EN中文

Journal of Resources and Ecology >

2026 , Vol. 17 >Issue 1: 275 - 290

DOI: https://doi.org/10.5814/j.issn.1674-764x.2026.01.022

Agricultural Ecosystem

The Effects and Mechanisms of Rural Digitalization and Agricultural Carbon Reduction

  • HUANG Longjunjiang , 1 ,
  • LI Lishan 1 ,
  • LIU Xiaojin , 2, *
Expand
  • 1. School of Business Administration, Zhongnan University of Economics and Law, Wuhan 430073, China
  • 2. School of Economics and Management, Jiangxi Agricultural University, Nanchang 330045, China
* LIU Xiaojin, E-mail:

HUANG Longjunjiang, E-mail:

Received date: 2025-10-29

  Accepted date: 2025-12-30

  Online published: 2026-02-02

Supported by

The Key Project of the Key Research Base for Philosophy and Social Sciences in Jiangxi Province(23ZXSKJD07)

Fold

Abstract

Investigating the impact of rural digitalization on agricultural carbon emissions contributes to achieving carbon neutrality goals and facilitates the green transformation of agriculture with enhanced efficiency. Based on panel data from 31 provincial-level regions in China spanning 2005 to 2022, this study employs a dynamic panel model to analyze the influence of rural digitalization on agricultural carbon emission intensity. Heterogeneity analysis, mechanism testing, and spatial effect examination are also conducted. The main findings are fourfold. (1) Rural digitalization effectively promotes the reduction of agricultural carbon emissions. (2) Heterogeneity analysis revealed that the effect of rural digitalization on lowering agricultural carbon emission intensity is particularly significant in production-marketing balanced regions. (3) The carbon emission reduction effect of rural digitalization is primarily realized through the scaling of agricultural operations, the accumulation of human capital, and the improvement of total factor productivity. (4) A positive spatial correlation exists in the agricultural carbon emission intensity across provinces, and the inhibitory effect of rural digitalization on agricultural carbon emission intensity exhibits spatial spillover effects. Therefore, to accelerate rural digitalization and advance agricultural carbon emission reduction, it will be essential to promote the scaling of agricultural operations, guide farmers in adopting advanced technologies, and enhance their ability to utilize digital tools.

Key words： rural digitalization; agricultural carbon reduction; green and low-carbon development

Cite this article

HUANG Longjunjiang , LI Lishan , LIU Xiaojin . The Effects and Mechanisms of Rural Digitalization and Agricultural Carbon Reduction[J]. Journal of Resources and Ecology, 2026 , 17(1) : 275 -290 . DOI: 10.5814/j.issn.1674-764x.2026.01.022

1 Introduction

Carbon emissions generated by human behavior are increasing day by day. Statistical evidence shows that global energy-related carbon dioxide emissions escalated by 410 million tons in 2023, representing a 1.1% year-on-year increase from 2022. The extreme climate change triggered by the greenhouse effect is posing profound challenges for global socioeconomic development, rendering carbon reduction a pressing concern for nations worldwide. As the world’s second-largest economy and the largest contributor to global carbon emissions (accounting for approximately 34% of the global total), China plays a pivotal role in international carbon governance. As an agricultural giant, promoting agricultural carbon reduction is a vital step toward achieving carbon neutrality in China. Concurrently, agriculture is a major source of greenhouse gas emissions (Xiong et al., 2016). According to the IPCC AR6 WGIII, the “Agriculture, Forestry and Other Land Uses (AFOLU)” sector accounts for 13%-21% of global anthropogenic greenhouse gas emissions. Within AFOLU, agriculture contributes approximately 89% of CH4 emissions and about 96% of N2O emissions. Nevertheless, China’s substantial emission baseline poses significant challenges for agricultural carbon reduction. A longitudinal analysis from 1980 to 2020 revealed that greenhouse gas emissions from China’s agricultural production exhibited an overall upward trend, growing by nearly 46% overall. Among the emissions, CH4 emerged as the primary greenhouse gas contributor, accounting for 47.33% of agricultural emissions. The emission sources are predominantly attributed to chemical fertilizer application, rice cultivation, and land tillage practices (Han et al., 2024). However, the pursuit of high yields has made it difficult for China’s agricultural operations to avoid using chemical fertilizers and pesticides. Transforming this high-input, low- efficiency production model is an urgent issue that must be addressed to promote sustainable agricultural development.
With the rise of big data, cloud computing, and other high-tech tools, the high-tech attributes of digitalization have exerted a significant influence on promoting technological green transformation, and this approach has emerged as a new avenue for achieving carbon neutrality. The agricultural sector, rural areas, and farmers have also enjoyed the dividends of digital economy development. Rural digitalization, as the outcome of the digital economy that empowers rural development, and its emerging technologies have been widely applied in agricultural production and rural livelihoods. These technologies can effectively mitigate the regional and fragmented constraints on agricultural production and enhance the stability of agricultural output (Zhao et al., 2023), and they have become a new engine driving the low-carbon, efficient transformation of agriculture. Digital inclusion can alleviate the barriers farmers encounter in adopting digital tools (Liu et al., 2022), thereby lowering the threshold for their use of green technologies. Meanwhile, the ability of farmers to apply green technologies has been enhanced through knowledge accumulation, thereby promoting agricultural carbon reduction (Guo and Zhang, 2023). However, few studies have investigated the impact of rural digitalization on agricultural carbon reduction, so several important questions remain. What are the specific effects of rural digitalization on agricultural development? Through what internal mechanisms are these effects achieved? Do the carbon reduction effects of rural digitalization exhibit spatial spillover? Answering these questions is of great significance for promoting the green transformation of agricultural production and achieving sustainable development.
From the academic perspective, studies on agricultural carbon emissions have primarily focused on three dimensions. The first is the measurement of agricultural carbon emissions. At present, scholars mainly employ the logarithmic mean divisia index method to measure agricultural carbon emissions (Ang, 2006; Han et al., 2018), and some scholars choose a decomposition method to analyze agricultural carbon emissions according to their own specific research goals (Yun and Zhang, 2014; Xiong et al., 2016). Some scholars have also measured the changing trend of carbon emissions and put forward corresponding policy implications. For instance, Huang et al. (2019) measured China’s agricultural carbon emissions from 1997 to 2019, which revealed an overall increase in emissions but a decline in emission intensity, with the spatial distribution characterized by high-high agglomeration. Some scholars have extended the investigation period and also came to the conclusion that carbon emissions are rising, and they suggested that the change of carbon emissions in China shows a “W” trend (Zhang et al., 2023). The second is the distribution and sources of carbon emissions. Su et al. proposed that the central and eastern carbon emissions are higher, while the western carbon emissions are lower (Su et al., 2023). Huang et al. (2025) found that eastern China’s emissions exceed those in the west due to rice field emissions, and southern China’s higher emissions than the north stem from agricultural input usage. In addition, the academic community has not yet reached a consensus on the major sources of carbon emissions. Some scholars argue that emissions are primarily derived from agricultural inputs like chemical fertilizers and pesticides (West and Marland, 2002), whereas others emphasize intestinal fermentation in livestock and unregulated agricultural waste discharge (Johnson et al., 2007). The third dimension focuses on influencing factors that encompass both internal agricultural development and external environmental factors. Internally, scholars have demonstrated an inverted U-shaped relationship between production efficiency and emissions (Zhu and Huo, 2022), while agricultural mechanization, scale, and specialization have been shown to drive emission increases (Wang et al., 2022; Guan et al., 2023). Conversely, agricultural branding and industrial structure upgrading are conducive to carbon reduction (Shi and Chang, 2023; Zhang et al., 2024a). External environmental studies mainly focus on policy and rural development characteristics. For example, Xia et al. (2024) found that environmental regulations can enhance agricultural carbon emission efficiency. Huang et al. (2024) proposed that the aging of the rural population can aggravate agricultural carbon emissions. The developmental trend of the digital age has prompted many scholars to focus on the issue of rural digital development empowering carbon reduction. While scholars generally agree that rural digital economy development can facilitate agricultural carbon reduction, the conclusions regarding the threshold, pathways, and regional heterogeneity of this effect are still divergent (Jin et al., 2024; Feng et al., 2025).
At present, scholars have conducted extensive analyses on agricultural carbon emissions, laying a theoretical foundation for analyzing the carbon reduction effect of rural digitalization. However, there are several key limitations in the existing studies. First, previous studies primarily focused on the impact of rural digital industry development on carbon emissions, such as digital finance and digital infrastructure (Li, 2023; Che et al., 2024), and there remains a lack of research that systematically examines a comprehensive enabling factor in influencing agricultural carbon—“rural digitalization”. Second, in terms of mechanisms, existing analyses tend to test intermediary variables such as technological innovation and human capital accumulation, but there are no in-depth dissections of the complex transmission pathways through which rural digitalization affects agricultural carbon intensity. Finally, discussions on the potential spatial spillover effects of rural digitalization’s carbon reduction impact are also insufficient.
Therefore, based on previous studies, this study employs panel data from 31 provincial administrative regions in China (2005-2022) to investigate the impact and mechanisms of rural digitalization on agricultural carbon intensity, thereby providing a theoretical basis for advancing rural digitalization and achieving carbon neutrality. The innovations of this study are threefold. First, it examines the effect of rural digitalization on agricultural carbon intensity, which adds empirical evidence to the agricultural sector regarding digital development’s carbon reduction effect. Second, this study introduces agricultural operation scale, agricultural total factor productivity, and human capital accumulation as mediating variables and constructs an integrated mechanism analysis framework. This approach can systematically reveal the multidimensional transmission pathways, which addresses the existing research gap in the exploration of its underlying mechanisms. Finally, this study investigates the spatial spillover effect of rural digitalization affecting agricultural carbon reduction, which provides theoretical support for realizing regional carbon reduction synergy and implementing policies according to local conditions.

