Climate policy uncertainty and agricultural digital-intelligent transformation in China: evidence from provincial panel data

1 Introduction

Global climate risks are increasing, and countries continue to introduce and adjust climate-related policies, which has led to climate policy uncertainty. The uncertainty of climate policy is reflected in the fact that it is difficult to form a stable expectation for the policy trend in the whole cycle from drafting, publicity, implementation to subsequent revision. It stems from the differences in decision-making orientation and implementation intensity of the competent authorities, and superimposes external disturbances such as geographical environment and regional implementation standard differentiation, which increases the difficulty of market players in predicting policy risks (Elie et al., 2022; Chen et al., 2024; Fei et al., 2025; Zhang et al., 2026). This kind of policy uncertainty constitutes an important external impact, while the investment cycle of agricultural digital-intelligent transformation field is long, the investment is irreversible, and it is highly sensitive to policy expectations. In this context, how climate policy uncertainty affects the digital transformation of agriculture has become a research proposition that needs to be tested.

China is located in a highly sensitive area of climate change and is significantly affected by climate fluctuations. In order to actively respond to various challenges brought by climate change, a number of targeted policies, governance measures, and special actions have been introduced at the national level. Local governments combine the current situation of regional development and regional characteristics, implement and promote relevant governance measures according to local conditions, and gradually build a climate governance system with a complete system and standardized operation (Zheng et al., 2023). As a basic industry supporting the development of the national economy, agriculture is the core foundation for ensuring national food security. The overall production and operation activities are highly sensitive to changes in climatic conditions and policy environment. That is why it is one of the sectors most directly affected by climate policy uncertainty. Climate uncertainty itself can hinder sustainable agricultural production and reduce the stability of agricultural systems (Tan et al., 2025). On the other hand, sound policy guarantees and long-term scientific planning can effectively mitigate the negative impacts of climate change on agricultural development (Abdo et al., 2025; Tang et al., 2025). However, in the process of actual implementation, the formulation and implementation of climate policy will be restricted by multiple conditions, such as the complexity of the climate problem itself, the demand of regional economic development, the level of social collaborative governance, and the overall scientific and technological support ability. Under the superposition of multiple factors, climate policies are prone to dynamic adjustment, which makes the implementation of policies, the intensity of implementation, and the pace of promotion have obvious uncertainties (Ma and Qin, 2022; Chen et al., 2024). This uncertainty in climate policy will further interfere with the layout planning of agricultural production and operation, the input of production factors, the selection of agricultural technology, and the long-term management decisions of agricultural subjects, and ultimately restrict the sustainable development of agriculture. Relevant studies have confirmed that climate policy uncertainty will significantly restrict the process of sustainable agricultural development. This negative effect will be reflected in the decline of land use efficiency, the obstruction of food structure upgrading and the disruption of carbon emission balance (Dai, 2025). The fluctuation and change of the global climate itself also continues to impact the traditional agricultural production mode. The risk resistance of the traditional agricultural system is weak, and the efficiency of the distribution and utilization of various agricultural resources also has obvious shortcomings. Under the dual squeeze of climate risk and climate policy uncertainty, its own development defects are further amplified, and transformation and upgrading have become the inevitable trend of high-quality agricultural development.

In the digital age, the rapid, iterative development of digital technology and artificial intelligence provides a new solution for the agricultural sector to resist climate risks and cushion the impact of policy uncertainty. At present, the agricultural development model is gradually moving away from the traditional extensive form and transforming into a large-scale, modern production model. The agricultural digital-intelligent transformation has thus become the core path for the agricultural field to adapt to the needs of climate governance and the development trend of the digital economy.

The Chinese government attaches great importance to the digital and intelligent upgrading of agriculture, and has successively issued a number of special policy documents to lay a solid policy foundation and provide clear development guidance for the digital-intelligent transformation of agriculture. Over the years, the No.1 Central Documents have repeatedly prioritized high-quality agricultural development and called for innovations in agricultural science and technology and digital technology adoption. The 15th Five-Year Plan (National People's Congress of the People's Republic of China, 2026) strengthens the strategic arrangement for agricultural modernization, offering universal guiding value and practical implications for rural and agricultural development across all provinces in China and building a sound policy environment for agricultural digital-intelligent transformation. The close introduction of relevant policies in the same period also provides clear policy guidance and institutional support for the transformation and development of agriculture: Digital Agriculture and Rural regions (2019–2025) (Ministry of Agriculture and Rural Affairs and Cyberspace Administration of China, 2026) clearly points out that the deep integration of digital technology and agricultural and rural economy is the key direction; The 14th Five-Year Plan for Digital Economic Development (2021) (State Council of the People's Republic of China, 2026a) emphasizes the need to promote the digital transformation of agriculture, innovate and develop smart agriculture, and vigorously improve the level of agricultural digitization; Key policy documents such as the Development of Smart Agriculture and the National Smart Agriculture Action Plan (2024–2028) (Ministry of Agriculture and Rural Affairs of the People's Republic of China, 2026) have been released one after another, which clearly points out the development direction of domestic smart agriculture construction, systematically plans the overall layout and key tasks of China's agricultural digital transformation in the next few years, and provides clear policy guidance and top-level design support for Industry digitalization and intelligent upgrading. In addition, the Opinions on Deepening the Implementation of the “AI+” Initiative (2025) (State Council of the People's Republic of China, 2026b) advocates strengthening the application of artificial intelligence in agricultural production and accelerating its deployment in smart agricultural machinery, agricultural drones, agricultural production management, and other key fields. Relying on the technical advantages of precise monitoring, intelligent decision-making and efficient resource allocation, agricultural digital-intelligent transformation can effectively enhance the adaptability of agricultural production to climate, optimize the effect of agricultural emission reduction, and improve the overall efficiency of emission reduction. However, the intelligent transformation of agricultural digitalization has the remarkable characteristics of a long investment cycle and high sunk cost. Early equipment deployment, technology landing, and system construction all need long-term sustained investment, which means that the steady progress of transformation work is highly dependent on stable policy expectations and perfect institutional guarantees. Once the policy environment fluctuates, it is easy to affect the enthusiasm of all kinds of agricultural operators to participate in digital transformation.

The frequent changes and increasing uncertainty of climate policies may lead to conservative investment decisions of agricultural business entities and digital intelligence service providers, slow down the pace of technology adoption, and deviate from the direction of integration, thus hindering the deep integration of agricultural digital intelligence with climate reduction and climate adaptation, and restricting the agricultural digital-intelligent transformation.

In this context, the impact of climate policy uncertainty on the development of agricultural mathematical intelligence has become an important issue that needs to be studied urgently. From the perspective of theory and practice, Foreign scholars Yeboah et al. (2026) proposed in related research that in the context of increasing climate variability, the promotion of climate-smart agriculture can effectively reduce the loss of agricultural productivity caused by climate change, stabilize agricultural production output, and play an important buffer effect against negative climate shocks. Domestic scholars believe that the digital transformation of agriculture is the result of the coordinated promotion of internal and external factors (Wang and Qu, 2024). It not only relies on the guidance and empowerment of external policies, but also stems from the internal reform and development needs of agriculture itself. The government's policy guidance and coordination are the key to promoting the transformation. Xiao Long and other scholars put forward that it is an important direction of agricultural transformation to integrate intelligent technology on the basis of rural digitalization and promote digitalization to intelligent upgrading relying on artificial intelligence (Xiao and Ma, 2024) Agricultural digital intelligence integrates the dual technical advantages of digitization and intelligence, and can rely on advanced technology to empower traditional agriculture to upgrade. This development model can optimize all kinds of agricultural management decisions from the production link, significantly improve the risk resistance ability of the agricultural production system, offset and buffer all kinds of negative shocks brought by the uncertainty of climate policy to a large extent, and provide solid technical support for the stable development of agriculture.

At present, the academic community has carried out a lot of discussions on climate policy uncertainty, covering macroeconomics, foreign trade, industrial development, enterprise operation, forestry, finance, energy, and other fields. Most of the literature points out that climate policy uncertainty as a whole shows significant negative constraints. From the macro perspective, the continuous rise of climate policy uncertainty will compress social and economic output, inhibit the vitality of private investment, reduce the level of market employment, and promote the rise of commodity prices and increase the cost of residents' consumption. In the field of foreign trade, this factor will obviously restrict the expansion of the foreign trade export scale (Fried et al., 2022; Wang et al., 2025). At the micro-enterprise level, climate policy uncertainty will hinder the process of industrial upgrading, drag down the improvement of total factor productivity, and greatly weaken the investment and operation efficiency of enterprises (Ren et al., 2022; Wang et al., 2024). In addition, in the energy market, climate policy uncertainty will aggravate the extreme risk spillover effect within the industry (Tong et al., 2019; Liu et al., 2023); In the financial sector, it has a significant positive risk spillover effect on domestic financial markets (Hu, 2024). According to the existing research, it can be found that the impact of climate policy uncertainty on the economy has been widely studied by academia. However, there are still obvious shortcomings in the existing research. Existing domestic studies mostly focus on the driving factors of agricultural digital transformation, the effect of technology application and the macroeconomic effect of climate policy. There is a lack of special discussion on the relationship between climate policy uncertainty and agricultural digital-intelligent transformation, and a systematic and rigorous empirical research system has not yet been formed. There is still an obvious research gap in the in-depth analysis of their internal mechanism, heterogeneous impact and transmission path. Based on this research gap, this paper takes the panel data of 31 provinces in China from 2013 to 2022 as the research sample, sets the climate policy uncertainty index (CCPU) and the agricultural digital intelligence index as the core variables, and relies on the two-way fixed-effect model to empirically test the specific mechanism of climate policy uncertainty on the transformation of agricultural digital intelligence, in order to provide an empirical basis and policy reference for the high-quality development of agricultural digitization.

