Influence Paths and Spillover Effects of Agricultural Agglomeration on Agricultural Green Development

Abstract

Analyzing the impact of agricultural industrial agglomeration (AIG) on agricultural green development (AGD) is of a great significance to realizing the sustainable and high-quality development of agriculture. Panel data of 31 provinces in China from 2009 to 2019 were analysed. For measuring efficiency, a non-parametric DEA approach in the presence of undesirable outputs, a slack-based measure (SBM) was used. From the perspective of the spatial spillover analysis and heterogeneity analysis, Moran’s I index and the Spatial Durbin Model (SDM) were used to empirically analyze the impact of AIG on AGD to alleviate conflicts between agricultural sustainable development and environmental pollution and further explore the regional heterogeneity of AIG on AGD-efficiency due to the vast territory of China. The mediation model is constructed to explore the paths of AIG affecting AGD. The results show that: (1) Chinese efficiency of AGD was raised continuously and the high efficiency was mainly located in the southeastern coastal areas. (2) AIG not only has a significant U-shaped impact on the AGD, but also has a nonlinear U-shaped spatial spillover effect in related regions, which shows that the “siphon effect” will be triggered in the early stage of AIG and the “diffusion effect” will be evoked in the later stage of AIG. (3) From the perspective of heterogeneity analysis, AIG significantly promotes the efficiency of AGD in the central region of mainland China. In the eastern region, the AIG has an inverted U-shaped effect on the efficiency of AGD from positive to negative. On the contrary, the AIG has a U-shaped impact on the efficiency of AGD from negative to positive in the western region. (4) The analysis of the mediation model plays a partial positive mediating role for AGD to persist in promoting technology innovation and increasing the speed of talent agglomeration. Accordingly, suggestions are provided to strengthen the coordination and cooperation in sustainable agricultural development among provinces, to drive the efficiency of science and technology through the scale knowledge spillover effect, and to conduct a scientific layout of agricultural industry development.

1. Introduction

1.1. Background and Research Motivation

Agriculture is the basic industry of the national economy, and its sustainable development is the core driving force of high economic and ecological quality [1]. Global agricultural production must double by 2050 to meet the projected demand of a growing population with improved living standards [2]. However, the industrialization of developing countries is causing widespread heavy pollution, such as the aggravation of the greenhouse effect crisis, agricultural non-point source pollution, heavy metal(loid)s in the soil environment, and so on [3]. There is a major threat to global food safety and food security, and the calls for low-carbon development and green sustainable development in all countries in the world are increasingly rising [4]. Since 1978, agricultural industry has developed rapidly in China. For example, the total output value of the planting industry was 111.75 billion yuan in 1978, reaching 6145.26 billion yuan in 2018. In 2019, China’s grain output reached 663.8 million tons, approximately 474 kg per capita, far exceeding the food security line [5]. However, the development of extensive agriculture with high input, high consumption, and high yield is achieved by sacrificing the ecological environment and the intensive use of fertilizers and pesticides. China has also faced many environmental problems through this traditional and unsustainable way [6]. AGD is an important part of comprehensively promoting the strategy of rural revitalization, the green economy, and the construction of a beautiful China. There is a need for resource-saving and environment-friendly production methods to improve agricultural sustainability and competitiveness. The Chinese government needs to further promote the AGD by creating an ecological pattern to match agricultural productivity with the carrying capacity of resources and environment to improve the rural living and production environment, to eliminate consumers’ concerns about the quality of agricultural products, and to ensure food security and ecological security. AIG can not only accelerate the process of agricultural modernization, increase employment, and improve the income of farmers, but also improve production efficiency and industrial competitiveness [7]. Therefore, it is important to clarify the relationship between AIG and the efficiency of AGD to accelerate the construction of ecological civilization.

1.2. Literature Review and Contribution

Research on the influence of AIG shows that it has two effects on sustainable agricultural development. From the perspective of positive effects, AIG can generate a scale effect that is conducive to promoting sustainable agricultural development [8]. This will affect the green total factor productivity (GTFP) directly. AIG can also realize large-scale management, resource sharing of agricultural infrastructure, and technological diffusion [3], thus promoting agricultural development. Furthermore, Li et al. (2021) held that AIG attracts the agglomeration of production factors such as people engaged in scientific and technological work, green mechanizations which can reduce agricultural non-point source pollution, and be useful for the innovation and promotion of green technology [9]. From the perspective of negative impact, excessive industrial agglomeration will cause crowding effects which may hamper the sustainable development of agriculture [10]. For example, Bartolini et al. (2019) showed that only industrial clusters in the growth stage can promote agricultural economic growth [11]. There is also excessive competition for limited resources among farmers, industries, and provinces along with the increase in agglomeration [12]. Many scholars showed that excessive agglomeration has “lock-in” and “crowding”, which are negative effects on technological innovation, industrial structure upgrading, infrastructure and service sharing, and agricultural non-point source pollution controlling [13].

The efficiency of AGD can reflect the capacity of achieving maximal agricultural output while consuming minimal resources and causing minimal environmental pollution after various agricultural essential factors are used under specific output conditions [14]. This is a substantial indicator of the sustainable development capacity of green agriculture. Many scholars have investigated the green agriculture development indicator system. For example, Bergius et al. (2018) established relevant indicators from the perspective of economical utilization of resources, environment-friendly agriculture, stable ecological systems, and high efficiency of green supply based on the green agriculture development goal [15]. Van et al. (2006) appraised the level of AGD from the perspectives of agricultural production, agricultural ecology, and economic development, without using an external environmental variable [16]. Typically, studies have used the entropy method, analytic hierarchy process (AHP) [17], and data envelopment analysis (DEA) [18]. Some studies have considered the external environment and used the SBM-Undesirable model to investigate the regional difference in Chinese green agriculture production efficiency and various influencing factors [19,20].