2 Theoretical analysis and research hypotheses

The theory of sustainable development seeks to achieve integrated and balanced development across environmental sustainability, economic viability, and social inclusivity, while emphasizing equity both within and across generations. Within the agricultural sector, carbon emissions serve as a key indicator for assessing sustainable agricultural development. Under this theoretical framework, rural digitalization can advance agricultural carbon reduction and thereby contribute to sustainable development from environmental, economic, and social perspectives. First, data-driven approaches enabled by rural digitalization allow for the optimized allocation of agricultural production resources, reducing the excessive use of chemical inputs such as fertilizers and pesticides. Second, digital platforms help lower information asymmetry and facilitate the transition toward a green value chain in agriculture. Third, rural digitalization helps establish collaborative networks that connect multiple stakeholders, reducing institutional barriers to individual participation in emission reduction. Guided by the three pillars of sustainable development, rural digitalization plays a significant enabling role in achieving agricultural carbon reduction.

2.1 The direct effect of rural digitalization on agricultural carbon emissions

Agricultural carbon emissions are primarily derived from inputs in the production phase and waste management in the post-production phase (West and Marland, 2002; Johnson et al., 2007). During the production process, rural digitalization influences agricultural carbon emissions through several pathways. First, rural digitalization empowers agricultural digital production. Through the application of digital technologies, farmers can enhance the precision of their agricultural inputs and monitor the entire agricultural production process, thus reducing inefficient input allocation. Second, rural digital development offers a platform for farmers to access market information, production knowledge, and new technologies, thereby promoting the green transformation of agricultural production technology (Zhang et al., 2024b). Finally, rural digitalization provides the foundation for rural financial development. The digital financial industry can avoid the infrastructure construction costs required by traditional finance to operate, and it also lowers the threshold for the use of financial products by farmers. The development of digital finance provides financial support for the green transformation of agricultural production, thereby achieving carbon reduction (Chang, 2022). In the post-production process, rural digitalization uses digital tools to publicize the concept of waste management to farmers, while simultaneously improving the accuracy and standardization of waste management, and ultimately contributing to agricultural carbon reduction (Kurniawan et al., 2022). Therefore, this study puts forward the following hypothesis:
Hypothesis 1: Improving the rural digitalization level can effectively reduce the agricultural carbon intensity.

2.2 The indirect effect of rural digitalization on agricultural carbon emissions

Rural digitalization mainly achieves agricultural carbon reduction through three paths. First, it achieves agricultural carbon reduction by promoting scale agricultural operation. The digital development of rural areas facilitates the adoption of advanced machinery and equipment, enhances production and operational efficiency, and drives the transition from fragmented agricultural production to scale operation (Zhao et al., 2024). At the same time, by relying on Internet technology, the rural land property rights transfer market has been further optimized, which has encouraged farmers to expand their business scale. Large-scale farmers can achieve agricultural carbon reduction by improving the efficiency of agricultural inputs like chemical fertilizers (Zhu et al., 2018). Second, rural digitalization advances agricultural carbon reduction through human capital accumulation. The proliferation of digital networks has broken the spatiotemporal constraints on farmers’ access to productive knowledge, establishing an integrated online-to-offline education and training system that enhances the human capital of producers. This accumulation of human capital helps farmers to understand and adopt new green production models, while also promoting collaborative innovation between farmers and research institutions to develop green technological outcomes that better align with actual agricultural production conditions (Danquah and Amankwah-Amoah, 2017; Song et al., 2023), ultimately contributing to carbon emission reduction. Third, rural digitalization has achieved agricultural carbon reduction by improving agricultural total factor production. It brings perfect infrastructure and advanced production equipment, enables precise agricultural operations, improves the efficiency in the use of various production factors, reduces inefficiencies, and ultimately achieves agricultural carbon reduction. Therefore, this study puts forward the following hypothesis:
Hypothesis 2: Rural digitalization can inhibit agricultural carbon intensity by promoting scale agricultural operation, human capital accumulation and agricultural total factor productivity improvement.

2.3 The spatial effect of rural digitalization on carbon reduction

Carbon emissions themselves have externalities, and digital economic development is not restricted by traditional regions, which can promote the free flow of production factors among regions. Consequently, the level of digitalization in adjacent regions may not only affect the local level of digitalization but also influence local carbon emissions. The specific paths by which rural digitalization impacts the spatial effect of carbon reduction include the peer effect, learning effect and diffusion effect. 1) Peer effect. Adjacent regions have both competitive and leading relationships. The digital leading area, as an agricultural carbon reduction paradigm, reduces trial and error costs in the surrounding areas (Xiong and Zhou, 2024). When the economic development levels of adjacent regions are similar, fierce competition arises. Due to political performance considerations, the carbon reduction performance of a region can drive surrounding regions to pay more attention to agricultural carbon reduction. 2) Learning effect. The measures and experiences gained in leading regions can spread to the agricultural sectors of neighboring areas. 3) Diffusion effect. Due to the high construction cost of digital infrastructure, only areas with a higher economic level can realize rural digitalization in the early stage of development, and these areas form digital demonstration areas. According to the logic of “center-edge”, the development experience and technology of the demonstration area will radiate to the surrounding areas, forming a synergistic effect of carbon reduction. Therefore, this study puts forward the following hypothesis:
Hypothesis 3: The inhibitory effect of rural digitalization on agricultural carbon intensity has a spatial spillover effect.
Based on the above theoretical analysis, the theoretical framework diagram of this study is shown in Figure 1.
Figure 1 Theoretical analysis framework

3 Research design

3.1 Data sources

Considering the availability and completeness of relevant statistical data, this study selected 2005-2022 as the research period and used panel data from 31 provincial administrative regions in China. Because the data of Hong Kong, Macao and Taiwan of China are not easily obtained, the inter-provincial panel data does not include them, it is limited to an academic treatment. The data were derived from the China Statistical Yearbook, China Rural Statistical Yearbook, China Animal Husbandry Yearbook, China Science and Technology Statistical Yearbook, China Population and Employment Statistical Yearbook, China Labor Statistical Yearbook, China Culture and Related Industries Statistical Yearbook, China Tertiary Industry Statistical Yearbook, China Regional Economic Statistical Yearbook, China Tax Yearbook, China Financial Yearbook, China Civil Affairs Statistical Yearbook, People’s Republic of China Administrative Division Statistical Table and relevant statistical yearbooks of various provinces. For some missing data, the interpolation method was used to fill in the balanced panel data, and all currency-related variables were deflated with 2005 as the base period.