The research contribution of this paper is mainly reflected in four aspects: first, it expands the research boundary of climate policy uncertainty, extends the Research scenario to the field of agriculture, and empirically tests its internal relationship with the transformation of agricultural digital intelligence. Second, build a multi-dimensional agricultural digital intelligence index system, improves the scientificity and accuracy of variable measurement, and provides a methodological reference for subsequent related research. Thirdly, it reveals the inhibitory effect of climate policy uncertainty on agricultural digital-intelligent transformation, and further classifies the regional heterogeneity characteristics based on the level of digital financial development and the division of food-producing areas, providing empirical support for the accurate implementation of policies in the subregion. Fourthly, this paper further explores the regulatory role of farmers' loans and the whole process of agricultural socialized services, confirms that they can effectively alleviate the inhibitory effect of climate policy uncertainty on agricultural digital-intelligent transformation, and provides an empirical basis for improving the relevant relief path.

2 Literature review, theoretical analysis, and research hypotheses

2.1 Literature review

2.1.1 Climate policy uncertainty and the climate policy uncertainty index

Climate policy is to formulate and lead the implementation of measures to deal with climate change at the national level, and to provide a clear overall direction for the formulation of climate policies by local governments. Climate policy uncertainty refers to the fact that the formulation and implementation of climate policy will be affected by multiple factors such as the subject of the system, the time of promulgation, and the strength of the policy, which leads to a lack of clarity and predictability of the climate-related policies of market players, and it is difficult to form an accurate judgment on the future policy direction and the risk of the external environment (Zhang et al., 2026). With the increasingly prominent global climate risks and frequent climate policies in various countries, climate policy uncertainty has become an important factor affecting social and economic development (Liu et al., 2023).

The factors affecting climate policy uncertainty are extensive and complex. From a climate point of view, Bian Lu and colleagues argued that extreme climate is frequent, and the continuous introduction of climate-related policies in various countries will lead to increased climate uncertainty; from the perspective of local governments (Bian et al., 2024). Fei Wei noted that due to factors such as climate conditions and economic entities in various places, different local governments have selective or incomplete implementation in the process of policy implementation, and different degrees of climate policy uncertainty in various places (Fei et al., 2025). At the world level, Chen Guorong and other scholars have concluded that international factors such as increased geopolitical conflicts, the prevalence of trade protectionism, and accelerated global climate governance actions may also further increase the uncertainty of China's climate policy. In addition, Bettarelli et al. started from economic and policy uncertainty and studied that uncertainty significantly improved the rigor of market-oriented climate policies, but had no significant effect on non-market or technical support climate policies, and the intensity of influence was significantly different in different political systems and economic cycles (Bettarelli et al., 2026).

In order to quantitatively study climate policy uncertainty, scholars have constructed the Climate Policy Uncertainty Index (CPU Index) to reflect the degree of climate policy uncertainty. There is much domestic and foreign research on the calculation of the climate policy uncertainty index.

In the existing academic research, the academic community has accumulated a variety of mature and feasible construction methods for the quantitative measurement of the policy uncertainty index. Generally speaking, the mainstream measurement methods can be divided into four categories, namely, Market Volatility Measurement (Sinha, 2016), composite index synthesis (Li et al., 2018), prediction error analysis (Wang and Zheng, 2020) and text mining analysis (Ma et al., 2023; Chen et al., 2024). In view of the research situation of climate policy uncertainty, gavriilidis carries out quantitative research based on news text data, and finally constructs a classic climate policy uncertainty index (CPU index) by counting the frequency of climate policy related keywords in U.S. news reports. Based on the keyword frequency of climate policy uncertainty in 13 mainstream newspapers in China from 2008 to 2023, Chen Guorong et al. used text mining techniques to construct the China Climate Policy Uncertainty Index (CCPU Index) based on the frequency of climate policy uncertainty keywords in 13 major Chinese newspapers from 2008 to 2023, which can more accurately reflect the changes and characteristics of China's climate policy uncertainty. Among them, text analysis is widely used.

In summary, existing literature has explored the socioeconomic consequences of climate policy uncertainty from multiple dimensions, generally verifying its inhibitory or risk-inducing effects on diverse economic activities and market operations. Regarding the configurational mechanism of technology adoption, Zhou employs FSQCA to unpack the adoption logic of artificial intelligence in environmental governance. Their study demonstrates that external institutional pressure does not independently shape technology deployment (Zhou, 2026). Instead, macro-environmental constraints interact with cognitive, technological, and infrastructural factors to jointly determine the uptake of intelligent technologies, offering an analytical framework for investigating how external policy pressures affect technology adoption. Inspired by their findings, this paper argues that climate policy uncertainty, as a macro-level external constraint, exerts no simple linear influence on agricultural digital-intelligent transformation. We therefore incorporate investment decisions of micro agricultural agents, alongside moderating effects from complementary conditions including financial support and agricultural socialized services. Nevertheless, few studies extend this line of inquiry to the agricultural sector, particularly concerning its implications for agricultural digital-intelligent transformation, a core pathway of modern agricultural development.

2.1.2 Agricultural digital-intelligent transformation

Digital-intelligent transformation is the deep integration of digitalization and intelligentization, restructuring production factors and innovating production modes. Digital and intelligent technological innovation has become a major driver of modern agricultural upgrading, stabilizing agricultural production and boosting production efficiency. Scholar Zhao Chun Jiang notes that intelligent agriculture integrates electronic and control technologies into agricultural equipment to improve the refined management of full-process agricultural production (Zhao, 2021).

Digital agriculture focuses on the digital transformation of the whole system, the whole process, and all elements of agricultural production. Agricultural digital intelligence is an organic integration of agricultural digitization and agricultural intelligence, and intelligent agriculture is its core development. At present, the definition of agricultural digital intelligence has not been unified in the academic community. From the perspective of technology application and effect, the agricultural digital-intelligent transformation can be regarded as a transformation process in the process of agricultural production, operation, and management, which can improve production efficiency, optimize resource utilization, and enhance risk resistance by introducing digital and intelligent technologies such as the Internet of Things, big data, and artificial intelligence.

By combining the existing research, some scholars have pointed out that agricultural mathematical intelligence has a positive effect on the level of agricultural green total factor productivity (Hao et al., 2024), the resilience of the agricultural industrial chain supply chain (Li, 2025), and the level of new quality productivity of agricultural enterprises (Su et al., 2025); Li Qiping and other scholars also studied the mechanism of food security from the perspective of agricultural equipment intelligence (Li and Li, 2023). In terms of the development path of agricultural digital intelligence, some scholars have pointed out that new quality productivity (Chen and Guo, 2025) and the first economy (Jiang et al., 2025) can empower the agricultural digital-intelligent transformation, Zuo Li and other scholars have carefully studied the digital intelligence path of agricultural green development under the background of 'double carbon' (Zuo et al., 2024). In addition, agricultural data intelligence has also promoted the improvement of agricultural production efficiency and the reduction of agricultural production management risks.

To sum up, the existing research mostly focuses on economic effects and discusses the enabling effect and development path of agricultural mathematical intelligence from the technology supply side. However, with the deepening of global climate governance and the gradual implementation of China’s “double carbon” goal, climate policy uncertainty has become a crucial macro factor affecting agricultural production decisions. However, only a few studies have systematically discussed its impact on the agricultural digital-intelligent transformation. From the perspective of realistic logic, the frequent adjustment of climate policy will change the market expectation of agricultural operators, reshape the income and risk pattern of the industry, and then profoundly affect the transformation rhythm and promotion effect of agricultural mathematical intelligence. Existing scholars have proposed that multi-party consensus can reduce the ambiguity of policy signals. Coordinating multi-party interest policy fluctuations will adjust the expectations of business entities and change the income risk structure (Zhou and Richardson-Barlow, 2026) Therefore, multi-agent collaborative governance may be able to alleviate policy uncertainty, stabilize market expectations, ensure the implementation of climate governance, hedge the impact of policy fluctuations, and promote the agricultural digital-intelligent transformation.

Accordingly, systematically exploring the driving logic and action path of climate policy uncertainty on agricultural digital-intelligent transformation can not only supplement and improve the theoretical system in the field of on agricultural digital-intelligent transformation, make up for the shortcomings and gaps of existing research, but also provide solid theoretical support and empirical reference for China to formulate differentiated and precise agricultural digital support policies and steadily promote high-quality agricultural development.

