In the context of digital transformation, DVC improves the overall security of the food system by reallocating informational elements and optimizing the institutional environment, which enhances information efficiency and the way resources are allocated, ultimately optimizing the operation logic of food production and supply systems. From the perspective of information economics, agricultural production generally suffers from information asymmetry and high information search costs, leading to distorted resource allocation and reduced production efficiency. Rural digitalization significantly improves production decision-making efficiency by reducing information acquisition and transmission costs and increasing information transparency (Aker, 2011).

First, agricultural informatization constitutes a core driving force in the evolution of modern agricultural production systems. By embedding digital technologies into production processes, it reshapes information transmission and decision making structures within agricultural production (Zhu et al., 2019; Han and Zhang, 2015). As a fundamental pillar of DVC, digital infrastructure functions not only as the physical carrier of information flows but also as a prerequisite for the integration of digital technologies into production systems. Improvements in rural network coverage, data collection terminals, and information sensing facilities enable digital technologies to be embedded throughout grain production, management, and circulation processes, facilitating a shift from experience based to data-oriented production. On the one hand, enhanced digital infrastructure substantially reduces information acquisition and transaction costs and mitigates the information asymmetry that is pervasive in agricultural production (Aker et al., 2016). This allows farmers to more promptly access information on climatic conditions, production environments, and market signals, thereby reducing uncertainty in production decisions, optimizing input allocation, and enhancing the stability and sustainability of grain production. On the other hand, the integration of digital technologies into production processes under improved infrastructure conditions increases precision and controllability, reduces resource waste and environmental pressure, and improves the stability of grain output while simultaneously accounting for ecological constraints and structural coordination.

Second, building upon the strengthened technical foundation provided by digital infrastructure, DVC further promotes FSL through digital services. Grain production inherently relies on the input of key factors, including labor, capital, and technology. By leveraging the inclusive effects of digital technologies, DVC expands access to credit resources and information channels for small scale producers (Yi, 2021), enabling farmers to dynamically adjust input structures during the production process, enhance their capacity to respond to external shocks, and improve the stability of the grain production system. Capital constitutes a critical support for grain production activities, and digital inclusive finance alleviates financing constraints faced by farmers by virtue of its digital and inclusive characteristics, thereby effectively safeguarding food security (Lin et al., 2022). Technology represents another important driving force, as agricultural technological innovation accelerates the transition toward green agricultural development, promotes resource saving and environmentally friendly production, and enhances the ecological security of grain production (Zhang et al., 2022). Through information feedback and service coordination mechanisms, digital services guide the rational allocation of production factors, mitigate risks of structural imbalance, and strengthen the overall resilience of the food system.

Taken together, DVC can reinforce FSL through a mechanism chain encompassing information efficiency enhancement, factor optimization, and improvements in system resilience. Accordingly, Hypothesis 1 is proposed.

DVC not only exerts a direct effect on FSL but also generates indirect effects by alleviating key constraints in grain production. In the production process, the achievement of FSL is often jointly constrained by technological limitations arising from insufficient farmers’ capacity to absorb technology and by factor constraints resulting from low efficiency in land factor allocation. By improving the information environment and the conditions for digital technology development, DVC provides critical support for relaxing these constraints. Based on this logic, this paper selects agricultural technology training coverage and land productivity as mediating variables to characterize the transmission pathways through which DVC influences FSL by easing technological and factor constraints.