3.2 Description of variables

(1) Explained variable: agricultural carbon intensity. Referring to the research of Wang et al. (2024) agricultural carbon intensity was measured by dividing the total agricultural carbon emissions of each province over the years by the actual total agricultural output value. The calculation of total agricultural carbon emissions referred to existing literature (Li et al., 2021; Guan et al., 2023) and the seven indicators of chemical fertilizer, pesticide, agricultural film, diesel oil, machinery, irrigation and plowing were selected for estimation. The corresponding indicators were multiplied by the emission coefficient, and the resulting values were summed to obtain the total agricultural carbon emissions. The emission coefficients of these sources are as follows: chemical fertilizer 0.8956 kg kg-1, pesticide 4.9341 kg kg-1, agricultural film 5.18 kg kg-1, diesel 0.5927 kg kg-1, machinery 0.18 kg kW-1, irrigation 20.476 kg ha-1, and tillage 312.6 kg km-2.
(2) Core explanatory variable: rural digitization level. The core explanatory variable of this study is the level of rural digitization. Referring to the existing research ideas (Chang et al., 2024), a digital index system was constructed from four dimensions: development environment, service capacity, application level and development potential, with a total of 32 indicators. The corresponding indicators were calculated by using the publicly available data obtained from the statistical yearbooks. Variables related to the output value were deflated using a GDP deflator (Setting the base year 2005 index to 100), and variables related to income and expenditure were deflated using a consumer price index for rural residents with a base period of 2005=100 to eliminate the impact of inflation. On this basis, this study used the entropy method to scientifically measure the comprehensive development level of rural digitalization, in which the index attribute of urban and rural residents’ income level comparison (rural residents = 1) is negative, while other indicators are positive. The total assignment weight calculated by entropy method is 1, and the specific index weights are shown in Table 1.
Table 1 Rural digitalization level index system
Primary indicator Secondary
indicator		 Indicator name and unit Property Indicator weight
Development
environment		 Economic
condition		 Per capita actual output value of agriculture, forestry, animal husbandry and fishery (104 yuan person-1) + 0.030
Per capita disposable income of rural residents (104 yuan person-1) + 0.038
Comparison of income levels between urban and rural residents (rural residents = 1) - 0.008
Length of rural postal delivery lines (km km-2) + 0.061
Educational investment Rural residents’ consumption expenditure on education and entertainment (yuan person-1) + 0.029
Proportion of rural residents’ consumption expenditure on education and entertainment (%) + 0.010
Service capacity Infrastructure Long-distance automatic switchboard capacity (road terminals (104 persons) -1) + 0.049
Local telephone office switchboard capacity (doors person-1) + 0.051
Mobile telephone switchboard capacity (households person-1) + 0.032
Length of long-distance optical cable line (km km-2) + 0.037
Business scale Per capita business volume of the telecommunications industry (104 yuan person-1) + 0.090
Number of fixed telephone users per capita (per person) + 0.032
Number of mobile phone users per capita (per person) + 0.021
Per capita mobile phone call time (104 minutes person-1) + 0.013
Per capita business volume of mobile text message (104 messages person-1) + 0.052
Average number of Internet users per capita (per person) + 0.028
Application level Investment situation Rural residents’ consumption expenditure on transportation and communication (yuan person-1) + 0.040
Proportion of rural residents’ consumption expenditure on transportation and communication (%) + 0.010
Terminal use Comprehensive population coverage rate of rural radio programs (%) + 0.005
Comprehensive population coverage rate of rural TV programs (%) + 0.005
Popularity of color TV in rural areas (units/100 households) + 0.008
Rural mobile phone popularity (departments/100 households) + 0.015
Computer popularity in rural areas (units/100 households) + 0.055
Number of Internet broadband access ports per capita (number per person) + 0.051
Proportion of rural broadband access users (%) + 0.023
Proportion of administrative villages with Internet broadband services (%) + 0.003
Development
potential		 Industrial level Proportion of main business income of the telecommunications industry to GDP (%) + 0.031
Number of employees as a percentage of total employed persons (%) + 0.066
Proportion of related investment to total social fixed asset investment (%) + 0.032
Proportion of tax revenue from the digital information industry (%) + 0.033
Rural demand Added value of the primary industry accounts as a proportion of regional GDP (%) + 0.025
Proportion of rural population to total population (%) + 0.017
(3) Control variables. Referring to previous studies (Xu et al., 2023; Feng et al., 2025), this study selected eight control variables. First, agricultural planting structure was measured as the proportion of grain sown area (103 ha) to total crop sown area (103 ha). Second, agricultural industrial structure was represented by the proportion of agricultural output value (108 yuan) to the total output value of agriculture, forestry, animal husbandry and fishery (108 yuan). Third, the natural disaster impact was conveyed by the proportion of the affected area (103 ha) to the total sown area of crops (103 ha). Fourth, urbanization level was measured by the proportion of urban population. Fifth, industrialization level was expressed by the proportion of industrial added value (108 yuan) to regional GDP (108 yuan). Sixth, the openness level was expressed by the proportion of the import and export amount of goods (108 yuan) to the regional GDP (108 yuan). The value of goods imported and exported was represented by the total value of goods imported and exported by location of the operating unit and converted into yuan. Seventh, land quality was measured by dividing the effective irrigated area (103 ha) by the total sown area of crops (103 ha). Eighth, agricultural mechanization level was obtained by dividing the total power of agricultural machinery (104 kW) by the number of employees in the rural primary industry (104 persons).
(4) Mechanism variables. First, referring to the research of Zhu and Wang (2023), scale agricultural operation was measured by using the ratio of the total sown area of crops (103 ha) to the number of employees in the rural primary industry (104 persons). Second, for human capital accumulation, referring to the existing literature (Li et al., 2022), the level of human capital was measured by the average educational level of rural residents. For the calculation method: 0 years was assigned to the population of rural residents who had not attended school, 6 years was assigned to primary school education, 9 years was assigned to junior high school education, 12 years was assigned to high school education, and 15 years was assigned to college and above. Third, agricultural total factor productivity was calculated by referring to the practice of Li et al. (2017).
This study employed the DEA-Malmquist index method based on input-oriented variable returns to measure the growth of agricultural total factor productivity. In terms of output, the actual total output value of agriculture, forestry, animal husbandry and fishery across 31 provinces from 2005 to 2022 (108 yuan) was selected as the expected economic output. In terms of inputs, this study selected four primary indicators and 11 secondary indicators as input variables, with the measurement indicators shown in Table 2 and Table 3.
Table 2 Measurement of agricultural total factor productivity indicators
Indicator type Primary indicator Secondary indicator Indicator meaning
Input indicators Land input Land input Total crop sown area (103 ha)
Labor input Rural labor input Number of employees in rural primary industry (104 persons)
Other labor inputs Number of employees in agriculture, forestry, animal husbandry and fishery in urban units (104 persons)
Capital input Mechanical input Total power of agricultural machinery (104 kW)
Draught animal input Year-end stock of large livestock (104 heads)
Fertilizer input Agricultural chemical fertilizer application amount calculated by the purity conversion method (104 t)
Pesticide input Pesticide usage (104 t)
Agricultural film input Agricultural plastic film usage (104 t)
Resource input Water resources input Effective irrigated area (103 ha)
Energy input Rural electricity consumption (108 kWh)
Fuel input Agricultural diesel consumption (104 t)
Output indicators Expected output Economic output Actual total output value of agriculture, forestry, animal husbandry and fishery (108 yuan)
Table 3 Descriptive statistics of the variables
Variable Variable code Sample size Average Standard deviation Minimum Maximum
Agricultural carbon intensity ACI 558 0.138 0.051 0.036 0.323
Rural digitization level RDL 558 0.251 0.083 0.118 0.564
Agricultural planting structure APS 558 0.654 0.136 0.328 0.971
Agricultural industrial structure AIS 558 0.521 0.086 0.302 0.746
Natural disaster impact NDI 558 0.182 0.145 0.000 0.936
Urbanization level UL 558 0.554 0.147 0.209 0.896
Industrialization level IL 558 0.360 0.102 0.068 0.536
Openness level OL 558 0.288 0.345 0.008 1.721
Land quality LQ 558 0.439 0.190 0.154 1.233
Agricultural mechanization level AML 558 4.461 2.379 0.739 13.689
Scale agricultural operation SAO 558 7.419 4.002 2.551 29.362
Human capital accumulation HCA 558 7.543 0.860 3.804 9.877
Agricultural total factor productivity ATFP 558 1.738 0.825 0.609 5.381