2.2 Theoretical analysis and research hypotheses

2.2.1 Climate policy uncertainty and agricultural digital-intelligent transformation

Scholars have pointed out that information and communication technology is the basic supporting condition for the digital-intelligent transformation of agriculture. Relying on precision agriculture mode, intelligent decision-making aids and perfect market information circulation network, this technology can effectively activate all kinds of agricultural production resources and promote the optimal allocation of agricultural production factors (Oyelami et al., 2022; Sethi et al., 2024; Zhao et al., 2026). The agricultural digital-intelligent transformation relies on digital technologies such as the Internet of Things, big data, and artificial intelligence to carry out intelligent system transformation of the whole chain of agricultural services. It has the characteristics of high investment, long cycle, irreversibility, and lagging returns. Its promotion is highly dependent on the stable long-term investment expectations and continuous factor input of agricultural business entities.

According to the real option theory, when there is a high degree of uncertainty in the external environment, the business entity will delay the irreversible investment to wait for more information, thereby inhibiting the current investment behavior (Huang, 2023). Scholars pointed out in the study of economic policy uncertainty restraining enterprise investment that the "waiting" value of investors is positively correlated with the uncertainty of future cash flow of investment objects, which can also explain that climate policy uncertainty may increase the investment uncertainty and risk of agricultural digital intelligence, thus restraining the agricultural digital-intelligent transformation (Li and Yang, 2015; Stroebel and Wurgler, 2021).

It is worth emphasizing that this paper makes a strict distinction between climate physical shocks and climate policy uncertainty. Specifically, climate physical shocks denote exogenous natural risks arising from climatic variations, including temperature changes, precipitation fluctuations, and extreme weather events. In contrast, climate policy uncertainty reflects unstable policy expectations triggered by the formulation, adjustment and revision of relevant rules covering climate governance, low-carbon agricultural development and environmental regulation. As a typical institutional-level uncertainty, it differs fundamentally from natural climate shocks in terms of impact mechanisms. Accordingly, this study mainly focuses on the institutional inhibition effect of climate policy uncertainty on agricultural digital-intelligent transformation. Climate policy uncertainty substantially elevates the uncertainty and risk of investment in agricultural digital intelligence, which functions through two core transmission mechanisms. On the one hand, fluctuations in climate policies intensify the ambiguity of the future policy environment. This will strengthen investors’ perception of climate risks and their risk aversion, trigger greater risk spillover effects in financial markets (Gao et al., 2024), raise the risk premium of agricultural digital investment, and thus drive business operators to delay or cut long-term irreversible investments, including the procurement of smart agricultural equipment and the development of digital service platforms. Meanwhile, frequent policy changes disrupt the allocation of agricultural production factors and the coordination of industrial chains, hinder the steady progress of agricultural digital and intelligent transformation, and reduce the practical effectiveness of transformation practices.

On the other hand, climate policy uncertainty increases the overall financing costs of agriculture-related markets (Bartram et al., 2022; Ruan and Yang, 2025). In line with the bounded rationality theory of market participants (Jacobs, 2015), policy instability further heightens investors’ risk sensitivity, exacerbates financing constraints on agricultural digital intelligence projects, and ultimately impedes the advancement of agricultural digital transformation. Based on this, this paper proposes the following hypothesis:

H1: Climate policy uncertainty will significantly restrain the agricultural digital-intelligent transformation.

Secondly, there are obvious differences in resource endowment, agricultural positioning, industrial structure, and digital financial environment in various regions of China, which makes the impact of climate policy uncertainty on the agricultural digital-intelligent transformation significant regional heterogeneity, and the development level of local digital finance will further adjust this impact effect.

From the perspective of digital finance, the structural differentiated advantages of rural digital credit can effectively ease producers’ credit constraints and improve capital allocation efficiency (Fu and Huang, 2018). Regions with a high level of digital financial development feature diversified agricultural financing channels and efficient capital allocation, which can effectively hedge the adverse impacts of climate policy uncertainty (Min et al., 2026). By contrast, areas with moderately developed digital finance are in a critical stage of industrial and digital transformation. Such regions suffer from unstable capital allocation and limited risk resistance capacity, making them the most vulnerable to policy fluctuations, where the inhibitory effect of policy uncertainty is particularly pronounced. Under such circumstances, the heterogeneous influence of farmers’ risk appetite on agricultural investment decisions becomes more prominent.

Economic entities with strong risk tolerance are more likely to break through credit constraints and allocate capital to high-return agricultural intelligent production activities. In comparison, farmers with low risk preference tend to cut productive investment, which hinders the progress of agricultural intelligent transformation. Furthermore, farmers’ individual risk preferences can either amplify or mitigate the above heterogeneous impacts through their credit access and capital allocation behaviors (Shen et al., 2026).

On the other hand, from the perspective of food functional zoning, significant systematic differences exist in policy dependence and marketization levels across different functional grain zones (Jiang and Zhu, 2015). Major grain-producing regions enjoy stable and predictable policy support, which weakens their sensitivity to individual climate policy shocks. Grain production and marketing balance regions feature diversified industrial structures and strong risk diversification capabilities, thereby suffering relatively mild adverse impacts.

By comparison, major grain-consuming regions present highly market-oriented and diversified agricultural production systems. These regions rely heavily on market factor allocation and external supply systems without rigid policy guarantees. Accordingly, climate policy fluctuations can be easily transmitted to digital agricultural investment through cost shocks, market trade changes, and expectation adjustments.

In short, the inhibitory effect of climate policy uncertainty on agricultural digital-intelligent transformation is not spatially homogeneous but presents obvious structural regional heterogeneity. Accordingly, this paper proposes a targeted, scenario-based heterogeneity hypothesis with clear directional implications:

H2: There is significant regional heterogeneity in the negative inhibitory effect of climate policy uncertainty on agricultural digital-intelligent transformation, which is more significant in regions with moderately developed digital finance and major grain-consuming regions.

2.2.2 Loans to agriculture, forestry, animal husbandry and fishery industries, full-process socialized agricultural services, and agricultural digital-intelligent transformation

Based on the theory of financing constraints and policy intervention theory, external funds and policy support can alleviate the inhibitory effect of uncertainty on investment behavior. As an important financial supply for agricultural production and management entities, loans to agriculture, forestry, animal husbandry and fishery industries can provide financial support and a risk buffer for various agricultural management entities. When the uncertainty of climate policy rises, a sufficient supply of loans for agriculture, forestry, animal husbandry and fishery can alleviate the financing constraints faced by business entities, reduce their concerns about risks, enhance their willingness and ability to invest, and promote the application of agricultural mathematical technology, thus hedging the negative impact of climate policy uncertainty. In addition, studies have shown that agricultural credit and insurance can enhance the resilience of the agricultural economy and enhance the risk resilience, recovery and adaptability of the agricultural system to cope with climate, market and policy shocks (Tendall et al., 2015; Fu and Li, 2026).

The policy of full-process socialized agricultural services takes the development of moderate-scale operations as the core. By cultivating diversified agricultural service markets, promoting mechanized and large-scale production management, and improving the agricultural production and operation system, we can enhance agricultural risk resistance, hedge the negative impact of climate policy uncertainty, play a regulatory role, continuously improve comprehensive agricultural income, and promote sustainable agricultural development. Meanwhile, this policy can also expand the scale of agricultural loans in pilot regions (Wang and Chen, 2026) and further produce an indirect regulatory effect. Based on the above analysis, this paper proposes the following two hypotheses:

H3: Loans to Agriculture, Forestry, Animal Husbandry and Fishery Industries have a positive moderating effect on the relationship between climate policy uncertainty and agricultural digital-intelligent transformation. The higher the level of loans, the weaker the inhibitory effect of climate policy uncertainty on the agricultural digital-intelligent transformation.

H4: Full-process socialized agricultural services have a positive moderating effect on the relationship between climate policy uncertainty and agricultural digital intelligence transformation.

3 Research design

3.1 Sample selection and data sources

This study is based on provincial panel data to carry out empirical analysis, and the sample interval and selection are based on the availability and credibility of data, combined with the key nodes of domestic climate policy, the layout of major agricultural modernization policies, and the characteristics of climate policy fluctuations. The sample period from 2013 to 2022 covers multiple rounds of climate fluctuations and agricultural development stages, which can effectively capture the lag effect of climate policy uncertainty on agricultural digital intelligence transformation. Taking 31 provinces in China as the research sample, this study builds the agricultural digital intelligence transformation index based on the provincial annual data and quantifies the uncertainty of climate policy (CCPU).