The existing literature provides a rich theoretical basis and experience reference for our research. However, the concepts of AIG and AGD are relatively new, and most studies mainly focus on two aspects: the construction and level measurement of the evaluation index system of AGD [21], and the driving or restrictive factors affecting on AGD [22]. The research on the influence of AIG on the AGD is still blank. Secondly, most of the literature focuses on the impact of agricultural mechanization and scientific and technological innovation on the AGD [23], and maintains focus on the improvement of regional R&D and innovation ability [24]. The practical path of influence of AIG on AGD, and the transmission mechanism of talent agglomeration and technological innovation and heterogeneity analysis, should be further studied. Finally, because the flow and agglomeration of factors under the guidance of policies have periodic characteristics, if we ignore the heterogeneity and spatial correlation, we cannot understand the inertia and endogeneity of AGD [25]. By incorporating spatial factors into the endogenous growth model, we can evaluate the impact of AIG on AGD in related areas with a spatial dynamic development of factors [26]. The results can provide important practical value for the government to realize the regional integration development of AGD.

Therefore, based on the existing research, this article first uses the SBM-Undesirable model and considers the undesired output to measure the efficiency of AGD based on the panel data of 31 provinces of China from 2009 to 2019. Then, based on theoretical analysis, we discuss the direct and indirect mechanism by exploring the relationship between AIG and AGD-efficiency; three hypotheses are also put forward. Compared with previous studies, the marginal contribution of this paper is mainly reflected in the following four aspects. First, this paper innovatively explores the impact of AIG on AGD-efficiency, which enriches the research topic in this field of sustainable agricultural development. Second, this paper uses the Spatial Durbin Model (SDM) to explore whether AIG can promote the improvement of AGD-efficiency and to analyze the nonlinear spatial spillover effect by adding the quadratic term of explanatory variables to the regression model. It will explain that AIG can promote the sustainable development of related regions through spatial spillover effect in the short and long term, which provides reasonable support for building a coordinated development mechanism of shared cooperation and growth in science and technology, and economical and sustainable development between adjacent regions. Thirdly, this paper further explores the research on the regional heterogeneity of AIG on AGD-efficiency due to the vast territory of China. This paper analyzes the impact of AIG on the AGD-efficiency of different development of different regions to find out whether AIG has a significantly different impact on efficiency. Finally, the mediation model is used to explore the influence mechanism of AIG on AGD-efficiency. This paper analyzes whether the level of talents accumulation and technological innovation have a moderating effect on the impact of AIG on AGD-efficiency. Local governments should make different decisions reasonably according to their own situation, rather than blindly developing AIG to promote rural sustainable development. The results provide illumination for formulating scientific and reasonable agricultural policies and promoting regional balance and sustainable development of green agriculture.

2. Theoretical Analysis

Under the law of biological growth, agriculture is an industry in which human beings obtain agricultural products through labor and economic investment by using water, soil, and other natural resources. Therefore, agricultural production not only depends on the natural environment but is also affected by economic and social conditions [27]. From a regional perspective, agricultural production is highly dependent on natural conditions such as terrain, sunshine, and temperature, which have regionality and seasonality. Neighboring provinces tend to converge in terms of agricultural production conditions, crop varieties, production mode, and development history in similar agricultural location conditions. With the rapid development of modern transportation and communication, the agricultural production relationship between adjacent areas is becoming closer, making the regional AIG and AGD have a spatial spillover effect [28]. Therefore, adjacent provinces will affect the green development efficiency of this province. Therefore, Hypothesis 1 proposed that:

Hypothesis 1 (H1).

AGD has spatial autocorrelation in the region.

2.1. Positive Effect Mechanism of AIG on AGD

The Positive Effect Mechanism of AIG on AGD is a theoretical analysis that includes three aspects: economies of scale effect, knowledge effect, and competition effect. On the one hand, AIG can produce the economies of scale effect. Geographically, AIG can improve the level of specialization of the agricultural industry, share infrastructure services and industry information, and reduce agricultural information costs, production costs, and transaction costs. It is beneficial to promote production efficiency and the upgrade of agricultural industrial structure. On the other hand, AIG can evoke the knowledge effect. The agglomeration can attract talents to participate in research and development (R&D) activities. Furthermore, the communication and learning of talents will help to spread advanced agricultural technology about production and management. It also can promote the innovation of green agricultural technology and form a positive circular cumulative effect of talent and industrial agglomeration through the information sharing and knowledge spillover effects among high-level talents. In addition, the competition will become more intense after the AIG. Agricultural operators encourage talents to participate in green agricultural technology innovation to increase the number of patent authorizations through the incentive of R&D investment. Only in this way can the operators ensure improved profits in the market and the utilization efficiency of agricultural resources. Finally, the province of AIG can be developed into a demonstration area to control pollution and ecological environmental protection. So, Hypothesis 2 proposed that:

Hypothesis 2 (H2).

AIG can improve the efficiency of AGD through talent agglomeration and agricultural scientific and technological innovation.

