Agriculture is characterized by long production cycles and high market uncertainty, which poses challenges for traditional rural financial products to serve agricultural production. The long-standing mismatch between supply and demand in rural finance has led to insufficient funding for agricultural production, severely constraining the development of large-scale and modern agriculture and creating a vicious cycle of mutual restraint. According to the endogenous financial growth theory, as the level of financial development improves, reductions in information asymmetry and transaction costs can effectively weaken the constraints on accessing rural financial services, thereby fostering a mutually beneficial relationship between rural finance and agriculture.

With the continuous improvement of rural digital infrastructure, digital technologies such as artificial intelligence and big data are enabling financial institutions to innovate and launch agricultural credit, insurance services, and other financial products. This enriches the rural financial service system and effectively enhances farmers’ investment capacity in agricultural production factors (Morshadul et al., 2020). Among these, rural credit and agricultural insurance have become effective institutional arrangements for alleviating financial constraints and balancing agricultural risks (Ai et al., 2023). At the same time, enhanced digital infrastructure helps financial institutions to accurately obtain real-time information on agricultural production and operations, which aids in evaluating and predicting risks associated with agricultural practitioners, thereby increasing the likelihood and efficiency of farmers accessing financial products. Digital inclusive finance is a new digital financial model that utilizes technology to lower the threshold for financial services and expand their scope (Hathroubi, 2019). As rural digital infrastructure improves, digital inclusive finance can achieve broader coverage of rural areas and agricultural market entities. Through digital financial platforms, farmers and agricultural enterprises can conveniently obtain loan services, which increases the availability of productive capital (Ozili, 2017). This provides strong financial support for expanding operational scales and for post-disaster recovery of grain production (Balogun et al., 2020), thus strengthening the recovery capacity of grain production when faced with external shocks. Accordingly, the following hypothesis is proposed:

Technological progress is a fundamental force driving sustained increases in grain yield and improvements in agricultural quality and efficiency (Gouvea et al., 2022). Innovation theory suggests that technological transformation and upgrading of existing production systems can promote the free flow of resource elements such as labor, capital, and information. This, in turn, leads to optimized resource allocation, industrial upgrading, and an overall increase in production efficiency (Das et al., 2020), which is conducive to nurturing new industries, new models, new driving forces, and enhancing the risk resistance of product markets.

As rural digital infrastructure is upgraded, the application scenarios for digital and intelligent technologies in rural areas are continually enriched. The application of rapidly iterating digital and mechanized products alleviates rural labor shortages, improving agricultural production efficiency while reducing dependence on traditional production factors, thereby effectively enhancing agricultural production resilience. Smart agriculture and Internet of Things (IoT) projects reduce the information-matching costs for land transfers, accelerate the orderly digital transfer of farmland, and increase farmers’ motivation to cultivate grain. The establishment of digital platforms such as “Yinong” information service stations and e-commerce service points effectively integrates smallholder farmers and connects them to larger markets. This not only facilitates the rapid sharing of production information among smallholders but also expands their sources of information for grain sales channels, thereby strengthening their grain production resilience. On this basis, the following hypothesis is proposed:

Digital infrastructure across China’s provinces is not synchronized and is often closely linked to regional economic development levels. Generally, cities with higher levels of economic development are capable of investing more financial, technical, and human resources into digital infrastructure. Their construction timelines are often ahead of less developed cities, and they can provide more comprehensive digital services to farmers, helping them to conduct agricultural production more effectively. In contrast, economically underdeveloped cities have relatively lagging digital infrastructure and fewer application scenarios for digitally empowering agricultural production. Based on the theory of latecomer advantage, this paper posits that the implementation of the “Broadband China” pilot policy will have a more significant effect on improving digital applications in less economically developed regions, and consequently, a stronger enhancement effect on grain production resilience. Therefore, the following hypothesis is proposed:

Compared to the plain areas in the north, the cultivated land in southern regions is characterized by steep slopes and fragmentation, which is unfavorable for both mechanized operations and large-scale farming. This results in a relatively weaker risk-resistance capacity for agricultural production in the south. The construction of digital infrastructure can enhance the disaster resistance and production efficiency of grain cultivation through precision meteorological monitoring, early warning systems, and smart agricultural technologies. It can also provide farmers with more accurate market information, comprehensively improving agricultural production resilience. Based on the latecomer advantage theory, this paper argues that because the effectiveness of digital infrastructure construction in southern regions is initially lower due to complex topography, the “Broadband China” pilot policy will have a more pronounced effect on improving its digital service level and, consequently, on enhancing grain production capacity. Based on this reasoning, the following hypothesis is proposed:

In summary, the following theoretical analysis framework is constructed (Figure 1).