First, by integrating information platforms with educational systems, DVC enhances both the breadth and effectiveness of agricultural technology training, thereby alleviating technological constraints in grain production. According to information economics theory, the diffusion of the internet facilitates information exchange among governments, research institutions, and social organizations, improving the efficiency of knowledge dissemination (Jiang et al., 2021). However, in developing countries, farmers commonly face knowledge gaps, particularly in pesticide application and pest and disease identification, and grain production decisions often rely on experience based judgment (Li et al., 2017; Khan and Damalas, 2015; Zhu et al., 2014), which increases production risk. Although agricultural information infrastructure has continued to improve, disparities persist in farmers’ capacity to access and utilize information, especially among emerging agricultural operators (Ruan et al., 2017). Farmer behavior theory suggests that technology adoption is primarily driven by expected benefit maximization, while the positive externalities associated with the digital attributes of DVC improve the decision environment for technology adoption by reducing information search costs and replication costs (Goldfarb and Tucker, 2019). Digital platforms not only lower training costs but also enable more precise matching of technical guidance. Existing studies show that improvements in farmers’ digital literacy significantly strengthen their sustainable cultivation practices (Liu et al., 2025). Taken together, DVC mitigates technological constraints in grain production by expanding agricultural technology training coverage, thereby enhancing FSL.

Second, DVC alleviates factor constraints in grain production by promoting the flow and intensive utilization of land factors, thereby improving land productivity. According to induced technical change theory, agricultural technological progress is not exogenously determined, but is jointly induced by factor endowment structures and market demands (Hayami and Vernon, 1985). When land factors are relatively scarce or large-scale operations are continuously advanced, technological progress will be directed towards improving land productivity. DVC accelerates the adoption of land-saving and efficiency-enhancing agricultural technologies by improving digital infrastructure and the technological application environment, allowing technological progress to better align with land resource constraints. In this process, the application of digital technologies in production enhances the precision and efficiency of land use. For example, automated machinery and autonomous operating systems improve field operation efficiency, reducing labor and resource consumption (Hu and Liu, 2025). At the same time, digital technologies accelerate the dissemination of agricultural technologies, enabling high-yield technologies and high-quality varieties to be applied more quickly in frontline production. Furthermore, the flow of information through the internet promotes rural land transfers, improving land use efficiency and the level of large-scale operations (Liu et al., 2021; He, 2016). Under the combined influence of digital technologies and the institutional environment, land factors are allocated more efficiently, thus strengthening food security.

Taken together, DVC indirectly promotes FSL through knowledge diffusion and factor reallocation effects. Accordingly, Hypothesis 2 is proposed.

In the process of DVC promoting FSL, government actions play a critical institutional moderating role. According to public economics and agricultural policy theory, government size is not only associated with public resource allocation capacity, but also affects the coordination efficiency between digital infrastructure and the grain production system. When the proportion of government fiscal expenditure to GDP is high, it indicates that the public sector has greater input and regulatory capacity in the agricultural and rural sectors. The government’s role in resource provision, institutional protection, and policy incentives will significantly amplify the marginal effects of DVC (Wang and Hou, 2024).

Under the same level of DVC, differences in government scale will influence the actual effectiveness of DVC implementation through fiscal input capacity. From a mechanism perspective, fiscal expansion enhances the construction of rural digital infrastructure, providing the necessary resource conditions for DVC to impact FSL. Studies have shown that government agricultural expenditure is positively correlated with agricultural output and food security (Iganiga and Unemhilin, 2011; Marson, 2025). In China, agricultural fiscal expenditure mainly focuses on farmland water conservancy, agricultural technology, and infrastructure construction (Gong and Wang, 2021), which helps improve digital infrastructure such as rural internet, meteorological monitoring, and data services. The promotion of agriculture and food security through digitalization is institutionally dependent, and its effectiveness partially depends on government fiscal capacity and governance levels (World Bank, 2021). Strengthened fiscal support lowers the barriers to the application of digital technologies, accelerating the extension of agricultural informatization to the grassroots level, thereby improving the implementation efficiency of DVC in the field of grain production. Through subsidies and other fiscal support for agriculture, government departments promote the substitution of labor by agricultural machinery, increasing mechanization levels and food production efficiency (Lv et al., 2015). A larger government scale can facilitate multi-party cooperation between research institutions, digital platforms, and farmers through policy subsidies and technology promotion, deepening the integration of digitalization and grain production, thereby amplifying the effects of DVC on food production efficiency and ecological sustainability. Conversely, when government resource provision is insufficient or fiscal support is weak, digital infrastructure investment is prone to regional gaps, limiting the diffusion and application of digital technologies in rural areas, and significantly weakening the impact of DVC on FSL. Therefore, government scale plays a positive moderating role in the process of DVC influencing FSL by enhancing resource provision capacity.