2.2. Negative Effect Mechanism of AIG on AGD

The impact of AIG on the efficiency of AGD is not always beneficial. The negative effect of agglomeration is reflected in three aspects: crowding effect, siphon effect, and confinement effect. First, the agricultural development space and resources in a province are limited. The input of essential resources such as pesticides and chemical fertilizers is also increasing with the gradual expansion of the scale of agricultural production in the province. As a result, the crowding effect increases the pressure on the agricultural ecological environment and inhibits the improvement of the efficiency of AGD in this province. Second, the agglomeration of agricultural industries will attract the agricultural production resources from the surrounding provinces into the central province. The siphon effect will lead to the shortage of production factors and the loss of high-quality agricultural resources in the surrounding provinces. Furthermore, the tendency of the spatial polarization will gradually widen the gap of AGD between the central province and adjacent provinces. Finally, according to the confinement effect, agricultural operators invest too much in the early stage of the agricultural industry, such as land, agricultural equipment, and infrastructure. If some agricultural managements are faced with low production efficiency and low profit, it is difficult to give up agricultural production and management. The only way is to continue consumption of agricultural production resources inefficiently eventually leading to the depression of AGD. Therefore, the impact of AIG on the efficiency of AGD is uncertain.

Hypothesis 3 (H3).

Agricultural industry agglomeration has an impact on agricultural green efficiency through the spatial spillover effect, but the effect is uncertain.

Accordingly, we divide the mediating effect and spatial spillover effect of AIG on AGD into two paths in this study: one of the mediator variables and a spatial spillover path is agricultural technology innovation, and the other is talent agglomeration, which deeply reveals the impact mechanism of AIG on AGD (Figure 1).

3. Materials and Methods

3.1. Base Regression Model

We selected the spatial model to analyze the impact of AIG on the efficiency of AGD. Concurrently, to test the possible nonlinear relationship, we further add the quadratic type of AIG, and set the model as follows:

$$\begin{array}{l}{AIG}_{i,t}={\lambda WADG}_{i,t}+{{\alpha }_{1}AIG}_{i,t}+{{\beta }_{1}WAIG}_{i,t}+{\alpha }_2{({AIG}_{i,t})}^2+{\beta }_{2}W{({AIG}_{i,t})}^2\\ +{\eta Control}_{it}+\sum Time+\sum Space+{\epsilon }_{it}\end{array}$$

(1)

(1) Dependent variable: ${AIG}_{i,t}$ represents the efficiency of AGD in province i during the t year. Based on the SBM-Undesirable model and undesired output considered [21], we select indicators from three perspectives, including the expected output, undesired output, and input variables. The expected output is the gross product of agriculture, forestry, animal husbandry, and fisheries (100 million yuan). The undesired output is the agricultural carbon emissions. Input variables include agricultural labor (ten thousand), total power of agricultural machinery (ten thousand kilowatts), the consumption of chemical fertilizers converted into a net amount (ten thousand tons), effective irrigation area (one thousand hectares), total sown area of crops (one thousand hectares), pesticide usage amount (ten thousand tons), and usage amount of agricultural plastic film (ten thousand tons) [29].

(2) ${AIG}_{i,t}$ is the core dependent variable and is calculated by the location quotient method. The Equation (2) is:

$${AIG}_{i,t}=\frac{\left(\raisebox{1ex}{$E_{ai}$}\!\left/ \!\raisebox{-1ex}{$E_{ei}$}\right.\right)}{\left(\raisebox{1ex}{$E_{ac}$}\!\left/ \!\raisebox{-1ex}{$E_{ec}$}\right.\right)}$$

(2)

$E_{ai}$ refers to the number of agricultural employees in the i region, $E_{ei}$ refers to the number of employees in all industries in the i region, $E_{ac}$ refers to the number of agricultural employees in the country, and $E_{ec}$ refers to the total number of employees in the country, for the further analysis of the nonlinear relationship between AIG and ADG using the square value of AIG into the model [5].

Control variables include agricultural financial investment rate (GOV, %), environmental protection investment rate (EPI, %), rate of disaster damage (DR, %), the urbanization rate (URB, 100%), and industrialization (IND, %) [23].

W is the spatial weight matrix including adjacency matrix (W1), geographic distance matrix (W2), and economic distance matrix (W3) [24].

3.2. Spatial Analysis Method

It is common to measure the global spatial autocorrelation with Moran’s I index before studying the spillover effects of the efficiency of AGD [29]. The global Moran index can reveal the similarity between AIG and the efficiency of AGD in adjacent provinces. The formula of the Moran’s I index is calculated in Equation (3) [30].

$$Moran’s I=\frac{n}{{{\displaystyle \sum _{i=1}^n(x_i-\overline{x})}}^2}\frac{{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{\omega }_{ij}(x_i-\overline{x})(x_j-\overline{x})}}}{{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{\omega }_{ij}}}}$$

(3)

The value of the Moran index is between −1 and 1. The value greater than 0 indicates positive autocorrelation; if it will close to 0, it indicates that the spatial distribution is random and there is no spatial autocorrelation. If the value is less than 0, it indicates negative autocorrelation. Here $\frac{{{}^{{\displaystyle \sum _{i=1}^n(x_i-\overline{x})}}}^2}{n}$ is the variance of the observation unit, $n$ is the number of observation units, $x_i$ denotes the observed value of the efficiency of AGD in province $i$; $x_j$ denotes the observed value of the efficiency of AGD in province $j$; and ${\omega }_{ij}$ denotes the spatial weight matrix. We used 0–1 adjacent space weight matrix (W1), distance spatial weight matrix (W2), and economic space weight matrix (W3) as the spatial weight matrices in this article [31]. The definition is shown in Equations (4)–(6):

$$W_{ij}^{01}={\{}_{0, otherwise\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}(i=j)}^{1, province i adjacent to province j\hspace{1em}(i\ne j)}$$

