excluding remote areas (Tibet and Xinjiang). The spatial lag term of the digital economy development level is significantly negative at the 1% significance level, indicating that the level of digital economy development in the neighboring regions suppresses the intensity of agricultural carbon emissions. The spatial autoregressive coefficient (rho) is also significant and positive, indicating that the agricultural carbon emissions in a region will affect those in other regions. This indicates that the conclusions of this study are still robust after excluding remote samples. The core regions, with their well-developed digital economy, form a strong radiation to its neighbors through the industrial chain and technology spillover. In contrast, remote areas, with a weaker digital economy foundation and lower industrial linkage, show minimal impact. Excluding these areas improves the data quality and reduces noise, highlighting the spatial spillover relationship of the digital economy on agricultural carbon emissions among the core regions. On the other hand, the spatial economic geographic weight matrix is used to construct the spatial matrix. While geographical or economic distance alone does not fully reflect the interactions between spatial units, the spatial economic geographic weight matrix combines geographic distance and economic characteristics, allowing for a more accurate portrayal of the comprehensive interaction effects between spatial units. It also more precisely captures the spatial characteristics of the digital economy’s impact on agricultural carbon emissions, thus improving the explanatory power and predictive ability of the model. Column (2) of Table 10 shows the robustness of the findings under the spatial economic geographic weighting matrix. The results remain consistent under different spatial weighting structures, further validating the stability and significance of the digital economy’s impact on agricultural carbon emissions, thus enhancing the credibility and reliability of the study’s conclusions.

4.6.3. Heterogeneity Analysis

Provinces were categorized based on terrain undulation [69], dividing regions into two types: steep and flat areas. To investigate the impact of topographical variability on the agricultural carbon emission intensity, a group regression analysis was conducted. The results, shown in Columns (1) and (2) of

Table 11. Results of heterogeneity analysis based on spatial spillover effects.

	SDM Steep		SDM Flat		SDM High Output Value		SDM Low Output Value
Digl	−0.155 ***	(0.0332)	−0.0889 **	(0.0413)	−0.344 ***	(0.0487)	−0.108 ***	(0.0395)
Ais	−0.510	(0.497)	2.709 ***	(0.641)	3.115 ***	(0.471)	−0.849	(0.542)
Agdp	−0.000428	(0.0241)	−0.0192 **	(0.00795)	−0.00680	(0.0124)	−0.0441 ***	(0.0101)
Asup	0.0243	(0.0550)	−0.243 ***	(0.0502)	2.026 ***	(0.447)	−0.270 ***	(0.0469)
Wp	0.000513	(0.0582)	−0.0383	(0.0433)	0.110**	(0.0449)	0.137 ***	(0.0363)
Lc	−1.133 ***	(0.170)	1.060 ***	(0.264)	0.409	(0.286)	−0.241	(0.202)
Er	0.228	(0.426)	−2.143 **	(1.093)	−2.546**	(1.269)	−0.709	(0.601)
W × Digl	−0.310	(0.205)	−0.573 ***	(0.208)	−1.546***	(0.264)	−0.386 *	(0.214)
W × Ais	−2.963	(2.374)	6.174 **	(3.111)	−0.532	(4.315)	−9.585 ***	(3.639)
W × Agdp	0.233 **	(0.106)	−0.162 ***	(0.0571)	0.118	(0.0761)	−0.120 **	(0.0476)
W × Asup	0.903 *	(0.501)	−0.455 **	(0.204)	7.217**	(2.874)	−0.720 ***	(0.236)
W × Wp	−1.123 ***	(0.264)	0.246	(0.193)	0.577	(0.398)	−0.275	(0.278)
W × Lc	−2.636 **	(1.141)	2.423	(1.845)	−1.040	(1.510)	−1.290	(0.903)
W × Er	0.763	(1.741)	−3.327	(5.118)	−1.169	(6.779)	1.363	(3.116)
Spatial rho	−0.661 **	(0.295)	0.404 ***	(0.152)	−0.350	(0.308)	0.232	(0.185)
Spatial lambda
Variance sigma2_e	0.00113 ***	(0.000150)	0.00266 ***	(0.000280)	0.000643 ***	(0.0000914)	0.00373 ***	(0.000364)
N	120		190		100		210
R2	0.190		0.212		0.023		0.305
Province Fixed Effects	Control		Control		Control		Control
Time Effects	Control		Control		Control		Control

Note: * p < 0.1, ** p < 0.05, *** p < 0.01.