This study employed provincial panel data from 31 provinces in China from 2011 to 2023 (excluding Hong Kong, Macau, and Taiwan) as the research sample. The data for the main variables are sourced as follows: rural digital inclusive finance index is taken from Peking University Digital Finance Research Center (2011–2023) agricultural patent data is sourced from the China National Knowledge Infrastructure (CNKI) patent database; agricultural technology training coverage is calculated and constructed using the method of Zhu and Ye (2024). Data for the remaining indicators are primarily obtained from the official website of the National Bureau of Statistics and the China Statistical Yearbook. Missing values for some indicators are supplemented and corrected based on the China Rural Statistical Yearbook and various provincial statistical yearbooks.

A two-way fixed effects (TWFE) model is employed to control for unobservable individual heterogeneity and macro-level temporal shocks that may bias the estimation results. The inclusion of fixed effects helps mitigate omitted variable bias and enhances the robustness of the estimates. Continuous variables with large standard deviations are logarithmically transformed. The baseline model is specified as follows:

In the above equation, FS i , t denotes the FSL of province i in year t , and DIG i , t represents the level of DVC, which is the core explanatory variable of this study. Control i , t includes all control variables. μ i denotes province fixed effects, and δ t denotes time fixed effects, which are used to control for unobserved time-invariant individual characteristics and macro-level shocks. ε i , t represents the random disturbance term. The estimated coefficient β 1 reflects the direction and magnitude of the direct impact of DVC on FSL.

Figure 2 displays the spatial distribution pattern of DVC and FSL levels across Chinese provinces. Overall, the level of DVC shows a spatial pattern of decreasing from east to west. Coastal provinces in the eastern region generally have higher levels, with an average index of 0.334, reflecting well-developed digital infrastructure and strong technological penetration. Most provinces in the central region are at a moderate level, with an index of 0.296, indicating relatively balanced digital development. The western and southwestern regions have significantly lower indices, with an index of 0.249. Some provinces are still in the early stages of DVC, and regional digital development is highly unbalanced.

The level of FSL shows a spatial pattern of “higher in major production areas, lower in non-major production areas,” with the average index in major production areas at 0.459, higher than the 0.241 in non-major production areas. The three northeastern provinces, the central grain production areas, and the Huang-Huai-Hai Plain generally fall into higher gradients, forming the core stable areas of China’s food security, while some southeastern coastal areas have relatively low FSL due to land resource constraints and a high degree of non-food crop development. Some western provinces are constrained by natural conditions, inadequate infrastructure, and limited technical support, resulting in generally weaker FSL.

From a spatial comparison, the distribution of DVC and FSL is not fully aligned. The eastern region has high levels of digitalization, but its FSL is not outstanding, while the central major production areas, despite having moderate levels of digitalization, show significantly higher FSL. The western region falls into lower gradients for both indicators. This discrepancy suggests that there may be regional dependencies and structural differences between DVC and FSL, providing important spatial context for the subsequent analysis.

Table 4 reports the estimated results of the impact of DVC on FSL. Columns (1) to (3) present the estimated results using ordinary standard errors, robust standard errors, and standard errors obtained through 1,000 bootstrap random sampling, respectively. The results show that, regardless of the standard error method used, the estimated coefficient of DVC is positive and significant at the 1% level, indicating that DVC has a significant positive impact on FSL. Specifically, holding other conditions constant, for every 1-unit increase in DVC, the FSL index increases by 0.083 units. This result suggests that the enhancement of rural digital infrastructure and digital service capacity can significantly improve the quantity, structure, and ecological security of food production by optimizing information transmission and resource allocation efficiency, thereby enhancing overall FSL. Hypothesis H1 is supported.