3.3 Model construction

(1) Dynamic panel model. To address endogeneity problems, the dynamic panel model was introduced in this study (Blundell and Bond, 1998). The major difference between the dynamic panel model and the static panel model is that the dynamic panel model introduces the explained variable with a lag of one period as the explanatory variable (Elhorst, 2012), so it can analyze the impact of agricultural carbon intensity on itself. Given that the evolutionary patterns of phenomena often exhibit inertia, integrating dynamic cumulative effects into the model enhances its realism. Therefore, this study constructed the dynamic panel model as follows:
${{Y}_{it}}={{\beta }_{0}}+{{\beta }_{1}}{{Y}_{it-1}}+{{\beta }_{2}}RD{{L}_{it}}+\sum\limits_{n=3}^{N}{{{\beta }_{n}}{{X}_{nit}}}+{{c}_{i}}+{{\eta }_{t}}+{{\varepsilon }_{it}}$
In the above formula, i represents each provincial region; t denotes the year; Yit indicates the agricultural carbon intensity; Yit-1 indicates the explained variable lagging by one period; RDLit represents the rural digitalization; Xit indicates the control variable, including a series of possible influencing factors; β is a coefficient matrix; ci and ηt represent region and time non-observed effects, respectively; and εit represents random disturbance terms.
(2) Mediating effect model. To further investigate the mechanism by which rural digitalization affects agricultural carbon intensity, this study referred to the research of Jiang (2022). To avoid endogeneity problems caused by mediating variables, the “two-step method” was employed to verify the action mechanism of rural digitalization on agricultural carbon intensity. Specifically, only the impact of the core explanatory variables on the mediating variables was empirically tested, while the influence of mediating variables on the explained variables was explained theoretically. This study constructed the following mediating effect model for analysis:
${{H}_{it}}={{\beta }_{0}}+{{\beta }_{1}}{{H}_{it-1}}+{{\beta }_{2}}RD{{L}_{it}}+\sum\limits_{n=3}^{N}{{{\beta }_{n}}{{X}_{nit}}}+{{c}_{i}}+{{\eta }_{t}}+{{\varepsilon }_{it}}$
where, Hit represents the mediating variable, and other variables have the same meanings as in the dynamic panel model formula.
(3) Spatial Dubin model. With the increasing inter-provincial economic exchanges, the agricultural carbon intensity of a single province may not only depend on the local economic and social development, but it also be closely related to the digital development of rural areas in neighboring areas, so there are spatial spillover effects. Before conducting an empirical analysis using spatial econometric models, a spatial correlation test was performed on the explained variable. According to Tobler’s First Law of Geography, “everything is related to everything else, but near things are more related to each other”. Drawing on existing research (Jiang and Chen, 2025), this study employed the Global Moran’s I index to examine whether there is spatial autocorrelation in agricultural carbon emission intensity across the regions. The Moran’s I index values range from -1 to 1. A positive value (Moran’s I>0) indicates positive spatial autocorrelation, while a negative value (Moran’s I<0) suggests negative spatial autocorrelation. The specific calculation formula is as follows:
$Morans\begin{matrix} {} \\\end{matrix}I=\frac{\underset{i=1}{\overset{N}{\mathop \sum }}\,\underset{j=1}{\overset{N}{\mathop \sum }}\,{{\omega }_{ij}}\left( {{y}_{i}}-\bar{y} \right)\left( {{y}_{j}}-\bar{y} \right)}{\underset{i=1}{\overset{N}{\mathop \sum }}\,\underset{j=1}{\overset{N}{\mathop \sum }}\,{{\omega }_{ij}}\underset{i=1}{\overset{N}{\mathop \sum }}\,{{\left( {{y}_{i}}-\bar{y} \right)}^{2}}}$
where, yi is the agricultural carbon intensity of province i; ȳ is the average value; and ωij represents the standardized spatial geographic weight matrix.
Referring to relevant studies (Halleck Vega and Elhorst, 2015), a spatial econometric model for further testing was selected on the basis of benchmark regression. After comparison with the spatial autoregressive model (SAR) and spatial error model (SEM), a spatial Durbin model (SDM) was constructed, and the spatial lag term of the explained variable and the spatial lag term and spatial error term of the explanatory variable were added to the model at the same time:
$\begin{align} & {{Y}_{it}}={{\beta }_{0}}+\rho \underset{j=1,j\ne i}{\overset{N}{\mathop \sum }}\,{{\omega }_{ij}}{{Y}_{it}}+{{\beta }_{1}}RD{{L}_{it}}+ \\ & \ \ \ \ \ \ \ \sum\limits_{n=2}^{N}{{{\beta }_{n}}{{X}_{nit}}}+\phi \underset{j=1}{\overset{N}{\mathop \sum }}\,{{\omega }_{ij}}{{Z}_{ijt}}+{{\mu }_{i}}+{{v}_{t}}+{{\varepsilon }_{it}} \\ \end{align}$
In the above formula, μi and νt represent the regional effect and time effect, respectively; and ρ represents the spatial autoregressive coefficient; $\phi $ represents the coefficient to be estimated. To simplify the spatial Durbin model formula, we used Z to represent all explanatory variables, which were used to add spatial lag terms. Finally, ωij represents the spatial weight matrix, and the other variables have the same meanings as in the dynamic panel model formula.

4 Empirical analysis

4.1 Benchmark regression

In the benchmark regression, a panel data model was used to examine the impact of rural digitalization level on agricultural carbon intensity at the national level. Using Stata18.0 software, the estimation results are shown in Table 4. In this study, the dynamic panel model was mainly used for analysis and interpretation, and the method of system generalized method of moments (SYS-GMM) was emphatically utilized (Xia, 2020). The estimation results of system GMM are shown in Model 5 in Table 4. For comparison, models 1-4 report the results from pooled OLS, fixed effects, random effects, and difference GMM estimations, respectively.
Table 4 Estimation results of rural digitalization level on agricultural carbon intensity
Variable Model 1
OLS		 Model 2
FE		 Model 3
RE		 Model 4
GMM		 Model 5
GMM
RDL -0.168*** -0.082* -0.078* -0.105*** -0.140***
(0.043) (0.042) (0.044) (0.012) (0.016)
APS 0.055*** -0.033 0.018 0.066*** 0.037
(0.013) (0.023) (0.021) (0.012) (0.041)
AIS 0.250*** 0.013 0.047* 0.058*** 0.051***
(0.017) (0.026) (0.026) (0.010) (0.013)
NDI 0.101*** 0.012 0.023*** -0.003 -0.002
(0.011) (0.008) (0.008) (0.002) (0.002)
UL -0.049** -0.379*** -0.307*** -0.130*** 0.038
(0.024) (0.028) (0.028) (0.016) (0.023)
IL 0.132*** -0.024 0.027 -0.084*** -0.038***
(0.018) (0.022) (0.022) (0.007) (0.006)
OL 0.032*** 0.016** 0.027*** 0.001 0.004**
(0.007) (0.007) (0.007) (0.002) (0.002)
LQ 0.062*** -0.009 0.014 0.024*** 0.026***
(0.010) (0.015) (0.014) (0.003) (0.005)
AML -0.001 0.000 -0.001 -0.003*** -0.003***
(0.001) (0.001) (0.001) (0.001) (0.001)
L. ACI 0.629*** 0.824***
(0.029) (0.030)
Constant term -0.059*** 0.389*** 0.267*** 0.106*** -0.001
(0.017) (0.031) (0.029) (0.014) (0.032)
R2 0.592 0.152 0.246
AR (1) 0.015 0.009
AR (2) 0.202 0.126
Sargan test 0.547 0.999
N 558 558 558 496 527

Note: ***, ** and * indicate significance at the statistical levels of 1%, 5% and 10%, respectively, and the values in parentheses are standard errors. The significance probability P-values were obtained using the Arellano-Bond and Sargan tests. The change in the sample size N is due to model specification rather than human modification. The same applies to the following tables.

