Introduction

Existing studies find that trade liberalization increases carbon emissions and accelerates global warming because rapid expansion in the scale of trade accelerates energy consumption (Burger et al. 2023; Coe et al. 1997; Keane and Neal 2020; Sbia et al. 2014). Moreover, studies focusing on the drivers of trade-embedded carbon emission increases indicate that such increases derive from expansions in the scale of trade (Dong et al. 2010; Onwachukwu et al. 2021; Wang et al. 2019; Yunfeng and Laike 2010). However, as shown in Fig. 1, in the past 30 years, with the continuous expansion of the foreign trade of China’s agricultural products, the carbon emissions associated with the output value per unit of agricultural product have shown a continuous downward trend (Li et al. 2020), which is obviously inconsistent with the existing research conclusion that trade expansions spur carbon emission increases. Therefore, while trade liberalization has brought about an increase in output and accelerated consumption of fossil fuel energy, has it also brought about positive externalities that help curb carbon emissions? An exploration of this issue can inform efforts to balance global trade liberalization and global climate change mitigation.

Fig. 1
figure 1

The magnitude comparison of foreign trade and carbon emissions in Chinese agricultural sector.

Full size image

An important reason for the trend shown in Fig. 1 is that through trade in agricultural products, China has strengthened its connection with the international market, come into contact with the experience of mature low-carbon agricultural development overseas, and absorbed risk spillovers of low-carbon technology through agricultural trade. For example, the introduction of soil testing and formula fertilization technology promoted by the United Nations has effectively reduced the use of chemical fertilizers and pesticides and curbed carbon emissions in agricultural production (Feng et al. 2018). In addition, China’s agricultural products have encountered various practical difficulties related to technical and green trade barriers overseas, which have forced domestic agricultural producers to adopt green and pollution-free business concepts and environmentally friendly production technologies, thereby reducing carbon emissions in the agricultural sector (Liu and Xu 2021).

Although existing studies have examined the relationship between international trade and carbon emissions under the division of labor in global value chains, the conclusions are inconsistent (Meng et al. 2018; Meng et al. 2018). Studies have suggested that international trade has the potential to increase carbon emission risk (Chichilnisky 2017); that international trade may inhibit such emissions (Braithwaite 2000); or that there may not be a significant relationship between the two (Kearsley and Riddel 2010). The conclusions of subregional studies are more consistent, as Europe and the Americas do not present a positive correlation between foreign trade and carbon emissions. However, other developing countries have presented a clear phenomenon of carbon increase accompanying foreign trade (Essandoh et al. 2020; Khan et al. 2020). Studies on the case of China have come to the same conclusion (Aslam et al. 2021). Other studies have used the environmental Kuznets curve (EKC) to determine whether there is a nonlinear relationship between foreign trade and carbon emissions, and the results for samples from developed countries are consistent with the inverted U-shaped EKC (Huang and Zhang 2023); however, the results for developing country samples are inconclusive (Zoundi 2017). Recent studies have argued that the carbon emission effect of foreign trade is closely related to the country’s position in the global value chain; for example, countries undertaking processing trade (Shi et al. 2022; Wang et al. 2022) and countries in the Global South undertaking marginal industrial transfers from northern countries (Han et al. 2018) face higher carbon emission risk from export trade. Furthermore, this carbon emission effect has country and industry heterogeneity. Specifically, developed countries and technology-intensive industries are more efficient in terms of carbon emissions than are developing countries and labor- and resource-intensive industries because of their improved position in the global value chain (Sun et al. 2019).

Studies of the case of China have also noted that the particular mode of value chain participation determines the carbon emission efficiency of export trade (Li et al. 2018; Wang et al. 2020). In particular, an increase in the degree of participation of pollution-intensive industries and manufacturing industries in global value chains will result in a more obvious carbon reduction effect (Wang et al. 2020). In addition, forward and backward participation in global value chains have opposite effects on carbon emissions from China’s export trade (Zheng et al. 2022). The same situation applies to the different regions within China in the context of the domestic value chain. As a result of interregional trade exchanges, developed provinces become net carbon emission inflow regions after industrial transfers, while developing regions that take on marginal industries become net carbon emission outflow regions (Zhang et al. 2021). Nevertheless, studies have argued that although China’s interregional trade weakens carbon emission reduction through carbon risk transfer, it also makes a positive contribution to carbon emission reduction through carbon substitution (Tian et al. 2014).

In addition, many studies have analyzed low-carbon technology from a variety of perspectives. Yin et al. (2023) analyzed the urban heat island effect caused by carbon emissions based on spatial agglomeration. Dar et al. (2022) assessed the impact of biofuels on financial development, including GDP and CO2, in ten major countries worldwide and concluded that the use of fossil energy, such as natural gas, does not necessarily lead to improved financial development or a decrease in CO2 emissions. Luo et al. (2023) proposed an intelligent decision-making scheme for predicting financial risk that incorporates multiple factors affecting the price of carbon emission trading rights. In addition, studies have focused on low-carbon technological innovation entities. Xu et al. (2023) used the Gini coefficient and the Atkinson-Kolm-Sen index to study innovation inequality among the three main related entities in China, namely, industries, universities and institutes, and found that the development trends of the three innovation circles in the north, south and west differed in the spatial dimension. Recently, digital economic transformation and socially responsible preferences have also become entry points for low-carbon emission reduction research. Li et al. (2023) developed an analytical framework for digital finance and green technology innovation and found that digital transformation facilitates the correlation between digital finance and green technology innovation by improving the efficiency of credit regulation and project evaluation. Wu et al. (2023) analyzed whether institutional investors’ common preferences for the environment, society and governance (ESG) are conducive to the low-carbon technology innovation of family enterprises. In addition, there are studies evaluating the effects of China’s low-carbon technology policy implementation. Zhao et al. (2023) treated China’s low-carbon city pilot policy (LCCP) as a quasinatural experiment, and the results showed that this low-carbon strategy promotes the digital transformation of pioneer cities but presents heterogeneity by firm, industry, and region.