(4)

$$W_{ij}^d=e^{-a{d}_{ij}}$$

(5)

The role of the spatial weight matrix is to analyze AGD spatial relationship with geographic and economic characteristics.

$$W_{ij}^e=W_{ij}^{d}diag(\overline{Y_1}/\overline{Y},\overline{Y_2}/\overline{Y,}\cdots ,\overline{Y_n}/\overline{Y})$$

(6)

$W_{ij}^d$ is the geographic distance matrix mentioned above; diag (…) is a diagonal matrix, in which $\overline{Y_i}$ is the average per capita gross domestic product (GDP) of the province i from 2009 to 2019; $\overline{Y}$ is the average per capita GDP of all provinces in the sample period. The main diagonal elements of the spatial weight matrix are 0 [32].

3.3. Mediation Models

The mediation model has been used to evaluate the influence paths of factors. Zhang et al. (2020) built five regression equations to investigate the impact of farm size, labor, and capital inputs on agricultural labor productivity [33]. Chen et al. (2020) used a mediator model to explore the influence paths of agricultural mechanization on green agricultural development [34]. To understand the mechanism of AIG affecting the efficiency of AGD, this study constructs equations based on two paths: talent aggregation and agricultural science, and technology innovation level. The mediation model that tests the mediation effect is as follows:

$${TA}_{it}={\beta }_0+{\beta }_1{AIG}_{it}+{\eta Control}_{it}+{\displaystyle \sum Year+{\displaystyle \sum City+}}{\epsilon }_{it}$$

(7)

$${AGD}_{it}={\alpha }_0+{\alpha }_1{AIG}_{it}+{\alpha }_2{TA}_{it}+{\eta Control}_{it}+{\displaystyle \sum Year+{\displaystyle \sum City+}}{\epsilon }_{it}$$

(8)

Here ${TA}_{it}$ refers to talent aggregation, represented by the number of undergraduate students per ten thousand people [35]. If we replace talent agglomeration with agricultural technological innovation level (INNO), INNO refers to the number of agricultural patents granted per ten thousand people by using logarithms [24,31]. α₁ represents the direct effect of AID on AGD based on controlling the mediator variable; ${\alpha }_2$ is the mediator effect of the variable ${TA}_{it}$ on AGD under controlling AID; other indicators are consistent with the base regression model.

3.4. Data Source and Description

Based on the operability of the data, we selected panel data from 31 provinces in China from 2009 to 2019. The data of AGD were obtained from “China Rural Statistical Yearbook”. The agricultural patent data were obtained from the patent database of the State Intellectual Property Office of China. The classification numbers representing agricultural patents were selected from the International Patent Classification System (IPC) to obtain relevant statistics on the number of agricultural invention patents and utility models granted. In addition, other types of data were obtained from the “China Statistical Yearbook”, “China Fiscal Yearbook” and the statistical yearbooks of various provinces and cities.

Table 1. Variable definitions.

Type	Variables	Symbols	Definitions	Source
Dependent Variable	The Efficiency of Green Agricultural Development	ADG	Based on the SBM-Undesirable model and considered about undesired output	[3,7,22,24]
Core Independent Variable	Agricultural Industrial Agglomeration	AIG	${AIG}_{i,t}=\frac{\left(\raisebox{1ex}{$E_{ai}$}\!\left/ \!\raisebox{-1ex}{$E_{ei}$}\right.\right)}{\left(\raisebox{1ex}{$E_{ac}$}\!\left/ \!\raisebox{-1ex}{$E_{ec}$}\right.\right)}$	[5,28]
Mediation Variables	Talent Aggregation	TA	Patents Granted Number (Ten Thousand)	[35]
	Technological innovation	INNO	Undergraduate Students’ Number (Ten Thousand People)	[23,24,31]
The Control Variables	Environmental Protection Investment Rate	EPI	Environmental Protection Financial Investment/Total Government	[3,7,22,36]
	Industrialization	IND	The Proportion of Industrial Added Value in Its Regional GDP	[3,7,22,36]
	Agricultural Financial Investment Rate	GOV	Agricultural Financial Investment/Total Government Expenditure	[3,7,22,36]
	Disaster Damage Rate	DR	Disaster Damage Area/Total Sown Area	[3,7,22,36]
	Urbanization Rate	URB	Urban Population/Total Population	[3,7,22,36]

shows the definitions of all variables and the specific descriptive statistical results are shown in

Table 2. Variable description and descriptive statistical results.

Variables	Min	Max	Mean	SE	Variables	Min	Max	Mean	SE
ADG	0.121	0.801	0.420	0.038	IND	0.263	0.435	0.379	0.082
AIG	0.005	1.714	1.143	0.583	GOV	0.026	0.051	0.089	0.028
INNO	2.723	8.803	6.243	1.266	FR	0.184	0.368	0.252	0.0636
TA	0.001	0.218	0.020	0.032	URB	0.139	0.896	0.519	0.144
EPI	0.026	0.051	0.089	0.028

Note: limited by the content and length of the article, the specific calculation results of green agricultural development efficiency and AIG are not listed.

.