The dynamic panel model adds the lagged dependent variable to the explanatory variable of the model, so it can better overcome the interference of endogeneity problems. After comparing autocorrelation test results across different lag orders of the dependent variable, the optimal lag order was determined as a first-order lag. That is, the first-order lag variable of the explained variable was used as the explanatory variable, and the second-order lag of the explained variable was employed as the instrumental variable for the first-order lag variable in the regression. In addition, in the dynamic panel model, the rural digitalization level with a one-period lag was set as the instrumental variable for the core explanatory variable to effectively address any endogeneity problems. In Model 5, the P-value of the Sargan’s test exceeds 0.1, indicating no over-identification, so the instrumental variables of lag terms are valid overall.
At the same time, based on the significance probability P from the Arellano-Bond test, the P-value of the AR(1) statistic test is less than 0.05, while the P-value of the AR(2) statistic test is greater than 0.1. This suggests that the random error term exhibits first-order autocorrelation but no second-order autocorrelation, implying that the system GMM passes the autocorrelation test and the estimation results are reliable. According to the estimation results of Model 5, the lag of agricultural carbon intensity has a significant impact, highlighting the necessity of considering the influence of the explained variable on itself in the model. Thus, the model specification is appropriate.
In Model 5, the estimated coefficient of the core explanatory variable (rural digitalization level) is -0.140 and statistically significant at the 1% level. This indicates that after fully addressing endogeneity issues, the level of rural digitalization development has a significant negative impact on agricultural carbon intensity. Specifically, for every unit increase in the rural digitalization level, agricultural carbon intensity decreases by an average of 0.14 units. Rural digital development can suppress agricultural carbon emissions, thereby promoting agricultural carbon reduction, and supporting Hypothesis 1. Rural digitalization facilitates agricultural carbon emission reduction by precisely identifying and targeting key sources of emissions. At the input stage, smart fertilization and irrigation systems lower the emissions linked to fertilizer production and machinery fuel consumption. During production, data-driven management of livestock manure suppresses methane release, while intelligent paddy water level control reduces methane generation. In the post-harvest stage, optimized cold-chain logistics cut down energy use in distribution. The core mechanism lies in replacing experience-based decisions with data-driven approaches, thus enabling systematic monitoring and the reduction of carbon emissions across the entire agricultural chain.
In terms of control variables, the coefficient of agricultural industrial structure in Model 5 is positive and significant at the 1% level, indicating that agricultural industrial structure has a positive impact on agricultural carbon intensity. Specifically, an excessively high proportion of the planting sector in the output value of agriculture, forestry, animal husbandry and fishery will lead to an increase in agricultural carbon intensity. The levels of industrialization and agricultural mechanization have a significantly negative impact on agricultural carbon emission intensity, which is consistent with previous research conclusions. Developing industries and improving the agricultural mechanization level can effectively promote the reduction of agricultural carbon intensity. The level of opening up to the outside world and land quality have significantly positive impacts on agricultural carbon emission intensity, indicating that resource advantages in agricultural carbon emissions have not yet formed a driving force for emission reduction. A good regional opening-up level and land quality are critical conditions for expanding agricultural production, but in this process, they also lead to an increase in agricultural carbon intensity. The estimated coefficients of agricultural planting structure, natural disaster impact degree and urbanization level are not significant. One likely reason is that China’s agricultural industry is facing further adjustment, transformation and upgrading, so the specific impact effects and action mechanisms need to be further investigated.

4.2 Robustness test

4.2.1 Replacing the explained variable

The explained variable of agricultural carbon intensity was replaced with agricultural carbon density for the robustness test. Agricultural carbon density is another important indicator for investigating agricultural carbon reduction. According to the research of Wu et al. (2023), the agricultural carbon density was calculated by dividing each province’s total agricultural carbon emissions by the total area planted with crops. The results are shown in Model 6 in Table 5. The improvement in the rural digital development level has a significant negative impact on agricultural carbon density, which is conducive to agricultural carbon reduction.
Table 5 Robustness test
Variable Model 6
Replacing the
explained variable		 Model 7
Increasing the
control variables		 Model 8
Removing outlier
provinces		 Model 9
Eliminating
outlier years		 Model 10
Winsorization
RDL -0.034*** -0.133*** -0.179*** -0.137*** -0.150***
(0.005) (0.014) (0.020) (0.017) (0.016)
Control variable Yes Yes Yes Yes Yes
FSL -0.100***
(0.028)
L. ACI 0.962*** 0.802*** 0.799*** 0.791*** 0.820***
(0.015) (0.040) (0.055) (0.019) (0.035)
Constant term 0.002 -0.014 0.070 -0.020 -0.009
(0.005) (0.043) (0.071) (0.016) (0.021)
AR (1) 0.031 0.008 0.022 0.019 0.001
AR (2) 0.256 0.110 0.215 0.072 0.365
Sargan test 0.999 0.999 0.999 0.998 0.999
N 527 527 459 434 527

4.2.2 Increasing the control variables

Financial support for agriculture was added as an additional control variable to validate the robustness of the results. The financial support level for agriculture is expressed by the proportion of fiscal expenditure on agriculture, forestry and water affairs (108 yuan) to the total expenditure of the general budget (108 yuan). Due to different statistical calibers, the data from 2005 to 2006 are expressed by the ratio of the sum of agricultural expenditure (108 yuan), forestry expenditure (108 yuan) and the agriculture, forestry, water conservancy and meteorology departments (108 yuan) to the total fiscal expenditure (108 yuan). The results are shown in Model 7 in Table 5.

4.2.3 Eliminating outliers and winsorization

Among the provincial administrative divisions in China, municipalities directly under the Central Government have certain atypical features. Accordingly, the observations for Beijing, Tianjin, Shanghai, and Chongqing were considered outliers and excluded. The regression results after eliminating municipalities directly under the Central Government are shown in Model 8 in Table 5. Since the sample time range includes 2020-2022, this period was objectively affected by the novel coronavirus epidemic, which had a certain degree of interference with China’s economic and social development and this study. Model 9 in Table 5 shows the regression results after excluding these outlier years. In addition, each variable in the model was winsorized by 5% to eliminate the influence of extreme data. The corresponding coefficient estimation results are shown in Model 10 in Table 5, which is largely consistent with the benchmark regression results, indicating that this study is robust.

4.3 Heterogeneity analysis

Regional differences are an important factor contributing to the heterogeneity of influencing effects. At present, 31 provinces in the mainland of China can be divided into a major grain producing area, a major grain sales area and a production-sales balanced area ① including Beijing, Tianjin, Shanghai, Zhejiang, Fujian, Guangdong and Hainan, which have a large amount of grain transferred, are the main sales areas; 11 provinces (autonomous regions and municipalities directly under the Central Government) including Shanxi, Guangxi, Guizhou, Yunnan, Chongqing, Tibet, Shaanxi, Gansu, Qinghai, Ningxia and Xinjiang are production and sales balanced areas.) based on grain output, per capita possession and commodity grain stocks. Based on the original variables, dummy variables were generated for each region (1=yes, 0=no) for the major grain producing area, major grain sales area, and production-sales balanced area. Subsequently, interaction terms between the core explanatory variables and regional variables were generated, and these interaction terms were separately included in the model for estimation. The regional heterogeneity of the influencing effects was analyzed according to the positive and negative coefficients and significance levels of the interaction terms. Models 11-13 in Table 6 show the system GMM model estimation results of increasing interaction terms in each of the three districts, respectively. The results show that rural digitalization has no significant regional effect on reducing agricultural carbon intensity in the major grain producing area or the main grain sales area, while in Model 13, the coefficient of interaction terms is significantly negative, indicating that rural digitalization has a stronger inhibitory effect on agricultural carbon intensity in the production-sales balanced area. This variation in regional effects stems from structural characteristics of agricultural systems and their compatibility with digital technologies. In major grain-producing areas, a long-term focus on yield maximization has led to technological path dependence, where digitalization is often prioritized for increasing output rather than reducing emissions, resulting in limited marginal abatement potential. In major grain-consuming areas, the smaller agricultural scale and a low carbon emission baseline make digital applications less economically viable, leading to negligible effects. In contrast, in production-marketing balanced regions, the relatively balanced agricultural structure allows digital tools greater flexibly to optimize resource allocation and production processes. Through precision management, these technologies simultaneously enhance efficiency and reduce emissions, thereby exerting a stronger inhibitory effect on agricultural carbon intensity.
Table 6 Regression results based on regional heterogeneity analysis
Variable Model 11
Main grain producing area		 Model 12
Major grain sales area		 Model 13
Production-sales balanced area
RDL -0.147*** -0.167*** -0.116***
(0.016) (0.012) (0.014)
Whether it is in the major grain producing area * RDL 0.000
(0.031)
Whether it is in the major grain producing area -0.005
(0.019)
Whether it is in the major grain sales area * RDL 0.046
(0.037)
Whether it is in the major grain sales area 0.002
(0.016)
Whether it is in the production and sales balanced area * RDL -0.065***
(0.024)
Whether it is in the production and sales balanced area 0.009
(0.012)
Control variable Yes Yes Yes
L. ACI 0.804*** 0.806*** 0.804***
(0.031) (0.030) (0.028)
Constant term -0.011 -0.018 -0.003
(0.020) (0.018) (0.018)
AR (1) 0.010 0.009 0.009
AR (2) 0.108 0.109 0.107
Sargan test 0.762 0.748 0.791
N 527 527 527

Note: “Whether it is in the major grain producing area * RDL” is the interaction term between “Whether it is in the major grain producing area” and “RDL”, representing the product of the two variables. The same applies to the similar variables in the table.