Following the existing research (Hao and Tan, 2022), this study constructs a comprehensive index system for grain production resilience based on three dimensions: Resistance, Recovery, and Transformability. This system comprises 6 subsystems and 16 elemental indicators (as shown in Table 1).

Resistance refers to the ability of the grain production system to mitigate and absorb internal and external risks and shocks. It includes two sub-dimensions: internal stability and supply-production robustness. Recovery denotes the speed of internal restoration and the degree of adaptation after the grain production system has been subjected to a shock. It includes two sub-dimensions: sustainability and recoverability. Transformability represents the capacity for self-adjustment and reorganization of the grain production system following a shock. It includes two sub-dimensions: diversity and collaboration and diversity and coordination.

To minimize bias in the measurement of the grain production resilience index, this paper refers to the study by Gao et al. (2024), and employs the entropy weight method for objective weighting. The specific calculation steps are as follows:

Step 1: Data Standardization. The range method is used to perform dimensionless and directional standardization on the positive and negative indicators within the evaluation system. The formula for positive indicators is equation (1), and the formula for negative indicators is equation (2):

Step 2: Calculate the proportion matrix. In equation (3), it involves calculating the proportion of the j indicator in the i year, where m is the number of years included in the evaluation.

Step 3: Calculate the information entropy and information entropy redundancy of the indicator. The formula for information entropy is equation (4):

Among them, k = 1 Inm . . The formula for information entropy redundancy of the indicator is equation (5):

Step 4: Calculate the indicator weights. The formula is equation (6):

Step 5: Calculate the comprehensive score. The comprehensive score for grain production resilience is calculated using the following formula:

In equation (7), S_ij is the standardized value of the j indicator for city i, x^′_ij is the original value of the j indicator for city i, W_j is the weight of the j indicator, and S_ij represents the score of the j indicator for city i. A higher score indicates a greater contribution of that indicator to the comprehensive grain production resilience score, while a lower score signifies a smaller contribution.

This paper uses the implementation of the “Broadband China” strategic pilot policy as a proxy variable for digital infrastructure construction. The variable did takes a value of 1 if city i has been selected as a “Broadband China” pilot city in year t, and 0 otherwise.

Agricultural Technological Innovation (inno). Following the study by Liu et al. (2021), this paper measures agricultural technological innovation by the sum of patents granted in four categories: agriculture, forestry, animal husbandry, and fishery.

Digital Inclusive Finance (fin). Drawing on the research of Du Yanan et al. (2023), this paper uses the Peking University Digital Financial Inclusion Index as the proxy variable for digital inclusive finance.

Previous studies have laid a foundation for our investigation into the impact of digital infrastructure on grain production resilience. Referencing relative researches (Hu et al., 2024; Zhang et al., 2024), this paper selects the following covariates that may influence grain production resilience: economic development level, industrial structure, education level, and openness to foreign investment. Specifically: Economic Development Level means the logarithm of per capita regional GDP. Industrial Structure means the proportion of the sum of the added value of the secondary and tertiary industries to the regional GDP. Education Level means the ratio of the number of students enrolled in regular institutions of higher education to the total year-end resident population. Openness to Foreign Investment means the ratio of foreign direct investment (FDI) to the city’s GDP. Government Intervention means the ratio of local general public budget expenditure to the regional GDP. The descriptive statistics for all variables are presented in Table 2.

This study uses the “Broadband China” pilot policy shows the digital infrastructure construction. Given that the “Broadband China” pilot policy was implemented in three batches in 2014, 2015, and 2016, the implementation times of the “Broadband China” pilot policy vary across different regions. Since the implementation timing of this policy varies across different regions, this study adopts the method from existing research (Tian and Zhang, 2022) to construct a staggered difference-in-differences (DID) model with two-way fixed effects. The model is specified as follows:

In Equation (8), F_it represents the grain production resilience index for city i in year t. DiD_it denotes the policy implementation status for city i in year t. Control X_it is a vector of other control variables that affect grain production resilience. μ_i represents city-level fixed effects, and ω t represents year fixed effects. α 0 , α 1 and α 2 are the coefficients to be estimated, and ε it is the random error term.