4. Empirical Results and Discussion

4.1. Calculation of AGD Level

To clearly describe the spatial distribution of Chinese green agricultural production efficiency, the green agricultural production efficiency in various provinces was divided into 5 levels: high (0.5~1.0), relatively high (0.4~0.5), medium (0.3~0.4), moderately low (0.2~0.3), and low (0.0~0.2) [3]. The spatial distribution of Chinese green agricultural development efficiency in 2009 and 2019 was mapped using ArcGIS 9.3 (Figure 2). In addition, variations in Chinese green agricultural development efficiency from 2010 to 2019 were reflected (Figure 3). According to these figures, the Chinese green agricultural development efficiency was raised. However, significant differences were found in various regions. Provinces achieving high efficiency were mainly located in the southeastern coastal areas, featuring high economic development levels, and the western provinces. In contrast, low efficiency provinces were found in the northern China plain, roughly increasing from central south China and north China.

4.2. Spatial Autocorrelation and Estimation Results

According to the efficiency of AGD calculated above, this study calculates the global Moran’s I of the efficiency of AGD in 31 provinces of China from 2009 to 2019 under W1, W2, W3, and uses the p-value to test its significance (

Table 3. Moran’s I index.

Space Weight Matrix	W1	W2	W3
the efficiency of AGD	0.331 **	0.059 ***	0.157 *
AIG	0.370 ***	0.016 **	0.158 **

Note: *** p < 0.01, ** p < 0.05, * p < 0.1.

). It shows a significant positive correlation between AGD and AIG in space. By comparing the coefficients, it is found that level of spatial correlation under three spatial weight matrices presented a significant positive of 0–1 adjacent space weight matrix > economic space weight matrix > distance spatial weight matrix. The most significant positive is the spatial correlation between AIG and AGD in adjacent areas. Therefore, there is a spatial correlation between the provincial level of AGD in China and Hypothesis 1 is true.

Before the quantitative analysis of the influence mechanism of AIG on the efficiency of AGD, we need to carry out the LM test to judge whether to choose the error model or the lag model according to LM-error and LM-lag results. Then, the Hausman test was used choose which one, fixed effect or random effect. Finally, the LR test and the Wald tests are used to ascertain if the SDM model can degenerate into SEM and SAR models. Combining the results of the Hausman test, the LM test, the LR test, and the Wald test in the

Table 4. Estimation results of the fixed-effect SDM.

Variable	W1	W2	W3	Variable	W1	W2	W3
	(1)	(2)	(3)		(4)	(5)	(6)
AIG	−0.2813 *** (−0.0920)	−0.0974 (−0.0773)	−0.1438 ** (−0.4902)	W × AIG	−1.1324 *** (−0.3761)	−0.0508 (−0.0833)	−1.0667 ** (−0.4156)
AIG2	0.1932 * (0.1713)	0.0803 (0.0539)	0.0843 (0.0516)	W × AIG2	0.2842 ** (0.1838)	0.2631** (0.1371)	0.2450 ** (0.1256)
EPI	0.5286 *** (0.5312)	0.3078 * (0.4530)	0.2788 * (0.4711)	W × EPI	0.4491 *** (0.8365)	0.4027 ** (0.6828)	0.2056 * (0.4240)
IND	−1.1305 *** (−1.0004)	−0.6328 (−0.0679)	−0.8321 * (0.0792)	W × IND	−1.1532 *** (−1.0039)	−1.0073 ** (−0.8951)	−0.8955 ** (−0.6982)
GOV	−0.1974 * (−0.1357)	−0.0420 (−0.0836)	−0.2871 * (−0.1896)	W × GOV	−0.0426 (−0.0722)	−0.0313 (−0.0409)	−0.0372 (−0.0420)
DR	−0.1774 ** (−2.4091)	−0.0562 (−1.0553)	−0.0079 (−0.0422)	W × DR	−0.3682 *** (−2.1192)	−0.0301 (−0.8256)	−0.0067 (−0.0041)
URB	0.7436 *** (0.3389)	0.7192 *** (0.3075)	0.0672 (0.0048)	W × URB	−0.8651 ** (−0.6672)	−0.8934 *** (−0.6792)	−0.6815 ** (−0.5902)
λ	−0.2067 *** (0.0892)	−0.1266 ** (0.674)	−0.2098 *** (0.0896)	$R-\mathrm{squared}$	0.0067	0.0944	0.0023
$LM-lag$	34.31 ***	32.11 ***	33.96 ***	LM – error	33.67 ***	28.16 **	31.95 **
RobustLM – lag	3.81 **	3.54 **	3.91 ***	LR_Spatial_error	3.07 *	2.93 *	3.11 *
LR_Spatial_lag	84.29 ***	85.12 ***	83.98 ***	RobustLM – error	83.23 ***	78.23 **	89.04 ***
Wald_Spatial_lag	91.28 ***	89.77 ***	90.26 ***	$Wald\_Spatial\_error$	90.32 ***	88.54 ***	90.21 ***
Observations	341	341	341	341	341	341	341

Notes: Numbers in the brackets denote t-values. * denotes p < 0.1. ** denotes p < 0.05. *** denotes p < 0.01; the same applies below.

, it can be seen that under the condition of the W1, W2, and W3, the time fixation and the individual fixation of SDM should be used to study the impact of AIG on the efficiency of AGD.

