The Impact of Agricultural Infrastructure on Carbon Reduction in Grain Production: A Comparative Study of Different Agricultural Infrastructure Types

Abstract

While extant literature has thoroughly investigated carbon mitigation in grain production and agricultural infrastructure’s yield effects, significant knowledge gaps remain regarding their synergistic pathways for emission reduction. This empirical study examines how agricultural infrastructure contributes to carbon emission reduction in grain production across 30 Chinese provinces from 2009 to 2023. Using two-way fixed-effects and mediation-effect models, we demonstrate that agricultural infrastructure significantly inhibits carbon emissions intensity, with effects varying by type of infrastructure: agricultural water infrastructure, digital infrastructure, agricultural power infrastructure and rural transportation infrastructure, in descending order. We identify three key mechanisms: planting structure optimization, technological progress, and disaster incidence reduction. Specifically, agricultural water infrastructure and digital infrastructure operate through structural improvement and technological advancement, while agricultural water infrastructure and rural transportation infrastructure function through disaster mitigation. Heterogeneity analysis reveals distinct regional patterns: northern regions benefit more from agricultural water infrastructure and rural transportation infrastructure, while southern regions show stronger effects from agricultural water infrastructure and digital infrastructure. In major grain-producing areas, agricultural water infrastructure and agricultural power infrastructure demonstrate significant emissions reduction, whereas non-core production regions rely more on agricultural water infrastructure and digital infrastructure. Additionally, infrastructure generates greater yield-enhancing effects for rice and wheat versus corn. Policy implications include strengthening investments in agricultural water infrastructure, promoting digital agriculture, and developing region-specific infrastructure strategies.

1. Introduction

In 2020, at the 75th UN General Assembly, China committed to peak carbon emissions by 2030 and achieve carbon neutrality by 2060. However, China’s agricultural sector faces significant challenges in meeting these goals. As agricultural modernization advances, the increased reliance on chemical fertilizers and pesticides has led to higher carbon emissions [1]. The China Agricultural Outlook Report (2021–2030) reveals that the farm sector’s annual net greenhouse gas uptake reached 590 million tonnes in 2020–2023, with 230 million tonnes classified as non-desired output [2]. Agricultural production now accounts for over 11–12% of global carbon emissions [3]. Grain production, the primary source of agricultural emissions, is particularly concerning. The Food and Agriculture Organization reports that China’s grain production activities generate 25-33% of global greenhouse gas emissions in this sector [4]. Overusing pesticides and fertilizers has led to surface pollution, while groundwater over-exploitation has created “funnel zones” [5]. These issues, along with declining soil fertility from over-cultivation of arable land, threaten both the quantity and quality of China’s grain production.

China’s grain production sector lacks a mature, systemic model that integrates production techniques, policy instruments, and management practices to achieve both high yield and low carbon emissions. There is insufficient synergy between low-carbon planting practices (e.g., conservation tillage, precision fertilization, integrated pest management, and water-saving irrigation) and efforts to improve grain production quality. The sector also lacks transformative technologies to simultaneously increase production and reduce emissions. In contrast, international experiences offer diverse, yet informative, approaches to reconciling productivity and emissions: The United States and Russia reduced emissions through intensive, large-scale development and increased grain yields, achieving CO₂ output reductions of 320,567 ktCO_2-e and 138,527 ktCO_2-e, respectively, from 2000 to 2014 [6]. Zealand has set a statutory target to reduce agricultural methane emissions by 14-24% below 2017 levels by 2050, backed by an investment of over NZ$400 million in methane-reduction technologies [7]. Brazil and India shifted carbon-intensive industries abroad, contributing to a reduction of 64.1% of global carbon emission reductions in recent years. While these pathways differ, they underscore the importance of foundational drivers, such as infrastructure and market mechanisms.

A common thread across many successful international approaches is significant investment in agricultural infrastructure. In recent years, China has prioritized agricultural infrastructure development. The 2016 initiative emphasized large-scale farmland water conservancy construction, while the 2023 policy expanded to include arable land protection and high-standard farmland construction. The 2024 directive further stressed modernizing key water sources, irrigation areas, and flood control zones. Therefore, agricultural infrastructure emerges as a fundamental factor worthy of focused study in the Chinese context. As a crucial prerequisite, agricultural infrastructure promotes large-scale operations and enhances technical capabilities in global grain production. Recent studies indicate that this infrastructure reduces the intensity of productive inputs like chemicals and machinery, thereby lowering carbon emissions while boosting production [8]. Major agricultural nations such as the United States, Australia, and France have actively promoted agricultural infrastructure development. This approach has not only strengthened their grain supply security but also achieved comparative advantages in input rates and cost-effectiveness [9]. Influenced by these global trends, China has also prioritized agricultural infrastructure development.

However, existing literature has separately examined the yield effects of agricultural infrastructure and the sources of agricultural carbon emissions. A critical conceptual gap remains in systematically linking the two and elucidating the specific theoretical pathways-such as scale expansion, efficiency enhancement, and risk mitigation-through which different types of infrastructure influence carbon emissions intensity. This study addresses two key questions: Can agricultural infrastructure contribute to carbon mitigation in grain production, and what are the mechanisms by which it affects carbon emission reduction? This study employs panel data from 30 Chinese provinces during the 2009–2023 period, constructing two-way fixed-effects models and mediation-effect models to examine the impact of agricultural infrastructure on grain production increase and its underlying mechanisms. The research aims to provide recommendations for reducing carbon emissions from grain production in China. To this end, this study conducts a rigorous empirical analysis using panel data and econometric models. The study makes several key contributions:

First, it examines the impact of agricultural infrastructure on carbon emission reduction in grain production, demonstrating its role in enhancing production sustainability.

Second, it confirms the scale expansion effect, efficiency enhancement effect, and risk mitigation effect of agricultural infrastructure on carbon emission reduction.

Third, it clarifies different infrastructure types, finding that agricultural water infrastructure has the largest impact, followed by digital infrastructure, with significant contributions from agricultural power infrastructure and rural transportation infrastructure. This ranking can guide prioritization in infrastructure investment.

Fourth, it reveals significant heterogeneity in infrastructure impacts, suggesting that effects vary across regions, functional zones, and crop types, demonstrating the need for tailored infrastructure development to optimize carbon reduction strategies.

The whole study is structured as follows: Section 2 reviews existing literature and identifies research gaps. Section 3 shares theoretical analysis and formulates hypotheses. Section 4 describes the material, method and data used in this study. Section 5 presents and discusses the empirical results. Finally, Section 6 concludes with key findings and policy recommendations.

2. Literature Review and Research Gaps

2.1. Literature Review

To situate this study, the paper reviews two interconnected strands of literature: “sources and mitigation of carbon emissions in agriculture” and “the role of agricultural infrastructure in grain production”. Synthesizing these strands reveals a gap in understanding how specific types of agricultural infrastructure influence carbon intensity through identifiable mechanisms.