From the analysis above, it can be concluded that most of the studies based on the national level analyze the impact of a country’s position in the global value chain on the carbon emissions of export trade, and few of them explore the carbon emission risk reduction of foreign trade with the help of technological spillover theory. In addition, studies at the regional level within a country also analyze the carbon emission risk transfer caused by interregional trade without considering the carbon emission reduction effect caused by the spatial diffusion of technology. Therefore, to fill these gaps in the current research, this paper aims to explore the carbon emission theory mechanism of the environmental effect of agricultural trade from the perspective of technology risk spillovers. On the one hand, based on products’ different attributes, we distinguish risk spillovers from the industrial capital goods trade and agricultural trade by technology type. On the other hand, in view of the tacit diffusion property of disembodied technology communication (DTC), we focus on the carbon emissions generated by technology risk spillovers in agricultural trade from a spatial perspective. Second, our paper expands the classical linear model of the technology risk spillover effects of foreign trade into a nonlinear spatial model. Existing studies have not examined the disembodied features of technology risk spillovers in agricultural trade or their spatial diffusion properties. To solve the endogeneity issue caused by the omission of unobserved factors associated with disembodied technologies with spatial attributes in the classical linear model, our paper extends the classical linear model into a spatial Durbin model (SDM) with a nonlinear structure. We expand the technology risk spillover index proposed by Coe and Helpman (1995) to fit the situation of developing countries such as China and thus to forecast the magnitude of carbon emissions from technology risk spillovers in agricultural trade across different regions. Third, this paper comprehensively estimates China’s agricultural carbon emissions; proposes an indicator system for calculating agricultural carbon emissions based on the three agricultural production processes of planting/breeding and production, input consumption, and waste disposal; and identifies 6 types of agricultural carbon emission activities, including 40 agricultural carbon emission sources. Finally, based on geographical distance and similarity of economic development, we map the relationships between different geographically adjacent and economically adjacent spatial units, considering the dual connotation of spatial diffusion in DTC.

The rest of the paper is organized as follows. The second section presents the literature review and our hypothesis. The third section discusses the research method, which expands the spatial dimension of the classical linear model to make it suitable for verifying the proposition in this paper; the fourth section describes the research results. The fifth section provides a discussion of the results, and the sixth section provides the conclusion and policy recommendations.

Theory

Magnitude of agricultural carbon emissions from technological progress

There is a general consensus that technological progress mitigates carbon emissions risk by reducing fossil fuel energy consumption (Samargandi 2017). The sustainable development of modern agriculture is also inseparable from low-carbon technology (Xiong et al. 2019). Some scientists believe that technological progress is an important reason for reducing carbon emission risk in agriculture (Ismael et al. 2018). Notably, the key factor that makes it difficult to improve the efficiency of carbon utilization in agriculture is technical inefficiency (Li et al. 2023). An important way to achieve low-carbon agriculture is through the advancement of resource reuse and recycling technologies (Fan and Wei 2016), which can promote the spatial risk spillover of low-carbon technology (Wu and Feng 2019). However, technological progress may also have a “rebound effect” on carbon emissions; namely, technological progress may promote energy efficiency while increasing total carbon emissions through substitution effects, income effects, and additional demand for energy due to increased output. Awaworyi Churchill et al. (2019) found that the increase in R&D intensity in G7 countries since the 19th century during specific periods has stimulated carbon dioxide emissions.

There are different opinions even on whether agricultural technological progress has brought about a carbon reduction effect in China. For example, Zhang et al. (2019) found that agricultural technological progress has increased the total amount of agricultural carbon emissions in China, especially in the eastern region, where the agricultural economy is the most developed, and that the carbon-increasing effect brought about by agricultural technological progress is far greater than that in the central and western regions. This is also confirmed by Tian and Yin (2021) research on the rebound effect of agricultural technological progress. Huang et al. (2022) also confirmed that technological progress triggers the rebound effect of agricultural carbon emissions and that while technology reduces the intensity of carbon emissions per unit of grain output, it leads to an increase in the total amount of agricultural carbon emissions by stimulating the growth of the agricultural economy and increasing the inputs of agricultural resources.

In addition, since Grossman and Krueger (1995) proposed the theory of the three environmental effects of foreign trade, many scholars have examined the environmental technology effects of foreign trade through technological risk spillovers, which could enhance energy efficiency (Assamoi et al. 2020). The results of Wan et al. (2018) also support the existence of technological risk spillovers. Furthermore, Coe and Helpman (1995) used the share of goods imported as a weight to quantify the technology risk spillovers from the international R&D capital stock to the import country. Coe et al. (2009) followed this method and came to the same conclusion in an examination of developing countries’ import trade in goods. Subsequent studies aimed to distinguish between two different types of technology risk spillovers from import trade. In addition to considering physical technology, represented by machinery and equipment, studies have considered the disembodied technology risk spillovers represented by intangible ideas, such as business concepts and skills, and yielded the same finding that risk spillovers from disembodied technology enhance production efficiency (Haider and Mishra 2021).

Magnitude of carbon emission from risk spillovers in agricultural trade

Risk spillovers from DTC in agricultural trade

Compared with industrial capital goods imports that bring physical technologies, imports of agricultural products are accompanied by more intangible disembodied technologies, such as green management concepts or green consumer ideas. Since most agricultural products function in human consumption or are used for livestock and poultry breeding, the safety risk and reliability of agricultural products have reached a global consensus. Since the reform and opening up, China’s traditional agricultural model has remained closed and backward. Its extensive growth model and large quantities of pesticides and fertilizers have exposed the quality and safety risks of China’s agricultural products to the international market. Accordingly, developed country markets in Europe and North America have begun to extensively detect excessive pesticide, heavy metal, and bacterial residues in Chinese agricultural products, covering a wide range and large quantity of agricultural products (Liu et al. 2019). By participating actively in the international market, importing high-quality agricultural products from overseas and adhering to international practices, it is possible to apply disembodied concepts and cutting-edge practices to green and pollution-free agricultural businesses, reduce chemical fertilizer and pesticide inputs, and achieve low-carbon agriculture. Specifically, the benefits of disembodied technologies can be summarized as follows.

First, brand awareness of agricultural products could help reduce agrochemical inputs. Quality and safety information such as pesticide residues, farmland contamination, and excessive heavy metals and hormones is difficult for consumers to use for direct identification (Hu et al. 2017). Driven by self-interest maximization, farmers overuse chemicals in the production process to ensure the desired scale of output, resulting in a private supply of high-quality agricultural products that is much lower than the optimal supply required by society. Furthermore, China’s agriculture fails to achieve large-scale and centralized production, which not only operates in a fragmented and extremely hidden manner but also results in high governmental regulatory costs (Song et al. 2014). The key to solving this problem still lies in the establishment of geographical indications certification for agricultural products, which gives farmers excess profits through brand premiums and ensures that they are engaged in compliant production guided by market prices, thus conveying the message of quality and safety of agricultural products to consumers (Van Rijswijk and Frewer 2012).

Second, a traceability system for agricultural product quality could control the pollution risk in the production and circulation stages. The traceability of agricultural products throughout their entire production process has become a basic requirement for food safety management in developed countries, and it is also an important means to address quality or safety emergencies involving agricultural products (Hobbs 2004; Jin and Zhou 2014). For example, EU Regulation No. 1760/2000 established a traceability process based on the labeling and identification of beef products. The International Standardization Organization (ISO) formulated a new food quality and safety traceability system in 2007, which has become an effective way to control the quality of agricultural products “from field to table”. By importing overseas agricultural products, the Chinese market can fully familiarize itself with internationally accepted quality traceability systems, which will encourage China to establish a similar system. For example, in 2006 and 2007, the Law of the People’s Republic of China on the Quality and Safety of Agricultural Products and the Measures for the Administration of Food Safety in the Circulation Field were promulgated to form a market access system for tracing all agricultural products in the market and effectively preventing the excessive use of chemicals.