The regression results are shown in Table 4. Columns (1)–(6) are the results of W1, W2, and W3 included in control variables. Furthermore, to figure out if two variables possibly have a nonlinear relationship, the quadratic form of AIG is put into the regression. The spatial lag term regression results of the double fixed-effect SDM under W1, W2, and W3 show that the coefficient of λ is −0.2067 at a significance level of 1%. It shows that there is a spatial correlation in ADG, which further indicates that this study is suitable for using the spatial econometric model and Hypothesis 1 is proved. From the perspective of the core explanatory variable, the primary coefficients of AIG are negative but the quadratic coefficients are positive. AIG has a U-shaped impact on AGD-efficiency, meaning that it has a negative impact in the early stage and then turns positive. The reason is that, during the development period of AIG, a large number of funds were invested in attracting talents, agricultural mechanization, and infrastructure development, and the government would increase agricultural expenditure continuously, such as the investment of R&D funds, the construction of agricultural research institutes, rewarding the promotion of green technology and providing subsidies for ecological agriculture, which is adverse to the sustainable development of agriculture [24]. From the perspective of the spatial lag term regression results, on the one hand, the AIG in this province has a negative effect on AGD in neighboring provinces. Specifically, AIG of spatial lag term regression correlated with AGD at a significance level of 5%, and the regression coefficients are 1.1324 and 1.0667 in columns (2) and (5), respectively, indicating that each one-unit increase in the two variables under W1 and W3 will have a negative effect of 1.1324 and 1.0667 percentage points on AGD respectively. On the other hand, the quadratic coefficient of AIG under the three weight matrices in columns (4), (5), and (6) are significantly positive, indicating that the AIG in this province has a U-shaped impact on other provinces of AGD from negative to positive. After comparing the autoregressive results of AIG and the intersection term of AIG under spatial matrices, the spatial spillover effect of AIG on the AGD mainly acts on the areas with similar economic development levels and the adjacent areas. On the whole, the early stage of AIG in one province will have a “siphon effect” on production resources in other provinces, which is detrimental to the AGD in other provinces. However, with the diffusion mechanism of production resources, knowledge spillover, and technology promotion, AIG has a positive spatial spillover effect on the AGD in other provinces.

4.3. Analysis of Spatial Spillover Results

The indirect effect reflects the spillover degree of changes in AIG to relevant provinces. It can be seen from

Table 5. Decomposition results of the fixed-effect SDM.

Variable	Direct Effect			Indirect Effect			Total Effect
	W1	W2	W3	W1	W2	W3	W1	W2	W3
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
AIG	−0.3658 ** (−0.0263)	−0.0394 (−0.0193)	−0.4172 ** (−0.0832)	−0.7694 *** (−0.1944)	−0.0198 (−0.0042)	−0.6911 *** (−0.1859)	−1.1352 *** (−0.3761)	−0.0592 (−0.0833)	−1.1083 *** (−0.4156)
AIG2	0.1032 (0.0382)	0.1026 (0.0331)	0.0672 (0.0084)	0.1769 ** (0.0527)	0.1602 ** (0.0502)	0.1765 ** (0.0520)	0.2801 *** (0.1824)	0.2628 *** (0.1326)	0.2437 *** (0.1231)
EPI	0.4201 *** (0.5425)	0.3053 ** (0.3892)	0.2717 ** (0.3085)	0.2161 * (0.2993)	0.2121 * (0.2816)	0.1353 (0.2091)	0.6362 *** (0.8321)	0.5174 *** (0.6820)	0.4071 *** (0.4393)
IND	−1.0435 ** (−0.8935)	−0.5592 * (−0.0509)	−0.4930 * (0.0492)	−0.1169 (−0.0037)	−0.5309 * (−0.0503)	−0.5992 * (−0.0547)	−1.1604 *** (−1.0531)	−1.0901 ** (−0.8890)	−1.0922 ** (−0.7102)
GOV	−0.0439 * (−0.0961)	−0.0432 * (−0.0878)	−0.0503 * (−0.1062)	−0.0157 (−0.0536)	0.0138 (0.0382)	0.0139 (0.0371)	−0.0596 * (−0.0722)	−0.0294 (−0.0382)	−0.0364 (−0.0417)
DR	−0.1801 ** (−0.8034)	−0.1921 ** (−0.8476)	−0.0035 (−0.0392)	−0.2292 ** (−1.2034)	−0.2711 ** (−1.2160)	−0.0037 (−0.0021)	−0.4093 *** (−2.1208)	−0.4632 *** (−0.8263)	−0.0072 (−0.0104)
URB	0.7102 *** (0.4910)	0.4212 ** (0.2015)	0.7008 *** (0.4722)	−0.3321 ** (−0.1862)	−0.1510 (−0.0582)	−0.3217 ** (−0.1980)	0.3781 ** (0.1804)	0.2702 (0.1026)	0.3791 ** (0.2083)

Notes: Numbers in the brackets denote t-values. * denotes p < 0.1. ** denotes p < 0.05. *** denotes p < 0.01.

that the total effects of AIG under the W1 and W3 are −1.1352 and −1.1083 in columns (7) and (9) respectively at a significance level of 1%, among the total effect, the direct effect and spillover effects are −0.3658 and −0.7694, respectively. These are significant at the level of 5%, indicating that the AIG on ADG has a spatial spillover effect, which is consistent with the previous analysis results. In addition, the indirect effect of AIG is significant under the W1, and the primary term coefficient is negative and the secondary term coefficient is positive, indicating that the improvement in the level of AIG in this province will have a U-shaped impact on AGD in neighboring provinces through spillover effects. Under the economic distance matrix of indirect effect, the primary term coefficient of AIG is significantly negative and the secondary term coefficient is positive, indicating that AIG has a significant negative impact on AGD in provinces with similar levels of economic development. The possible reason is that provinces with similar levels of economic development are in similar stages of agricultural development, urban development, and economic development. All of them have similar demand for production factors and fierce regional competition due to the phenomenon of the “Zero-Sum Game” [37]. As a result, the “siphon effect” will develop into the “diffusion effect”, and thus, through system connection, market linkage, and service connection, the adjacent or similar provinces will build a space for sharing with innovative talents and advanced agricultural green technology to comprehensive advances and promote the regional balance of AGD [38].