2.1.1. Sources and Mitigation of Carbon Emissions in Agriculture

From existing research, scholars mainly discuss carbon emissions from grain production. Current evidence indicates that grain systems contribute 26–37% of global greenhouse gas emissions [10]. At the consumption level, menu restructuring that clusters high-carbon dishes increases selection of plant-based options, reducing weekly dietary carbon footprints by 30–31% [11]. Protein substitution from beef to legumes cuts emissions by 99% regionally, though outcomes depend on local dietary patterns [12]. On the production front, optimized nutrient management in Chinese potato farming reduces synthetic nitrogen inputs by 34.4%, lowering carbon emissions by 34.1% [13]. Novel cellular agriculture in Finland decreases climate impacts by 52–77% with renewable energy. For supply chains, electric refrigeration with route optimization cuts transport emissions through energy efficiency, while differentiated localizing strategies in mega-cities lower annual emissions by 2–5% [14].

2.1.2. The Role of Agricultural Infrastructure in Grain Production

In addition, recent scholarly discourse has evolved to encompass not only the role of agricultural infrastructure in promoting grain production but also its environmental implications—particularly in relation to agricultural carbon accounting, digital agriculture, and the infrastructure-environment nexus. Scholars generally define agricultural infrastructure as the physical and digital assets that support agricultural production and distribution, encompassing four key types: agricultural water infrastructure, rural transportation infrastructure, agricultural power infrastructure, and digital infrastructure [15]. Recent studies can be analyzed from the following perspectives: First, traditional agricultural infrastructure can directly promote grain production by expanding cultivation scale. It improves crop soil structure, optimizes sowing patterns among various grain crops, and enhances arable land quantity and quality [16]. For example, agricultural water infrastructure helps consolidate fragmented land into high-quality fields, mitigating water and soil constraints and thereby supporting grain output and quality [17]. Similarly, rural transportation infrastructure expands effective planting space through road hardening and improved soil carrying capacity, strengthening the scale effect of grain production [18]. Second, traditional agricultural infrastructure enhances grain production efficiency [19]. Agricultural water infrastructure provides essential conditions for cultivation—irrigation during droughts and drainage during floods—which strengthens crop resilience [20]. Rural transportation infrastructure also facilitates the input of production factors and the use of mechanized equipment, thereby improving efficiency in both production and resource distribution [21]. However, earlier studies on these linkages often had methodological limitations, such as overlooking systemic carbon emission impacts or failing to integrate digital governance dimensions into environmental assessments. For instance, while digital infrastructure promotes intensified and specialized farming [22]—leading to more efficient land allocation and economies of scale—recent debates emphasize its dual role in enabling smart farming and reducing carbon footprints through precision agriculture. Yet, the environmental outcomes of digital agriculture, especially in terms of net carbon accounting and its interaction with traditional infrastructure, remain under-explored in the existing literature. Furthermore, the infrastructure-environment relationship has often been approached fragmentedly, with limited cross-sectoral analysis linking water, transport, and digital systems to integrated carbon and sustainability metrics.

2.2. Research Gaps

Collectively, the extant literature provides a comprehensive examination of carbon mitigation in grain production and the impact of agricultural infrastructure on yield outcomes, offering foundational insights that critically inform this study’s integrated analysis of emission-efficient infrastructure synergies. However, existing research still has the following limitations: First, current studies have failed to consider their impact on carbon emissions generated during the grain production process [23]. Second, the measurement of carbon emission reduction in grain production mostly relies on macro-level emission factor methods or coefficient decomposition approaches [24], without conducting detailed accounting of specific carbon emissions from major grain crops such as rice, wheat, and soybeans during their production processes.

To address the gaps identified in existing research, this study constructs two-way fixed-effects models and mediation-effect models, utilizing panel data from 30 Chinese provinces and municipalities from 2009 to 2023, to empirically examine the impact of agricultural infrastructure on carbon emissions from grain production and its underlying mechanisms. The aim is to provide a scientific basis for formulating differentiated carbon emission reduction policies for grain production in China.

3. Theoretical Analysis and Hypothesis Formulation

According to Figure 1, this section analyzes the effects of agricultural infrastructure, scale expansion, and efficiency enhancement on carbon reduction in grain production, based on current research and practical applications.

3.1. Impact of Agricultural Infrastructure on Carbon Reduction in Grain Production

A strong synergy exists between agricultural infrastructure and low-carbon emission reduction in grain production. The development of agricultural infrastructure plays a vital role in advancing green agricultural production. Through measures such as land consolidation, ditch construction, and water-saving irrigation, it enhances the ecological protection capacity of farmland and reduces soil erosion, thereby helping to resolve the conflict between increasing grain output and environmental conservation [25]. Specifically, farmland water conservancy facilities leverage the agglomeration characteristics of grain production to strengthen carbon mitigation by promoting technological progress. By ensuring reliable irrigation and drainage, they improve agricultural water use efficiency and enhance the disaster resistance of cultivated land. For instance, such infrastructure enables the adoption of water-saving and emission-reducing practices like Alternate Wetting and Drying (AWD) for paddy rice, which has been shown to reduce methane emissions by 52–55% [26]. Well-developed rural transport infrastructure cuts the cost and time required for moving production inputs [27]. It enables timely and effective application of fertilizers, pesticides, and other inputs, raising carbon reduction efficiency in grain production. Moreover, it shortens the distance from field to market, further cutting carbon emissions during the distribution of grain products. Agricultural power facilities facilitate the widespread use of mechanized farming equipment. By reducing the carbon emissions from irrigation systems and agricultural machinery [28], this facilitates the replacement of diesel generators with distributed solar PV and other clean energy systems in the field, reducing carbon emissions at the source [29]. Digital infrastructure accelerates the transition toward smart agriculture, breaking down information barriers in the sector. It speeds up the dissemination of production techniques and market demand information in rural areas, lowering transaction costs and information search costs in grain production [30], which in turn reduces carbon emissions. Furthermore, digital infrastructure can directly reduce fossil fuel consumption in field operations [31]. However, the direct deployment and operation of infrastructure (e.g., construction emissions, energy use for digital systems) may also generate immediate carbon costs, creating a potential trade-off with their long-term mitigation benefits. In summary, this study proposes Hypothesis 1:

Hypothesis 1.

Agricultural infrastructure can contribute to carbon reduction in grain production.