Trimming extreme values: Considering that the composite indices of DVC and FSL, calculated using the entropy weight method, may be influenced by outliers, and that some provinces or years exhibit large distributional differences, this study applies a 5% winsorization at both ends to reduce the impact of extreme values. The results in column (1) of Table 5 show that, after trimming the outliers, the regression coefficient for DVC remains significantly positive at the 1% level, confirming the robustness of the baseline regression results.

Lagging variables by one period: Given that the effects of DVC on FSL may exhibit temporal lags, this study lags the DVC variable by one period and re-estimates the model. The results in column (2) of Table 5 show that the lagged coefficient remains significantly positive, consistent with the baseline regression. This suggests that the promoting effect of DVC on FSL is not only immediate but also persistent and cumulative over time.

Excluding the impact of the COVID-19 pandemic: The exogenous shocks induced by the COVID-19 pandemic may have substantially affected both FSL and DVC. To address this concern, observations from 2020 onward are excluded, and the model is re-estimated. The results in column (3) of Table 5 show that the coefficient of DVC remains positive and significant at the 1% level, with only minor changes in magnitude. This indicates that the findings are not driven by short-term pandemic-induced fluctuations and that the positive effect of DVC on FSL remains robust across different economic and social environments.

In the identification of the impact of DVC on FSL, potential endogeneity issues arise. On one hand, reverse causality may lead to endogeneity bias, as regions with higher FSL tend to have better agricultural infrastructure, stronger policy support, and are more likely to promote DVC. On the other hand, omitted variable bias cannot be ignored, as factors such as institutional environment, local governance capacity, and policy implementation strength affect both DVC and FSL, potentially leading to biased OLS estimates. To address this, this paper employs the Two-Stage Least Squares (2SLS) method for robustness checks. Following the methods of Zhao et al. (2020), He and Zhang (2022) and Zhang et al. (2019), this paper selects the number of post and telecommunication bureaus in each province in 1984 and the distance between each city and Hangzhou as instrumental variables. Since the 1984 postal density and geographic distance are cross-sectional data, the paper constructs interaction terms between the national average DVC index and the number of post offices and geographic distance to generate panel instrumental variables.

Regarding the number of post and telecommunication bureaus in each province in 1984, on one hand, the telecommunication infrastructure in the 1980s was a crucial reflection of regional information dissemination and communication conditions. The internet is an extension and development of traditional communication technologies, and the historical telecommunication infrastructure in the region influenced the subsequent application of internet technology through factors such as technological level and usage habits (Zhao et al., 2020). Empirical evidence also shows that DVC is strongly correlated with the existing level of telecommunication infrastructure, satisfying the relevance requirement for the instrumental variable. On the other hand, traditional communication media, represented by post and telecommunication bureaus, are used less frequently today, and their influence on the socio-economic environment has gradually diminished (Lin and Zhu, 2022), with no direct causal link to the current FSL, thus meeting the exclusion restriction for the instrumental variable.

Regarding geographic distance, on one hand, DVC is influenced by the radiation effect of core digital hubs. Hangzhou, as a digital economy hub, has a technological diffusion effect that diminishes with increasing geographic distance. The distance from provincial capitals to Hangzhou is related to the ease of digital technology diffusion to rural areas. Areas closer to Hangzhou generally have higher DVC advancement efficiency, thus meeting the relevance requirement for the instrumental variable. On the other hand, the distance from provincial capitals to Hangzhou is a historically established geographic endowment, which does not directly interfere with core aspects such as crop cultivation and production efficiency but only indirectly affects DVC through the diffusion of digital technology. Therefore, both variables satisfy the relevance and exogeneity assumptions for instrumental variables.