, indicate that the digital economy significantly lowers agricultural carbon emissions in both steep and flat regions. In addition, the dampening effect of digital economy development in neighboring regions on the agricultural carbon emission intensity is particularly pronounced in flat regions. The spatial autoregressive coefficient (rho) is significantly positive at the 1% level in flat areas, indicating that in flat areas, agricultural carbon emissions in one flat region will have a positive spatial spillover effect on agricultural carbon emissions in neighboring regions. This difference may be attributed to the complex topography of steep areas, which limits agricultural development and reduces the rural population’s dependence on agriculture. In contrast, flat areas, characterized by low topographic relief and easy access to transportation, tend to exhibit stronger economic linkages. Neighboring provinces in these areas experience more frequent economic activity, resource flows, and technological exchanges. These strong economic linkages accelerate the diffusion and spread of the digital economy, allowing its inhibitory effects to spread more quickly to neighboring provinces. As a result, spatial spillover effects are more pronounced in flat areas.

To distinguish the level of agricultural economic progress among regions, agricultural carbon emissions are significantly influenced by grain production functional zones. These zones differ in aspects such as crop planting structures, production processes, and pesticide usage. In particular, provinces with higher agricultural gross output values tend to have planting structures more focused on grain crops, increasing the proportion of grain production, which in turn alters the usage of chemicals, fosters high-carbon production models, and ultimately affects agricultural carbon emissions. To further explore this phenomenon, we ranked 31 provinces based on their agricultural gross output value. The top 10 provinces were classified as high-agricultural-output-value regions, while the remaining 21 provinces were categorized as low-agricultural-output regions. Regression analyses were then performed separately for these two subsamples.

As shown in columns (3) and (4) of Table 11, the development of the digital economy greatly reduces the agricultural carbon emission intensity in both high- and low-agricultural-output regions, with a more obvious suppression effect in high-output regions. However, the spatial lag term of the level of digital economy development in high-value regions is significantly negative at the 1% significance level, indicating that the spatial spillover effect is more prominent. This may be ascribed to the high-output regions’ efficient agricultural production, advanced technology, and rich management experience, which facilitate economic and technological spillover effects. These regions are often the focus of policy innovation and implementation, receiving more policy support and resource investment, thereby driving green agricultural development and carbon emission reductions more effectively. In contrast, low-output regions face challenges such as outdated technology and extensive management. In the initial stages, resistance from certain “vulnerable groups” may exacerbate the “digital divide” within society, hindering the widespread adoption and practical application of digital technologies and thus making it difficult to establish an effective regional coordinated emission-reduction mechanism [70]. However, as digital technologies continue to be promoted and applied, the carbon-reduction effects in low-output regions are gradually becoming apparent. However, the spatial autoregressive coefficients (rho) of the two regions are not significant. In the high-agricultural-output regions, these coefficients are negative, and the correlation between the variables tends to be negative as the spatial distance decreases, which indicates that the relationship between geographically close variables is weaker.

5. Research Conclusions and Suggestions

As a basic industry of the national economy, agriculture’s carbon emission is not only related to its own sustainable development of agriculture but also affects global climate change and the ecological environment. Based on the above research, the following conclusions are drawn: first, the core conclusion is that the level of digital economy development significantly inhibits the intensity of agricultural carbon emissions. Second, in terms of the mechanism of action, the development of the digital economy will promote the labor force transfer, and this transfer of the rural labor force will lead to the improvement of the level of agricultural mechanization and changes in the proportion of crop cultivation, which indirectly inhibit the intensity of agricultural carbon emissions. Third, the development of the digital economy presents a significant spatial impact, which can not only effectively reduce the intensity of agricultural carbon emissions in the region but also significantly inhibit emissions in the neighboring regions through its spatial spillover effect. Fourth, the development of the digital economy in flat areas and areas with high agricultural production value has the greatest significant inhibitory effect on agricultural carbon emissions in the neighboring areas through its spatial spillover effect, followed by areas with low agricultural production value. In contrast, the development of the digital economy in steep areas does not significantly inhibit the intensity of agricultural carbon emissions in the neighboring areas through its spatial spillover effect. Differences in the difficulty of agricultural production under different terrain conditions may affect farmers’ subsequent occupational choices and further influence the agricultural carbon emission intensity.

Building on the research findings, the following suggestions are put forth in this paper:

The first is to seek development amidst opportunities. Given that the development of the digital economy significantly inhibits the intensity of agricultural carbon emissions, it is important to seize the opportunity presented by the digital era and technological revolution to accelerate the agricultural digitization and industrialization. The government should increase policy and financial support for “issues relating to agriculture, rural areas, and rural people”, strengthen the rural digital infrastructure, promote equal access to educational services, and enhance the inclusiveness of rural finance. The advantages of the digital economy will be amplified, superior agricultural development will be encouraged, farmers’ incomes will rise, and rural rejuvenation will be supported. For example, the government of Xuecheng District, Zaozhuang City, Shandong Province, has strongly supported the development of the Zero Carbon Digital Industrial Park, which, through the integration of photovoltaic power generation, intelligent water and fertilizer integration systems, and other digital means, has achieved an annual power generation of 8.61 million kWh, saved 2535 tons of standard coal, and reduced carbon dioxide emissions by 4780 tons. The government should also encourage agricultural operators to embrace green, low-carbon digital technologies as soon as possible, develop economies of scale in technical reserves, and fully utilize the time advantages of digital technology. Additionally, the government must encourage the broad use of digital technology in agriculture by way of financial expenditures, policy recommendations, and other initiatives. It should also encourage enterprises and cooperatives to adopt digital management tools to lower carbon emissions and increase industrial efficiency. While reaping the benefits of digital development, the government needs to strengthen the monitoring of agricultural carbon emissions, focus on their dynamic changes, improve the carbon trading mechanism [71], strictly control emission totals, and advance the “dual-carbon” goals, contributing to global emission-reduction efforts [72].