This study adopts the sample period from 2013 to 2022 rather than extending the dataset to 2025 for three main reasons. First, this ten-year timeframe fully captures the process of climate policy enforcement and the progression of agricultural digitalization from pilot exploration to large-scale application, effectively addressing the limitations of short time spans and outdated sample data in existing literature. Second, constrained by data accessibility, the indicators of agricultural digitalization and climate policy uncertainty (CCPU) involve multidimensional, subdivided data with an official release lag of one to three years. Provincial-level data from 2023 to 2025 remain insufficiently transparent and incompletely disclosed, failing to meet the continuity and accuracy requirements for panel data quantification and thus compromising empirical reliability. Third, in terms of impact mechanisms, agricultural production features prominent long-cycle attributes. The influences of climate policies on agricultural digital transformation are significantly lagged, as the cumulative effects of policy implementation and technological empowerment require time to materialize. A ten-year research span is capable of capturing the dynamic lag effects of climate policy uncertainty and mitigating estimation biases stemming from short-term data fluctuations. Overall, the time window of 2013–2022 balances policy stability, data availability, and lagged variable characteristics, making it the optimal sample interval for this research.

This paper combines the China Statistical Yearbook, the China Agricultural and Rural Yearbook, the China Agricultural Statistical Yearbook, the China Environmental Statistical Yearbook, the Scientific Research Statistical Bulletin, and the Rural Financial Service Report, relying on the statistical yearbooks of the Ministry of Agriculture and Rural Affairs, the Ministry of Industry and Information Technology, and various provinces, and matching the CAMA database, the China Meteorological Administration, the economic policy uncertainty website, and the CSMAR database, to sort out and obtain the relevant data of agricultural digital intelligence indicators and control variables. The missing data in some provinces are supplemented by the linear interpolation method.

3.2 Variable selection

3.2.1 Dependent variable

The agricultural digital-intelligent transformation index is the explanatory variable of this study. At present, there is no unified standard for the index system of agricultural digital intelligence in academic circles. This paper comprehensively refers to the construction methods of the index system related to agricultural digitization by Li Shengzhu, and other scholars (Li et al., 2024), combined with the characteristics of agricultural digital-intelligent transformation and the availability of data at the provincial level, and constructs the index system of agricultural digital intelligence from the three directions of digital intelligence infrastructure, numerical environment and digital intelligence benefits. This study uses the entropy weight method to calculate the weight coefficients and comprehensive evaluation value of the agricultural digital intelligence index of 31 provinces in China from 2013 to 2022.

This study constructs a comprehensive evaluation index system for the agricultural digital-intelligent transformation, including three core dimensions: the digital-intelligent infrastructure, digital environment and digital-intelligent benefits. Based on the entropy weight calculation method, panel data of 31 provincial-level administrative regions in China from 2013 to 2022 are selected to quantitatively calculate the weight of each sub-index and the comprehensive index of agricultural digital-intelligent transformation in each province. The results of each sub-index and corresponding weight calculation are shown in Table 1.

Table 1. Agricultural digital-intelligent transformation indicator system.

The transformation of agricultural digital intelligence discussed in this paper is the deep integration of traditional digital construction with intelligent development. It differs markedly from the earlier stage of agricultural digitization, which centered on the popularization of information and communication hardware and chiefly aimed at the electronization of agricultural operations and the coverage of basic information infrastructure. By contrast, digital-intelligent transformation expands the application of artificial intelligence technologies—such as intelligent recognition and intelligent decision-making—within agricultural scenarios on top of the established digital foundation.

Drawing on sociotechnical transition theory, Selgas-Cors argues that artificial intelligence technologies should not be treated merely as tools for raising production efficiency. Intelligent-driven changes are deeply embedded in macro-institutional contexts, and the effectiveness of technology diffusion is conditioned by institutional factors such as the policy environment and the interplay among multiple actors (Selgas-Cors, 2025). The technical system and the external institutional environment remain in a state of dynamic interaction; thus, technological development cannot be evaluated in isolation from its institutional background. This conclusion provides a theoretical foundation for our indicator construction and subsequent empirical analysis: the level of agricultural digital-intelligent transformation is not determined solely by hardware facilities. The external institutional environment—for instance, climate policy—shapes the investment expectations and technology adoption decisions of agricultural business entities. The effectiveness of regional digital-intelligent transformation is consequently the joint outcome of technological endowments, the external environment, and institutional conditions.

Based on this, the reasons for the selection of the three dimension indicators in this paper are as follows.

First, the digital-intelligent infrastructure. Digital infrastructure is the material carrier of agricultural digital intelligence development. Whether it is traditional digital tools or future-oriented intelligent transformation, it must rely on basic resources such as networks and hardware. There are significant differences in the basis of agricultural development among provinces in China. Some areas are still in the stage of completing the short board of digital hardware. The completeness of infrastructure directly determines whether all kinds of digital and intelligent technologies can penetrate into agricultural production scenarios. Therefore, infrastructure is taken as the underlying dimension of the whole evaluation system.

Second, the digital-intelligent environment dimension. As the preceding theoretical discussion suggests, technological transformation is not merely a technical matter. The soft environment—encompassing talent reserves, market conditions, and supporting industrial chains—constitutes an indispensable social foundation for the implementation of these technologies. Even with complete hardware facilities, farmers and other agricultural business entities can hardly apply digital and intelligent tools to actual production and operation if the requisite human capital and industrial ecosystem are lacking. This dimension captures the external socio-economic conditions underpinning agricultural digital-intelligent development. It reflects the interaction between technology and society and prevents overestimating a region's true level of digital-intelligent development on the basis of hardware indicators alone.

Third, the digital-intelligent benefit dimension. Digital-intelligent transformation must ultimately take root in agricultural production practice, manifesting in tangible outcomes such as higher production efficiency and optimized business models. This dimension evaluates the level of digital-intelligent development from the real outputs generated by the transformation. In doing so, it objectively reveals the extent to which digital technologies are converted into agricultural production value, while identifying the "heavy hardware investment, light practical application" problem. Consistent with the sociotechnical framework emphasizing that technology should be assessed by its actual implementation outcomes, this dimension provides a balanced measure of agricultural digital-intelligent transformation across provinces.

In summary, this paper combines the three levels of basic hardware conditions, social support environment and actual transformation benefits, taking into account both China’s provincial heterogeneous agricultural foundation and the sociotechnical attributes of agricultural digital-intelligent transformation, to construct a multi-dimensional evaluation index system for provincial agricultural digital-intelligent development.

3.2.2 Key explanatory variable

Climate policy uncertainty index (CCPU) is the core explanatory variable of this paper. The climate policy uncertainty index in this paper adopts the climate policy uncertainty index calculated by Yan Ran Ma and others in 2023. Ma and other scholars use the six mainstream official, authoritative and professional newspaper sources of People's Daily, Guangming Daily, Economic Daily, Global Times, Science and Technology Daily, and China News Agency as data sources (Ma et al., 2023). The selected samples were manually audited, and the MacBERT model of deep learning method was used to automatically identify the text content. The number of news articles containing words related to climate policy uncertainty in a specific period was calculated, and it was compared with the total number of articles in that period. After data standardization processing, China's climate policy uncertainty index was obtained. This paper uses the provincial annual climate uncertainty index constructed by the Macbert deep learning algorithm to measure the uncertainty level of climate policy in each province, and then carries out subsequent empirical analysis.

3.2.3 Control variables and moderating variables

Based on the research and analysis of Li Qiping, Guo Chengcheng, and other scholars on agricultural digital intelligence (Li and Li, 2023; Guo and An, 2025), this paper selects the control variables affecting agricultural digital intelligence (Table 2): agricultural production infrastructure (API), rural electricity consumption efficiency (RECE), rural education (REA), land output efficiency (LOE), total power of agricultural machinery (AMP) and employment in the primary industry (EPI). Loans to agriculture, forestry, animal husbandry and fishery industries (AGIL) and the full-process socialized agricultural services (DID) are selected as the regulatory variables of this study.

Table 2. Description of control and moderating variables.

3.3 Model specification

In order to explore the impact of climate policy uncertainty on agricultural digital-intelligence transformation and consider its lag characteristics, this paper constructs a two-way fixed effect measurement model with climate policy uncertainty as the core explanatory variable and agricultural digital-intelligent transformation as the explained variable as shown in Equation 1. The model is as follows:

ADI i , t = α 0 + α 1 L . CCPU i , t + α n X i , t + μ i + δ t + ε i , t (1)

Where ADI_i,t refers to the agricultural digital-intelligent transformation index, and L.CCPU_i,t indicates the one-period lagged climate policy uncertainty. Specifically, i stands for the province and t denotes the year. X_{i, t} is a set of control variables. α₀ and α₁ represent the constant term and core regression coefficient, while α_n denotes the coefficients of control variables. μ_i and δ_t represent provincial fixed effect and annual fixed effect, respectively. ε_{i, t} is the random disturbance term. Before determining the two-way fixed model, in order to ensure the validity and accuracy of the model setting, the data were standardized. After the data were standardized, the preliminary F test and Hausman test were carried out. The results were significantly passed, indicating that the establishment of a fixed effect model was feasible.

Before the formal empirical analysis, the variable standardization transformation is performed on the full sample panel data to eliminate the interference caused by the dimensional differences of each index, so that the variables of different orders of magnitude can be compared horizontally to optimize the quality of model parameter estimation. After the data preprocessing is completed, the F test and the Hausman test are carried out in turn. The results of the two tests are significant at the common significance level. Based on this, it is reasonable to use the two-way fixed effect to carry out the subsequent regression at the statistical level.