4.4 Mechanism study

To further analyze the impact path of rural digitalization on agricultural carbon emissions, this study conducted a mechanism test, and the empirical results are shown in Table 7. Model 14 shows that rural digitalization can effectively promote scale agricultural operation, and this result passes the 1% significance level test. Model 15 shows that rural digitalization can promote human capital accumulation at the 1% significance level. Model 16 confirms the positive impact of rural digitalization on agricultural total factor productivity. Combined with the previous theoretical analysis and the research of other scholars, the influences of mediating variables on agricultural carbon emissions was verified (Zhu et al., 2018; Huang et al., 2021), thus confirming Hypothesis 2.
Table 7 Regression results based on mechanism studies
Variable Model 14
Scale agricultural operation		 Model 15
Human capital accumulation		 Model 16
Agricultural total factor productivity
RDL 8.347*** 1.473*** 3.749***
(0.524) (0.313) (0.491)
Control variable Yes Yes Yes
L. SAO 0.860***
(0.015)
L. HCA 0.533***
(0.039)
L. ATFP 0.914***
(0.020)
Constant term 0.876* 1.574*** -0.471
(0.513) (0.475) (0.298)
AR (1) 0.005 0.009 0.014
AR (2) 0.129 0. 698 0.240
Sargan test 0.651 0.588 0.695
N 527 527 527

4.5 Spatial effect analysis

Before conducting an empirical analysis using the spatial econometric model, this study performed a spatial correlation test on the explained variables. If the agricultural carbon intensity passes the significance test of spatial autocorrelation, then it is suitable for spatial effect analysis. Common geographical weight matrices include the 0-1 adja-cency matrix, geographical distance weight matrix, economic distance weight matrix, and economic geography nested matrix. After comparing regression results, this study selected the 0-1 adjacency matrix as the primary matrix and used the economic distance weight matrix as a control to ensure robustness. Based on the provincial agricultural carbon intensity data from 2005 to 2022, combined with the calculation formula described above, the Moran’s I and its significance over the years were obtained, as shown in Table 8. According to the Moran’s I results over the years, all of them passed the significance test at levels of at least 10%, and the values are greater than 0, indicating that there is a significant spatial positive correlation between the agricultural carbon intensities in various provinces, that is, there is a spatial spillover effect. Specific regional identification shows that China’s agricultural carbon intensity has significant local spatial aggregation characteristics, and the spatial pattern is primarily low-low aggregation and high-high aggregation.
Table 8 Global Moran’s I index table
Year Moran’s I Year Moran’s I Year Moran’s I Year Moran’s I Year Moran’s I Year Moran’s I
2005 0.190** 2008 0.122* 2011 0.151** 2014 0.184** 2017 0.246*** 2020 0.330***
2006 0.179** 2009 0.171** 2012 0.200** 2015 0.198** 2018 0.265*** 2021 0.399***
2007 0.189** 2010 0.170** 2013 0.192** 2016 0.237*** 2019 0.301*** 2022 0.360***
Based on the spatial correlation analysis results, the spatial weight matrix was introduced into the model, and an appropriate spatial econometric model was selected for estimation. The selection criteria of the spatial lag model, spatial error model and spatial Durbin model drew lessons from existing research ideas. The pre-estimation test used the spatial Lagrange Multiplier (LM) test to determine whether there are spatial error effects and spatial lag effects. If they have both, then the spatial Durbin model should be selected. The post-hoc test was divided into three steps. The first step was to judge whether the spatial regression model is suitable for fixed effects or random effects by the Hausman test. Second, the likelihood ratio (LR) test was conducted. Assuming the use of a spatial Durbin model, pairwise comparisons are made to determine whether the SDM degenerates into a spatial error model (SEM) or a spatial lag model (SAR). The third step was to judge whether the SDM is degenerated into a SEM or a SAM by the Wald test. The type of spatial econometric model suitable for the data sample was determined by comprehensively comparing different test results. The results for spatial econometric model selection are presented in Table 9. The LM test and Robust-LM test results show that there are spatial error and spatial lag effects. Under the premise of confirming spatial correlation between variables, the spatial Durbin model was initially selected. The Hausman test statistic is significant at the 1% level, suggesting the adoption of a fixed-effect model. The post-hoc test further verified the applicability of the SDM with spatial lag and spatial error terms. In Table 9, the results of the LR test and Wald test both support the selection of the SDM model, that is, indicating that it does not degenerate into the SAR or SEM. Therefore, this study selected the spatial Durbin model for subsequent spatial effect analysis.
Table 9 Test results of spatial econometric model selection
Test method Statistical value P-value
LM test, no spatial error 294.551*** <0.001
Robust LM test, no spatial error 98.633*** <0.001
LM test, no spatial lag 196.053*** <0.001
Robust LM test, no spatial lag 0.135 0.713
Hausman 52.080*** <0.001
LR Lag 129.180*** <0.001
LR Err 123.520*** <0.001
Wald Lag 27.350*** 0.001
Wald Err 23.770*** 0.005
According to the relevant test results, the SDM with two-way fixed effects was used for spatial econometric regression, and the results are shown in Table 10. Model 17 employed a 0-1 adjacency matrix, while Model 18 used an economic distance weight matrix. The significance of the spatial autocorrelation coefficients indicates the presence of spatial autocorrelation and a spatial spillover effect, meaning that the agricultural carbon intensity of one province affects the values of surrounding provinces. In Model 17, the estimated coefficient of rural digitalization is -0.214 and significant at the 1% level. For every 1 unit of increase in the level of rural digitalization, the agricultural carbon intensity decreases by 0.214 units, confirming that considering the spatial effect, rural digitalization significantly inhibits agricultural carbon intensity. The estimated results remain robust when the weight matrix is replaced in Model 18. Further decomposition of the direct and indirect effects using the bidirectional fixed effects estimates of the SDM yields values for the direct effect, indirect effect, and total effect, as also shown in Table 10. For the core explanatory variable, both the direct and indirect effect coefficients of rural digitalization are negative, indicating that rural digitalization in both local and neighboring provinces has a negative impact on local agricultural carbon intensity. Digital transformation is characterized by breaking spatial constraints, so it facilitates the coordination of multiple resources to improve allocation efficiency, fully unleashes the cross-regional driving effect, and jointly promotes agricultural carbon reduction—validating Hypothesis 3 that the inhibitory effect of rural digitalization on agricultural carbon intensity has spatial spillover effects. In addition, the estimated coefficient of the total effect of rural digitalization is significantly negative, consistent with the directions of the direct and indirect effects. The total effects value typically represents the average impact of a province’s rural digitalization level on the overall explained variable within its region. A larger total effect implies a stronger inhibitory effect of rural digitalization in one region on agricultural carbon intensity in both itself and surrounding areas, highlighting the growing importance of inter-regional communication and cooperation.
Table 10 Regression results and decomposition of the spatial Durbin model
Variable Model 17
0-1 Adjacency
matrix		 Model 18
Economic distance
weight matrix
RDL -0.214*** -0.249***
(0.057) (0.056)
Control variable Yes Yes
Wx -0.129 -0.451***
(0.097) (0.150)
ρ -0.169*** -0.169**
(0.061) (0.082)
σ²_e 0.000*** 0.000***
(0.000) (0.000)
Direct effect -0.209*** -0.234***
(0.060) (0.059)
Indirect effect -0.081 -0.361***
(0.092) (0.137)
Total utility -0.290*** -0.595***
(0.082) (0.134)
R2 0.157 0.208
N 558 558