The validity of the staggered DID model relies on the satisfaction of the parallel trends assumption. This assumption requires that, prior to the implementation of the “Broadband China” policy, the treatment and control groups exhibited a consistent trend in the evolution of their grain production resilience. Drawing on the work of Jacobson et al. (1992), this study employs an event study approach to conduct the parallel trends test. The model is specified as follows:

In Equation (9), D i , t 0 + k is a dummy variable that corresponds to the timing of the policy implementation. As the base year is 2016 and the sample period is 2012–2022, the value of k ranges from −4 to 6. X it represents the control variables. The coefficient α t captures the difference in grain production resilience between the treatment and control groups in the k year relative to the policy’s implementation.

The theoretical analysis suggests that digital infrastructure construction may enhance grain production resilience by promoting agricultural technological innovation and developing digital inclusive finance. To test these mediating mechanisms, we refer to the research (Jiang, 2022), and specify the mediation model as follows:

In Equation (10), M_it represents the mediating variables, which include digital inclusive finance and agricultural technological innovation. DiD_it denotes the policy implementation status for city i in year t. Control X_it is the vector of control variables, μ_i represents city-level fixed effects, and ω t represents year fixed effects. β 0 , β 1 and β 2 are the coefficients to be estimated, and ε it is the random error term. Based on our theoretical analysis and the general consensus in the academic literature that digital inclusive finance and agricultural technological innovation significantly enhance grain production resilience, we can infer that a mediation effect exists if the coefficient β 1 is statistically significant.

To ensure the robustness of the baseline regression results, this paper conducts a parallel trends test, with the year prior to policy implementation (k = −4) set as the reference period. As shown in Figure 2, for all periods before the implementation of the “Broadband China” policy (k < 0), the estimated coefficients fluctuate around zero and their 95% confidence intervals include zero, indicating no statistically significant difference in grain production resilience between pilot and non-pilot cities prior to the policy rollout. In contrast, the regression coefficients become statistically significant in all periods following policy implementation, suggesting that digital infrastructure construction significantly enhanced grain production resilience in pilot cities after the policy took effect. Moreover, the positive effect exhibits a strengthening trend over time. These results confirm that the parallel trends assumption holds for the sample used in this study.

To address potential biases from unobservable factors in the baseline estimates, this study performs a placebo test following the approach of Chetty et al. (2009). The sample was randomly reshuffled without stratification, with treatment status and policy timing assigned randomly across observations to construct a pseudo-policy shock variable. This variable was then incorporated into Model (1) for 1,000 repeated regressions. The distribution of the p-values of the estimated coefficients from these regressions is shown in Figure 3. The results indicate that the estimated coefficients of the randomly generated pseudo-policy variable are concentrated around 0 and follow an approximately normal distribution. This allows us to largely rule out the possibility that the positive effect of the “Broadband China” pilot policy on grain production resilience is driven by unobservable factors, thereby confirming the reliability of our research conclusions.

To mitigate the potential influence of systematic differences between pilot and non-pilot cities, this study re-estimates the model using the Propensity Score Matching-Difference-in-Differences (PSM-DID) method. To ensure consistency with the original full sample (including control variables), we use the control variables from the baseline regression as covariates and employ the nearest-neighbor caliper matching method to match treatment and control group samples on a year-by-year basis. As shown in Table 4, after matching, the absolute standardized mean differences (SMD) of all covariates fall below 0.1, and t-tests for between-group differences become statistically insignificant, indicating that the balancing assumption is satisfied. The subsequent regression results, presented in column (1) of Table 5, show that even after controlling for sample-selection bias, the “Broadband China” pilot policy continues to exert a significant positive effect on grain production resilience. This finding further supports H1.

Considering that the six provincial capitals (Taiyuan, Nanchang, Zhengzhou, Hefei, Changsha, and Wuhan) have relatively advanced digital infrastructure, their inclusion in the regression sample could potentially inflate the overall estimated coefficients. Therefore, we exclude these six cities and re-run the empirical test. The results in Table 5, column (2), show that the implementation of the “Broadband China” pilot policy still significantly enhances grain production resilience.

The implementation of “Smart City” policies has driven urban digital transformation and intelligent development, which has also spurred the upgrading of agricultural productive services (Jia, 2023), playing a positive role in ensuring China’s food security. To exclude the potential impact of the “Smart City” policy and the “National Big Data Comprehensive Pilot Zone” on grain production resilience, we include dummy variables for these two policies as additional controls in a robustness test. The results, shown in Table 5, column (3), indicate that the “Broadband China” pilot policy continues to significantly promote grain production resilience, once again verifying H1.