4 Analysis of empirical results

4.1 Descriptive analysis

Table 3 summarizes the descriptive statistical results of all the variables used in the empirical study. There are a total of 310 observations in the explained variable, the provincial agricultural digital-intelligent transformation index ADI. The average value of the sample is 0.204, and the corresponding standard deviation is 0.082. The index interval falls between 0.081 and 0.448. From the extreme value and dispersion degree, it can be seen that the progress of agricultural digitization and intelligent construction in different provinces and regions is uneven.

Table 3. Descriptive statistics.

The single-period lag climate policy uncertainty index (L.CCPU) has a total of 279 effective observation values, with an average value of 2.258, a standard deviation of 0.561, and a range of 0.891 to 3.978, indicating that the uncertainty of climate policies in various regions is significantly different. The control variables also have regional heterogeneity, reflecting the development differences in agricultural infrastructure, rural electricity efficiency, climate risk, rural education, cable facilities and land output efficiency. The above descriptive statistics clearly present the characteristics of sample distribution, providing basic support for the subsequent empirical analysis.

4.2 Analysis of baseline regression results

The above descriptive statistical results intuitively show the sample distribution law of each variable, which proves the validity of the research data, but there are obvious differences in the values of each variable. In order to avoid data bias and ensure the accuracy and reliability of follow-up regression analysis, all variables were standardized in this study.

In order to accurately identify the impact of lagging climate policy uncertainty (L.CCPU) on agricultural digital-intelligent transformation (ADI), this paper uses the two-way fixed effect panel model to control the fixed effect of provinces and years to carry out empirical analysis, and the benchmark regression results are shown in Table 4. Through the gradual introduction of control variables, the study verified that the core variable coefficient has good stability under different model settings.

Table 4. Baseline regression results.

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

The regression results are shown in Table 4, the baseline regression results are obtained using the lagged climate policy uncertainty (L.CCPU) variable without standardization and the regression coefficient of L.CCPU in each column is significantly negative. Column (1) shows the benchmark regression results without control variables, and the L.CCPU coefficient is −0.0515, which is significant at the level of 5%, which preliminarily verifies the research hypothesis H₁, that is, the uncertainty of climate policy significantly inhibits the transformation of agricultural digital intelligence. Column (2) two control variables, agricultural production infrastructure (API), rural electricity consumption efficiency (RECE), were further introduced to carry out more perfect regression analysis.

The L.CCPU coefficient fell back to −0.0509, and the symbol and significance level of the coefficient did not change, indicating that after controlling for the interference of the above factors, there was no substantial deviation in the benchmark empirical conclusions, which initially confirmed that the benchmark regression results were robust and reliable. Column (3) further includes the rural educational attainment (REA), land output efficiency (LOE), employment in the primary sector (EPI), and Agricultural machinery total power (AMP). The coefficient of L.CCPU is−0.0608 and remains significant at the 1% level. Therefore, Column (3) is used as the benchmark specification.

In terms of economic meaning, the result indicates that higher climate policy uncertainty is associated with a lower level of agricultural digital-intelligent transformation. Based on the coefficient in Column (3), a one-unit increase in L.CCPU is associated with a 0.0608-unit decrease in ADI, holding other factors constant. Based on the original standard deviation of L.CCPU, the specific economic meaning is explained. For every standard deviation (0.561) added to L.CCPU, the level of agricultural digital-intelligent transformation (ADI) will decrease by about 0.034 units. All variables in this paper are standardized before regression, so that the measurement results can truly reflect the actual role of the core variables and improve the interpretation of the empirical results.

This model includes the provincial and annual two-way fixed effects, which can effectively avoid the interference of provincial non time varying characteristics and annual macro impacts on the regression results. From the point of view of model fitting degree, the adjusted R-squared of the initial model is 0.953, which increases to 0.962 when it reaches the third column, and the fitting effect continues to optimize, indicating that the new control variables can effectively enhance the model's ability to explain the changes in the transformation of agricultural digitalization and intellectualization. The overall regression results show that climate policy uncertainty has a significant negative inhibitory effect on the transformation of agricultural digitalization and intellectualization, and this core conclusion remains stable and reliable in the process of gradually introducing various covariates.

4.3 Endogeneity tests

4.3.1 Instrumental variables method

There is a potential endogenous relationship between lagging climate policy uncertainty (L. CCPU) and agricultural digital-intelligent transformation index (ADI). Climate policy uncertainty will distort the expectations and investment decisions of agricultural operators and hinder the transformation toward agricultural digitalization and intelligence; conversely, the development of agricultural digitalization and intelligence may also affect the formulation and adjustment of climate policy. To address the endogenous bias arising from two-way causality and ensure the robustness of the empirical results, this study employs the instrumental variable method.

Climate policy adjustments and associated uncertainty are closely tied to large-scale climate anomalies. As a major large-scale ocean-atmosphere oscillation signal, the El Niño-Southern Oscillation (ENSO) can alter regional temperature and precipitation patterns via atmospheric teleconnection, generating regionally heterogeneous impacts on crop yields and agricultural disaster risks across China (Li et al., 2019; Sun et al., 2020). Li et al. (2019) document that ENSO-induced climate anomalies significantly change agricultural production conditions and reshape farmers’ production decisions, which raises public and governmental awareness of climate risks and facilitates adjustments to climate-related regulatory and policy arrangements, further triggering fluctuations in climate policy uncertainty. Following the IV construction strategy proposed by Wang et al. (2025), this study adopts the Niño index to construct the instrumental variable. The two-stage IV regression specification is shown in Equation 2. As a natural climatic phenomenon originating from tropical Pacific sea surface temperature anomalies, ENSO is not affected by China’s agricultural digital-intelligent transformation or domestic climate (Wang et al., 2025). Policy adjustments, satisfying the exogeneity requirement. Meanwhile, ENSO-induced climate anomalies can affect climate policy uncertainty, which meets the relevance condition for a valid instrument. To satisfy the exclusion restriction, we further control provincial agricultural disaster conditions and crop yield levels in the regression, so as to rule out the direct impact of ENSO on agricultural production and isolate its effect channel operating through climate policy uncertainty.

Because climate policy uncertainty is influenced to some extent by climate change, this study draws on the instrumental-variable construction method used by Wang et al. and selects the Niño index as the basis for constructing the instrument.

The El Niño index is a global climate observation index that is objectively related to the uncertainty of climate policy. From the perspective of exogenous constraints and the exclusive constraints of instrumental variables, the index is dominated by the natural evolution of the atmosphere and ocean. The progress of agricultural mathematical intelligence in domestic provinces cannot reverse affect the index value, and there is no possibility of reverse interference by the explained variables. From the analysis of the transmission path, the El Niño climate phenomenon forced the revision of domestic climate governance policies through extreme meteorological disasters, which in turn caused fluctuations in climate policy uncertainty. Relying on this chain path, it indirectly affects agricultural production activities and does not directly act on the process of agricultural digitization and intelligent construction in various provinces. In addition, the Nino index is based on the calculation of global climate change and has strong exogenous attributes for China's provincial research samples, which is in line with the exogenous premise required for the selection of instrumental variables. Meanwhile, the model in this paper further incorporates the provincial fixed effect, the year fixed effect and a set of regional-level control variables, which further rules out the possibility that the instrumental variable influences the explained variable through other unobserved factors, thereby satisfying the instrumental variable assumptions. Specifically, this paper selects the central Pacific El Niño index (NiñOCP) and combines the longitude data of each province to construct the instrumental variables.

IV = NinoCP × provincial longitude (2)

Table 5 shows the regression test results of the two-stage least squares (2SLS) method.

Table 5. Endogeneity test results.

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

Column (1) of Table 5 reports the first-stage regression results of the instrumental variables. The regression coefficient of the instrumental variable is 0.0658, which is significant at the level of 1% statistical significance, indicating that there is a significant positive correlation between the instrumental variable and L.CCPU, which meets the correlation requirements of the instrumental variable. Column (2) of Table 5 reports the two-stage regression results with ADI as the explained variable. The L.CCPU coefficient is −0.2770, which is significant at the 1% level. This shows that after properly handling the endogenous problems, the uncertainty of climate policy still significantly inhibits the transformation of agricultural digital intelligence.

The diagnostic test results verify the effectiveness of the tool variables in this paper. Among them, the weak instrumental variable test statistic is 22.18, p value is 0.000, which is significant at the level of 1%, and there is no weak instrumental variable problem. The weak identification metric reaches 20.125, substantially exceeding the conventional threshold of 10, thereby confirming the absence of weak bias in the instrumental variables. Given that this model employs exact identification, it is not feasible to utilize overidentification tests to verify the exogeneity of the instrumental variables. Combined with the theoretical analysis of the exogenous and exclusive mechanism of instrumental variables, it can be determined that the instrumental variables selected in this paper are reasonable at both the theoretical and statistical levels, and effectively overcome the endogenous problem of the model.

In conclusion, the estimation results of instrumental variables further confirm that the uncertainty of climate policy will significantly restrict the promotion of agricultural digital-intelligence transformation.

