Does Digital Transformation Promote Agricultural Carbon Productivity in China?

Abstract

Against the background of global climate change and the rapid rise of the digital economy, the digital transformation of agriculture is profoundly changing the agricultural production and operation mode with the help of digital technology, becoming a new driving force for low-carbon and sustainable development of agriculture. However, previous studies rarely examined the impact of agricultural digital transformation on agricultural low-carbon transformation from the perspective of carbon productivity. To fill this gap, this study attempts to build a theoretical analysis framework for the impact of agricultural digital transformation on agricultural carbon productivity (ACP). By using a set of panel data from 30 provinces (cities) in China from 2011 to 2019, this study explores the impact of agricultural digital transformation on ACP, as well as its conduction mechanism and the non-linear mechanism. Empirical results show that the transformation of agricultural digitalization is conducive to the promotion of ACP. A series of robustness analyses support this conclusion. The main transmission mechanisms for digital transformation to affect ACP include agricultural industrial structure upgrading, and the agricultural scale operation. In addition, with the improvement of urbanization level and rural human capital, the impact of agricultural digital transformation on ACP presents a “U” type non-linear feature of inhibition first and promotion later. Furtherly, heterogeneity analysis shows that the impact of digital transformation on ACP will vary greatly depending on the levels of ACP, the geographical location of the studied area and whether it is a main grain-producing area. This study provides a theoretical and empirical basis for the improvement of China’s agricultural carbon productivity from the perspective of the digital economy.

1. Introduction

Agriculture is a huge carbon sink system and an important global source of greenhouse gas emissions. Greenhouse gas emissions from agricultural production activities account for about 1/3 of global emissions [1]. Compared with industrial sectors, carbon emission sources of agriculture are complex, which is one of the important reasons for the high total agricultural greenhouse gas emissions, threatening the low carbon transformation of agriculture and global sustainable development. Carbon productivity is a key indicator for balancing economic growth and carbon emissions [2,3]. Improving carbon productivity has become an effective measure for the industry to reduce carbon emissions and increase efficiency [4], as well as cope with global warming [5]. In China, in 2019, the greenhouse gas emissions led by carbon dioxide in the agricultural sector reached about 17% of the total national greenhouse gas emissions [6]. In order to achieve the low-carbon transformation of agriculture [7], in 2021, the Chinese government issued the document “Opinions on the Complete, Accurate and Comprehensive Implementation of the New Development Concept to Do a Good Job in Carbon Peak and Carbon Neutrality”, which clearly proposed that to achieve agricultural carbon sequestration and efficiency enhancement is to improve agricultural carbon productivity (ACP) in essence. Therefore, a study on how to improve agricultural carbon productivity has important theoretical and practical significance for China’s agricultural low-carbon transformation and sustainable economic development.

In recent years, with the rapid rise of new generation information technologies such as 5G, big data, artificial intelligence, cloud computing, and the Internet of Things, the digital economy with modern information networks as the carrier and digital technology, digital infrastructure, digital knowledge and information as the production factors [8] has shown strong economic resilience. Therefore, many countries around the world have formulated a series of policies to develop digital technology and the digital economy. For instance, in 2010, the United States proposed the “National Broadband Plan”, and the EU released the “European Digital Agenda”. In 2014, Singapore proposed the “Smart Country” strategy [9]. In China, the “Outline of the 14th Five-Year Plan (2021–2025) for National Economic and Social Development and Vision 2035 of the People’s Republic of China” issued by the Chinese government in 2021 pointed out that China should “accelerate the construction of digital villages and promote the digital transformation of industries”. In addition, according to “Development Plan for Digital Agriculture and Rural Areas (2019–2025)”, in 2018, the contribution rate of agricultural digital transformation to the added value of agricultural output reached 7.3%, which exerted a profound impact on China’s agricultural development [10]. Thus, does the transformation of agricultural digitalization contribute to the improvement of agricultural carbon productivity? If the answer is yes, what is the conduction mechanism? Is there any non-linear characteristic and impact heterogeneity? Studying this issue is conducive to realizing the low-carbon and high-quality development of China’s agriculture under the dual influence of global warming and science and technology development, and it will also have some reference significance for other countries in the world to develop low-carbon agriculture in a high-quality way.

At present, previous studies mainly focused on the measurement and influencing factors of agricultural carbon productivity, carbon emission effects of the digital economy, etc. In terms of agricultural carbon productivity calculations, the measurement methods of agricultural carbon productivity include single-factor carbon productivity and total-factor carbon productivity. The single factor agricultural carbon productivity is defined as the inverse of agricultural carbon emission intensity, implying gross agricultural output value produced per unit of agricultural carbon emissions [11,12], which explores the amount of agricultural economic growth taking carbon emissions into consideration. The total-factor agricultural carbon productivity follows the TFP analysis paradigm [13,14], which regards carbon emissions as the unexpected output and can fully explore agricultural development quality. In terms of the appraisement method of agricultural total factor productivity, data envelopment analysis (DEA) and stochastic frontier analysis (SFA) methods are widely used by scholars [15,16]. In terms of the influencing factors, land, capital, machinery, and other production factors can exert a direct influence on agricultural carbon performance [17,18]. Additionally, agricultural, industrial structures, planting structures, rural power consumption, rural human capital, and agricultural disaster all have significant correlations with agricultural carbon productivity [19,20]. Agricultural policies, technological innovation and urbanization level are also vital influencing factors that cannot be ignored [11]. In terms of the carbon emission effect of the digital economy, some studies believe that the development of digital technology and the digital economy can reduce the carbon emission intensity [21], and some others believe that the development of digital technology and the digital economy has aggravated carbon emissions [22,23], which is not conducive to the reduction of carbon emissions per capita [24]. Additionally, it is also believed by some scholars that the impact of the digital economy on carbon emission intensity is characterized by an inverted “U” curve that increases first and then decreases [25,26], and the spatial spillover effect [27,28]. The improvement in carbon emission performance indicates the reduction of carbon emissions. Some studies find that the digital economy improves urban carbon emission performance by changing the intensity of energy use, the scale of energy consumption and urban greening. Under different energy consumption structures, government intervention, etc., the impact between the two also shows a non-linear feature [29]. However, previous studies have not thoroughly and carefully investigated the emission reduction and efficiency increase consequences of agricultural digital transformation. Some scholars built an indicator system covering agricultural production factors and gross agricultural output value to investigate the impact of agricultural digital transformation on green agricultural development, and found that agricultural digital transformation can promote the improvement of agricultural green development level, which shows a non-linear feature of “marginal efficiency increases” [30]. However, this study failed to empirically analyze the transmission mechanism of the impact between the two and failed to investigate the non-linear characteristics of the impact between the two based on external factors.