Finally, regional indicators of the geographical origin of agricultural products could reduce the cost of supervising farmers’ illegal production risk. A regional reputation premium for agricultural products built around indicators of geographical origin could motivate farmers to comply with production standards and provide the market with quality and safe agricultural products (Mancini 2013). Brand premiums for a single agricultural enterprise, which are quite common among large agricultural enterprises, are conducive to compliant production on the part of the focal enterprise only. Since small farmers cannot benefit from a brand premium, they may provide the market with agricultural products with hidden safety hazards. One way to address the lack of a reputation premium for small farmers is to organize them in a certain area and encourage them to provide quality and safe agricultural products by establishing a regional reputation mechanism, thereby upgrading individuals’ weak credit to more reliable credit at the group level. Some countries use geographical indicators to emphasize the connection between the quality, reputation and characteristics of agricultural products and their geographical origin within a specific area (Fernández-Barcala et al. 2017), as in the case of the appellation of origin law in France and the Administrative Measures for Geographical Indications of Agricultural Products in China. This regional reputation mechanism greatly helps prevent the risk of illegal use of fertilizers and pesticides to obtain short-term profits due to farmers’ decentralized operations.

Attributes of spatial risk spillovers of DTC

Previous work has noted that the study of technological risk spillovers needs to move from the micro level for the firm to the macro level of the region. On the one hand, a decisive relationship between technological innovation and output has not been found at the firm level but rather at the regional level (Audretsch and Feldman 1996); on the other hand, technology is rooted in the individual, and the free movement of individuals between regions creates spatial risk spillovers of technology (Los and Verspagen 2000).

However, owing to their disembodied attributes, agricultural trade technological risk spillovers are rooted in individuals, which means that they can be exchanged only at the individual level (Fallah and Ibrahim 2004). These “face-to-face” activities, such as flows of personnel, contact with customers or informal meetings, have become an important means of diffusing disembodied technologies (Desrochers 2001), thus contributing to collaborative research (Dong et al. 2020) and technology services (Cai et al. 2016) to excel on a spatial scale. For example, following the U.S. Department of Commerce, which put China’s Huawei on its “Entity List” in May 2019, the ARM chip company banned its employees from meeting with Huawei’s employees because meeting them would be a way of leaking their technology (Shattuck 2021). Thus, the spatial extent of the disembodied technology that accompanies agricultural imports depends on the degree of “face-to-face” communication between agricultural entities.

Geographical adjacency is an important factor influencing the “face-to-face” communication of disembodied technologies in agricultural trade. DTC is also rooted in the sociocultural and institutional environment of a particular region, and geographical adjacency tends to exploit both of these aspects, which makes disembodied technological risk spillovers more localized and unable to exist independently from specific individual or regional situations, with high localization characteristics of spillovers (Xu et al. 2022). Even risk spillovers across administrative boundaries usually follow the first law of geography, which states that spatial spillovers tend to diminish with geographic distance (Liu et al. 2024).

The spatial risk spillovers of disembodied technologies from agricultural trade share the same economic adjacency. Although disembodied technology risk spillovers are highly localized, they can also occur over long distances through trade and economic exchanges (Basile et al. 2012). Even if the geographical distance is far away, economic adjacency provides a more similar technological context, which in turn makes it easier to achieve disembodied technology communication (Caragliu and Nijkamp 2012). Regions with comparable levels of agricultural economic development exhibit stronger economic adjacency in terms of the carbon reduction effects of technology risk spillovers from agricultural trade.

Carbon emission stage in agricultural trade

Based on the above analysis, we divide the spatial analysis of carbon emissions from technology risk spillovers in agricultural trade into three stages as follows.

The first stage involves local risk spillovers of mixed DTC and embodied technology brought about by the import of overseas agricultural products. Promoting brand-building awareness, quality traceability systems, and geographical indicators in agricultural product import sectors help foster local low-carbon agriculture, but certain technologies introduced by agricultural product imports may result in a rebound effect. For example, the introduction of embodied technologies such as agricultural mechanization equipment can easily cause scale effects, thereby accelerating agricultural carbon emissions. Therefore, in the first stage of agricultural trade, the expected impact of local technology risk spillovers on agricultural carbon emissions in importing regions is uncertain and depends on the gap between the carbon emission reduction from technology risk spillovers and the rebound effect.

The second stage involves global risk spillovers of disembodied technologies from imports of overseas agricultural products to other regions beyond the import area. On the one hand, DTC is carried out by individuals, which means that agricultural trade produces cross-regional risk spillovers of DTC through flows of people. On the other hand, due to its uncodified character, the risk spillovers of DTC are highly localized. Therefore, the global spatial risk spillovers of disembodied technologies brought about by agricultural trade affect adjacent areas beyond import areas and produce carbon reduction effects.

The third stage involves risk spillovers of disembodied technologies to economically adjacent areas. Individuals in regions with similar levels of agricultural economic development have highly similar cognitive abilities and technical backgrounds affected by multidimensional adjacency. Even if the geographical distance is high, the global spatial risk spillovers of disembodied technologies through imports of agricultural products can affect economically adjacent areas beyond the import region.

Data and methods

Data sources and data preprocessing

Data sources

The data sample used in this paper consists of panel data for 31 provinces in mainland China from 1981 to 2019. To make the results more robust, we expand the source countries of technology risk spillovers in agricultural trade from the G7 countries, which are considered in existing studies (Awaworyi Churchill et al. 2019), to 16 OECD countries, namely, Ireland, Portugal, Finland, Denmark, Belgium, Austria, the Netherlands, Spain, Canada, Italy, Italy, the UK, France, Korea, Germany, Japan, and the USA. The raw data used in the calculation of the model variables are presented in Online Appendix A1.

Data processing

Due to changes in the statistical caliber and the different base years of the statistical data released by different statistical yearbooks, there are missing values for some indicators. In Online Appendix A2, we present our data preprocessing procedure in detail.

Variables

Dependent variable: agricultural carbon emissions

Calculation method

Life cycle assessment (LCA) is widely used in calculating carbon emissions due to its feasibility and applicability (Dyer et al. 2010; Johnson et al. 2007). The first step is to identify different carbon emission activities, and the second step is to find the number of different carbon emission sources and the corresponding coefficient. Calculations of China’s agricultural carbon emissions usually cover three major agricultural production processes: planting/breeding production, use of energy and agricultural materials, and agricultural waste disposal. Then, the carbon emission coefficients are used to calculate the emissions of greenhouse gases, such as carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O). We present the use of the LCA method to estimate agricultural carbon emissions (\(C\)) in the Online Appendix A3 in detail.