In order to further verify that the instrumental variables meet the exclusion constraints, referring to the research method of from Scientific and Technological Talents, Technology Diffusion and Economic Transformation: The Influence of Overseas Students on the Industrial Development of Modern China (Liu et al., 2025), the sub-samples of the core explanatory variable L.CCPU in the top 25 quantiles were selected for regression test. In this sub-sample, the value level of L.CCPU is low, which means that the indirect transmission channel of climate policy uncertainty through L.CCPU is greatly weakened. If the instrumental variable can only act on the agricultural digital-intelligent transformation through L.CCPU, the direct impact of IV on ADI should not be statistically significant in this subsample.

The regression results of Table 6 show that the estimated coefficient of instrumental variable IV is-0.0013 under the first 25 quantile samples of L.CCPU, which does not pass the statistical significance test at the 10% level. This result offers preliminary support for the exclusion restriction.

Table 6. Exclusive test of instrumental variables.

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

4.3.2 Lagged control variables

Drawing on Liu and Chen (2025), this study further conducts a regression in which the control variables are lagged by one period to mitigate potential bidirectional causality and omitted variable bias.(Liu and Chen, 2025) Table 7 reports the corresponding regression results of this study.

Table 7. Regression results with lagged control variables.

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

According to the regression results in Table, the L.CCPU coefficient is −0.0512, which is significant at the 5% level. This result is consistent with the benchmark regression conclusion, which shows that the uncertainty of climate policy still significantly hinders the agricultural digital-intelligent transformation under the setting of various explanatory variables and one-stage lag control variables. In addition, the coefficients of lag control variables L.RECE and L.EPI were significantly negative and passed the significance test of 1%.The L.REA coefficient was significantly positive, meeting the 5% significance requirement, while the L.API and L.LOE also showed positive significant characteristics and was established at the 1% statistical level. The coefficient of L.AMP is not statistically significant. Overall, these results provide further support for the robustness of the baseline conclusion.

4.4 Propensity score matching

This study applies propensity score matching (PSM). The core explanatory variable, L.CCPU, is divided according to the global mean into a high climate policy uncertainty group (L.CCPU ≥ global mean, treat = 1) and a low climate policy uncertainty group (L.CCPU < global mean, treat = 0). This paper employs the conventional 1-to-1 nearest-neighbor matching to perform regression analysis on the matched samples.

Table 8 reports the covariate balance test before and after matching (U = unmatched, M = matched). Table 9 presents the overall balance test. The results show that covariate balance is substantially improved after sample matching. The mean standardized bias declines from 17.4% to 7.5%, and the median standardized bias decreases from 18.4% to 7.1%. The pseudo-R² drops from 0.043 to 0.007, and the LR chi-square test becomes statistically insignificant (p = 0.756). The overall standardized bias (B = 20.3 < 25%)) and the variance ratio (R = 0.95) lie within the recommended range of [0.5,2].

Table 8. Balance tests before and after matching.

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

Table 9. Overall balance test.

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

The standardized deviation of most covariates falls below 10%, satisfying the conventional criterion for PSM balance. The p-values for all covariates exceed 0.05, suggesting no statistically significant mean differences between the treated and control groups after matching. Collectively, the covariate distributions of the two groups are well balanced after matching, and the sample selection bias arising from observable characteristics is effectively mitigated.

Figure 1 presents the kernel probability density plots before and after matching. The right panel shows a higher degree of overlap between the treatment and control groups after matching, further indicating that the PSM procedure improves comparability between the two groups.

Table 10 reports the regression results for the matched sample. The coefficient of L.CCPU is −0.0105 and is significant at the 10% level. This result is consistent with the baseline conclusion, although the magnitude and significance level are weaker than those in the baseline regression. Overall, the matched-sample results further support the conclusion that climate policy uncertainty inhibits agricultural digital-intelligent transformation.

Table 10. Regression results for the matched sample.

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

On the whole, after using a series of methods to deal with endogeneity problems, we can still conclude that climate policy uncertainty inhibits the development of agricultural digital-intelligent transformation, which shows that the estimation results of this paper are robust.

4.5 Robustness tests

This paper carries out a number of robustness tests to verify the reliability of the benchmark regression results, and effectively avoid the bias problem in the process of empirical estimation. The test methods used in this study include two-way elimination, regression analysis based on Provincial cluster robust criteria, addition of control variables, adjustment of research samples, replacement of core explanatory variables, and replacement of the original lag value with the current value of core independent variables. The regression results of all robustness tests are unified in Table 11.

Table 11. Additional control variables.

4.5.1 Two-sided trimming

Considering the moderate sample size in this study, excessive sample elimination would lead to the loss of valid observations. Therefore, this paper adopts a bilateral 1% winsorization for L.CCPU to eliminate the interference of extreme outliers on empirical results. The results in Column (1) of Table 12 show that the coefficient of the core variable is −0.0608, which is significantly negative at the 1% level. Consistent with the benchmark regression findings, this result confirms that the inhibitory effect of climate policy uncertainty on agricultural digital-intelligent transformation is not driven by extreme sample values.

Table 12. Robustness test results.

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

4.5.2 Clustered robust standard errors

To test the robustness of the benchmark regression results, this paper adopts province-level clustered robust standard errors in the regression analysis to eliminate the interference of heteroskedasticity and ensure the reliability of the estimated results. As shown in Column (2) of Table 12, the coefficient of L.CCPU remains −0.0608 and is significant at the 1% level, confirming that the baseline conclusion remains stable after adjusting the standard errors.

4.5.3 Adding control variables

To rule out possible bias induced by unaccounted confounding factors, four newly supplemented covariates are introduced into the regression, namely rural wired infrastructure (QWI), computer ownership among rural residents (COMR), provincial average precipitation (PAP), and Climate risk index(CPRI). Detailed definition of these indicators is listed in Table 11. From Column (3) of Table 12 L.CCPU is calculated at −0.0731 and is significant at the 1% cutoff following the inclusion of extra controls. It can be concluded that the restraining influence of climate policy uncertainty on agricultural digital-intelligent transformation remains statistically significant after accounting for more contextual factors.

4.5.4 Adjusting the sample

Considering the potential effects of the COVID-19 pandemic on economic activity and climate policy implementation, this study excludes observations from 2020 and 2021 and re-estimates the model. As shown in Column (4) of Table 12, the coefficient of L.CCPU is −0.0651 and is significant at the 1% level. This result suggests that the negative effect of climate policy uncertainty on agricultural digital-intelligent transformation remains significant after excluding the pandemic period.

4.5.5 Replacing the dependent variable

We used the entropy method to calculate new digitalization and intelligentization indicators for agriculture to replace the original ones.The results are shown in Column (5) of Table 12. The regression coefficient is −0.0840 and is significant at the 1% level, further confirming the inhibitory effect of climate policy uncertainty on agricultural digital-intelligent transformation.

4.5.6 Using the current-period core explanatory variable

In the long run, the impact of climate policy on agricultural digital intelligence often has a cyclical lag, so the uncertainty of climate policy lagging behind one period is selected as the core variable in the benchmark regression of this paper, but it may still have a certain impact in the short term. The core variable of the current period is introduced into the robustness test to supplement the test of short-term immediate impact. According to the regression results of Ccolumn (6) in Table 12, the coefficient of CCPU in the current period is −0.0396, which is significant at the statistical level of 5%. The results show that the negative constraint of climate policy uncertainty on the agriculture digital-intelligent transformation has both long-term and short-term effects, and the long-term inhibition effect is more prominent, which further proves the robustness and reliability of the benchmark regression conclusion.

In this paper, the robustness test is carried out by eliminating extreme values, adjusting cluster standard errors, expanding control variables, changing sample intervals, reconstructing explanatory variable indicators, and correcting lags of core variables. Under different model settings, climate policy uncertainty significantly inhibits the agricultural digital-intelligent transformation, and there are sustained negative effects, indicating that the benchmark regression results have good robustness.

5 Heterogeneity analysis

In order to explore the differentiated impact and applicable boundary of climate policy uncertainty on the construction of agricultural digital intelligence, this paper carries out multi-dimensional heterogeneous regression analysis.To avoid the empirical errors caused by subjective grouping, the sample split strictly refers to the regional development planning defined by the state and the authoritative industry classification criteria, and selects the two dimensions of regional digital financial development degree and food function zoning to complete the grouping test. Combined with the data of Table 13, it can be seen that the inhibitory effect of climate policy uncertainty is only statistically significant in provinces with medium-level digital financial development; based on the regression results of Table 13, it can be found that the negative impact only significantly affects the major grain-consuming areas, while no significant association is observed in the other grain-production groups.

Table 13. Heterogeneity by level of digital finance development.

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

5.1 Heterogeneity analysis based on the level of digital finance development

This part takes the development degree of digital finance as the starting point, carries out the heterogeneity research at the regional level, and discusses the different effects of climate policy uncertainty on the agriculture digital-intelligent transformation under different digital financial development. The sample is classified according to the 2020 Peking University Digital Inclusive Finance Index rankings (Guo et al., 2020), and divides all the sample provinces into three categories: advanced digital finance developed, moderately developed and underdeveloped regions.