3.2. Scale Expansion Effect and Efficiency Enhancement Effect on Carbon Reduction in Grain Production

Building on the direct effect posited in Hypothesis 1, we further propose that agricultural infrastructure influences carbon emissions intensity through three interconnected channels, which may operate simultaneously or sequentially:

3.2.1. Scale Expansion Effect

This aligns with the classic economic concept of economies of scale, where infrastructure reduces the average cost (here, carbon intensity per unit of output) as the scale of operation expands. Agricultural infrastructure directly facilitates the expansion of cultivation scale, thereby increasing sown areas of grain crops, lowering carbon emissions intensity per unit of output, and securing yield growth. Well-established infrastructure continuously enhances the scale and mechanization of grain production [32], which raises production efficiency and contributes to lower carbon emissions, thus supporting increased output. Among these, farmland water conservancy works—through land consolidation—create contiguous, well-equipped, and productive fields. This mitigates the negative impact of fragmented land holdings on soil quality, reduces carbon sequestration loss caused by inefficient land use, lowers the carbon emission rate of grain production, and improves expected grain output. Meanwhile, developed agricultural power facilities enable the use of mechanized equipment for large-scale farming. Such scaled operations expand cultivated areas, help curb the overuse of chemical fertilizers and pesticides, and reduce energy consumption in agriculture, thereby decreasing total carbon emissions from grain production. Digital infrastructure further supports a shift toward grain-oriented planting structures, enlarging the scale of grain production. By enabling interregional information connectivity, it amplifies the diffusion and scale effects of digital technologies in cutting carbon emissions [33]. This encourages farmers to choose grain crops and fosters sustained engagement in low-carbon production, contributing to both stable yields and supply. Nevertheless, the scale expansion effect may carry a rebound risk: if the enlarged production area or intensified cultivation leads to a net increase in the use of fertilizers, pesticides, or machinery, the total carbon emissions could rise despite a lower emission intensity per unit. In summary, this paper proposes Hypothesis 2:

Hypothesis 2.

Agricultural infrastructure contributes to carbon reduction in grain production through the scale expansion effect.

3.2.2. Efficiency Enhancement Effect

Agricultural infrastructure can improve the technical level in the agricultural production process. This channel resonates with induced innovation theory, which posits that changes in factor endowments (e.g., improved infrastructure) drive technological advancements aimed at saving increasingly scarce or costly resources (e.g., reducing input waste and associated emissions). Through technological progress, it brings about an increase in grain production efficiency and, to a certain extent, reduces the carbon emissions intensity of grain production, promoting carbon emission reduction in grain production. The improvement of farmland water conservancy facilities can promote the development of the agricultural economy in an ecological direction, thereby reducing the carbon emissions intensity in grain production and promoting carbon emission reduction in grain production; rural transportation facilities, through the transformation of field roads and determining the width and density of roads according to local terrain conditions, optimize the operating environment for agricultural machinery and logistics transportation, providing the possibility for the reduction of chemical fertilizers and pesticides, and thus promoting carbon emission reduction in grain production; rural power facilities can directly accelerate agricultural technological progress. Through the mechanization of agricultural tools, they effectively ensure that agricultural machinery efficiently enters the fields and that fields are efficiently drained and irrigated [34], improving the realization of the carbon emission reduction effect of agricultural infrastructure in grain production. Digital infrastructure, through real-time monitoring and differentiated management of the sowing, growth, irrigation, fertilization, and harvesting stages of grain crops [35], achieves refined management of energy consumption and carbon emission reduction issues in grain production. A critical consideration here is the potential for efficiency gains to lower the marginal cost of production, which could stimulate further output expansion and partially offset the carbon savings—a phenomenon known as the rebound effect. In summary, this paper proposes Hypothesis 3:

Hypothesis 3.

Agricultural infrastructure promotes carbon reduction in grain production through the efficiency enhancement effect.

3.2.3. Risk Mitigation Effect

Agricultural infrastructure contributes to cost reduction through disaster prevention and mitigation by minimizing losses. This mechanism is grounded in the theory of production risk management, where infrastructure investment is viewed as a form of self-insurance that stabilizes expected output and, by extension, stabilizes the emission intensity under volatile climatic conditions. Specifically, farmland water conservancy facilities enhance the capacity to “provide irrigation during droughts and drainage during floods,” effectively addressing the uneven spatiotemporal distribution of precipitation. This expands the effective irrigated area, strengthens the disaster resilience of cultivated land, and mitigates the exacerbating effect of drought and flood disasters on carbon emissions in grain production. Investment in digital infrastructure reinforces the digital empowerment of grain production. Through technologies such as big data platforms and the internet, it facilitates farmers’ access to market information, enables supply-demand matching between grain production and markets, and guides farmers in optimizing their production structure. This helps curb overinvestment of resources in grain cultivation and ensures the reduction of carbon emissions from grain production [36]. Furthermore, the advancement of digital infrastructure enhances the application of digital technologies and intelligent monitoring in predicting drought and flood disasters and other related fields, thereby further reducing carbon emissions in grain production. A critical consideration here is the potential for efficiency gains to lower the marginal cost of production, which could stimulate further output expansion and partially offset the carbon savings—a phenomenon known as the rebound effect. In summary, this study proposes Hypothesis 4:

Hypothesis 4.

Agricultural infrastructure promotes carbon reduction in grain production through the risk mitigation effect.

These three mechanisms are not mutually exclusive and may operate simultaneously or sequentially. For instance, scale expansion might initially increase input use, but efficiency gains can offset this through better management. Meanwhile, risk mitigation stabilizes outputs, securing the long-term benefits of both scale and efficiency improvements. It is hereby clarified that the subsequent empirical test of these hypotheses is based on measuring the carbon emissions most directly influenced by infrastructure, namely, those from in-season crop cultivation and key input use (see Section 4.2.1). According to Figure 1, the dominant pathway may vary by infrastructure type and regional context.

4. Material and Methods

To test the hypotheses derived from our theoretical framework (Figure 1), we employ the following econometric models. The baseline model (Equation (1)) estimates the overall effect (Hypothesis 1). Subsequently, mediation models (Equation (2)) are used to examine the proposed channels of planting scale (Hypothesis 2), technological progress (Hypothesis 3), and disaster incidence (Hypothesis 4).

4.1. Model of Estimation

4.1.1. Baseline Regression Model

To examine the impact of different types of agricultural infrastructure on carbon emissions from grain production, this paper establishes the following two-way fixed-effects model:

$$lnGC{E}_{it}={\alpha }_0+{\alpha }_{1}Infr{a}_{m,it}+{\alpha }_{2}Controls+{\mu }_i+{\nu }_t+{\epsilon }_{it}$$

(1)

In Equation (1): lnGCE_it represents the logarithmically transformed carbon emissions intensity of grain production for the i-th province in year t; Infra_m,it denotes the variables of different types of agricultural infrastructure for the i-th province in year t, where m = 1, 2, 3, 4 corresponds to farmland agricultural water infrastructure, agricultural power infrastructure, rural transportation infrastructure, and digital infrastructure, respectively; Controls refers to other control variables influencing grain production; $\mu$_i represents province fixed effects, $\nu$_t represents time fixed effects, $\epsilon$_it is the random error term; α₀ is the constant term, while α₁ and α₂ are coefficients to be estimated.