To sum up, the measurement and influencing factors of agricultural carbon productivity, and the carbon emission effect of the digital economy have been fully emphasized and studied. However, few studies focus on the agricultural sectors or discuss the impact of the digital economy on agricultural carbon productivity. Therefore, this study attempts to use a set of panel data from 30 provinces in China from 2011 to 2019 to investigate the impact of agricultural digital transformation on agricultural carbon productivity to make up for the loopholes in existing research. The possible marginal contributions of our study are as follows: (1) Our study focuses on the agricultural field, provides a new vision of agricultural digital transformation for agricultural low-carbon transformation, and enriches relevant research on digital economy and high-quality agricultural development. (2) We reveal two important transmission mechanisms of agricultural digital transformation affecting agricultural carbon productivity: industrial structure upgrading and the scale operation of agriculture, opening a “black box” between the two. (3) This study explores the non-linear characteristics of the impact of agricultural digital transformation on agricultural carbon productivity under the constraints of urbanization and human capital. (4) We also reveal the differences in the impact of agricultural digital transformation on agricultural carbon productivity from the perspectives of agricultural carbon productivity level, geographical location, and whether it is a main grain-producing area.

2. Theoretical Analysis and Hypothesis Formulation

2.1. The Direct Impact of Agricultural Digital Transformation on ACP

Agricultural carbon productivity refers to the ratio of total agricultural output to all agricultural production factors (land, capital, labor, resources, etc.), taking into account agricultural carbon emissions and other unexpected outputs [31]. It requires less input from agricultural production factors and lowers agricultural carbon emissions to obtain larger agricultural economic output, which directly reflects the improvement of agricultural production and operation efficiency and the reduction of carbon emissions. As digital economy is characterized by high innovation, strong penetration and full network [30], agricultural digital transformation can promote the deep integration of agricultural digital knowledge and information with traditional agriculture by virtue of digital infrastructure and digital technology, optimize production and operation processes, change the traditional agricultural development model based on extensive resources, and improve agricultural carbon productivity. With the integration of digital knowledge, information and traditional agricultural production factors, on the one hand, the digital transformation of agriculture can accurately and effectively embed science and technology, talents, information, capital and other factors into the agricultural industry business chain, thus reducing the information asymmetry between the supply and demand of various agricultural operators’ production factors, and improving the allocation efficiency and production and operation efficiency of production factors [32]. On the other hand, the transformation of agricultural digitalization can give full play to the inherent advantages of information technology, realize information transmission and technology sharing across time and space, thus reducing unnecessary activities of agricultural labor, and helping save more energy and reduce carbon emissions [33]. At the same time, with the help of communication and network technology, automation technology, blockchain, big data, etc., agricultural digital transformation realizes the accurate proportion and effective use of chemical fertilizers [34], pesticides [35], seeds [36] and other production factors, and conducts accurate and dynamic monitoring of agricultural resources and environment [37], so as to reduce unnecessary chemical pollution and agricultural carbon emissions. In addition, the agricultural digital transformation makes it convenient for farmers to use digital communication devices, such as smart phones to obtain green production information and green agricultural technologies publicized by the government [38], guide farmers to form a low-carbon concept, and carry out low-carbon production and operation. Therefore, we propose the following hypothesis:

Hypothesis 1 (H1):

The transformation of agricultural digitalization has a positively promoting effect on ACP.

2.2. Conduction Mechanisms of Agricultural Digital Transformation Promoting ACP

Agricultural digital transformation indirectly promotes ACP may through two action mechanisms, including agricultural industrial structure upgrading and agricultural scale operation. Figure 1. shows the conduction mechanisms of agricultural digital transformation affecting ACP.

First, studies have confirmed that the upgrading of agricultural industrial structure improves ACP through inter-industry technology spillover, diffusion and agricultural division of labor [39]. The digital economy can accelerate industrial transformation and upgrading by promoting rational allocation of resources, transforming traditional industries, and helping generate new industries [40], and reducing urban carbon emissions [41]. Digital transformation of agriculture can help transform and reconstruct the entire agricultural industry chain and industrial cluster by the use of cutting-edge digital technology, so as to achieve a reasonable allocation of agricultural resources and promote the transformation of agricultural industrial structure to a high added value, low cost and low carbon model [42].

Second, compared with the fragmented and decentralized agricultural operation mode, the agricultural scale operation can scientifically and reasonably optimize and match production factors [43]. For example, it can help realize scientific management and utilization of land and soil, and can accurately match pesticides and fertilizers. In addition, compared with ordinary farmers, farmers engaged in large-scale operations also have more green and low-carbon agricultural production and management ideas, and tend to adopt green and low-carbon technologies for agricultural production and management. Moreover, previous studies have implied that agricultural digital transformation has a significant positive role in promoting agricultural scale operation. On the one hand, the popularization of communication networks and the use of smartphones by farmers help reduce the transaction cost of agricultural land transfer and facilitate land transfer in rural China, which contributes to the formation of family farms and scale operation [44]. On the other hand, informatization and networking help reduce communication costs, broaden social networks, and improve the bargaining power of farmers to obtain means of production, thus reducing agricultural production costs, and promoting agricultural scale operation. Furthermore, the construction of the e-commerce platform facilitates the sales of agricultural products, increases the sales income of agricultural products and, in turn, encourages farmers to carry out agricultural scale operation [45].