Among them, the developed areas of digital finance include Shanghai, Beijing, Zhejiang, Jiangsu, Fujian, Guangdong, Tianjin, Hubei, Anhui and Shandong; Moderately developed areas cover Chongqing, Hainan, Shaanxi, Henan, Jiangxi, Sichuan, Hunan, Liaoning, Shanxi and Guangxi; Underdeveloped areas include Hebei, Yunnan, Tibet, Ningxia, Inner Mongolia, Qinghai, Jilin, Guizhou, Heilongjiang and Gansu.

Grouped regression results are presented in Table 13. For provinces with a medium level of digital finance development, the L.CCPU coefficient is −0.0770, significant at the 10% level, suggesting that climate policy uncertainty significantly impedes the digital and intelligent transformation of local agriculture. In contrast, the coefficients are −0.0403 for advanced digital finance regions and 0.0130 for underdeveloped ones; neither estimate is statistically significant. These findings confirm notable heterogeneity in the impact of climate policy uncertainty on digital finance development across regions.

This paper sorts out its internal mechanism matching empirical findings. Regions with advanced digital finance own sound financial systems and sufficient factor resources to offset adverse shocks from climate policy uncertainty. Areas lagging in digital finance lack financial support and slow agricultural digital upgrading, weakening the correlation between core variables and yielding insignificant regression outcomes. Regions with medium digital finance development possess basic financial facilities yet poor risk resistance. Shifting climate policies disrupt farmers’ expectations and digital investment plans, greatly restraining local agricultural digital-intelligent transformation.

In summary, the development level of regional digital finance influences the degree to which climate policy uncertainty affects the digital and intelligent transformation of agriculture.

5.1.1 Heterogeneity analysis based on grain production layout

Combined with the strategic framework of national food security, we rely on the grain functional areas designated by the Ministry of agriculture and rural areas to carry out empirical tests of heterogeneity. According to the official planning and unified classification standards, 31 provinces in China are divided into major grain-producing areas, grain production–marketing balanced areas, and major grain-consuming areas. Regional classification comprehensively refers to quantitative indicators such as total grain output, per capita grain holdings and commercial grain reserves, which can truly reflect the actual situation of agricultural production in various places.

Thirteen provinces (autonomous regions) in Hebei, Inner Mongolia, Jilin, Heilongjiang, Liaoning, Jiangsu, Anhui, Jiangxi, Shandong, Henan, Hubei, Hunan and Sichuan were identified as the major grain-producing areas; Seven provinces and municipalities with large grain inflows are defined as major grain-consuming areas: Beijing, Tianjin, Shanghai, Zhejiang, Fujian, Guangdong, and Hainan;11 provinces (autonomous regions and municipalities directly under the Central Government) including Shanxi, Guangxi, Guizhou, Yunnan, Chongqing, Xizang, Shaanxi, Gansu, Qinghai, Ningxia and Xinjiang are grain production–marketing balanced areas.

Table 14 presents an analysis of heterogeneity across grain-producing regions. The results indicate that climate policy uncertainty yields a significant negative effect only in major grain-consuming areas, with significance at the 5% level. No significant results were observed in major grain-producing regions or grain production–marketing balanced regions, suggesting that climate policy uncertainty primarily inhibits the agriculture digital-intelligent transformation in major grain-consuming areas.

Table 14. Heterogeneity by grain production layout.

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

Because the industrial structure of the major grain-consuming areas is dominated by the secondary and tertiary industries, the proportion of agriculture in the regional economy is low, and the dependence of local grain supply on foreign countries is high. The agricultural digital-intelligent transformation is highly dependent on the external policy environment and highly sensitive to policy fluctuations. The rising uncertainty of climate policy will weaken the long-term investment confidence of agricultural operators, restrain the investment and adoption of digital intelligent technology, and then slow down the overall transformation of agriculture. As the core guarantee area of national food security, agricultural development in the main grain-producing areas has been in line with the policy orientation for a long time, and continues to enjoy stable agricultural support policies. These sustained policy advantages effectively hedge the risk of uncertainty brought about by short-term climate policy changes and offset adverse policy shocks, so climate policy uncertainty has not significantly restrained the transformation of agricultural digitalization and intelligence in major grain producing areas.

Although a balanced production and marketing region needs to consider the fluctuation of local food supply and market policy, the foundation of agricultural mathematical intelligence is weak, and the transmission effect of policy uncertainty is difficult to fully emerge, so it has not had a significant impact.

The above empirical results show that climate policy uncertainty has a stronger inhibitory effect on the transformation of agricultural digitalization and intelligence in major grain-consuming areas. This finding shows that the core grain producing areas assume the responsibility of national food security, and in order to continue to promote the digital and intelligent transformation of regional agriculture, we must establish a stable and orderly climate policy environment.

Due to the limited sample size in this study, grouping by grain functional zones leads to substantial sample size disparities across groups. To ensure the robustness of the heterogeneity analysis, we conduct bootstrap tests for between-group coefficient differences. The bootstrap is implemented with 1,000 replications, and the test results are presented in Table 15.

Table 15. Test of coefficient difference between groups.

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

The coefficient difference between major grain-producing areas and major grain-consuming areas is significant at the 10% statistical level, and the coefficient difference between major grain-consuming areas and grain production-marketing balanced areas is significant at the 5% statistical level. In contrast, the coefficient difference between major grain-producing areas and grain production-marketing balanced areas fails to pass the significance test. These findings demonstrate that the impact of climate policy uncertainty on agricultural digital-intelligent transformation exhibits pronounced heterogeneity across grain functional zones. The inhibitory effect in major grain-consuming areas is statistically distinct from that in the other two types of regions.

6 Further analysis

To further study the relationship between climate policy uncertainty and agricultural digital-intelligent transformation, this part adds Loans to agriculture, forestry, animal husbandry and fishery industries (AGIL) and the pilot policy for full-process socialized agricultural services (DID) as regulatory variables, and constructs an interactive term to verify its regulatory effect on the relationship between climate policy uncertainty and agricultural digital intelligence transformation. The moderating effect is estimated using Equation 3. Construct the following regression model:

ADI i , t = β 0 + β 1 L . CCPU i , t + β 2 mod i , t + β 3 ( mod i , t × L . CCPU i , t ) + γ X i , t + μ i + λ t + ε i , t (3)

Mod_i,t denotes the moderating indicators, including Loans to agriculture, forestry, animal husbandry and fishery industries (AGIL) and the policy DID variable. X_i,t represents the set of control covariates. μ_i and λ_t denote provincial fixed effects and year fixed effects, respectively, and ε_i,t refers to the stochastic error term of the model. All empirical results of the moderation analysis are reported in Table 16.

Table 16. Moderation effect results.

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

6.1 Moderating effect of loans to agriculture, forestry, animal husbandry and fishery industries

Combined with the regression results of the moderating effect of Table 16, it can be seen that loans to agriculture, forestry, animal husbandry and fishery industries (AGIL) can play a significant regulatory role. Column (1) is the benchmark regression result, and the L.CCPU coefficient is −0.0608, which passes the 1% significance test. Column (2) after the interaction item between agriculture, forestry, animal husbandry and fishery industries and climate policy uncertainty is included, the L.CCPU coefficient is still significantly negative (− 0.0588, 1% level), and the AGIL × L.CCPU coefficient of the interaction item is 0.0255, which is significant at 10% level.

The significant positive interaction coefficient shows that loans to agriculture, forestry, animal husbandry and fishery industries have a positive regulatory effect on the relationship between climate policy uncertainty and agricultural digital-intelligence transformation. The higher the level of loans to agriculture, forestry, animal husbandry and fishery industries, the weaker the inhibitory effect of climate policy uncertainty on the agricultural digital-intelligent transformation. This shows that loans to agriculture, forestry, animal husbandry and fishery industries can alleviate the main financing constraints, ensure the continuous investment of digital intelligence technology, and effectively hedge the negative impact of climate policy uncertainty.

6.2 Moderating effect of full-process socialized agricultural services

This study employs the DID analytical framework for policy evaluation, following the research approach of Yuan et al. Based on the Guiding Opinions on the Implementation of the Pilot Program for Full-Process Agricultural Production Socialized Services issued by the Ministry of Finance in 2013, the policy variable is constructed as DID = Treat × Post. The first batch of pilot provinces in 2013 included Hebei, Jiangsu, Anhui, Jiangxi, Shandong, Henan, Hubei, and Hunan. Inner Mongolia, Jilin, Heilongjiang, Liaoning, Zhejiang, Fujian, Guangxi, Chongqing, and Sichuan were added as pilot provinces in 2015. Thus, the final sample includes 17 pilot regions.

Coulmn (3) of Table 16 shows the regression results of the adjustment effect of the full-process socialized agricultural services (DID). After introducing the interaction term of DID and L.CCPU into the model, the regression coefficient of the core variable L.CCPU is −0.0911 which is significant at the statistical level of 1%. At the same time, the coefficient of the interaction term DID × L.CCPU was 0.0590, which passed the 5% significance test. The significant positive coefficient shows that the whole-process socialized service of agriculture can effectively weaken the negative impact of climate policy fluctuations on the digital-intelligent transformation of agriculture and play a significant positive buffering role.