Sources of agricultural carbon emissions

We included seven sources of agricultural carbon emissions in this paper, namely, the number of crops planted, the number of livestock and poultry bred (IPCC, 2006), the amount of nitrogen fertilizer applied (Zhang et al. 2010), the amount of fossil fuel consumed (Chen and Chen 2018), the amount of electricity consumed, the embodied carbon of agricultural materials (Zhang et al. 2009), and the amount of crop straw burning (Bi et al. 2009). More details are presented in Online Appendix A3.

Independent Variables

Technology risk spillovers from agricultural trade

Following the method of Coe and Helpman (1995) for indexing international R&D capital stock risk spillovers, this paper calculates the technological risk spillovers from agricultural trade at the interprovincial level in China as follows:

$${F}_{{it}}=\frac{\mathop{\sum}\nolimits_{q=1}^{7}{I}_{{itq}}}{\mathop{\sum}\nolimits_{q=1}^{7}{I}_{{tq}}}\times \mathop{\sum}\nolimits_{j=1}^{16}\left(\frac{{E}_{{jt}}}{{{GDP}}_{{jt}}}\times {{RD}}_{{jt}}\right)$$
(1)

The calculation is performed in two stages. The first stage is based on the calculation method of Coe and Helpman (1995) and uses the proportion of the export value of agricultural products (\({E}_{{jt}}\)) from technology risk spillover country j in year t to China’s gross domestic product (GDPjt) to multiply the R&D capital stock RDjt of the technology risk spillover country, which indicates how much of the R&D capital stock of the technology risk spillover country has spilled over to China through the export of agricultural products. The second stage subdivides the technology risk spillovers that China acquired from overseas through the import of agricultural products to the 31 focal provinces in China. The China Statistical Yearbook and China Rural Statistical Yearbook record the import quantities (tons) of 7 main agricultural products in the 31 provinces. Therefore, we use the proportion \(\frac{\mathop{\sum}\nolimits_{q=1}^{7}{I}_{{itq}}}{\mathop{\sum}\nolimits_{q=1}^{7}{I}_{{tq}}}\) of the import quantities of these 7 kinds of agricultural products in each province in total national imports to multiply the technological risk spillovers obtained through foreign trade in agricultural products at the national level in China to obtain the technology risk spillovers \({F}_{{it}}\) acquired from overseas through agricultural trade at the provincial level.

China’s R&D capital stock

This paper employs the perpetual inventory method to calculate the R&D capital stock (\({D}_{{it}}\)) of the 31 provinces in mainland China:

$${D}_{{it}}=\left(1-{\delta }_{{it}}\right)\times {D}_{{it}-1}+\left(1-0.5\times {\delta }_{{it}}\right)\times {I}_{{it}}$$
(2)

where \({D}_{{it}}\) and \({D}_{{it}-1}\) represent the current and previous R&D capital stock, respectively. \({I}_{{it}}\) represents the current R&D capital investment, and δ is the depreciation rate. The R&D capital investment is converted into constant prices with the capital investment price index (\({{PI}}_{{it}}\)). The detailed parameters used are described in Online Appendix A4.

R&D capital stock in OECD countries

The R&D capital stock of the 16 OECD countries is also calculated using the perpetual inventory method in Eq. (2). The R&D capital investment (\({I}_{{it}}\)) in the equation is measured by “GDP expenditure on R&D at 2015 prices” in Online Appendix A1.

Control variables

Control variables are introduced in the empirical model to correct potential endogeneity issues and thus prevent biased estimates. The control variable selection procedure in this paper is based on three steps (more detailed explanations are presented in Online Appendix A5). First, we follow the initial specifications of the classical model (C-H) proposed by Coe and Helpman (1995) to select the control variable. To control for endogeneity, we include the domestic R&D knowledge capital stock, namely, D, in Eq. (3). Second, because the spatial risk spillover effects of independent variables can be theoretically elucidated, new control variables are not considered in spatial expansion modeling. Third, based on the tradeoff between the unbiasedness of the estimation and the validity of the statistical inference, additional control variables are excluded.

Spatial weight matrix

The spatial weight matrix is used to represent whether there is a correlation between different regions or between the strengths of the correlations. This paper uses a variety of spatial weight matrices to support the robustness of the empirical results.

Geographical adjacency matrix

Since the administrative regions are not geographically completely regular, geographical adjacency can be defined in different ways. This paper adopts two methods: a spatial contiguity weight matrix and a nearest-neighbor spatial weight matrix. The detailed explanations are presented in Online Appendix A6.

Economic adjacency matrix

Residents of regions with similar levels of economic development have similar cognitive abilities in understanding tacit knowledge, and disembodied technologies can thus be transmitted to economically adjacent regions even when they are separated by large geographical distances. The economic adjacency matrix reflects not only whether different regions are related but also how strong the correlation is.

Modeling

Spatial modeling to correct for endogeneity

This paper expands the linear model proposed by Coe and Helpman (1995) (the C-H model) to identify the technological risk spillover effects of foreign trade. The model improvements are presented in Online Appendix A7.

Then, we follow the C-H model of Coe and Helpman (1995) for examining technology risk spillovers from foreign trade:

$$C={\beta }_{0}+{\beta }_{1}\times F+{\beta }_{2}\times D+u$$
(3)

The dependent variable \(C\) is the target variable for technological risk spillovers from foreign trade, \(F\) represents the international capital stock of R&D knowledge that spills over to the importing country through overseas trade, \(D\) represents the domestic capital stock of R&D knowledge in the importing country, and \(u\) is the error term. Coe and Helpman (1995) applied the above model to test the technology risk spillover effect of trade in tangible goods to the importing region. Since the hidden disembodied technology (\(T\)) cannot be quantified as a dependent variable in the model, the subsequent spatial diffusion effect of the disembodied technology risk spillover in the import region to the adjacent regions beyond the import region is not accounted for; thus, the C-H model is a classical linear model that satisfies the assumptions of OLS estimation.

Given the disembodied technology risk spillovers from foreign trade to the import region and the diffusion of these spillovers through various adjacent regions beyond the import region, the specification of the initial model by Coe and Helpman (1995) is incomplete. Since it cannot be quantified as a dependent variable, there is a clear correlation between disembodied technology (\(T\)), which is hidden in the error term \(u\), and the independent variable \(F\), which represents the risk spillover effect of foreign trade in the model, thus violating the “zero conditional mean” premise assumption of OLS estimation and leading to endogeneity in the estimation of the linear model (Liu et al. 2022).