Therefore, we propose the following hypothesis:

Hypothesis 2 (H2).

Agricultural industrial structure upgrading, and agricultural scale operation are two effective conduction mechanisms for agricultural digital transformation to promote ACP.

2.3. The Non-Linear Characteristics of the Impact of Agricultural Digital Transformation on ACP

Although it is known that the digital transformation of agriculture brings profound changes to agricultural production, there are still great difficulties in the popularization of digital technology, especially in developing countries [33]. According to Lio and Liu [46], rural human capital is the key factor in bridging the digital divide. Wang and Ran [47] argued that in areas with a low level of rural human capital, due to the low educational level of local rural labor, they correspondingly lack the ability to apply rural digital resources and digital technologies, which leads to the weak role of agricultural digitalization. Besides, some scholars adopted a panel threshold model to study the role of human capital in the causal connection between the digital economy and carbon emissions [29], and proposed that with the enhancement of human capital, the role of the digital economy will become greater. Meanwhile, studies have proved that the level of urbanization also has an impact on the role of digital agriculture [48]. The promotion effect of agricultural carbon productivity through agricultural digital transformation cannot be achieved without the support of talent, capital and technology. The improvement of the urbanization level plays a radiating role in promoting rural development, which helps the industry to provide feedback on agriculture and provides corresponding talents and technical support for agricultural development. Obviously, the impact of agricultural digital transformation on agricultural carbon productivity will also be constrained by the level of urbanization.

Based on the above analysis, we propose the following hypothesis:

Hypothesis 3 (H3).

With the improvement of urbanization level and rural human capital level, the impact of agricultural digital transformation on ACP presents a “U” type non-linear characteristic of inhibition first and promotion later.

3. Research Design

3.1. Model Setting

In order to test Hypothesis 1, we construct the benchmark regression model:

$$AC{P}_{it}={\beta }_0+{\beta }_{1}Di{g}_{it}+{\beta }_2 X_{it}+{\lambda }_i+{\mu }_t+{\epsilon }_{it}$$

(1)

where$AC{P}_{it}$refers to agricultural carbon productivity. Dig_it is the agricultural digitalization transformation index of area i in year t, and X_it is a set of control variables affecting ACP. λ_i is the regional fixed effect. μ_t is the time-fixed effect. ε_it is the random perturbation term.

In order to verify Hypothesis 2, referring to Baron and Kenny [49], on the basis of Equation (1), we establish the following models:

$$Z_{it}={\alpha }_0+{\alpha }_{1}Di{g}_{it}+{\alpha }_2 X_{it}+{\lambda }_i+{\mu }_t+{\epsilon }_{it}$$

(2)

$$AC{P}_{it}={\rho }_0+{\rho }_{1}Di{g}_{it}+{\rho }_2 Z_{it}+{\rho }_3 X_{it}+{\lambda }_i+{\mu }_t+{\epsilon }_{it}$$

(3)

where Z_it refers to an intermediary variable, indicating the conduction mechanism. X_it is a vector including a series of control variables. λ_i is the regional fixed effect. μ_t is the time-fixed effect. ε_t represents the random perturbation terms.

To verify Hypothesis 3, according to Hansen [50], we build a panel threshold model. The panel threshold model with a single threshold is as follows:

$$AC{P}_{it}={\theta }_0+{\theta }_{1}Di{g}_{it}\cdot I({\gamma }_{it}\le {\eta }_1)+{\theta }_{2}Di{g}_{it}\cdot I({\gamma }_{it}\ge {\eta }_1)+{\theta }_3{X}_{it}+{\lambda }_i+{\mu }_t+{\epsilon }_{it}$$

(4)

where γ_it is the threshold variable. I() is an indicative function. θ₀ refers to the global intercept, and ε_it refers to the idiosyncratic error term. Equation (4) also can be extended to the multi-threshold case.

3.2. Variable Selection

3.2.1. Explained Variable

Referring to the practice of Liu et al. [31], we build an ACP indicator system from the perspective of total factor productivity (

Table 1. ACP index system.

Index Category	Index Name	Evaluating Indicator
Input index	Labor input	The number of primary industry employees
	Capital input	Agricultural fixed capital stock
	Land input	Total crops sown area
	Other input	Application amount of agricultural chemical fertilizer
		Pesticide input
		Total power of agricultural machinery
		The input of plastic mulch
Output index	Expected output	Total output value of agriculture–forestry–stockbreeding–fishery
	Unexpected output	Agricultural carbon emissions

). In addition, we combine the research methods of Tone and Tsutsui [51], Pastor and Lovell [52], and Oh [53], use the variable return to scale, super-efficient EBM (epsilon-based measure) model and GML (Global Malmquist Luenberger) index, namely, the EBM-GML index to measure ACP.

3.2.2. Explanatory Variable

So far, the construction of indicators for agricultural digital transformation has not been unified. Drawing on Mu and Ma [54], and Zhang and Bai [55], the digital transformation of agriculture in this study includes four basic parts: the construction of agricultural information infrastructure, agricultural digital service supply, agricultural digital transactions, and output from agricultural digitalization. First, the core of digital infrastructure construction is the construction of information infrastructure, including Internet penetration rate in rural areas, the use of smartphones, the construction of rural smart weather stations and the investment of fixed assets. Second, the measurement method of agricultural digital service supply is reflected by the supply of rural information services. Third, agricultural digital transactions include the scale of digital transactions in the sale of agricultural products, the popularity of online payment and the establishment of e-commerce platforms. Finally, we use the contribution of the digital economy in the added value of the primary industry to measure its output. The index system of agricultural digital transformation is shown in

Table 2. Index system of agricultural digital transformation.