4.4. Heterogeneity Analysis

There are great differences in climate, land, geographical location, and economy between different regions in China, so the impact of AIG on the AGD-efficiency may be heterogeneous. Therefore, the article constructs the SDM with double fixed time and individual, and 31 provinces and cities are divided into three areas for analysis: the eastern, central, and western areas. The estimated results are shown in

Table 6. Estimation results of SDM from three areas in China.

Variable	Eastern		Central		Western
	(1)	(2)	(3)	(4)	(5)	(6)
AIG	0.0209 (0.0331)	0.2311 *** (0.0814)	0.0412 (0.0309)	0.2638 *** (0.0817)	0.0982 (0.0633)	−0.0304 (0.0144)
AIG2		−0.1374 *** (0.0441)		0.2297 ** (0.0743)		0.1615 *** (0.0581)
EPI	0.0744 * (0.0683)	0.0637 * (0.0597)	0.0481 * (0.0361)	0.0184 * (0.0214)	0.1149 *** (0.0249)	0.1045 ** (0.0502)
IND	−0.1733 ** (0.0892)	0.2554 *** (0.0617)	−0.0469 * (0.0791)	−0.2336 *** (0.0569)	0.1300 (0.1086)	−0.2371 (0.2293)
GOV	−0.0913 (0.0800)	0.0635 (0.0431)	−0.5032 ** (0.0489)	0.0658 *** (0.0257)	−0.0894 * (0.0539)	−0.0792 (0.1218)
DR	0.0100 (0.0120)	0.0100 (0.0080)	−0.0258 ** (0.0113)	0.0037 (0.0010)	−0.0700 *** (0.0182)	−0.0674 * (0.0336)
URB	−0.3389 *** (0.1145)	−0.3029 *** (0.0743)	0.4121 *** (0.0739)	0.2176 ** (0.0548)	−0.0285 (0.1369)	0.0073 (0.1422)
_cons	−0.5561 * (0.564)	−0.5704 * (0.4912)	0.2523 (0.4125)	0.2271 (0.5367)	0.2340 (0.3742)	−0.0864 (0.5547)
R-squared	0.0905	0.1681	0.2701	0.2194	0.2944	0.2352
Time Fixation	YES	YES	YES	YES	YES	YES
Regional Fixation	YES	YES	YES	YES	YES	YES
Observations	132	132	99	99	110	110

Notes: Numbers in the brackets denote t-values. * denotes p < 0.1. ** denotes p < 0.05. *** denotes p < 0.01.

.

(1) In the eastern area, the AIG’s coefficient of the first-order term is significantly positive and the AIG’s coefficient of the square term is significantly negative. This indicates that the AIG has a shape of an inverted U curve on the efficiency of AGD from positive to negative. The possible reasons are that the eastern area is economically developed, the level of agricultural mechanization is high, wide application of agricultural green technology, and excessive industrial agglomeration leads to unreasonable competition.

(2) In the western area, the AIG’s coefficient of the first-order term is negative and not significant, and the AIG’s coefficient of the square term is significant and positive. This indicates that the AIG has a U-shaped impact on the efficiency of AGD. The western area has a low economy and invested more resources for industrial agglomeration in the early stage [39]. The agglomeration of agricultural resources promotes the green transformation of agriculture through expanding market scale, reducing production costs, sharing infrastructure, and technology R&D [3].

(3) In the central area, the coefficient is significantly positive, showing that AIG significantly promotes the AGD in the area. The central area is an important grain producing area in China, and the level of agricultural production is relatively developed. Moreover, the central area undertakes the overflow of talents, equipment, science and technology, and other resources from the adjacent eastern area, so it will improve the efficiency of green agricultural development after agricultural industrials agglomeration [40].

5. Test of the Mediation Models

Table 7. Estimation results of the mediation models.

Variables	ADG	ADG	TA	ADG	ADG	INNO
	(1)	(2)	(3)	(4)	(5)	(6)
AIG	0.0576 (0.0522)	0.0643 (0.1148)	0.1950 ** (0.1793)	0.0305 (0.0165)	0.0354 (0.0273)	0.1433 * (0.0455)
EPI		0.3154 * (8.5297)	0.4378 ** (7.3908)		0.5637 *** (8.9144)	0.3611 * (8.7026)
IND		0.3776 *** (0.0576)	−0.0697 (−0.0311)		0.0728 ** (0.0302)	0.4525 *** (0.0553)
FR		0.3121 (0.4568)	0.1290 ** (0.0802)		−0.0367 (−0.0141)	0.6319 *** (0.1153)
URB		0.3191 (0.4821)	0.7290 *** (0.2857)		−0.0363 (−0.1368)	0.6177 *** (0.1256)
GOV		0.0392 (0.3371)	0.5927 ** (0.2763)		−0.0310 (−0.3389)	1.1464 *** (0.7833)
TA	0.1371 *** (0.0215)	0.0715 * (0.1902)
INNO				0.2159 *** (0.1069)	0.3487 *** (0.0718)
Constant	2.0075 *** (0.1327)	−0.1893 (−0.5215)	0.8756 * (0.4312)	2.1957 *** (0.1673)	1.3151 ** (0.6265)	−0.3276 ** (−0.8115)
Time Fixation	NO	YES	YES	NO	YES	YES
Regional Fixation	NO	YES	YES	NO	YES	YES
R-squared	0.5962	0.2464	0.2417	0.8342	0.2740	0.2334
N	341	341	341	341	341	341