The solution to the endogeneity problem caused by the above OLS estimation is to accommodate the double-technology risk spillover effect of foreign trade to the import region and its adjacent regions at the same time in the specification of the model, which means that the linear model in which the independent variable can only affect the dependent variable in the same region can be extended to a nonlinear model in which the independent variable can affect the dependent variable in all regions. Thus, such an extension of the specification of the nonlinear mode in spatial dimensions requires the help of spatial econometric modeling theory.

In response to the endogeneity problem of the classic linear model of Coe and Helpman (1995) in terms of OLS estimation due to the omission of disembodied technologies, the solution of spatial econometric modeling theory is to use a spatial autoregressive process such as \(T\) = \(\rho \times W\times T+\varepsilon\) to characterize the disembodied technology brought about by foreign trade with cross-regional diffusion properties, thus deriving \(T\) = \({\left(I-\rho \times W\right)}^{-1}\times \varepsilon\) from the error term (\(u\)) and iteratively processing it to obtain the most general spatial Durbin model in the spatial econometrics model family:

$${C=\rho W\times C+{\alpha }_{0}+\beta }_{1}\times F+{\beta }_{2}\times D+{\alpha }_{1}W\times F+{\alpha }_{2}W\times D+h$$
(4)

where \(C\), \(F\), and \(D\) are still consistent with (1) and W is a spatial weight matrix used to describe whether different regions are adjacent. If the value of the matrix unit is not 0, the two regions are related. Geographical adjacency or economic adjacency can be used to quantify the degree of interregional correlation, which is the key to the subsequent cross-regional diffusion of disembodied technology risk spillovers of foreign trade from the import region to other regions beyond the import region; \(\rho\) is the spatial autocorrelation coefficient, representing the correlation between different regions; \({\alpha }_{0}={\beta }_{0}\times \left(I-\rho \times W\right)\), \({\alpha }_{1}=\) −\({\beta }_{1}\times \rho\), \({\alpha }_{2}=\) −\({\beta }_{2}\times \rho\), \(h=\varepsilon +\left(I-\rho \times W\right)u\). Thus, the nonlinear model obtained from this spatial dimension extension is able to simultaneously measure the double risk spillover effects of foreign trade on the import region and its adjacent regions.

Robustness check of spatial modeling

The construction of the SDM in this paper is based on the spatial diffusion properties of risk spillovers of DTC in agricultural trade, but to assess whether the spatial model specification matches the data-generating process (DGP) of the sample, it is necessary to implement a robustness check of the model (Bao et al. 2023). Following the method proposed by Elhorst (2011), we apply the following steps to check the robustness of Eq. (4). The first step involved a robustness check of the extension of the linear model to a spatial model; the second step involved a robustness check of the spatial Durbin model. More details are presented in Online Appendix A8.

Parametric interpretation of spatial modeling

The spatial econometric model relaxes the strict assumption of the classic linear model that section units are not correlated, allowing independent variables to affect not only the dependent variable of the section units locally but also the dependent variables of other section units. Then, the OLS parameter interpretation of the classical linear model cannot be followed, and the parameter estimates (\({\hat{\beta }}_{1}\), \({\hat{\beta }}_{2}\), \({\hat{\alpha }}_{1}\), \({\hat{\alpha }}_{2}\)) in Eq. (4) of the spatial econometric model can no longer represent the cross-sectional average effect of the independent variables on the dependent variables. In this paper, the partial matrix proposed by LeSage and Pace (2010) is used to define the direct and indirect effects of the respective variables on the dependent variable in Eq. (4), and the mathematical definitions are detailed in LeSage and Pace (2010). Thus, we obtain the effects of the foreign trade technology risk spillover variable (F) and domestic R&D intellectual capital stock (D) on the dependent variable C in the import region, namely, the direct effects \({\bar{M}\left(F\right)}_{{direct}}\), \({\bar{M}\left(D\right)}_{{direct}}\), as well as the effects on the dependent variable of the adjacent regions beyond the import region, namely, the indirect effects \({\bar{M}\left(F\right)}_{{indirect}}\), \({\bar{M}\left(D\right)}_{{indirect}}\). When the direct and indirect effects are summed, we also obtain the total effects \({\bar{M}\left(F\right)}_{{total}}\) and \({\bar{M}\left(D\right)}_{{total}}\), respectively. More details are presented in Online Appendix A9.

Results

Descriptive analysis

We use panel data on agricultural carbon emissions (10,000 tons of CO2e), technology risk spillovers in agricultural trade (million US dollars), and the domestic R&D capital stock (10,000 CNY) from 1981 to 2019 to run Eq. (2). Table 1 presents the descriptive analysis of the variables.

Table 1 Descriptive analysis.
Full size table

Heterogeneity analysis of technology risk spillovers

Our hypotheses depict that DTC spillover from agricultural international trade not only affects agricultural carbon emissions in local regions but also affects agricultural carbon emissions in adjacent areas due to its intangible characteristics. Benefiting from the Chinese policy dividend since the reform and opening up and its own location advantage near the sea, the eastern provinces are the most developed regions in terms of foreign trade, and they hold an unparalleled advantage over the middle and western provinces in terms of using agricultural international trade to absorb low-carbon technologies from overseas. Therefore, it is also more likely that the subsequent secondary risk spillover effects of DTC in agricultural trade in Chinese inland regions will originate more from eastern regions to the inland middle and western regions. To this end, we analyze the heterogeneity of agricultural carbon emissions and technological risk spillovers of agricultural trade in the three major regions of China. Based on the heterogeneity analysis in Table 2 and Fig. 2, it can be concluded that the carbon emissions per unit of agricultural output and technology risk spillover of agricultural trade in the three major regions show significant regional heterogeneity, laying a foundation for checking the subsequent secondary risk spillover effect of agricultural trade disembodied technologies in inland China in the following subsection. More explanations are presented in Online Appendix A10.

Table 2 Agricultural carbon emissions and technology risk spillover of agricultural trade in the three major regions of China.
Full size table
Fig. 2
figure 2

Scatterplot of carbon emissions per unit of agricultural output and technology risk spillover of agricultural international trade in 31 provinces.