The second is to capitalize on labor strengths for development. Because labor transfer is a crucial mechanism through which the digital economy impacts the agricultural carbon emission intensity, it is necessary to scientifically plan the transfer of rural labor while following the principle of adapting to local conditions. While transferring rural labor can help reduce agricultural carbon emissions, it should be implemented cautiously to ensure that the supply of basic agricultural labor is not compromised. Therefore, it is of necessity to optimize the development environment for agriculture and rural areas, improve the labor market, increase fiscal investment and tax incentives for agriculture and rural areas, and enhance infrastructure and public services. These measures will improve the attractiveness and competitiveness of rural areas. To facilitate rural communities’ and agriculture’s sustainable growth, efforts should be made to build a team of professionals proficient in agricultural knowledge, passionate about rural life, and committed to caring for farmers’ welfare. This reservoir of the “Three Rural” talent pool is set to furnish robust human resource backing for the prosperity of agriculture and rural sectors. Additionally, there should be a greater emphasis on vocational education and skills training to enhance the overall quality and skill level of the labor force, optimize the employment structure, and create more non-agricultural job opportunities for general agricultural workers. Furthermore, the government should promote innovation in green agricultural technologies by establishing special funds and rewards to attract enterprises and research institutions to develop and promote low-carbon agricultural technologies. Demonstration bases should be established to provide practical training and the latest technologies to farmers, enhancing their ability to cope with climate change.

Thirdly, the development of the digital economy has demonstrated significant spatial influence, highlighting the need to strengthen regional synergy and promote information sharing and resource complementarity in its development. Cross-regional cooperation mechanisms should be built to jointly advance the digital economy and reduce the agricultural carbon emission intensity, particularly in places where spatial spillover effects are large. To fully increase the degree of informatization in rural areas, specific goals should be established for the development of digital infrastructure, with an emphasis on improving communication networks and raising the acceptance of the internet and mobile devices in these areas. In order to lessen the inequality in the creation of the digital agricultural infrastructure, focus should also be made on encouraging the coordination of the digital economy within areas. A high-quality spatial framework for the interconnection of information elements should be built, with the goal of maximizing the diffusion and scale effects of digital technologies to facilitate agricultural carbon reductions.

The fourth is to adopt localized control measures. Given the heterogeneity of the spatial spillover effect of the digital economy’s development on the intensity of agricultural carbon emissions, future efforts should focus on implementing targeted agricultural carbon-emission-reduction policies tailored to the actual situation, while advocating for resource sharing. Boost regional collaboration and the division of labor, and create a forum for exchanging the results of the expansion of the digital economy. Reasonably plan the development of the digital economy, enhancing its natural integration with the green, low-carbon transformation of agriculture, taking into account the regional features and agricultural output value of each region. Encourage the concurrent development of the digital economy and green transformation in high-output areas, leading by example. Reduce agricultural carbon emissions by leveraging spatial spillover effects and provide policy support to areas that are lagging behind in digital economic growth, forging a low-carbon agricultural development path suitable to their specific circumstances. Relying on existing resources, adopt clean energy, maintain agricultural ecosystems, and build a new energy system. This is conducive to boosting the coordination of ecology and industry, improving rural living standards, and protecting the ecological environment of agriculture, actively and steadily advancing the “carbon peak and carbon neutrality” goals in the agricultural field.

While this paper makes a modest contribution to research in this area, it has several limitations. First, due to the lack of publicly available data on carbon emissions at the municipal and county levels, valid data cannot be obtained. Therefore, for a more in-depth analysis, future studies should refine the data granularity and focus on the municipal or county-level analysis. Second, this study has not yet delved into the nonlinear relationship or heterogeneity of labor transfer between the digital economy and agricultural carbon emission reductions. In future research, the authors will further separately explore the possible heterogeneous relationship between labor outflow and advanced labor inflow based on the relationship between the digital economy and agricultural carbon emission reductions to obtain more precise results. Finally, due to the complexity of multiple influencing elements, future studies could benefit from using machine learning and other techniques to improve the performance.