Notes: Numbers in the brackets denote t-values. * denotes p < 0.1. ** denote p < 0.05. *** denotes p < 0.01.

reports the results of the mediation model. First, the estimated coefficient of AIG in column (3) is significant: 0.1950 at the 5% level, which indicates that AIG does accelerate talent agglomeration and brings the effect of human capital accumulation. Second, the AIG regression coefficients in columns (1) and (2) are 0.0576 and 0.0643, respectively. Comparing the base regression result (Table 4), the core explanatory variable coefficients are significant. However, after adding the moderator variable, the results become insignificant and this is a complete moderator effect. It indicates that the increasing rate of talent agglomeration will lead to a more direct promoting effect on the AGD. Then, talent agglomeration is replaced by the level of science and technological innovation and agricultural technology innovation which have a significant promoting effect on AGD. Specifically, the regression coefficients are both positively correlated with AGD at a significance level of 1% in columns (4) and (5). We also found that the estimated coefficient of AIG in column (6) is significantly positive, 0.1433, and the regression coefficients of AIG in columns (4) and (5) are not significant (0.0305 and 0.0354, respectively). Similarly, compared with the base regression results (Table 4), it can be seen that AIG indirectly raises the efficiency of AGD by promoting the level of agricultural technological innovation. This is because agricultural technological innovation will bring about new technology, new equipment, and new products. The improvement of green agricultural production efficiency will promote the upgrading of agricultural industrial structures [41]. Finally, it is beneficial to popularize new techniques and produce green and high-quality products. So, persisting in promoting technology innovation and increasing the speed of talent agglomeration play partial positive mediating roles between AIG and AGD, and Hypothesis 2 is proven. Furthermore, it should be pointed out that the article only gives two factors to test the influence paths through the mediation models, and other possible influence paths need to be further studied.

6. Conclusions and Suggestions

Based on panel data of 31 provinces in China from 2009 to 2019, this paper analyzed the overall and regional impact of AIG on AGD-efficiency from the perspective of spatial spillover analysis and heterogeneity analysis [42], and it further explored the level of technology innovation and talent agglomeration on the impact of AIG on AGD-efficiency [43], and the following conclusions were obtained: (1) there is a U-shaped nonlinear relationship between AIG and AGD-efficiency, which indicates that AIG has two stages of impact on the efficiency. In the early stage, due to the low level of rural economical development, the deficient human resources, backward infrastructure, and other factors, AIG has difficulty in improving the efficiency of AGD. With the increasing development of AIG, a large amount of capital input and attracting scientific and technological talents in rural areas will improve the efficiency of agricultural technology and enhance the ability of green innovation in agriculture, and the scale effect promotes the reduction of regional agricultural costs and pollutant emissions. The upgrading of the agricultural industrial structure further accelerates talent and industrial agglomeration, forms a positive cycle cumulative effect, and thus promotes the improvement of AGD-efficiency [44]. (2) AIG has spatial spillover effects and heterogeneous effects on AGD-efficiency in different regions of China. From the perspective of spatial spillover analysis, AIG not only affects the local agricultural development, but also has a significant nonlinear U-shaped spatial spillover effect on the regions related to adjacent geography and similar economic level. Specifically, there is a “siphon effect” in the early stage, resulting in the lack of production factors in other regions, and a “diffusion effect” in the later stage. From the perspective of heterogeneity analysis, AIG significantly promotes the efficiency of AGD in the central region of mainland China. In the eastern region, the AIG has an inverted U-shaped effect on the efficiency of AGD from positive to negative. On the contrary, the AIG has a U-shaped impact on the efficiency of AGD from negative to positive in the western region. (3) In the mechanism analysis, by introducing the level of talent accumulation and technological innovation as moderating variables. it is found that talent accumulation and technological innovation significantly enhances the promotion effect of the AIG on AGD-efficiency [45].

Based on the above conclusions, the following policy implications are proposed:

(1) To grasp the spatial correlation law of provincial agriculture and coordinate regional agricultural development. To strengthen the coordination and cooperation among regions, give full play to the radiation and driving role of strong agricultural provinces, strengthen the exchange and learning of advanced production technology and management experience among regions, promote the innovation and promotion of agricultural green technology, and realize the green and coordinated development of regional agriculture. (2) To conduct a scientific layout of agricultural industry development, appropriately guide agricultural industry agglomeration within the reasonable range allowed by regional resource and environmental carrying capacity, optimize the allocation of agricultural resources, and overcome the crowding effect, siphon effect, and confinement effect caused by agricultural industry agglomeration. (3) To promote the flow of agricultural scientific and technological talents among regions, to make full use of science and technology for preventing and controlling natural disasters, and to realize the promotion and development of agricultural green technology. AIG will promote the efficiency of AGD through the scale knowledge spillover effect and demonstration effect, so as to achieve the goal of agricultural sustainable development.

Due to the short time of putting forward AGD, most regions are still in the exploratory stage, and the research results are not sufficient. In the future, we can try to use the dynamic spatial Durbin model to research the impact of factors on AGD in the short and long term. Furthermore, the paper only gives the mediating effect of two paths on AGD. How to accurately identify other influencing factors and possible other mechanisms needs to be further studied.