Full size image

Spatial modeling robustness check results

Robustness check results on the extension of the linear model into a spatial model

Table 3 shows the results of checking whether the classical linear model (3) equation omits spatial interaction effects; from these results, it can be determined that the linear model needs to be extended to a nonlinear spatial econometric model. First, the LM test values for the spatial lag and spatial error effects in the three specifications of Pooled, Spatial Fixed, and Time-Period Fixed all pass the 1% significance level. In addition, the confidence level of the LM test values remains unchanged whether the spatial adjacency matrix (adjW), the spatial nearest-neighbor matrix (W5-W7), or the economic adjacency matrix (ecoW) is used. Thus, the results suggest that the classical linear model, whose specification is too specific, omits spatial interaction effects. Accordingly, this approach needs to be extended to a spatial econometric model with a generalized specification to match the way the data are generated in the empirical sample. Second, the goodness-of-fit (R2) of the linear model under the three specifications is too low, with the maximum value occurring under the setting form of cross-sectional fixed effects but only 0.4883, which implies that the model specification is incomplete and that more important explanatory variables may have been omitted. Thus, this evidence supports the extension of the classical linear model to a spatial econometric model to accommodate more explanatory variables. More explanations are presented in Online Appendix A11.

Table 3 The specification check from linear model to spatial model.
Full size table

Robustness check results on the specification of the spatial Durbin model

In this step, we employ four different spatial weight matrices to check the robustness of the SDM. First, the SDM is estimated under fixed effect and random effect specifications. Second, the Wald and likelihood ratio (LR) tests are performed with parameter constraints. The results are presented in Table 4, which implies that our study should adopt two spatial interaction effects (SLM and SEM) simultaneously in the spatial model to depict the attributes of spatial risk spillovers of DTC in agricultural trade. More details are presented in Online Appendix A10.

Table 4 The specification check of spatial Durbin model.
Full size table

Spatial modeling results

Table 5 presents the estimated results for the direct, indirect and total effects. Since the P value corresponding to the Hausman test in Table 4 passes the significance test at the 1% level, the null hypothesis is rejected. Therefore, the interpretation of the results in Table 4 requires fixed effect specifications.

Table 5 The estimated results of direct, indirect and total effect.
Full size table

Local technological risk spillovers from agricultural trade decrease agricultural carbon emissions locally. Local technological risk spillovers from China’s own R&D capital stock have no impact on local agricultural carbon emissions. When we use the four spatial weight matrices, the estimates of the direct effect of technology risk spillovers from agricultural trade are not only negative but also pass the significance test, indicating that technology risk spillovers from agricultural trade significantly reduce agricultural carbon emissions in the local region. This inhibitory effect is approximately −0.0079 ~ −0.0116 under the fixed effect specification; that is, every 1% increase in the technological risk spillover from agricultural trade results in a 0.79~1.16% decrease in agricultural carbon emissions in the local region.

Global risk spillovers of disembodied technologies in agricultural trade decrease agricultural carbon emissions in adjacent areas. Global technology risk spillovers from China’s R&D capital stock increase agricultural carbon emissions in geographically adjacent areas but decrease agricultural carbon emissions in economically adjacent areas. Under the fixed effect specification, regardless of which spatial weight matrix is used, the estimates of the indirect effect of technology risk spillovers from agricultural trade are negative, and all pass the significance test at 5%, indicating that disembodied technologies generated by agricultural trade decrease agricultural carbon emissions not only locally but also geographically and economically adjacent areas. For the four geographical adjacency matrices, the estimates of the indirect effect are between −0.0285 and −0.1157; that is, for every 1% increase in technological risk spillovers from agricultural trade, the global risk spillovers of disembodied technologies reduce agricultural carbon emissions by 2.85 to 11.57% in geographically adjacent areas. Similarly, this indirect effect is estimated to be −0.0490 in the economic adjacency matrix. When we use the four geographic adjacency matrices, the estimate of the indirect effect of China’s R&D capital stock is not only positive but also significant at the 1% level, indicating that global technology risk spillovers from China’s R&D capital stock boost agricultural carbon emissions in geographically adjacent areas, resulting in a rebound effect. When we use the economic adjacency matrix, the indirect effect of China’s R&D capital stock is estimated to be −0.0930, with a significance level of 5%, indicating that global technology risk spillovers from China’s R&D capital stock decrease agricultural carbon emissions in economically adjacent areas.

Technology risk spillovers from agricultural trade significantly decrease agricultural carbon emissions, while technology risk spillovers from the R&D capital stock increase agricultural carbon emissions. Regardless of which spatial weight matrix is used, the estimated total effect of technology risk spillovers from agricultural trade is negative and passes the significance test at 5%, indicating that every 1% increase in technology risk spillovers from agricultural trade reduces agricultural carbon emissions in both the local region and geographically adjacent regions by 3.46~12.36% and reduces agricultural carbon emissions in both the local region and economically adjacent regions by 5.83%. When we use the four geographic adjacency matrices, the total effect estimates of China’s R&D capital stock are not only positive but also pass the significance test at 1%, indicating that the technology risk spillovers from the R&D capital stock increase agricultural carbon emissions. When we use the economic adjacency matrix, the total effect is no longer statistically significant, indicating that technological risk spillovers from the R&D capital stock have no effect on agricultural carbon emissions.

Discussion

Why do domestic and foreign R&D capital technology risk spillovers strongly contrast in terms of their impacts on China’s agricultural carbon emissions? For the OECD countries, whether we consider local technology risk spillovers or global risk spillovers from the R&D capital stock in agricultural trade, agricultural carbon emissions are reduced in both the local region and adjacent regions. However, in China, technology risk spillovers from the domestic R&D capital stock have led to a significant rebound effect, which has increased carbon emissions in agricultural production. The reason lies in the different preferences for the orientation of technological progress due to differences in the level of economic development. Most of the OECD countries are mature market economies, and they have entered the postindustrial stage. The steady increase in per capita income has led to a shift in consumer demand preferences toward environmentally friendly products. With the incentives of strict environmental regulations, OECD countries’ preference for technological innovation has also gradually shifted from innovations enhancing output scale to those enhancing resource conservation, which has made these countries sources of innovation in global clean production technology and exports from the environmental protection industry (de la Tour et al. 2011; Wan et al. 2018). Especially in the field of agriculture, OECD countries have introduced very strict food quality and safety laws, such as the positive list system of agricultural chemical residues in food in Japan, the general food law regulation in the EU, and hazard analysis and key point control in the United States. These regulations promote progress in the development of green and pollution risk control technology in agricultural production in OECD countries, and with the export of agricultural products, the concept of environmentally friendly agricultural production is gradually passed on to developing countries.

In this vein, we construct an environmental Kuznets curve (EKC) model of Chinese agricultural carbon emissions, per capita agricultural GDP and its square using the panel data of the 31 provinces in mainland China from 1981 to 2019 as a sample. We aim to identify the turning point in the inverted U-shaped curve based on the estimated results and then judge whether the current agricultural economic growth in China has reached a new stage characterized by a preference for progress in low-carbon technologies. The estimates are presented in Table 6. More explanations are presented in Online Appendix A12. According to the results, we should choose random effect estimation to improve the validity of the statistical inference.