4.1.2. Mediation Model

Following Preacher and Kelley (2011) [37] mediation model testing procedure, this study will test the effects of the mediating variables:

$$M_{it}={\alpha }_0+{\alpha }_{1}Infr{a}_{m,it}+\phi Control{s}_{it}+{\beta }_i+{\gamma }_t+{\epsilon }_{it}$$

(2)

In Equation (2): M_it represents the mediating variables for the i-th province in year t, namely planting scale, technological progress, and disaster incidence rate; α₀ is the constant term, and α₁ is coefficient to be estimated; $\phi$ denotes the estimated coefficients of the control variables, β_i represents province fixed effects, ${\gamma }_t$ represents time fixed effects, ${\epsilon }_{it}$ is the random error term, and all other variables are consistent with those in Equation (1).

4.2. Variable Definitions

4.2.1. Explained Variables

This study adopts the carbon emissions intensity of grain production as the explained variable. Drawing on the accounting scope defined by Qi et al. (2024) [38], the carbon emissions from grain production defined in this study include CH₄ and N₂O gases generated during the planting processes of five major crops: rice, corn, wheat, soybeans, and potatoes. Furthermore, referencing the carbon emission factor method applied by He et al. (2020) [39], the total carbon emissions and carbon emissions intensity of grain production are calculated to characterize the carbon emission level of grain production. The specific calculation formulas are as follows (Equations (3) and (4)):

$$C_{it}={\displaystyle \sum C_{nit}}={\displaystyle \sum T_{nit}}\times {\theta }_n$$

(3)

$$GC{E}_{it}={\displaystyle \frac{C_{it}}{GC{P}_{it}}}$$

(4)

Equation (3) represents the formula for calculating carbon emissions from grain production. Here, C_it denotes the total carbon emissions from grain production for the i-th province in year t, measured in tons; C_nit represents the total carbon emissions from carbon source n, where n denotes the carbon sources (rice, corn, wheat, soybeans, potatoes); T_nit indicates the quantity of carbon source n, and θ_n is the carbon emission coefficient for carbon source n. Equation (4) expresses the calculation formula for the carbon emissions intensity of grain production. Here, GCE_it denotes the carbon emissions intensity of grain production for province i-th in year t, measured in tons per 10,000 yuan; GCP_it represents the gross grain output value of the region, measured in 10,000 yuan. The gross grain output value is calculated as: Agricultural Gross Output Value × (Agricultural Gross Output Value/Gross Output Value of Agriculture, Forestry, Animal Husbandry, and Fishery) × (Grain Planting Area/Agricultural Planting Area), all other variables remain consistent with those in Equation (3).

Furthermore, referencing perspectives from both recent scholars as well as documents from the Intergovernmental Panel on Climate Change (IPCC), this study adopts the carbon emission coefficients for grain crops as shown in

Table 1. Main carbon emission factors for grain crops.

Carbon Sources	Carbon-Emitting Gases	Carbon Emission Factor	Reference Sources
Rice	CH4	210 kg CH4/hm2	IPCC
	N2O	0.240 kg N2O/hm2
Wheat	N2O	2.050 kg N2O/hm2	[40]
Corn	N2O	2.530 kg N2O/hm2
Soybeans	N2O	0.770 kg N2O/hm2
Potato	N2O	0.950 kg N2O/hm2

:

In order to standardize the measurement units, this study draws on the research of Tian and Zhang (2013) [41] to convert various greenhouse gases into standard carbon (C) equivalents, where 1 ton of N₂O is equivalent to 81.2727 tons of C, and 1 ton of CH₄ is equivalent to 6.8182 tons of C.

In addition, this study chooses to focus the baseline carbon emission measurement on direct biological emissions (CH₄ and N₂O) from cultivation is based on three primary considerations. First, conceptual directness: these emissions are direct products of crop physiology and field-level microbiological processes, intrinsically tied to the planting activity itself. Second, data consistency and comparability: biological emission factors based on the IPCC methodology are relatively stable across provinces and over time, making the results less susceptible to exogenous fluctuations in supply chains or energy mixes, which facilitates spatiotemporal comparison. Third, policy relevance: mitigating these emissions depends more directly on field management practices (e.g., water management, fertilization techniques), aligning closely with the production links that agricultural infrastructure may affect. We fully acknowledge the importance of life-cycle emissions from agricultural inputs and have addressed this through a robustness check in Section 5.2.2 using an expanded carbon intensity measure, ensuring our core findings are not contingent on this specific accounting boundary.

4.2.2. Core Explanatory Variables

While previous studies have often used the agricultural infrastructure investment price index or perpetual inventory method to represent agricultural infrastructure [42], this study adopts the alternative variable method to represent agricultural water infrastructure, agricultural power infrastructure, rural transportation infrastructure, and digital infrastructure:

① Agricultural water infrastructure (lnIrri): Agricultural water infrastructure is crucial for strengthening agricultural resilience and reducing carbon emissions in grain production. Following the methodology of previous research [43], the effective irrigated area is used to represent agricultural water infrastructure.

② Rural transportation infrastructure (lnRoad): Rural transportation infrastructure enhances the flow of agricultural inputs, reduces production costs, and promotes the spread of new technology. Following a related work [44], this study uses the ratio of rural road mileage to the cultivated land area for rural transportation infrastructure.

③ Agricultural power infrastructure (lnElec): Agricultural power infrastructure ensures a stable energy supply for grain production and can enhance the resilience of sustainable agricultural development. Drawing on relevant research [45], this study uses total rural electricity consumption to represent agricultural power infrastructure. However, as this indicator includes non-agricultural electricity use, the total rural electricity consumption variable may introduce measurement error. Therefore, the coefficient for the total rural electricity consumption variable in the subsequent regressions should be interpreted as an estimate of the “net effect” of relevant agricultural power infrastructure inputs, after controlling for other factors.

④ Digital infrastructure (Digi): Digital infrastructure, built on technologies such as the internet and big data platforms, helps reduce resource inputs in grain production. Drawing on relevant research [46], this study characterizes digital infrastructure using information service facilities, convergence infrastructure, and mobile communication facilities. Meanwhile, this study uses the entropy value method to measure the level of digital infrastructure construction (

Table 2. Description of the Digital Indicator System Construction.

Variable Indicators	Formula	Unit	Causality
Information service facility	Rural internet broadband access port	Size	+
	Rural internet broadband access users	Ten thousand households	+
Converged infrastructure	Total postal operations	Billions	+
	Rural delivery routes	Kilometer	+
Mobile communications facility	Mobile phone subscribers	Ten thousand households	+
	Length of long-haul fiber-optic cable routes	Kilometer	+

); regarding the specific calculation process of the entropy method and the resulting indicator weights, please refer to Appendix A.