Index Category	Index Name	Evaluating Indicator
Digital infrastructure construction	The internet penetration rate in rural areas.	$\frac{Number of rural broadband access households }{Number of rural households}$
	Penetration rate of smartphones in rural areas.	Annual mobile phone ownership of rural residents per 100 households
	Number of rural smart weather stations.	Number of regional agrometeorological observation stations
	Investment scale of agricultural fixed assets.	$\frac{Investment in fixed assets of agriculture, forestry, animal husbandry and fishery }{Total regional investment in fixed assets }$
Digital service supply	Provision of rural information technology services.	Average service population of rural postal outlets
Agricultural digital transactions	Digital transaction scale of agricultural products.	E-commerce sales and procurement turnover in rural areas
	Network payment level in rural areas.	Rural digital finance inclusive index
	Construction of e-commerce platform in rural areas.	Number of “Taobao” villages
Output from agricultural digitalization	Scale of agricultural digital output value.	Added value of the digital economy in primary industry

. On this basis, this study uses the entropy method to measure the agricultural digital transformation index.

3.2.3. Conduction Mechanism Variable

(1)

Agricultural industrial structure upgrading (Ais): According to Jin and Jin [39], this study uses the ratio of the output value of productive agricultural services to the output value of the primary industry to characterize the upgrading of the agricultural industrial structure.

(2)

Agricultural scale operation (Scale): We adopt agricultural scale efficiency de composed by comprehensive agricultural technical efficiency to represent agricultural scale operation.

3.2.4. Threshold Variable

(1)

Rural human capital (Edu): Rural human capital is an important factor in agricultural economic growth and sustainable development. The higher the educational level of farmers, the easier they are to accept new technologies, and promote agricultural productivity. Therefore, we set Edu not only as a threshold variable but also as a control variable. We adopt the average years of education for the rural labor force to measure the human capital in rural areas.

(2)

Urbanization level (Urb): According to previous theoretical analysis, urbanization level is another threshold variable, and it is measured by the ratio of the urban population to the total regional population.

3.2.5. Other Control Variable

(1)

Fiscal support (Fis): The development of agriculture depends on financial support to a certain extent [56]. The extension of agricultural green low-carbon technology is closely related to fiscal support. Therefore, fiscal support measured by the proportion of agricultural fiscal expenditure in total fiscal expenditure in each region is included in the control variables.

(2)

Natural Disaster (Dis): Agricultural productivity is often affected by natural disasters. Therefore, we take the natural disaster into account and use the area of crops affected in each region to measure Dis.

(3)

Agricultural mechanization level (Machine): Agricultural mechanization is the symbol of modern agricultural development [57]. The substitution of machinery for the labor force can effectively improve agricultural production efficiency. The machine is measured by agricultural machinery power per unit planting area.

(4)

Agricultural electric power input (Elec). Electric power facilities are related to the use of energy and are the main source of carbon emissions [58]. We use the proportion of rural electricity consumption scale in regional electricity consumption to measure Elec.

(5)

Agricultural industrial agglomeration (Agg): Agg is another main element to influence ACP. For one thing, it can exert technology spillover and scale effect, which will help promote agricultural carbon productivity. For another, it may also produce a “crowding” effect, which is not conducive to the improvement of agricultural carbon productivity [59]. Agg is calculated by using the location entropy method.

(6)

Average temperature (Tem): The temperature affects the growth cycle of crops, thus affecting the greenhouse gas emissions produced by the planting industry. We convert the daily average temperature to the year as the proxy variable of Tem.

The definitions of all variables are given in

Table 3. Variable definition.

	Variable Name	Symbol	Variable Definition
Dependent variable	Agricultural carbon productivity	ACP	The cumulative index of agricultural carbon productivity in Table 1
Independent variable	Agricultural digital transformation	Dig	A comprehensive index of agricultural digital transformation in Table 2
Control variables	Rural human capital	Edu	Average education year of rural labors
	Fiscal expenditure on agriculture	Fis	The proportion of agricultural fiscal expenditure in total regional fiscal expenditure
	Natural disasters	Dis	Affected area of crops
	Agricultural machinery power	Machine	$\frac{ Total power of agricultural machinery}{Total planting area}$
	Rural electric power facilities	Elec	$\frac{Village electricity consumption}{Total regional electricity consumption }$
	Average temperature	Tem	Convert the daily average temperature to the year
	Agricultural industrial agglomeration	Agg	Calculated using the location entropy method
Mechanism variables	Agricultural industrial structure upgrading	Ais	The ratio of the output value of productive agricultural services to the output value of the primary industry
	Agricultural scale operation	Scale	Agricultural scale efficiency decomposed by agricultural comprehensive technical efficiency
Threshold variables	Rural human capital	Edu	Average education year of rural labors
	Urbanization	Urb	Ratio of urban population to regional total population

.

3.3. Data Source and Descriptive

In order to analyze the impact of agricultural digital transformation on ACP, after considering the data availability, this study uses the panel data from 30 provinces (cities, autonomous regions) in China from 2011 to 2019 to conduct the empirical tests. All data above comes from China Statistical Yearbook (2012–2020), China Rural Statistical Yearbook (2012–2020), China Environment Statistical Yearbook (2012–2020), China Population and Employment Statistical Yearbook (2012–2020), China Agricultural Machinery Yearbook (2012–2020), EPS database and Carbon Emission Accounts and Datasets.

Table 4. Descriptive statistics.