5 Discussion

As an important source of carbon emissions, the agricultural sector is confronting the urgent task of achieving green transformation to meet the carbon neutrality goal. The distinctive advantages of digital technology present an opportunity for agricultural carbon reduction. Therefore, this study analyzed the impact of rural digitalization on agricultural carbon intensity, and the results show that rural digitalization development can significantly reduce agricultural carbon intensity, which is consistent with certain existing studies (Jin et al., 2024). Technologies such as intelligent monitoring, agricultural machinery emission reduction, and water-saving irrigation contribute to precise agricultural production management and directly reduce agricultural carbon emissions. In addition, data-driven farmer decision-making and agricultural circular models play an important role in the low-carbon transformation of the agricultural sector. By optimizing the entire agricultural industry chain, these approaches help to achieve ecological synergies in agricultural production.
To provide better policy implications for agricultural carbon reduction, this study further analyzed the action mechanism through which rural digitalization influences agricultural carbon emissions. These findings demonstrate that by encouraging scale agricultural operation, accumulating human capital, and raising total factor productivity, rural digitization can lower the agricultural carbon intensity. Many scholars have conducted empirical research on the mediating impacts of scale agricultural operation and human capital formation, which supports the findings of this study (Zhao et al., 2023). Despite differences in agricultural structures and policy environments across countries, the findings of our study still offer significant policy implications for many nations currently undergoing transformation toward agricultural modernization. For countries with substantial potential for scale agricultural operation, the policies should pay attention to the “scale operation” pathway, leveraging digital platforms to facilitate land transfer and consolidation while promoting the adoption of precision agriculture technologies. This approach aims to ensure yield stability while optimizing the use of agricultural inputs. For countries dominated by smallholder farmers and in urgent need of human capital enhancement, such as Kenya, Tanzania and others, we can apply the “human capital accumulation” pathway. This involves using inclusive digital education platforms and mobile services to lower the barriers to learning and applying green production technologies. In Africa, practices such as the “ICow” and “FarmDriver” digital platforms, which have helped some farmers increase their incomes, have demonstrated the transformative power of rural digitalization.
Considering that the data used in this study represent a national sample of China, there are differences in the economic development levels and natural conditions in different regions, so this study conducted a heterogeneity analysis. The results show that the inhibitory effect of rural digitalization on agricultural carbon emissions is only significant in the production-sales balanced area, but not in the major grain producing area or the major grain sales area. This conclusion is consistent with the findings of Chen and Li (2024), who argued that the carbon reduction effect of digital transformation is not significant in major grain-producing areas. They attributed this primarily to the characteristics of high carbon emissions and low digitalization levels in the central, western, and northeastern regions where these areas are often located, which consequently suppresses the potential emission reduction benefits of digitalization. Building on this existing research, the present study further categorized the regions into major grain-producing areas, major grain-consuming areas, and production-marketing balanced areas, thereby extending and corroborating prior findings. We propose that the underlying reasons for these results stem not only from regional developmental disparities but also from the distinct functional roles and production pressures inherent to each region. In major grain-producing areas, under policy constraints, digital investments are prioritized for yield-increasing technologies; in major grain-sales areas, more digital elements are invested in secondary and tertiary industries, marginalizing the agricultural sector; while in production-sales balanced areas, digital technology serves the dual goals of quality improvement and emission reduction. Furthermore, based on regional differences, a spatial effect test was carried out in this study. The results illustrate that there are spatial spillover effects in both carbon emissions and rural digitalization, which is consistent with previous scholarly research findings (Guo, 2024; Jin et al., 2024). Moreover, the Chinese government and relevant departments should not limit their carbon reduction policy design to “local management”, but rather shift toward spatial collaborative governance. Meanwhile, the international community should actively encourage technology demonstration, knowledge sharing, and infrastructure connectivity at both regional and national levels to build a collaborative governance network while promoting digital agriculture for emission reduction.
This study makes three main marginal contributions. First, unlike some scholars who approached the issue solely from a digital perspective (Zhao et al., 2023; He et al., 2025), this study explored the impact of rural indicators on agricultural carbon reduction with a stronger correlation. In the process of China’s agricultural development, there are differences in agricultural development in different regions, and the agricultural industry is mainly developed in rural areas. Therefore, exploring the impact of rural digitalization on agricultural carbon emissions is of more practical significance, and it also provides agricultural evidence for the carbon reduction effect of digital economy development. Second, although some scholars have analyzed the impact path of rural digitalization on agricultural carbon emissions, most of them focus on human capital accumulation, technological innovation and social services (Jin et al., 2024). In this study, scale management and agricultural total factor productivity innovation were introduced into the model, and their mediating effect was analyzed, which made up for some gaps in the current research. Finally, unlike the group regression of samples based on geographical regions and economic development levels (Jin et al., 2024), this study conducted a heterogeneity analysis from the perspective of food security, according to major grain producing area, major grain sales area, and production-sales balanced area, so as to deeply understand the differences in the impact of rural digitalization on carbon reduction in different regions.
Although this study offers an innovative perspective, it also has some limitations. First, this study conducted a qualitative analysis of the impact of rural digitalization on agricultural carbon intensity. However, it did not examine whether this relationship is nonlinear; in other words, it could not determine whether the reduction effect varies across different thresholds of rural digitalization or agricultural carbon emissions. Second, this study used the total amount of carbon emissions when measuring agricultural carbon emissions. If carbon emissions could be further divided into planting and breeding, we could better understand which sub-industry in the agricultural field is the most affected by rural digitalization. Finally, agricultural production conditions vary in multiple aspects across different regions. Limited by data availability, this study could not cover all these differences. Therefore, the use of macro-level data may have inevitable defects. Our team will supplement more micro-data in the future to analyze the impact of rural digitalization on carbon emissions and threshold effects of different agricultural industries.

6 Conclusions

In the context of the digital age, this study constructed indicators for rural digitalization and agricultural carbon emissions, and then employed a dynamic panel model to analyze the relationship between rural digitalization and agricultural carbon emissions. A mediating effect model was further used to explore the different pathways through which rural digitalization influences agricultural carbon emissions. The specific conclusions are fourfold. First, rural digitalization can effectively promote agricultural carbon reduction, and this result passed the robustness test. Second, the heterogeneity analysis demonstrated that rural digitalization has no significant effect on reducing agricultural carbon emission intensity. However, in production-sales balanced areas, rural digitalization demonstrates a stronger inhibitory effect on agricultural carbon emission intensity. Third, the mechanism test results reveal that the carbon reduction effect of rural digitalization is achieved through scale agricultural operation, human capital accumulation and the improvement of total factor productivity. Fourth, the spatial spillover effect analyzed the spatial correlation of agricultural carbon emissions, and the results show that there is a positive spatial correlation between the agricultural carbon intensities of different provinces. Specific regional identification indicates that the spatial pattern of agricultural carbon emission intensity in China is currently dominated by low-low agglomeration and high-high agglomeration. Meanwhile, the inhibitory effect of rural digitalization on agricultural carbon intensity also has a spatial spillover effect, since the rural digitalization of both local and neighboring provinces exhibits negative impacts on local agricultural carbon intensity.

References

Publishing order | Descend order by publishing year | Descend order by cited within
[1]
Ang B W. 2006. Monitoring changes in economy-wide energy efficiency: From energy-GDP ratio to composite efficiency index. Energy Policy, 34(5): 574-582.

DOI

[2]
Blundell R, Bond S. 1998. Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics, 87(1): 115-143.

DOI

[3]
Chang J. 2022. The role of digital finance in reducing agricultural carbon emissions: evidence from China’s provincial panel data. Environmental Science and Pollution Research, 29(58): 87730-87745.

DOI

[4]
Chang Y L, Lv Y, He C Y. 2024. Measure and comparison of China— ASEAN digital economic development and cooperation. Journal of Nanning Normal University (Philosophy and Social Sciences Edition), 45(1): 115-136. (in Chinese)

[5]
Che S, Wen L, Wang J. 2024. Global insights on the impact of digital infrastructure on carbon emissions: A multidimensional analysis. Journal of Environmental Management, 368: 122144. DOI: 10.1016/j.jenvman.2024.122144.

[6]
Chen Y, Li M. 2024. How does the digital transformation of agriculture affect carbon emissions? Evidence from China’s provincial panel data. Humanities and Social Sciences Communications, 11(1): 1-17.

DOI

[7]
Danquah M, Amankwah-Amoah J. 2017. Assessing the relationships between human capital, innovation and technology adoption: Evidence from sub-Saharan Africa. Technological Forecasting and Social Change, 122: 24-33.

DOI

[8]
Elhorst J P. 2012. Dynamic spatial panels: Models, methods, and inferences. Journal of Geographical Systems, 14(1): 5-28.

DOI

[9]
Feng L, Yang W, Hu J, et al. 2025. Exploring the nexus between rural economic digitalization and agricultural carbon emissions: A multi-scale analysis across 1607 counties in China. Journal of Environmental Management, 373: 123497. DOI: 10.1016/j.jenvman.2024.123497.

[10]
Guan N, Liu L, Dong K, et al. 2023. Agricultural mechanization, large-scale operation and agricultural carbon emissions. Cogent Food & Agriculture, 9(1): 2238430. DOI: 10.1080/23311932.2023.2238430.

[11]
Guo H. 2024. The agricultural carbon reduction effect of digital rural construction under the dual carbon target. Plos One, 19(4): e299233. DOI: 10.1371/journal.pone.0299233.

[12]
Guo Z, Zhang X. 2023. Carbon reduction effect of agricultural green production technology: A new evidence from China. Science of the Total Environment, 874: 162483. DOI: 10.1016/j.scitotenv.2023.162483.

[13]
Halleck Vega S, Elhorst J P. 2015. The SLX model. Journal of Regional Science, 55(3): 339-363.

DOI

[14]
Han G, Xu J, Zhang X, et al. 2024. Efficiency and driving factors of agricultural carbon emissions: A study in Chinese state farms. Agriculture, 14(9): 1454. DOI: 10.3390/agriculture14091454.

[15]
Han H, Zhong Z, Guo Y, et al. 2018. Coupling and decoupling effects of agricultural carbon emissions in China and their driving factors. Environmental Science and Pollution Research, 25: 25280-25293.