4.2.3. Control Variables

The control variables are selected based on established literature to isolate the infrastructure effect. Factors like irrigation energy use and fertilizer subsidies are excluded as their influence is either captured by the core infrastructure variables or reflected in the calculated input-based emissions. Drawing on existing research [47], the control variables selected are as follows: ① Land input (lnLand), characterized by the logarithmically transformed grain planting area; ② Rural economic development level (lnEco), measured by the logarithmically transformed per capita disposable income of rural residents; ③ Machinery input (lnMach), measured by the logarithmically transformed total power of agricultural machinery; ④ Government financial expenditure on agriculture (Gfe), calculated as the expenditure on agricultural water affairs/general fiscal expenditure; ⑤ Human capital level (Hc), measured by the average years of education in rural areas; ⑥ Industrial structure (Is), calculated as the output value of the primary industry/regional GDP.

4.2.4. Mechanisms Variables

① Planting Structure (Stru): The “tends to grain” planting structure reflects the proportion of grain crops sown, which has been steadily increasing, indicating a shift in China’s arable land use towards grain production. In selecting indicators for a “grain-oriented” planting structure, Guo and Lyu (2024) [48] use the proportion of grain planting area relative to the total crop area. Rice and wheat are crucial for the country’s grain security, making them highly representative. Therefore, this study employs the proportion of (rice sown area + wheat sown area) to the total sown area of cultivated land as the measurement for the mediating variable “planting structure”.

② Technological Progress (Tech): To measure the rate of grain technological progress, the traditional approach involves using the DEA model to calculate the technological progress index representing grain technological progress. However, this method may not comprehensively account for all relevant input indicators when calculating the output of grain technological progress. In this way, Yang et al. (2025) [49] found that the number of agricultural patents can serve as a proxy for agricultural technological progress. With reference to the aforementioned studies, this paper adopts the logarithmically transformed number of agricultural technology patents as the measurement for the mediating variable “technological progress”.

③ Disaster Incidence Rate (Disa): Drought and flood disasters increase the energy consumption and material inputs required per unit of grain yield, thereby indirectly elevating carbon emissions from grain production. Considering that a reduction in the proportion of affected crop area to total sown area can lower energy consumption in grain production and improve agricultural resource utilization efficiency, this study draws on the research of Hu et al. (2023) [50] by calculating the crop disaster incidence rate as the ratio of affected crop area to total sown area. This metric is used to represent the mediating variable “disaster incidence rate”.

4.3. Data Sources

This study utilizes panel data from 30 provinces in China from 2009 to 2023 (Tibet, Hong Kong, Macao, and Taiwan were excluded due to incomplete data) as the research sample to investigate the impact of different types of agricultural infrastructure on carbon emissions from grain production. Data on agricultural gross output value, expenditure on agriculture, forestry, and water affairs, value-added of the primary industry, and regional GDP were sourced from the “China Statistical Yearbook”. Indicators for calculating carbon emissions intensity of grain production, effective irrigated area, rural electricity consumption, total power of agricultural machinery, grain sown area, and grain disaster-affected area were obtained from the “China Rural Statistical Yearbook” and “China Statistical Yearbook”. Detailed descriptive statistics are presented in

Table 3. Descriptive statistics of variables.

	Variable	Obs	Mean	Std. Dev.	Min	Max
Explained Variables	lnGCE	450	4.052	0.763	1.786	5.742
Core Explanatory Variables	lnIrri	450	7.276	1.033	4.694	8.785
	lnRoad	450	3.713	0.489	2.546	5.085
	lnElec	450	4.813	1.306	1.308	7.606
	Digi	450	0.596	0.161	0.127	0.965
Control Variables	lnEco	450	7.738	1.233	3.840	9.594
	lnMach	450	9.382	0.513	8.082	10.669
	Gfe	450	4.481	2.285	0.943	13.689
	Hc	450	0.124	0.074	0.030	0.764
	Is	450	8.480	1.502	4.521	12.782
Mechanisms Variables	Stru	450	0.101	0.054	0.002	0.277
	Tech	450	0.306	0.142	0.058	0.624
	Disa	450	9.184	1.544	4.394	12.813

.

5. Results and Discussion

5.1. Baseline Regression

Prior to the baseline regression, to avoid the potential impact of multicollinearity on estimation results, we calculated the variance inflation factor (VIF) for all explanatory variables prior to regression. The VIF values for all variables were well below the threshold of 10, suggesting the absence of severe multicollinearity in the model. In addition, this research conducted model diagnostic tests to ensure robust estimation. Tests indicate significant cross-sectional dependence (Pesaran CD test, p < 0.01) and serial correlation (Wooldridge test, p < 0.01). This confirmed that it is important to use the “Robustness test” to control for potential issues of heteroskedasticity, autocorrelation, and cross-sectional dependence. For both Model Specification Tests, refer to Appendix B.

Table 4. Baseline Regression Results.

Variables	(1)	(2)	(3)	(4)	(5)
lnIrri	−0.594 ***	−0.496 ***
	(0.155)	(0.147)
lnRoad	−0.045 ***		−0.046 ***
	(0.014)		(0.014)
lnElec	−0.038 ***			−0.040 ***
	(0.014)			(0.014)
Digi	−0.363 ***				−0.370 ***
	(0.090)				(0.091)
Cons	11.402 ***	14.469 ***	12.933 ***	13.436 ***	12.973 ***
	(4.057)	(3.609)	(3.824)	(3.731)	(3.954)
Control	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes	Yes
N	450	450	450	450	450
R2	0.930	0.928	0.921	0.918	0.917

Note: *** denotes significance at the 1% levels, respectively. The t-statistics are reported in parentheses.

presents the baseline regression results of the impact of different agricultural infrastructure on carbon emissions from grain production. Column (1) shows the regression results of farmland water conservancy facilities, rural transportation facilities, agricultural power facilities, and digital infrastructure on carbon emissions from grain production. Although the effects of these four types of agricultural infrastructure on carbon emissions vary, all of them negatively affect carbon emissions from grain production at the 1% significance level, demonstrating that the construction of agricultural infrastructure can reduce the carbon emissions intensity of grain production in China. Column (2) reports the regression results of farmland water conservancy facilities on carbon emissions from grain production, indicating that farmland water conservancy facilities negatively affect the carbon emissions intensity of grain production at the 1% significance level, with an estimated coefficient of −0.496. Column (3) displays the regression results of rural transportation facilities on carbon emissions from grain production, showing that rural transportation facilities negatively affect the carbon emissions intensity of grain production at the 1% significance level, with an estimated coefficient of −0.046. Column (4) provides the regression results of agricultural power facilities on carbon emissions from grain production, revealing that agricultural power facilities negatively affect the carbon emissions intensity of grain production at the 1% significance level, with an estimated coefficient of −0.040. Column (5) presents the regression results of digital infrastructure on carbon emissions from grain production, indicating that digital infrastructure negatively affects the carbon emissions intensity of grain production at the 1% significance level, with an estimated coefficient of −0.370.