Variable	N	Mean	S.D.	Min	P25	P50	P75	Max
ACP	270	1.410	0.390	0.830	1.160	1.330	1.550	3.430
Dig	270	0.140	0.090	0.060	0.100	0.120	0.150	0.910
Edu	270	8.030	0.520	6.600	7.760	8.060	8.300	9.910
Dis	270	0.150	0.120	0.000	0.070	0.130	0.210	0.620
Fis	270	0.500	0.680	0.130	0.240	0.290	0.420	5.710
Machine	270	6.350	2.350	2.640	4.480	5.680	7.780	13.860
Elec	270	0.120	0.120	0.010	0.050	0.080	0.120	0.700
Tem	270	2.530	0.480	0.950	2.250	2.720	2.850	3.240
Agg	270	1.180	0.600	0.060	0.800	1.180	1.460	3.460
AIS	270	0.040	0.020	0.010	0.030	0.040	0.0500	0.120
Scale	270	0.830	0.150	0.480	0.720	0.810	1.000	1.000

reports descriptive statistics for the main variables.

4. Empirical Analysis

4.1. Temporal and Spatial Variation of ACP

According to the calculation results of ACP in China, we use ArcGIS to draw a spatial-temporal evolution map of the ACP values from China’s 30 provincial administrative regions for the years 2011, 2014, 2017 and 2019, as displayed in Figure 2.

In order to identify the temporal and spatial distribution pattern of ACP in China, we divide the estimated values of ACP level in the samples into five levels: [0.8285–1.3486], [1.3486–1.8687], [1.8687–2.3888], [2.3888–2.9089] and [2.9089–3.4290]. It can be seen from Figure 2 that in 2011, the ACP levels of all provinces (cities) in China were relatively low, and they were all in the lower-level range of [0.8285–1.3486]. In 2014, the levels of ACP in nine provinces (cities) of Yunnan, Sichuan, Guizhou, Shaanxi, Henan, Beijing, Shandong, Jiangsu, and Inner Mongolia presented a significant increase and took the lead in entering the second level of [1.3486–1.8687]. In 2017, the ACP of Jiangsu province became the third level of [1.8687–2.3888] and which in Guizhou and Heilongjiang province has become the fourth level of [2.3888–2.9089]. By 2019, the ACP in western provinces is coming into the highest level of [2.9089–3.4290]. There are eight provinces with ACP at the third level of [1.8687–2.3888]. In addition, only Inner Mongolia and Jilin provinces still have the lowest level of ACP. From the spatial-temporal variation map of these four years, it is found that the ACP in China shows an obvious growth trend and which in Southwest China is relatively higher. Obviously, ACP in different provinces almost presents a rising tendency and there are apparent regional differences among provinces.

4.2. Benchmark Regression Results

Table 5. Benchmark regression results.

Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Dig	2.2059 ***	1.4613 ***	1.3913 ***	1.4395 ***	1.4718 ***	1.4173 ***	1.3036 ***	1.4665 ***
	(8.699)	(5.758)	(5.502)	(5.736)	(5.840)	(5.520)	(5.057)	(5.926)
Edu		0.7252 ***	0.6488 ***	0.5558 ***	0.5549 ***	0.5483 ***	0.5042 ***	0.3953 ***
		(7.082)	(6.113)	(4.978)	(4.976)	(4.912)	(4.512)	(3.643)
Dis			−0.4438 **	−0.4251 **	−0.4058 **	−0.4048 **	−0.4185 **	−0.3660 **
			(−2.426)	(−2.347)	(−2.235)	(−2.230)	(−2.331)	(−2.138)
Fis				0.1209 **	0.1046 **	0.1177 **	0.1037 **	0.1485 ***
				(2.456)	(2.055)	(2.253)	(1.997)	(2.957)
Machine					0.0231	0.0244	0.0254	0.0224
					(1.248)	(1.316)	(1.385)	(1.282)
Elec						−0.5458	−0.5545	−0.6345
						(−1.102)	(−1.132)	(−1.361)
Tem							0.7770 **	0.2300
							(2.526)	(0.736)
Agg								0.6054 ***
								(5.028)
Constant	1.0992 ***	−4.6173 ***	−3.9257 ***	−3.2494 ***	−3.3881 ***	−3.2777 ***	−4.8705 ***	−3.3516 ***
	(27.587)	(−5.715)	(−4.624)	(−3.675)	(−3.806)	(−3.661)	(−4.481)	(−3.110)
Fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observation	270	270	270	270	270	270	270	270
R2	0.5278	0.6100	0.6195	0.6290	0.6314	0.6333	0.6431	0.6781
R2-adjust	0.4686	0.5592	0.5681	0.5771	0.5781	0.5785	0.5879	0.6268
F-value	75.6648	70.6909	50.0573	39.8483	32.2652	27.1146	24.6867	27.0115

Notes: t values are in parentheses. **, and *** indicate the significance at the 5%, and 1% levels, respectively.

shows the benchmark regression results of the impact of agricultural digital transformation on ACP. Column (1) shows the regression results without any control variables. Columns (2)~(8) show the regression results when control variables such as Edu, Dis, Fis, Machine, Elec, Tem, Agg, etc., are gradually added. It can be concluded that whether control variables are added or not, the transformation of agricultural digitalization significantly promotes the improvement of ACP. According to column (8), for every 1% increase in agricultural digital transformation, ACP will increase by 1.4665%. This result confirms the reliability of Hypothesis 1 proposed above. In addition, taking control variables into account, Edu is significantly positively correlated with ACP, indicating that the higher the education level of agricultural labor, the more conducive to improving ACP. The estimated coefficient of Dis is −0.3660, with a significance level of 1%, indicating that natural disasters have a negative impact on ACP, which is consistent with previous studies [60] The estimated coefficient value of Fis is 0.1485, with a significance level of 1%, showing that fiscal support for agriculture can significantly promote ACP. In addition, Agg contributes to the improvement of ACP, implying that Agg is conducive to agricultural technology spillovers and knowledge dissemination, and is conducive to the low-carbon transformation of agriculture. Finally, Elec, Machine and Tem have no significant effect on ACP.