The selection of “Broadband China” pilot cities could be influenced by pre-existing factors such as the city’s original infrastructure level. Therefore, using this policy as a proxy for digital infrastructure to study its impact on grain production resilience may be subject to endogeneity from omitted variables or reverse causality. To address this issue, this study follow the approach of Huang et al. (2019) and Nunn and Qian (2014) by using an instrumental variable (IV). The IV is constructed as the interaction between the lagged number of national internet users and the number of post offices per million people in each city in 1984. A two-stage least squares (2SLS) approach is employed to capture the net effect. The distribution of post offices in 1984 was primarily determined by administrative planning during the planned-economy era and is not directly linked to contemporary grain production resilience. As early communication infrastructure, however, the historical distribution of post offices may have influenced subsequent internet penetration, thereby relating indirectly to the intensity of “Broadband China” policy implementation. The lagged number of internet users at the prefecture-level city level reflects regional trends in technology diffusion, which can affect local Broadband deployment but is unlikely to be reversely influenced by grain production resilience in any single city, thus satisfying the exclusion restriction of the IV. The first-stage results are reported in column (4) of Table 5, where the estimated coefficient of the IV on the “Broadband China” policy is statistically significant at the 1% level, confirming the relevance of the instrument. Column (5) presents the second-stage results, which indicate that the “Broadband China” policy continues to have a significant positive effect on grain production resilience. These findings suggest that endogeneity does not substantially undermine the main conclusions of the study.

A city’s level of economic development is a core support for digital infrastructure construction and an enabler for ensuring grain production resilience. In this study, the sample is divided into “economically developed” and “economically less developed” groups based on per capita GDP. The results of the grouped regression (columns 1–2 in Table 7) show that while the “Broadband China” pilot policy enhances grain production resilience in both, the effect is only statistically significant in the group of economically less developed cities.

As shown in Table 8, before the policy implementation, cities selected as “Broadband China” pilot sites exhibited significantly lower levels of economic development (mean = 0.285) than non-pilot cities (mean = 0.689), indicating that the policy has covered more areas with initially weaker economic conditions. In these “low-starting-point” regions with limited existing digital infrastructure, the pilot policy has generated a notable “catch-up effect.” By rapidly narrowing the digital divide, it has substantially unlocked the potential for empowering agricultural production systems, thereby significantly strengthening grain production resilience. In contrast, economically developed cities already possessed relatively complete digital infrastructure and mature smart-agriculture applications prior to the policy. For these cities, the “Broadband China” policy offered limited scope for further digital infrastructure improvements, resulting in a weaker effect on enhancing grain production resilience. These findings support H4.

This study categorizes sample cities into northern and southern regions based on agro-geographical conditions and dominant grain crop types for subgroup analysis. The regression results, reported in columns 3–4 of Table 7, indicate that the “Broadband China” pilot policy significantly enhances grain production resilience in southern cities, whereas its effect in northern regions is not statistically significant. This heterogeneous impact stems primarily from differences in resource endowments and production risks between the two regions. Compared with the plains-dominated north, southern China is characterized by steep, fragmented arable land with extensive mountainous and hilly terrain. This topography limits the application of traditional large-scale mechanization and inherently weakens the risk-resistance capacity of agricultural production. Prior studies have shown that land fragmentation is more severe in mountainous and hilly areas than in plains, raising costs for labor, fertilizer, and seeds while constraining the use of machinery and other productive inputs (Wang et al., 2019; Zhang X. et al., 2023; Zhang D. J. et al., 2023).

Digital infrastructure—particularly the technologies it enables, such as precision meteorological monitoring, early-warning systems, remote sensing, and smart agricultural IoT—can effectively mitigate these topographic constraints. As argued by Zhao (2021), digital-enabled smart agriculture represents a crucial pathway for addressing land fragmentation and advancing large-scale agricultural management. For example, high-precision spatial information supports optimized soil-water management and enables precise input allocation across scattered plots, reducing resource waste while enhancing disaster resilience and productivity of small-scale farmland. Consequently, the marginal benefits of digital technologies are more pronounced where traditional farming models face greater constraints. At the same time, the complex terrain in the south raises the cost of deploying digital infrastructure. The “Broadband China” policy, through concentrated investment, effectively improves digital service levels in these regions, making its role in strengthening grain production resilience more evident. In contrast, the northern plains are more suited to large-scale mechanized operations, where existing production systems already possess relatively strong stability and risk resistance. Although digital technologies—such as smart machinery scheduling and big-data market analysis—can bring further optimization, their marginal contribution to overall resilience is comparatively limited. These findings support H5.