In summary, the impact of agricultural infrastructure development on the carbon emissions intensity of grain production is significantly negative overall. To interpret the economic magnitude, the coefficient for agricultural water infrastructure is −0.496 (Column 2 of Table 4). This implies that a 10% increase in the effective irrigated area is associated with a 4.96% reduction in carbon emissions intensity. Digital infrastructure also shows a strong effect, with a coefficient of −0.370 (in Column 5 of Table 4). A one-standard-deviation increase in the digital index is linked to an approximate 6.0% decrease in intensity. In contrast, the effects for rural transportation (coefficient: −0.046) and agricultural power facilities (coefficient: −0.040) are more modest; a 10% improvement in these proxies correlates with reductions of only 0.46% and 0.40%, respectively (both from Column 1 of Table 4). This demonstrates that the effects vary substantially, with farmland water conservancy facilities having the most substantial impact, followed by digital infrastructure, which is largely consistent with Hypothesis 1.

5.2. Robustness Test

In order to control for potential issues of heteroskedasticity, autocorrelation, and cross-sectional dependence in the baseline regression results discussed above, the study will conduct the following robustness tests (

Table 5. Results of the robustness test.

	(1)	(2)	(3)	(4)
	lnGCE	lnCCE	lnGCE	lnGCE
	Endogeneity Test	Recalculated Explained Variable	PCSE Method	Two-Tailed Winsorization
LlnGCE	−0.316 ***
	(0.010)
lnIrri	−4.458 ***	−4.826 ***	−2.104 **	−0.614 ***
	(0.341)	(1.073)	(0.986)	(0.096)
lnRoad	−0.581 ***	−0.933 ***	−0.594 ***	−0.258 ***
	(0.112)	(0.300)	(0.195)	(0.047)
lnElec	−0.069 **	−0.510 ***	−0.374 *	−0.033 **
	(0.042)	(0.044)	(0.226)	(0.015)
Digi	−0.379 *	−2.723 **	−1.765 **	−0.374 *
	(0.227)	(1.340)	(0.860)	(0.226)
Cons	7.764 ***	10.11 ***	15.108 *	11.507 **
	(1.949)	(1.073)	(7.896)	(5.626)
Control	Yes	Yes	Yes	Yes
AR (1)	0.000
AR (2)	0.388
Sargan est	0.621
Province FE	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes
N	450	450	450	450
R2		0.949	0.929	0.921

Note: *, ** and *** denote significance at the 10%, 5% and 1% levels, respectively. The t-statistics are reported in parentheses.

):

5.2.1. Endogeneity Test

Considering the potential bidirectional causality between the proxy variables for various types of agricultural infrastructure and the variables measuring carbon emissions from grain production, which may lead to endogeneity issues, this study employs the Generalized Method of Moments (GMM) for re-estimation to examine the unbiasedness and consistency of the baseline regression results. The results in Column (1) of Table 5 indicate that agricultural water infrastructure, rural transportation infrastructure, agricultural power infrastructure, and digital infrastructure continue to negatively affect the carbon emissions intensity of grain production at the 1% significance level. This demonstrates that the baseline regression conclusions are unbiased and consistent.

5.2.2. Recalculated Explained Variable

Considering that the study initially accounted only for CH₄ and N₂O emissions from five major crops (rice, corn, wheat, soybeans, and potatoes) while overlooking carbon emissions from grain production inputs such as fertilizers, pesticides, agricultural films, diesel for machinery, and irrigation electricity, this study further tests the robustness of the baseline regression results by recalculating the carbon emissions intensity of grain production in China. The results in column (2) of Table 5 demonstrate that agricultural water infrastructure, rural transportation infrastructure, agricultural power infrastructure, and digital infrastructure continue to exhibit a statistically significant negative impact on the carbon emissions intensity of grain production at the 1% level. This confirms the robustness of the baseline regression conclusions even under an alternative measurement of the explained variable.

In addition, the recalculated Explained Variable is performed using the carbon emission factor method applied earlier and the emission coefficients listed in

Table 6. Carbon emission factors for major grain production inputs.

Carbon Sources	Carbon Emission Factor	Reference Sources
Fertilizer	0.8956 kg C/kg	ORNL
Pesticides	4.9341 kg C/kg	ORNL
Agricultural film	5.18 kg C/kg	IREEA
Diesel oil	0.5927 kg C/kg	IPCC
Irrigation	266.48 kg/hm2	[51]

, thereby modifying the measurement approach of the explained variable.

5.2.3. PCSE Method

Considering potential issues such as heteroskedasticity and auto-correlation in panel data regression, this study draws on the research of Zhu et al. (2025) [52] and employs the panel-corrected standard errors (PCSE) method for re-estimation. The results in column (3) show that agricultural water infrastructure, rural transportation infrastructure, agricultural power infrastructure, and digital infrastructure continue to exhibit a statistically significant negative impact on the carbon emissions intensity of grain production at the 1% level. This confirms that the baseline regression conclusions remain robust and are not affected by heteroskedasticity or auto-correlation.

5.2.4. Two-Tailed Winsorization

Considering that the presence of outliers may adversely affect the empirical results, this study applies a 1% winsorization to both tails of the variables related to grain production carbon emissions intensity, the four types of agricultural infrastructure, and other control variables. Regression is subsequently performed using the winsorized variables. The results in column (4) show that agricultural water infrastructure, rural transportation infrastructure, agricultural power infrastructure, and digital infrastructure continue to exhibit a statistically significant negative impact at the 1% level. This confirms that the baseline regression conclusions remain robust and are not affected by the bilateral winsorizing procedure.

5.3. Mediation Mechanism Analysis

The baseline regression results confirm the overall reducing effect of agricultural infrastructure on carbon emissions. To move beyond this aggregate finding and uncover the specific pathways through which this effect operates, this study examines the three mechanisms proposed in our theoretical framework (Figure 1): Planting Structure, Technological Progress, and Disaster Incidence Rate. This mediation analysis directly tests Hypotheses 2, 3, and 4. The results are presented in columns (1)–(3) of

Table 7. Mechanism Analysis.

	(1)	(2)	(3)
	Stru	Tech	Disa
lnIrri	0.215 ***	1.106 **	−0.469 **
	(0.056)	(0.471)	(0.209)
lnRoad	0.071 **	0.289 ***	−0.127 *
	(0.030)	(0.102)	(0.067)
lnElec	0.060 **	0.026 *	−0.052 ***
	(0.024)	(0.014)	(0.012)
Digi	0.201 **	3.739 ***	−0.005 ***
	(0.097)	(0.821)	(0.002)
Cons	1.846 ***	5.099 ***	5.008 ***
	(0.471)	(1.141)	(1.762)
Control	Yes	Yes	Yes
Province FE	Yes	Yes	Yes
Time FE	Yes	Yes	Yes
N	450	450	450
R2	0.973	0.982	0.984

Note: *, ** and *** denote significance at the 10%, 5% and 1% levels, respectively. The t-statistics are reported in parentheses.