4.3. Robustness Analysis

In order to test the reliability of the results of agricultural digital transformation to improve ACP in column (8) of Table 5, we adopt the following four methods to examine the robustness of the benchmark results. First, the single factor carbon productivity (agricultural added value of unit carbon emissions) is used to measure ACP. The regression results are shown in column (1) of

Table 6. Robustness test of the benchmark regression results.

Variable	(1)	(2)	(3)	(4)
Dig	3.4005 ***	1.3782 ***	2.8382 ***	0.3721 ***
	(7.134)	(6.312)	(6.043)	(2.742)
L.ACP				0.9252 ***
				(29.902)
Controls	Yes	Yes	Yes	Yes
Constant	−5.1531 **	−3.5740 ***	−7.5309 ***	−0.5360 ***
	(−2.483)	(−3.759)	(−3.682)	(−3.006)
Fixed effect	Yes	Yes	Yes	Yes
Observation	270	270	270	240
R2	0.8500	0.6904	0.6432	-
R2-adjust	0.8260	0.6410	0.5863	-
F-value	43.0216	30.5560	23.6052	10,491.99

Notes: t values are in parentheses. **, and *** indicate the significance at the 5%, and 1% levels, respectively.

. Second, ACP is measured by the super-efficient EBM model with constant returns to scale and GML index. The regression results are shown in column (2) of Table 6. Third, ACP is measured again by using the super-efficient SBM model with variable returns to scale and GML index. The regression results are demonstrated in column (3) of Table 6. Fourth, the dynamic panel model has been used to re-estimate the results. Column (4) of Table 6 shows the estimation results. It can be concluded that the transformation of agricultural digitalization can still significantly promote ACP. These results imply that the estimation result of agricultural digital transformation improving ACP in column (8) of Table 5 is reliable. Hypothesis 1 has been confirmed again.

4.4. Conduction Mechanism Regression Results Analysis

Table 7. Conduction mechanism regression results.

Independent Variable	Industrial Structure Upgrading		Agricultural Scale Operation
	(1)	(2)	(3)	(4)
Dig	0.0381 ***	1.1927 ***	0.4494 ***	1.2712 ***
	(5.016)	(4.684)	(5.487)	(4.873)
Ais		7.1849 ***
		(3.437)
Scale				0.4346 **
				(2.209)
Tech
Controls	Yes	Yes	Yes	Yes
Constant	−0.1020 ***	−2.6188 **	−0.0102	−3.3472 ***
	(−3.083)	(−2.437)	(−0.029)	(−3.132)
Fixed effect	Yes	Yes	Yes	Yes
Observation	270	270	270	270
R2	0.8852	0.6938	0.7438	0.6848
R2-adjust	0.8669	0.6434	0.7030	0.6330
F-value	15.2509	26.4423	17.1062	24.9538
Intermediary effect	0.2737		0.1952
$\frac{\mathrm{Intermediary}\mathrm{effect}}{\mathrm{Total}\mathrm{effect}}$	18.66%		13.31%
Sobel Test	2.835		2.049
	[0.0045]		[0.0404]

Notes: t values are in parentheses. **, and *** indicate the significance at the 5%, and 1% levels, respectively.

shows the regression results of the two conduction mechanisms, including agricultural industrial structure upgrading and agricultural scale operation of agricultural digital transformation promoting ACP. Columns (1) and (2) of Table 7 are the regression results of the action mechanism of agricultural industrial structure upgrading. It can be found that agricultural digital transformation significantly promotes the upgrading of agricultural industrial structure, and Ais in column (2) can significantly promote ACP, and the regression coefficient of Dig decreases compared with the regression coefficient of Dig in column (8) of Table 5, which indicates that agricultural digital transformation can help improve ACP by promoting the upgrading of agricultural industrial structure. Columns (3) and (4) in Table 7 are the regression results of the conduction mechanism of agricultural scale operation. It is concluded that the agricultural digital transformation is conducive to promoting agricultural scale operation, the Scale in column (4) significantly promotes ACP, and the regression coefficient of Dig is smaller than the estimated coefficient of Dig in column (8) of Table 5, indicating that the agricultural digital transformation improves ACP by promoting agricultural scale operation.

Hypothesis 2 is proved to be true here. In addition, the Sobel test results in Table 7 furtherly confirm the existence of these two conduction mechanisms, and the intermediary effects of agricultural industrial structure upgrading, and agricultural scale operation are 0.2737 and 0.1952, respectively, accounting for 18.66% and 13.31%of the total improvement effect.

4.5. Non-Linear Regression Results Analysis

Table 8. Threshold value estimation results.

Threshold Variable	Number	Threshold Value	F-Stat	Prob
Urban	Single	0.3648	31.57 *	0.0880
	Double	0.3648	26.32	0.1960
		0.4227
Edu	Single	7.2517	46.86 ***	0.0020
	Double	7.2517	21.97	0.1200
		7.9626

Notes: *, and *** indicate the significance at the 10%, and 1% levels, respectively.

and

Table 9. Threshold regression results.