The regression results in column (1) of Table 6 show that the instrumental variables are statistically significant, and their direction of effect is consistent with theoretical expectations, satisfying the relevance condition. The second-stage regression results show that the estimated coefficient for DVC is 0.223, which is significantly positive at the 1% level, indicating that, after considering potential endogeneity bias, the instrumental variable estimates align with the direction of the baseline regression, providing supplementary evidence for the impact of DVC on FSL. Furthermore, the Durbin–Wu–Hausman (DWH) test yields a p-value of 0.010, rejecting the exogeneity hypothesis at the 1% level, suggesting that DVC is an endogenous variable. The Cragg-Donald Wald F-statistic is 31.96, which is greater than the standard critical value of 10, indicating no weak instrument problem. The p-value of the overidentification test is 0.0733, and the null hypothesis of instrument exogeneity is not rejected at the 5% level. However, given that the p-value is close to the significance level, caution should be maintained, and the exogeneity of the instrument is considered to be robust.

Building on the previous theoretical analysis, this paper further examines the causal mechanisms through two mediating pathways: agricultural technology training coverage and land productivity, to explore how DVC impacts FSL. The mechanism analysis follows the mediation effect approach commonly used in economics, as proposed by Jiang (2022). This approach primarily tests the causal relationship between the core explanatory variable and the mediating variables, while ensuring the theoretical directness of the effect of the mediating variable on the dependent variable, without relying on the coefficient decomposition of traditional stepwise methods, thus avoiding bias from the endogeneity of the mediating variable. Therefore, the mechanism analysis in this paper aims to provide empirical evidence at the pathway level, rather than precisely decomposing the size of the mediation effect. The regression results are presented in Table 7.

The regression results in column (1) of Table 7 show that the coefficient for the impact of DVC on agricultural technology training coverage is 0.237, which is statistically significant at the 1% level. This indicates that for every 1-unit increase in DVC, agricultural technology training coverage increases by approximately 0.24 units. This suggests that DVC has a significant practical impact on improving the coverage of agricultural technology training, thereby providing human and technical support for FSL. Rural digitalization enables governments and research institutions to more efficiently disseminate agricultural information and technical knowledge, providing farmers with timely training resources and decision-making guidance. Previous studies have pointed out that one of the main reasons for excessive pesticide and fertilizer use is the lack of technical information and guidance. However, farmers who have received systematic training can effectively reduce fertilizer usage and improve field management (Jia et al., 2013; Ying and Zhu, 2015). The acceptance of new knowledge and technologies by farmers helps to increase crop yields (Hua et al., 2013). Therefore, expanding agricultural technology training coverage helps enhance the sustainability of the food security system from both production efficiency and ecological security perspectives, and DVC provides effective conditions for this.

The regression results in column (2) of Table 7 show that DVC significantly increases land productivity at the 1% level. This result suggests that DVC may enhance production efficiency through improved information precision and optimized factor allocation. Increasing the efficiency of cultivated land use is widely regarded as a key strategy for ensuring China’s food security (Liu et al., 2008). Under the context of DVC, the improvement of information and communication infrastructure and digital service capacity helps enhance the precision of market information supply and production management, thereby improving the efficiency of labor and land factor allocation (Ogutu et al., 2014). Furthermore, the improvement in digital conditions provides the foundation for production entities to more effectively access land use information, production environment data, and technical services, allowing them to better optimize production structures and increase land productivity (Chandio et al., 2024; Peng and Huang, 2024) Thus, the increase in land productivity supports FSL.

To enhance the credibility of the conclusions, this paper further incorporates the mediating variables into the regression model. The results, shown in columns (3) and (4) of Table 7, indicate that after controlling for agricultural technology training coverage and land productivity, the impact of DVC on FSL remains statistically significant and positive, with the direction and significance remaining stable. These results provide supportive empirical evidence that DVC influences FSL through the enhancement of agricultural technology training coverage and land productivity, thereby partially supporting Hypothesis H2.