.

Whether control variables are included or not, the mediation coefficients can be translated into practical effect sizes. For the planting structure channel, the coefficient of 0.215 for agricultural water infrastructure (Column 1) indicates that a 10% expansion in irrigation infrastructure is associated with a 2.15% increase in the grain-oriented planting area share. Regarding technological progress, the coefficient of 3.739 for digital infrastructure (Column 2) suggests that enhancing the digital index by one standard deviation could be linked to over a 60% increase in agricultural patent grants. For disaster mitigation, the coefficient of −0.127 for rural transportation infrastructure (Column 3) implies that a 10% improvement in rural road density is associated with a 1.27% reduction in the crop disaster incidence rate. Among them, Agricultural Water Infrastructure and Digital Infrastructure more significantly reduce carbon emissions intensity by enhancing the Planting Structure and Technological Progress, while Agricultural Water Infrastructure and Rural Transportation Infrastructure more significantly reduce carbon emissions intensity by lowering the Disaster Incidence Rate.

Based on the theoretical analysis, the expansion of planting scale, improvement in Technological Progress, and reduction in Disaster Incidence Rate are important mechanisms for suppressing carbon emissions from grain production. Therefore, the construction of agricultural infrastructure can expand planting scale, drive Technological Progress, and reduce Disaster Incidence Rate, thereby lowering carbon emissions from grain production. This validates Hypotheses 2, 3, and 4.

In addition, it is important to note that the results of the mediation analysis above should be interpreted as evidence of statistical associations rather than definitive causal mechanisms. The standard mediation model relies on the strong assumption of ‘sequential ignorability’. In our context, mediators such as ‘technological progress’ (patent count) and ‘disaster incidence’ could be influenced by factors other than infrastructure investment (e.g., regional innovation policies, climatic anomalies) and may even have a reciprocal relationship with infrastructure. While our analysis controls for multiple covariates and leverages panel data, we cannot fully rule out such potential endogeneity biases. Therefore, the results in Table 7 provide strong and consistent empirical support for the hypothesized channels—that infrastructure influences carbon intensity through planting structure optimization, technological progress, and disaster mitigation—yet the interpretation of the precise causal links within these pathways should remain cautious. Future research employing more exogenous policy shocks or natural experiment designs would help further identify these mechanisms.

5.4. Heterogeneity Analysis

Given the significant variations in grain production patterns between northern and southern China, the strategic delineation of grain production functional zones, and the diverse characteristics of different grain crop types—all of which may lead to differential effects of agricultural infrastructure on carbon emissions from grain production—this study extends the baseline regression analysis to examine the heterogeneous impacts of agricultural infrastructure on carbon emissions:

5.4.1. Heterogeneity in North–South Rain Layout

Due to differences in climatic conditions and economic development orientations, significant disparities exist in grain crop varieties between northern and southern China. To examine the heterogeneous effects of the four types of agricultural infrastructure on carbon emissions from grain production in these regions, this study draws on the approach of Huang et al. (2023) [53] by dividing the research sample into northern and southern regions using the Qinling-Huaihe Line as the boundary for heterogeneous regression analysis. According to columns (1)–(2) of

Table 8. Heterogeneity analysis in north–south rain layout and grain production functional zones.

	(1)	(2)	(3)	(4)
	North	South	Major Producing Areas	Non-Major Producing Areas
lnIrri	−1.587 ***	−1.081 ***	−1.363 ***	−1.097 ***
	(0.487)	(0.235)	(0.548)	(0.260)
lnRoad	−0.600 *	−0.065 ***	−0.189 ***	−0.127 *
	(0.336)	(0.020)	(0.081)	(0.073)
lnElec	−0.189 *	−0.048 *	−0.431 ***	−0.066 **
	(0.096)	(0.028)	(0.005)	(0.031)
Digi	−0.421 ***	−0.770 ***	−0.338 ***	−0.334 *
	(0.123)	(0.156)	(0.084)	(0.195)
Cons	3.081 ***	2.490 ***	3.627 **	3.312 **
	(1.011)	(0.602)	(1.755)	(1.673)
Control	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Time FE	Yes	Yes	Yes	Yes
N	225	225	195	255
R2	0.919	0.960	0.941	0.948

Note: *, ** and *** denote significance at the 10%, 5% and 1% levels, respectively. The t-statistics are reported in parentheses.

, the estimated coefficient for the overall construction of agricultural infrastructure is larger in the northern region than in the southern region. Among the infrastructure types, Agricultural Water Infrastructure and Rural Transportation Infrastructure exhibit more prominent carbon emission reduction effects in the northern region, whereas Agricultural Water Infrastructure and Digital Infrastructure demonstrate greater effects in the southern region compared to the other two types of agricultural infrastructure.

The reasons for this heterogeneity lie in regional characteristics. In the northern region, large-scale agricultural production and concentrated contiguous farmland provide a solid foundation. Here, Agricultural Water Infrastructure and Rural Transportation Infrastructure effectively contribute to carbon emission reduction by ensuring stable grain output and reducing energy consumption in the circulation of agricultural production resources. In contrast, the southern region features complex terrain and highly fragmented farmland. In this context, Digital Infrastructure and Agricultural Water Infrastructure play a critical role in reducing carbon emissions by minimizing fertilizer and pesticide residues in the soil.

5.4.2. Heterogeneity in Grain Production Functional Zones

Considering the varying functional positioning of grain production across provinces, the carbon emission reduction effects of different types of agricultural infrastructure may differ. To examine the heterogeneous impacts of the four types of agricultural infrastructure on carbon emissions from grain production in different functional zones, this study draws on the research of Lu et al. (2024) [54] by dividing the sample of 30 provinces into major grain-producing areas and non-major grain-producing areas for heterogeneity regression analysis. According to the regression results in columns (3) and (4) of Table 8, the estimated coefficient for the overall construction of agricultural infrastructure is larger in major grain-producing areas than in non-major grain-producing areas. Among them, Agricultural Water Infrastructure and Rural Transportation Infrastructure exhibit more prominent carbon emission reduction effects in major grain-producing areas, while Agricultural Water Infrastructure and Digital Infrastructure demonstrate greater effects in non-major grain-producing areas. The reasons for this heterogeneity lie in regional characteristics. Major grain-producing areas possess conditions for large-scale production, where Agricultural Water Infrastructure and Agricultural Power Infrastructure effectively achieve carbon emission reduction in grain production by improving the supply efficiency of agricultural resources. In contrast, non-major grain-producing areas have limited agricultural resources and a functional orientation focused on stabilizing production and ensuring supply. Here, carbon emission reduction primarily relies on Agricultural Water Infrastructure to reduce water consumption and Digital Infrastructure to minimize the application of pesticides and fertilizers. As a result, the effects of these infrastructure types on increasing grain production are lower in non-major grain-producing areas compared to those in major grain-producing areas.