Table 6 The estimated results of EKC.
Full size table

We present the above results in Fig. 3. Then, we further explain the EKC estimation results under the random effect specification. The direction of the estimates of per capita agricultural GDP and its square in the fourth column of Table 6 indicate that the EKC displays the classic inverted U shape and that the metrics associated with the turning points of the curve, namely, per capita agricultural GDP and maximum agricultural CO2 emissions, can be calculated accordingly. The turning point value is 4427.381 CNY (measured in constant 1980 prices), indicating that only once per capita agricultural GDP crosses this threshold will the agricultural economy enter the stage of clean production dominated by low-carbon technologies. Reviewing the sample data on per capita agricultural GDP, we find that as of 2019, most agricultural economic growth in the 31 provinces had not yet crossed the turning point of this inverted U-shaped curve. Only Hainan (in 2013), Heilongjiang (in 2018), and Jilin and Jiangsu (in 2019) crossed the turning point, indicating that agricultural economic growth in these four provinces has begun to increase during the low-carbon production stage, while the per capita agricultural GDP of the remaining 27 provinces still falls short of the turning point to different degrees. The environmental effects of agricultural economic growth are still in the stage of increasing carbon emissions. This may explain why China’s accumulation of R&D capital stock is currently still inclined more toward technologies that can enhance the output scale than toward innovation in low-carbon production technology.

Fig. 3
figure 3

The magnitude of carbon emission reduction from spatial risk spillovers.

Full size image

After the EKC in Table 6 is estimated, we also assess whether there is a true inverse U-shaped relationship between agricultural carbon emissions and agricultural GDP. Table 7 presents the results for the existence of the inverse U-shaped curve. First, the minimum and maximum values are 154.3564 and 6453.277, respectively, while the extreme point is 4427.647, which falls perfectly within the data sample interval, indicating that the empirical sample data completely cover both sides of the inverse U-shaped curve. There is no extreme concentration on one side of the inverse U-shaped curve. Second, the slopes calculated from the minimum and maximum values of the sample data are 2.7202 and −1.2894, respectively. The positive and negative values are also completely consistent with the slope conditions on both sides of the inverse U-shaped curve. Finally, the significance check results suggest that the t values corresponding to the above two slope values are 23.5736 and −4.8259, respectively, and the p values are reported to be 0 for four decimal places, which indicates that the probability of making mistakes in rejecting the null hypothesis of a “monotone or U shape” is extremely small. Therefore, the alternative hypothesis should be accepted; that is, the inverse U-shaped curve is valid. Furthermore, the overall check t value for the presence or absence of an inverse U shape curve is 4.83, corresponding to a p value of 7.86e-07, which confirms that the data sample is consistent with the inverse U shape curve trend.

Table 7 Checking results for the existence of the inverse U shape curve.
Full size table

Conclusions and implications

Conclusions

Based on the existing studies on technology risk spillovers from foreign trade and the classification of spillover technologies, this paper distinguishes the types of technology risk spillovers between agricultural trade and industrial capital goods trade and suggests that risk spillovers of disembodied technologies (such as the concept of green and pollution risk mitigation of agricultural production) mainly come from agricultural trade, given their character as nonintermediate inputs. We note that spillovers of disembodied technologies can achieve regional diffusion through face-to-face contact between people in different regions and propose a framework to analyze the magnitude of carbon emission reduction from technology risk spillovers in agricultural trade from a spatial perspective.

Building on the classic model proposed by Coe and Helpman (1995), we apply a spatial econometric modeling technique to solve the endogeneity issue caused by the omission of the unobserved factor of DTC from the classic model of Coe and Helpman (1995). We expand the linear structure into an SDM with a nonlinear structure in the spatial dimension.

Local technology risk spillovers from agricultural trade reduce agricultural carbon emissions in the local region. Moreover, the global spatial spillovers of overseas DTCs that accompany the imports of agricultural products also reduce agricultural carbon emissions in geographically and economically adjacent regions.

Local technology risk spillovers from the Chinese R&D capital stock do not reduce agricultural carbon emissions locally; rather, global spatial spillovers contribute to an increase in agricultural carbon emissions in geographically adjacent regions and reduced agricultural carbon emissions in economically adjacent regions.

Considering both local and global spillovers, technology risk spillovers from agricultural trade reduce Chinese agricultural carbon emissions, but technology risk spillovers from the Chinese R&D capital stock contribute to an increase in agricultural carbon emissions.

In China, agricultural economic growth has not yet crossed the turning point for pollution reduction in the inverted U-shaped EKC, and the accumulation of R&D capital stock is currently still inclined more toward technologies that can boost the output scale than toward the innovation of low-carbon production technology.

In reality, the most representative case is the long-term cooperation between China and the UK in the field of agricultural technology. As early as 2008, the Chinese Ministry of Agriculture and Rural Affairs and the UK Department for Environment, Food and Rural Affairs established the China-UK Sustainable Agriculture Innovation Network (SAIN), which subsequently implemented technology transfer in water conservation, climate change, greenhouse gases, and agricultural nutrients, with the aim of stimulating the sustainable development of agriculture in China. In 2019, the Sino-British Smart Farm Science and Technology Innovation Cooperation was launched, and by introducing the technical advantages of the UK in terms of sensors and smart farm management, the overall smart farm solution of “field-facility-livestock integration” was eventually developed in China. The introduction of these low-carbon emission reduction technologies has largely regulated the improper application of excessive chemical fertilizers and pesticides in China’s agricultural production, which is conducive to the prevention and control of agricultural nonpoint surface pollution as well as the reduction in agricultural carbon emissions. Moreover, the introduction of these technologies cannot be separated from the continued growth of bilateral trade in agricultural products between China and the United Kingdom. Sun et al. (2019) constructed a precise agricultural technology diffusion mechanism to analyze the promotion and application of soil testing and fertilizer application technology in China and found that different publicity methods adopted by the government had a significant impact on whether farmers adopted fertilizer application technology.

Implications

Based on the above analysis, this paper offers the following conclusions.

First, the agricultural sector should be opened to the outside world, and the closed and backward pattern of rural areas should be broken. The agricultural sector should be guided by agricultural industrialization and modernization elsewhere and should acquire technologies from the international agricultural market and the global agricultural industry chain, employ new types of agricultural business entities as carriers, and seek opportunities for exchanges of agricultural technology among domestic regions, departments, and industries. This is evident from the regional heterogeneity analysis in Section 4.2, which demonstrates that the eastern region not only has significantly lower carbon emissions per unit of agricultural output than the central and western regions but also has significantly better carbon emissions because it absorbs advanced overseas technologies through agricultural trade. Given the verification of our hypothesis, we provide strong empirical support for breaking down various economic, trade, social, and transportation barriers between Chinese regions and accelerating the flow of disembodied technologies for agricultural trade in China’s inland regions. Second, the central government should break local administrative monopolies and lower the threshold of administrative approval for new clean technology projects, such as pollution prevention and environmental risk adaptation services. Barriers to the flow of production factors, such as distortions in factor market allocation, should be removed. The economic connections and coordination among the eastern, central, and western regions should be strengthened, and the free diffusion of green and low-carbon technology elements developed in China should be promoted in a wider space. On the one hand, this approach can help unlock scale effects in clean agricultural production. On the other hand, the entire agricultural industry chain can be upgraded to achieve cleaner production through ripple effects from industrial interaction.