The research results show that loans to agriculture, forestry, animal husbandry and fishery industries as a key financial tool in rural regions, can provide stable financial security and risk resistance support for agricultural production entities. Under the background of increasing climate policy uncertainty, sufficient credit for farmers can effectively break through the financial constraints of agricultural production, reduce the risk concerns of business entities, enhance their enthusiasm for investing in digital and intelligent agricultural technology, and effectively offset the inhibitory effect of policy uncertainty on the digital-intelligent transformation of agriculture.

The full-process socialized agricultural services can hedge the impact of climate policy fluctuations and help agricultural transformation. This kind of service cultivates appropriate scale operation, enriches the supply of agricultural services, promotes mechanization, intensification and green production mode, and continues to promote the construction of agricultural modernization. A sound public service system can enhance agricultural risk resistance, weaken the disturbance caused by policy changes, and maintain the stable development of agriculture.

In conclusion, loans to agriculture, forestry, animal husbandry and fishery industries and the full-process socialized agricultural services have obvious regulatory effects. Abundant rural financial supply and standardized agricultural socialized services can alleviate the inhibitory effect of climate policy uncertainty on the agricultural digital-intelligent transformation.

7 Conclusion

7.1 Discussion

This paper empirically examines the association between climate policy uncertainty and agricultural digital-intelligent transformation using provincial-level panel data covering 31 Chinese provinces from 2013 to 2022. Drawing on a self-constructed agricultural digital-intelligent index (ADI), this study applies two-way fixed-effects estimation, instrumental-variable regression, propensity-score matching, heterogeneity and moderation tests to unpack this relationship.

The baseline results document that higher climate policy uncertainty is negatively associated with agricultural digital-intelligent transformation. Fluctuating climate policy environments shake the expectation of agricultural operators and raise perceived investment risks. Since agricultural digital-intelligent upgrading requires long-lived, sunk-cost investment in machinery, digital platforms, data infrastructure and technical services, unstable policy expectations discourage market entities from committing resources to digital-intelligent projects. This finding aligns with real-option logic: heightened uncertainty increases the value of “waiting-to-invest” and suppresses irreversible long-term investment in agriculture.

Heterogeneity analyses reveal that this negative association does not hold uniformly across regions. The correlation is statistically significant only in provinces with moderately developed digital finance and major grain-consuming areas. Regions with advanced digital finance possess sufficient capital and risk-buffering capacity to absorb policy-related shocks. By contrast, regions with under-developed digital finance lack the foundational conditions for large-scale agricultural digital-intelligent investment, so climate-policy-related variation exerts limited observable influence. For moderately developed digital-finance regions, basic digital infrastructure exists while risk-bearing capacity remains weak; local agricultural actors are highly sensitive to policy swings, hence digital-intelligent investment is notably restrained. In terms of grain functional zoning, major grain-consuming areas rely heavily on external market and policy signals for agricultural development, making local digital-intelligent investment vulnerable to policy turbulence, whereas major grain-producing zones benefit from persistent, stable agricultural support policies that partly offset uncertainty-driven adverse effects. Grain production-marketing balanced areas show no significant correlation, due to their diversified agricultural structure and weak transmission of policy-uncertainty shocks.

Further moderation tests confirm that agricultural-related loans and full-process socialized agricultural services can mitigate the adverse correlation between climate policy uncertainty and agricultural digital-intelligent transformation. Sufficient agricultural credit eases financing constraints for farm operators and sustains digital-intelligent investment even amid policy swings. The full-process socialized agricultural-service pilot optimizes production organization, lowers individual operational risks, and forms an institutional buffer against policy volatility. Taken together, stable policy expectation, rural financial supply and agricultural service systems jointly shape the pace of agricultural digital-intelligent transition.

7.2 Policy implications

Derived strictly from the above empirical findings rather than general-purpose arguments, targeted policy recommendations are proposed as follows: First, for provinces with moderately-developed digital finance, local governments should prioritize consolidating digital-agriculture hardware facilities and launch special agriculture-oriented credit products. It is also necessary to strengthen the transparency and communication of climate governance rules to stabilize long-term investment expectations of agricultural market participants.

Second, implement differentiated expectation-management policies targeting major grain-consuming areas. Considering their high sensitivity to climate policy swings, authorities need to publish clear implementation timelines and assessment criteria for climate-related measures, and allocate special resources for local agricultural digital-intelligent infrastructure construction.

Third, give full play to the risk-buffering function of agriculture-related credit. Financial institutions can develop credit products tailored to smart-farming equipment and digital-agriculture projects. Meanwhile, improve risk-sharing mechanisms to cut the actual financing cost of agricultural digital-intelligent investment for operating entities.

Fourth, expand the implementation scope of full-process socialized agricultural services. Popularize mature pilot experience across more regions. By providing outsourced professional services, lower the threshold for small- and medium-scale agricultural operators to adopt digital-intelligent technologies.

Fifth, pay attention to multi-policy coordination and build a long-term guarantee mechanism. The effect of a single policy is easily affected by the fluctuation of policy uncertainty, and multi-policy coordination can form a joint force. A collaborative evaluation mechanism of climate policy, financial policy and agricultural industrial policy can be established to avoid policy conflicts. Improve the policy information sharing and interpretation mechanism, timely transmit authoritative information, and reserve policies to adapt to the transition period; improve the marketization mechanism and enhance the coping ability and long-term confidence of business entities.

7.3 Theoretical and practical contributions

This study offers several extensions to existing research. Theoretically, this paper introduces the macro-institutional factor of climate policy uncertainty into the analytical framework for agricultural digital-intelligent transformation. From a sociotechnical-transition perspective, agricultural AI-driven upgrading is regarded as more than a pure technical output; its development trajectory is jointly shaped by external institutional contexts such as climate-policy regimes. This paper adds empirical evidence regarding how policy uncertainty transmits to long-term investment behavior within agriculture. It also tests the risk-buffering effect of multi-agent supporting systems, revealing the function of multi-party coordination under institutional volatility, and expands analytical perspectives at the intersection of institutional environment and agricultural technological transition.

Practically, the multi-dimensional comprehensive evaluation index constructed in this paper provides a quantifiable tool for measuring provincial agricultural digital-intelligent development. The identified regional heterogeneous effects deliver empirical references for formulating location-specific, differentiated policies under the national “dual-carbon” goals and digital-agriculture strategy.

7.4 Limitations and future research directions

This study also has several noticeable limitations.

First, the Agricultural Digital-Intelligence Transformation Index (ADI) constructed in this study has certain inherent limitations, which are primarily restricted by the availability of provincial public statistical data. At present, there is no specialized and refined statistical caliber for artificial intelligence in agriculture. Existing public indicators cannot accurately isolate pure agricultural AI inputs and inevitably cover general digital infrastructure construction and partial agricultural mechanization content, which weakens the pertinence of the current ADI measurement. Essentially, numerous refined provincial-level indicators that can directly reflect the investment, application and promotion of artificial intelligence in agriculture have not been systematically counted and publicly released. This makes it impossible for this study to adopt ideal targeted variables for index construction, and the current indicators can only provide a limited characterization of agricultural digital intelligence transformation. In future research, we will adopt micro-level data of agricultural enterprises and farms to further improve and refine the evaluation system, develop more targeted indicators, and thereby accurately depict the process of agricultural digital-intelligence transformation.

Second, the sample ends in 2022; newer post-2022 data cannot be incorporated due to data accessibility limits. A small number of provincial-level statistical indicators, policy-related statistics and climate-relevant datasets are not yet publicly released by official authorities. Some targeted indicators that would enrich our empirical analysis are therefore not adopted in the current study. Subsequent research can update the sample once official statistics become available and re-examine the robustness of the documented associations.

Third, the instrumental variable constructed as NiñoCP × provincial longitude cannot fully eliminate all potential direct climate-shock channels, even though a subsample exclusive test is performed. Readers should interpret IV results with caution; the paper identifies correlative evidence with mitigated endogeneity concerns rather than definitive causal proof. Future research may explore alternative identification strategies to further disentangle climate-shock effects from climate-policy-uncertainty effects.

Fourth, this study also has limitations regarding model fit. The adjusted (R2) values in most two-way fixed-effects specifications exceed 0.95. Such high goodness-of-fit statistics should not be interpreted as strong evidence that the core explanatory variable can explain most variations in agricultural digital-intelligent transformation. Instead, the large adjusted (R2) is primarily driven by the two-way fixed effects, which absorb persistent province-specific characteristics and common time trends across years. While the core findings remain qualitatively stable across alternative empirical settings, readers should be cautious when interpreting the magnitude of model fit. This high R2 reminds us that provincial panel data with strong unit and time trends tend to inflate the overall explanatory power of the regression model.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fsufs.2026.1909691/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Matthew Ayamga, Wageningen University and Research, Netherlands

Reviewed by: Cheng Zhou, Nanjing Normal University, China

Yijiu Ding, Shandong Vocational University of Foreign Affairs, China