5.4.3. Heterogeneity in Types of Grain Crops

The carbon emission reduction effects of various types of agricultural infrastructure on grain production may vary due to differences in grain crop varieties. To examine the heterogeneous impacts of the four types of infrastructure on carbon emissions across different crops, this study draws on the research of Qiu et al. (2024) [55] and applies the carbon emissions intensity calculation formula established in Section 4.2.1 to separately compute the carbon emission intensities of wheat, corn, and rice, followed by empirical testing. According to the regression results in columns (1)–(3) of

Table 9. Heterogeneity analysis in types of grain crops.

	(1)	(2)	(3)
	lnRCE	lnWCE	lnPCE
lnIrri	−1.590 ***	−1.520 ***	−1.067 **
	(0.614)	(0.396)	(0.483)
lnRoad	−0.572 ***	−0.490 ***	−0.364 ***
	(0.314)	(0.143)	(0.064)
lnElec	−0.207 **	−0.149 ***	−0.125 **
	(0.093)	(0.086)	(0.055)
Digi	−0.783 **	−0.728 **	−0.687 ***
	(0.364)	(0.282)	(0.234)
Cons	4.181 **	4.030 ***	1.653 ***
	(1.971)	(1.270)	(0.259)
Control	Yes	Yes	Yes
Province FE	Yes	Yes	Yes
Time FE	Yes	Yes	Yes
N	450	450	450
R2	0.977	0.957	0.967

Note: ** and *** denote significance at the 5% and 1% levels, respectively. The t-statistics are reported in parentheses.

, the inhibitory effects of the four types of agricultural infrastructure on the carbon emissions intensity of different grain crops vary, with the effectiveness ranking as follows: rice, wheat, and corn. These results are statistically significant at the 5% and 1% levels, indicating that agricultural infrastructure construction has a stronger carbon emission reduction effect on rice and wheat production. The reason for this lies in the fact that corn, as a dryland crop, has carbon emissions during its production process that are highly dependent on fertilizer inputs and mechanical operations. As a result, the effect of the four types of agricultural infrastructure on the carbon emissions intensity of corn is relatively weaker.

6. Conclusions and Policy Implications

6.1. Conclusions

This study aimed to assess the impact of agricultural infrastructure on carbon reduction in grain production across 30 Chinese provinces from 2009 to 2023. Using two-way fixed-effects model and mediation-effect model, the study evaluated different types of agricultural infrastructure and their effects on carbon reduction in grain production. The main conclusions are as follows:

(1)

Agricultural infrastructure construction significantly inhibits carbon emissions from grain production. Overall, it markedly reduces the carbon emissions intensity of grain production. However, the magnitude of this negative effect varies by infrastructure type, in the following order: agricultural water infrastructure, digital infrastructure, agricultural power infrastructure, and rural transportation infrastructure. This conclusion remains robust after a series of rigorous tests.

(2)

Agricultural infrastructure facilitates carbon emission reduction in China’s grain production through three mechanisms: optimizing the planting structure, promoting technological progress, and reducing the disaster incidence rate. Specifically, agricultural water infrastructure and digital infrastructure lower carbon emissions intensity by improving the planting structure and advancing technology, while agricultural water infrastructure and rural transportation infrastructure contribute to emission reduction mainly by lowering the disaster incidence rate.

(3)

The inhibitory effect of agricultural infrastructure on carbon emissions exhibits significant heterogeneity. In terms of the north–south grain production pattern, agricultural water infrastructure and rural transportation infrastructure play a more prominent role in reducing emissions in the northern region, whereas agricultural water infrastructure and digital infrastructure have stronger inhibitory effects in the southern region. From the perspective of functional zoning, agricultural water infrastructure and agricultural power infrastructure show significant emission reduction effects in major grain-producing areas, while agricultural water infrastructure and digital infrastructure are more effective in non-major grain-producing areas. Regarding crop-specific heterogeneity, compared to corn, agricultural infrastructure construction demonstrates greater effects in rice and wheat.

6.2. Limitations and Future Perspectives

This study contributes to the literature by establishing and empirically testing a unified framework that links agricultural water, rural transportation, agricultural power and digital infrastructure to carbon intensity reduction through the specific, hypothesized channels of scale expansion, efficiency enhancement, and risk mitigation. Furthermore, it provides comprehensive evidence on the differential effects of these infrastructure types across northern/southern regions, grain production zones, and major crop varieties, offering a nuanced basis for policy. However, several limitations qualify the findings. The mediation analysis remains susceptible to endogeneity, and the use of proxy variables for concepts like digital infrastructure and agricultural power, for example, total rural electricity consumption, which may capture non-agricultural use and also directly constitute an emission source, along with the exclusion of emissions from manure management, reflects inherent data constraints. The explanation for regional heterogeneity, while provided, could be further deepened. Building on this, promising research directions include: employing more robust causal identification strategies to validate the mechanisms; utilizing higher-resolution data to improve measurement accuracy; and extending the analysis to assess long-term outcomes and potential rebound effects at the micro level.

6.3. Policy Implications

Based on these findings, this research has found agricultural infrastructure in China must be adapted to local circumstances. Several policy implications emerge:

First, policies for agricultural water infrastructure must be regionally tailored. In northern China and major grain-producing areas, the focus should be on expanding water-saving irrigation systems, such as drip and sprinkler irrigation, supported by modernized canals and smart water-allocation mechanisms. In southern regions and non-major production zones, development should prioritize smart irrigation districts and integrated drainage-irrigation facilities, utilizing automated controls for pumps and sluice gates to enhance resilience against floods and droughts.

Second, a differentiated strategy is needed for deploying digital infrastructure. In southern regions and non-major grain-producing areas, investment should target agricultural IoT networks and intelligent decision-support systems, deploying field sensors and smart irrigation devices for precise resource management. Conversely, in core grain-producing areas, efforts should center on establishing comprehensive big-data platforms for production while scaling up smart agricultural equipment like UAV-based plant-protection systems to optimize inputs and reduce field operations.

Third, the modernization of agricultural power infrastructure should be strategically focused. In major grain-producing areas, priorities include upgrading in-field power grids and promoting photovoltaic-agriculture projects to supply clean energy for farming. In other regions, the emphasis should shift to deploying agricultural clean-energy micro-grids and refining grid-integration protocols for photovoltaic systems, thereby accelerating the national transition to sustainable agricultural power.

Fourth, policies governing rural transportation infrastructure must reflect regional variations in impact. In the north and major grain belts, development should strengthen grain-storage facilities, logistics hubs, and field road networks, alongside integrated cleaning, drying, and storage facilities at production sites. Nationally, integrating existing logistics resources—such as postal and supply chain networks—can reduce energy consumption across the grain circulation system, supporting its low-carbon transformation through optimized logistics and coordinated resource use.