Variable	Threshold Value: Urban	Threshold Value: Edu
	(1)	(2)
Threshold variable	Urban	Edu
Threshold value	0.3648 *	7.2517 ***
Dig #Regime1	−9.7180 ***	−5.3169 ***
	(−4.565)	(−4.584)
Dig #Regime2	1.4793 ***	1.4677 ***
	(6.316)	(6.358)
Controls	Yes	Yes
Constant	−3.2216 ***	−2.9437 ***
	(−3.158)	(−2.922)
Fixed effects	YES	YES
Observation	270	270
R2	0.5381	0.5514
R2-adjust	0.4621	0.4776
F-value	29.9038	31.5506

Note: t values are in parentheses.; *, and *** indicate the significance at the 10%, and 1% levels, respectively.

are the non-linear mechanism tests under the constraints of urbanization and human capital of ACP affected by the agricultural digital transformation. Table 8 shows the threshold test results of urbanization and human capital, and Table 9 shows the threshold regression results of the impact of agricultural digital transformation on ACP. It can be seen from Table 8, urbanization and human capital all have a single threshold effect in the causal impact of agricultural digital transformation on agricultural carbon productivity, and the threshold values are 0.3648 and 7.2517, respectively. Table 9 shows that when urbanization is less than 0.3648, the estimated coefficient of Dig is −9.7180, which is significant at the level of 1%; When urbanization exceeds 0.3648, the coefficient of Dig is 1.4793, which is significant at the 1% level. When the human capital is less than 7.2517, the estimated coefficient of the impact of agricultural digital transformation on ACP is −5.3169, which is significant at the level of 1%; When the estimated value of Edu exceeds 7.2517, the coefficient of Dig is 1.4677, passing the significance level of 1%. This shows that only when Urb and Edu span 0.3648 and 7.2517, respectively, can agricultural digital transformation promote ACP. The impact of agricultural digital transformation on agricultural carbon productivity shows a “U” type non-linear characteristic of inhibition first and promotion later. Here, Hypothesis 3 is verified.

4.6. A Further Analysis: Impact Heterogeneity

The above comprehensively and systematically examined the impact of agricultural digital transformation on ACP and its two important intermediate mechanisms. In addition, under the constraints of urbanization and human capital, the non-linear characteristics of the impact of agricultural digital transformation on ACP are also examined. This part will furtherly reveal the differences in the impact of agricultural digital transformation on ACP from the perspectives of agricultural carbon productivity level, geographical location, and whether it is the main grain-producing area.

4.6.1. Heterogeneity of Different ACP Level

We divide ACP into five quantiles: 10%, 25%, 50%, 75% and 90%, and then regress the benchmark model again under different ACP quantiles. The regression results are shown in

Table 10. Regression results when ACP is at different levels.

Variable	10%	25%	50%	75%	90%
	(1)	(2)	(3)	(4)	(5)
Dig	1.2007 ***	1.2958 ***	1.4174 ***	1.6208 ***	1.8929 **
	(3.164)	(4.598)	(5.430)	(3.402)	(2.130)
Controls	Yes	Yes	Yes	Yes	Yes
Fixed effects	Yes	Yes	Yes	Yes	Yes
Observation	270	270	270	270	270

Note: t values are in parentheses; **, and *** indicate the significance at the 5%, and 1% levels, respectively.

. It is found that at different quantile levels of 10%, 25%, 50%, 75% and 90%, the estimation coefficients of Dig are 1.2007, 1.2958, 1.4174, 1.6208 and 1.8929, respectively, which are all significant at the 1% level. This shows that with the improvement of the ACP level, the promotion effect of agricultural digital transformation is more obvious.

4.6.2. Location Heterogeneity

Table 11. Regression results of regional heterogeneity.

Variable	Geographical Location				Main Grain Producing Areas or Not
	Eastern	Central	Western	Northeastern	Yes	No
	(1)	(2)	(3)	(4)	(5)	(6)
Dig	1.1794 ***	10.4809 ***	7.8823 ***	2.6791	2.1622 ***	1.4196 ***
	(4.900)	(7.076)	(4.543)	(1.511)	(5.281)	(4.493)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
Fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
observation	90	54	99	27	108	162
R2	0.7416	0.8311	0.7671	0.9547	0.7638	0.6823
R2-adjust	0.6806	0.7762	0.7147	0.9264	0.7128	0.6239
F-value	15.6104	19.2982	16.0979	15.2203	13.8745	17.7857

Note: (1) The eastern region includes 10 provinces and cities (Beijing, Tianjin, Hebei, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Guangdong and Hainan). The central region includes six provinces (Shanxi, Anhui, Jiangxi, Henan, Hubei, and Hunan), the western region includes 12 provinces (Inner Mongolia, Guangxi, Sichuan, Chongqing, Guizhou, Yunnan, Tibet, Shaanxi, Gansu, Qinghai, Ningxia, and Xinjiang) and the northeastern region covers 3 provinces (Heilongjiang, Jilin and Liaoning). (2) The main grain-producing area includes 13 provinces (Liaoning, Hebei, Shandong, Jilin, Inner Mongolia, Jiangxi, Hunan, Sichuan, Henan, Hubei, Jiangsu, Anhui and Heilongjiang). (3) t values are in parentheses and *** indicates the significance at 1% level.

shows the regression results of regional heterogeneity. First of all, this study divides 30 provinces (cities and autonomous regions) in China into four regions: the eastern, the central, the western and the northeastern regions, and examines the impact of agricultural digital transformation on ACP in each region. The regression results are shown in columns (1) to (4) of Table 11.

From columns (1) to (4), the transformation of agricultural digitalization significantly promotes the ACP in the eastern, central and western areas, but fails to promote ACP in the northeastern region. The estimated coefficient of Dig in the central region is the biggest.

Columns (5) and (6) of Table 11, respectively, show the impact of agricultural digital transformation on ACP in main grain-producing areas and non-major grain-producing areas. We can find that the transformation of agricultural digitalization can improve ACP in both major and non-major grain-producing areas. Compared with the non-main gain-producing areas, the agricultural digital transformation in the main grain-producing areas plays a greater role.