Based on the global Moran’s I test results, FSL exhibited significant positive spatial autocorrelation between 2012 and 2023, with the Moran’s I values passing the test at the 1% significance level. The spatial autocorrelation of DVC was not significant in the earlier period of the sample but gradually became significantly positive starting from 2016, passing the test at the 10% significance level and above. This indicates that the spatial agglomeration characteristics of DVC began to emerge in the later period of the sample. The results show that both variables exhibit varying degrees of spatial dependence (see Tables 8, 9).

Building on the significant spatial correlation of FSL, this paper applies the testing framework proposed by Elhorst (2014), using LM tests, Hausman tests, SDM simplification tests, and fixed effects structure tests. The Hausman test results reject the random effects model at the 5% significance level. Both the Wald test and LR test reject the null hypothesis of the spatial Durbin model (SDM) being degenerate to the spatial lag model (SAR) and the spatial error model (SEM) at the 1% significance level, confirming the appropriateness of using the SDM model. Regarding the fixed effects structure, the LR test further rejects models that control only for time effects or individual effects, thus confirming the use of the two-way fixed effects SDM.

The results in column (3) of Table 10 show a significant positive relationship between DVC and local FSL, with the regression coefficient being positive at the 1% significance level, indicating that improvements in digital infrastructure and digital service capacity promote local FSL. The spatial lag term for DVC is significantly negative, suggesting that digital progress in neighboring areas does not drive local FSL improvements. In fact, it may generate a certain crowding-out effect through mechanisms such as factor mobility, resource competition, or regional development imbalance, thereby weakening the positive spatial spillover effect between regions. The spatial autoregressive coefficient (ρ) in the two-way fixed effects model does not pass the significance test, indicating that after incorporating the explanatory variables and their spatial lag terms, the spatial linkage effect of FSL itself is not significant. Further decomposition of effects reveals that the direct effect of DVC on FSL is significantly positive, mainly promoting local FSL through improved digital infrastructure and service capacity. However, the indirect effect is significantly negative, indicating that the digital improvement between regions has certain limitations, showing a negative impact of spatial lag effects on cross-regional spillovers, leading to the total effect not passing the significance test.

Building on the regional correlation evidence provided by the spatial model, this paper further examines the differentiated effects of DVC from the perspectives of location and functional positioning.

The process of DVC empowering FSL requires corresponding policy and fiscal support, with local fiscal expenditure on agricultural, forestry, and water affairs serving as a key indicator of government size in the agricultural sector. Government size not only affects the fiscal capacity to support digital infrastructure construction and technology promotion but also influences the depth of digital technology penetration and implementation effectiveness within the agricultural system. To identify the moderating role of government size in the process of DVC impacting FSL, this paper constructs an interaction term between DVC and government size for regression analysis.

The regression results in Table 13 show that the coefficient of the interaction term is 0.0849, and it is statistically significant at the 1% level, meaning that, ceteris paribus, for every 1-unit increase in government size, the marginal empowering effect of DVC on FSL increases by an additional 0.0849 units. This indicates that government size plays a significant positive moderating role, with larger government size leading to stronger fiscal and policy support, effectively lowering the implementation threshold for DVC and amplifying its positive empowering effect on FSL. Specifically, at the hardware level, larger government size corresponds to stronger fiscal mobilization and resource allocation capabilities, which can mobilize more funds to core areas of DVC, accelerating the deployment and upgrading of digital infrastructure, such as Internet of Things (IoT) monitoring devices, smart irrigation systems, and big data platforms for food production. This promotes the deep integration of digital facilities with the entire food production process and alleviates the barriers to the rural implementation of digital technologies. At the software level, sufficient fiscal expenditure ensures the provision of public services for food production, supporting targeted digital technology training to improve farmers’ digital skills and technological absorption capabilities. Furthermore, relying on fiscal subsidies and policy guidance, it enhances complementary services such as digital finance, lowering the adoption threshold and risks for farmers. Overall, the larger the local government size, the stronger its fiscal and resource coordination capacity, which helps bridge the gap between digital technology and food production, thereby amplifying the positive empowering effect of DVC. Hypothesis H3 is supported.