However, these policy measures may encounter obstacles in the process of practical implementation. First, the openness of the agricultural field to the outside world should be further expanded, with a focus on the economically underdeveloped central and western regions and remote rural areas. However, these regions are the most deeply influenced by China’s 5000 years of traditional farming culture and are the most difficult regions in which to realize a breakthrough. The mentality of smallholder farming and agricultural self-sufficiency has made these regions the slowest to accept foreign concepts of modern agricultural management. The backward cultural level of farmers and the deep-rooted mentality of relying on the sky for food have made it difficult for foreign low-carbon emission reduction technologies accessed through agricultural trade to be popularized on a large scale in these backward, landlocked areas. Second, there may be a serious principal-agent problem in the “green GDP” assessment mechanism formulated by the central government for local governments. Since the administrative, economic, and tax systems of China’s provinces are under the high-level jurisdiction of the central government, regional discretionary power is very weak. In addition, China’s regional economic, social and political development is extremely unbalanced. Therefore, if a single green assessment is implemented uniformly and without discrimination, local governments will likely try to meet the central government’s green performance assessment by falsifying the implementation of low-carbon emission reduction technologies in agriculture and ignoring the central government’s environmental protection requirements for the sake of the jurisdiction’s fiscal and tax revenues alone. Finally, local protectionism plays an important role in the promotion of local officials in China. To stand out in the “political promotion tournament” among other officials at the same level, local officials may try their best to hinder the economic development of rivals in other regions, which is manifested by setting up artificial geographic obstacles and industry access barriers to hinder the free flow of economic factors within competitors’ jurisdictions at the transregional level, thus cutting off the channel of technological spatial diffusion. Considering the difficulties faced in the practical implementation of these measures, we believe that the central government’s efforts in the three areas of further opening up to the outside world, enforcing strict green performance assessments on localities, and breaking up the monopoly of administrative regions will, on the contrary, result in the overconsumption of the central government’s fiscal funds, stimulate the rise of a new round of protectionism in various regions of China, and thus affect the pace of integrating China’s traditional agriculture into the global economy.

In addition, given the current changes in global trade policy, as well as the central government’s recent proposal that “a new development pattern will gradually be created whereby domestic and foreign markets can boost each other, with the domestic market as the mainstay”, it is worth pondering whether the strategy of utilizing the technological spillovers of agricultural products in foreign trade to reduce China’s agricultural carbon emissions will be effective in the long run. Since the implementation of the reform and opening up policy, China’s economic development strategy of focusing on the use of outward foreign direct investment and the development of foreign trade has provided an excellent window of opportunity for China to draw on advanced technologies from overseas markets, thereby continuing to promote the modernization and technologization of China’s agricultural production. However, following the trade war between the United States and China, the growth of China’s utilization of outward foreign direct investment (OFDI) and foreign trade has begun to take a turn for the worse, especially considering the bilateral tariff increases on each country’s products, which has led to the shrinking of agricultural imports originating from the United States into China. Accordingly, this necessitates the substitution of agricultural products from other developing countries, which has a negative impact on China’s access to low-carbon emission reduction technologies from developed countries with agrarian economies. In addition, the Chinese government has begun to propose a development strategy based on the domestic market as the mainstay, which is bound to further shrink import demand, thus affecting the sustainability of China’s acquisition of low-carbon emission reduction technologies from developed countries through agricultural trade. In the long run, this will have negative impacts on China’s agricultural sector and the environment. Because agricultural low-carbon technology originates from developed countries with rich experience in modern agricultural production, the fully competitive market and strong awareness of ecological environmental protection give developed countries a complete mechanism for investment in low-carbon technology research and development and protection of intellectual property rights, which can ensure that the application of agricultural low-carbon production technology is in line with the needs of both agricultural production and ecological environmental protection. However, China’s market economy started late, state-owned monopolies and local administrative protection are extremely entrenched, and there is no healthy and competitive market environment. Moreover, the awareness of intellectual property protection is relatively low, which makes it difficult to activate the enthusiasm of enterprises for research and development of low-carbon agricultural production technology in China. Due to the loss of the excellent window of access to foreign technology through agricultural trade, domestic agricultural production is likely to revert to smallholder economic production methods, resulting in the solidification of China’s agricultural industrial structure, the overconsumption of natural resource endowments, and severe ecological and environmental pollution pressures.

Directions for future research

The key contribution of this study is to construct different spatial weight matrices, thus associating different spatial regions in various possible forms to examine the properties of the spatial diffusion of DTC. Although this paper theoretically analyses the carriers of DTCs in terms of spatial diffusion and diffusion modes, there may be various other ways in which such diffusion occurs. For example, modern risk communication technology provides a convenient mode of face-to-face exchange among carriers of DTCs, with the implication that the spatial correlation between different regions may be greatly reduced incorrectly by the constraints of traditional geographical adjacency. Addressing this issue requires us to break through the framework set by traditional spatial econometrics when we construct the spatial weight matrix. Moreover, although DTCs have more convenient spatial transmission modes than other types of technology, whether they can produce a corresponding effect in a destination area depends largely on whether that area can provide an environment suitable for absorbing the technology. For instance, existing research highlights the important impacts of social capital on the dissemination and absorption of DTC (Maurer et al. 2011; Olcott an`d Oliver 2014). These issues can be further explored in the construction of spatial weight matrices in future research.

It is also worth noting that China, as a large country with a large population and high demand for agricultural products, has a certain advantage in absorbing low-carbon emission reduction technologies from foreign countries through trade in agricultural products in terms of market size, human capital and rural infrastructure construction. In particular, the new socialist countryside construction vigorously advocated by the Chinese government in the past two decades has given Chinese agriculture both software and hardware advantages in absorbing and utilizing foreign advanced technologies. Therefore, whether this approach of promoting the spatial diffusion of low-carbon technologies through agricultural trade can be applied to other developing countries deserves close attention. Thus, future studies should mainly focus on how developing countries with small domestic markets, weak agricultural human capital accumulation and insufficient investment in communication infrastructure in rural areas can find an effective way to absorb and utilize low-carbon emission reduction technologies suitable for their own contexts through participation in global agricultural trade.