DOI

[16]
He D, Deng X, Gao Y, et al. 2025. How does digitalization affect carbon emissions in animal husbandry? A new evidence from China. Resources, Conservation and Recycling, 214: 108040. DOI: 10.1016/j.resconrec.2024.108040.

[17]
Huang C, Zhang X, Liu K. 2021. Effects of human capital structural evolution on carbon emissions intensity in China: A dual perspective of spatial heterogeneity and nonlinear linkages. Renewable and Sustainable Energy Reviews, 135: 110258. DOI: 10.1016/j.rser.2020.110258.

[18]
Huang J, Lu H, Du M. 2025. Regional differences in agricultural carbon emissions in China: Measurement, decomposition, and influencing factors. Land, 14(4): 682. DOI: 10.3390/land14040682.

[19]
Huang X, Xu X, Wang Q, et al. 2019. Assessment of agricultural carbon emissions and their spatiotemporal changes in China, 1997-2016. International Journal of Environmental Research and Public Health, 16(17): 3105. DOI: 10.3390/ijerph16173105.

[20]
Huang Y, Elahi E, You J, et al. 2024. Land use policy implications of demographic shifts: Analyzing the impact of aging rural populations on agricultural carbon emissions in China. Land Use Policy, 147: 107340. DOI: 10.1016/j.landusepol.2024.107340.

[21]
Jiang T. 2022. Mediating effect and moderating effect in empirical research on causal inference. China Industrial Economics, (5): 100-120. (in Chinese)

[22]
Jiang W, Chen C. 2025. The impacts of rural digitization on agricultural carbon emission efficiency: Evidence from 30 provinces in China over 2011-2022. Frontiers in Sustainable Food Systems, 9: 1593986. DOI: 10.3389/fsufs.2025.1593986.

[23]
Jin M, Feng Y, Wang S, et al. 2024. Can the development of the rural digital economy reduce agricultural carbon emissions? A spatiotemporal empirical study based on China’s provinces. Science of the Total Environment, 939: 173437. DOI: 10.1016/j.scitotenv.2024.173437.

[24]
Johnson J M, Franzluebbers A J, Weyers S L, et al. 2007. Agricultural opportunities to mitigate greenhouse gas emissions. Environmental Pollution, 150(1): 107-124.

PMID

[25]
Kurniawan T A, Maiurova A, Kustikova M, et al. 2022. Accelerating sustainability transition in St. Petersburg, Russia through digitalization-based circular economy in waste recycling industry: A strategy to promote carbon neutrality in era of industry 4.0. Journal of Cleaner Production, 363: 132452. DOI: 10.1016/j.jclepro.2022.132452.

[26]
Li H. 2023. Digital inclusive finance, agricultural green technology innovation and agricultural carbon emissions: Impact mechanism and empirical test. Plos One, 18(10): e288072. DOI: 10.1371/journal.pone.0288072.

[27]
Li H, Zhuge R, Han J, et al. 2022. Research on the impact of digital inclusive finance on rural human capital accumulation: A case study of China. Frontiers in Environmental Science, 10: 936648. DOI: 10.3389/fenvs.2022.936648.

[28]
Li N, Jiang Y, Yu Z, et al. 2017. Analysis of agriculture total-factor energy efficiency in China based on DEA and Malmquist indices. Energy Procedia, 142: 2397-2402.

DOI

[29]
Li S, Yang Y, Wei Y. 2021. Study on agricultural carbon emission estimation and grading evaluation—Taking Chengdu as an example. IOP Conference Series: Earth and Environmental Science: 719(4): 042035. DOI: 10.1088/1755-1315/719/4/042035.

[30]
Liu W, Zhou W, Lu L. 2022. An innovative digitization evaluation scheme for spatio-temporal coordination relationship between multiple knowledge driven rural economic development and agricultural ecological environment—Coupling coordination model analysis based on Guangxi. Journal of Innovation & Knowledge, 7(3): 100208. DOI: 10.1016/j.jik.2022.100208.

[31]
Shi H, Chang M. 2023. How does agricultural industrial structure upgrading affect agricultural carbon emissions? Threshold effects analysis for China. Environmental Science and Pollution Research, 30(18): 52943-52957.

DOI

[32]
Song W, Meng L, Zang D. 2023. Exploring the impact of human capital development and environmental regulations on green innovation efficiency. Environmental Science and Pollution Research, 30(25): 67525-67538.

DOI

[33]
Su L, Wang Y, Yu F. 2023. Analysis of regional differences and spatial spillover effects of agricultural carbon emissions in China. Heliyon, 9(6): e16752. DOI: 10.1016/j.heliyon.2023.e16752.

[34]
Wang D, Chen C, Zhu N, et al. 2024. Impact of agricultural digitalization on carbon emission intensity of planting industry: Evidence from China. Heliyon, 10(10): e31215. DOI: 10.1016/j.heliyon.2024.e31215.

[35]
Wang R, Zhang Y, Zou C. 2022. How does agricultural specialization affect carbon emissions in China? Journal of Cleaner Production, 370: 133463. DOI: 10.1016/j.jclepro.2022.133463.

[36]
West T O, Marland G. 2002. A synthesis of carbon sequestration, carbon emissions, and net carbon flux in agriculture: Comparing tillage practices in the United States. Agriculture, Ecosystems & Environment, 91(1-3): 217-232.

DOI

[37]
Wu H, Yan W, Zheng X, et al. 2023. Carbon-emission density of crop production in China: Spatiotemporal characteristics, regional disparities, and convergence. Agriculture, 13(5): 951. DOI: 10.3390/agriculture 13050951.

[38]
Xia Q. 2020. Manufacturing servitization and export dual margin under intra-product specialization: An empirical study based on system GMM. South China Journal of Economics, (3): 53-72. (in Chinese)

DOI

[39]
Xia Y, Guo H, Xu S, et al. 2024. Environmental regulations and agricultural carbon emissions efficiency: Evidence from rural China. Heliyon, 10(4): e25677. DOI: 10.1016/j.heliyon.2024.e25677.

[40]
Xiong C, Yang D, Huo J. 2016. Spatial-temporal characteristics and lmdi-based impact factor decomposition of agricultural carbon emissions in Hotan Prefecture, China. Sustainability, 8(3): 262. DOI: 10.3390/su8030262.

[41]
Xiong W, Zhou P Y. 2024. Evaluation of agricultural green development level and analysis on spatial-temporal differences in the Yangtze River Basin. Ecological Economy, 40: 119-128.

[42]
Xu Y, Li C, Wang X, et al. 2023. Digitalization, resource misallocation and low-carbon agricultural production: Evidence from China. Frontiers in Environmental Science, 11: 1117086. DOI: 10.3389/fenvs.2023.1117086.

[43]
Yun T, Zhang J. 2014. Research on spatial-temporal characteristics and driving factor of agricultural carbon emissions in China. Journal of Integrative Agriculture, 13(6): 1393-1403.

DOI

[44]
Zhang H, Guo K, Liu Z, et al. 2024a. How has the rural digital economy influenced agricultural carbon emissions? Agricultural green technology change as a mediated variable. Frontiers in Environmental Science, 12: 1372500. DOI: 10.3389/fenvs.2024.1372500.

[45]
Zhang S, Wen X, Sun Y, et al. 2024b. Impact of agricultural product brands and agricultural industry agglomeration on agricultural carbon emissions. Journal of Environmental Management, 369: 122238. DOI: 10.1016/j.jenvman.2024.122238.

[46]
Zhang X, Zhang J, Yang C. 2023. Spatio-temporal evolution of agricultural carbon emissions in China, 2000-2020. Sustainability, 15(4): 3347. DOI: 10.3390/su15043347,

[47]
Zhao L, Rao X, Lin Q. 2023. Study of the impact of digitization on the carbon emission intensity of agricultural production in China. Science of the Total Environment, 903: 166544. DOI: 10.1016/j.scitotenv.2023.166544.

[48]
Zhao S, Li M, Cao X. 2024. Empowering rural development: Evidence from China on the impact of digital village construction on farmland scale operation. Land, 13(7): 903. DOI: 10.3390/land13070903.

[49]
Zhu W, Wang R M. 2023. Study on the impact of technological progress and operation scale on agricultural carbon emissions. Agricultural Economy, 2(12): 13-15.

[50]
Zhu Y, Huo C. 2022. The impact of agricultural production efficiency on agricultural carbon emissions in China. Energies, 15(12): 4464. DOI: 10.3390/en15124464.

[51]
Zhu Y, Waqas M A, Li Y, et al. 2018. Large-scale farming operations are win-win for grain production, soil carbon storage and mitigation of greenhouse gases. Journal of Cleaner Production, 172: 2143-2152.

DOI

Options
Outlines

/

〈 〉