5. Discussion

5.1. Research Conclusions

Low carbon transformation of agriculture is an important part of reducing global carbon emissions and coping with global warming. At the same time, digital technology is profoundly changing the way of agricultural production and operation and injecting new momentum into the low-carbon and sustainable development of agriculture. In the context of global warming and the rise of a new generation of information technology, this study, based on the theoretical analysis of the impact of agricultural digital transformation on ACP, empirically tests the impact of agricultural digital transformation on ACP and its indirect mechanism using China’s 2011–2019 provincial panel data, and investigates the non-linear characteristics of the impact of agricultural digital transformation on ACP based on different levels of urbanization and human capital. In addition, this study furtherly reveals the differences in the impact of agricultural digital transformation on ACP from the perspectives of agricultural carbon productivity level, geographical location and whether it is the main grain-producing area. A series of useful conclusions have been drawn, which can provide China’s theoretical basis and practical reference for other countries around the world to effectively use the new generation of information technology to promote the low-carbon transformation of agriculture. The research conclusions of this study are as follows:

First, the transformation of agricultural digitalization can effectively improve ACP. Whether it is to replace the evaluation and measurement methods of ACP or to use dynamic panel models to control endogenous problems, this conclusion is still robust, which verifies Hypothesis 1 above is reliable. In recent years, as the Chinese government has incorporated the digital economy into its “national strategy”, China’s agricultural digital transformation has achieved unprecedented results and has formed a digital agriculture model dominated by market demand, guided by government policies, and participated by all kinds of subjects. This model not only improves agricultural production and operation efficiency and extends the agricultural industry chain but also promotes the low-carbon transformation of agriculture. It truly realized “controllable production, traceable quality and measurable environment” [30].

Second, agricultural digital transformation can promote ACP through upgrading of agricultural industrial structures and realizing agricultural scale operation. This conclusion validates Hypothesis 2 above. The improvement of agricultural carbon productivity is reflected in the improvement of production and operation efficiency, and the reduction of carbon emissions, which are realized through the improvement of factors utilization efficiency and the adoption of low carbon technology, respectively. The transformation of agricultural digitalization can achieve the improvement of factory utilization efficiency and agricultural low-carbon technology progress through the upgrading of agricultural industrial structures and realizing agricultural scale operation. Different from previous studies [30], this study reveals and demonstrates these two important conduction mechanisms.

Third, only when urbanization and human capital respectively cross the threshold of 0.3648 and 7.2517, can agricultural digital transformation promote ACP. When they are lower than the corresponding threshold, agricultural digital transformation is not conducive to ACP. This conclusion verifies Hypothesis 3. Due to the different levels of urbanization and rural human capital in different countries, the threshold values of urbanization and rural human capital in different countries are also different; this should be considered by policy makers.

Fourth, when the level of ACP is relatively high, the promotion of agricultural digital transformation on ACP is more obvious. Compared with the eastern, western and northeastern regions, the agricultural digital transformation can promote the ACP of the central region more. Compared with the non-main grain-producing regions, the agricultural digital transformation can promote the ACP of the main grain-producing regions more. Possible reasons are with the improvement of ACP level, agricultural production and operation will enter into a stage of low-carbon and efficient development. At this stage, agricultural digital technology will penetrate multiple links of the agricultural industry chain, and the substitution effect on traditional agricultural production factors is more obvious, which is more conducive to the improvement of ACP. This practical law should be paid attention to. In addition, many provinces with large agricultural scales are concentrated in the central region, such as Henan, Anhui, Jiangxi, etc. These provinces adopt the strategy of giving priority to agricultural development, and the scale of agricultural fiscal expenditure in these provinces is large, which provides strong support for the promotion of agricultural digital technology. Meanwhile, there are relatively more agricultural practitioners in the central region, and the government attaches importance to the skills training of the agricultural labor force and the promotion of agricultural digital technology. In terms of the northeastern region, in recent years, the innovation of the agricultural service industry in northeastern China has been insufficient, and the economic dividend brought by the digital transformation of agriculture has not yet appeared. Therefore, the digital transformation fails to promote ACP in northeastern provinces. Furtherly, the agricultural operation scale in the main grain-producing areas is larger, and the demand for agricultural digital technology is also greater, which can play a more effective role in reducing emissions of agricultural digital technology. This is consistent with the regression results of the central region, where most provinces are the main grain-producing regions.

5.2. Policy Implications

Based on the above conclusions, we propose the following policy recommendations:

(1)

The Chinese government should continue to strengthen the strategic layout of the digital economy, focus on the agricultural field, increase financial support for agricultural digital infrastructure, and attach importance to the construction of agricultural land circulation information platform, agricultural product sales electronic platform, and agricultural production electronic monitoring platform. Meanwhile, it is crucial to strive to increase the institutional supply of data-sharing mechanisms.

(2)

It is crucial to strengthen the in-depth implementation of the green and low-carbon concept of farmers, strengthen the cultivation of farmers’ digital literacy through vocational education and skills training, and promote rural human capital in different regions to cross the threshold as soon as possible.

(3)

In China, it is necessary to further break down the institutional barriers to the two-way flow of urban and rural elements, promote the construction of urban-rural integration and urban-rural integration, and promote the coordinated progress of new-type urbanization and rural revitalization.

(4)

The construction of digital agriculture should also be carried out step by step according to the level of low-carbon agricultural development and geographical location. It is vital to summarize and refine the successful experience of the central region and the main grain-producing areas and promote them in a classified, reasonable and orderly manner.

5.3. Future Directions

This study systematically and deeply investigates the impact of agricultural digital transformation on agricultural carbon productivity at the provincial level and draws some valuable conclusions. If data permit, this study can be expanded from the following four aspects:

(1)

It can be expanded from a more micro level of cities, enterprises, and individual farmers.

(2)

The case study can focus on the production electronic monitoring platform of digital agriculture.

(3)

We can also focus on a certain type of digital technology in agriculture, such as using agricultural drones to spread seeds, to investigate the impact of digital transformation on agricultural low-carbon development.

(4)

The economic, social and environmental effects of agricultural digital transformation can be expanded to industrial and service fields, and then its impact on sustainable economic and social development can be furtherly comprehensively investigated.