Agricultural Industry Agglomeration and Carbon Emission Intensity: A Systems Perspective on Resilience and Digital Development

Highlights

Please indicate how your work links to systems science via your contributions to systems practice, theory, and/or methodology.

• Vertical agglomeration connects agricultural production, processing, and services, while its coupling with horizontal production agglomeration forms the agricultural industry agglomeration subsystem.
• The subsystem, agricultural industry resilience, and digital village development form an integrated analytical framework, with agricultural carbon emission intensity representing environmental performance.

What are the main findings and/or the implications of the main findings?

• Evidence from the indirect association is consistent with agricultural industry resilience serving as one pathway linking system coordination to carbon performance.
• The positive association between agricultural industry agglomeration and agricultural industry resilience is larger at higher levels of digital village development.

Abstract

Decarbonizing agriculture depends not only on the efficiency of individual production activities but also on coordination among production, processing, and service activities. From a systems perspective, this study treats provincial agriculture as an interconnected industrial system and examines the relationship between agricultural industry agglomeration and agricultural carbon emission intensity. Moving beyond conventional measures of production concentration, agricultural industry agglomeration is measured as coordination between horizontal agglomeration in agricultural production and vertical agglomeration across production, processing, and agricultural services. Based on a balanced panel of 390 observations across 30 Chinese provinces from 2011 to 2023, the study uses province and year fixed effects, mediation and moderation analyses, robustness checks, and an endogeneity check. Results show that agricultural industry agglomeration is significantly associated with lower agricultural carbon emission intensity. The mediation results are consistent with a partial intermediary role for agricultural industry resilience in this relationship. The interaction and simple slope results show that the positive association between agricultural industry agglomeration and agricultural industry resilience is larger at higher levels of digital village development. These findings connect subsystem coordination, adaptive capacity, digital conditions, and carbon performance, thereby extending research on agricultural transformation toward lower carbon intensity from a systems perspective.

1. Introduction

Climate change and rising food security risks pose two major challenges to global agriculture. In 2025, the global mean temperature was approximately 1.43 °C above the 1850 to 1900 average [1]. An estimated 645 million people experienced hunger, representing 7.8% of the global population [2]. In China, input-intensive and resource-intensive production has supported agricultural output growth. It has also increased the environmental pressure associated with the transition toward greener agriculture [3]. Agricultural carbon emission intensity measures carbon emissions per unit of agricultural output and is an important indicator of low-carbon performance [4]. From a systems perspective, provincial agriculture comprises interdependent production, processing, and agricultural service activities. Products, production factors, information, and technology connect these activities [5]. Horizontal concentration within agricultural production and vertical linkages across production, processing, and services represent two organizational dimensions of this industrial system. Production agglomeration may improve agricultural environmental efficiency through economies of scale and technology spillovers [6]. However, the relationship between coordination across these organizational dimensions and carbon emissions per unit of output remains insufficiently understood.

Existing research documents nonlinear and negative associations between agricultural industry agglomeration and environmental efficiency, green development, agricultural carbon emissions, and carbon emission intensity [3,7,8,9,10]. Rural industrial integration is also associated with lower agricultural carbon emissions [11]. Differences in the reported patterns may arise from variations in outcome variables, model specifications, spatial scales, and agglomeration measures. A common approach measures agglomeration using location quotients based on agricultural employment or output shares [3,12]. These indicators primarily describe regional specialization in agricultural production. Additional dimensions are therefore needed to characterize coordination among production, processing, and agricultural services. Manufacturing research further distinguishes horizontal from vertical agglomeration and indicates that the two dimensions convey different structural information [13]. Drawing on this distinction, this study measures agricultural industry agglomeration through the coordination between horizontal production agglomeration and vertical agglomeration across production, processing, and agricultural services.

A second strand examines how agglomeration and digitalization relate to the adaptive capacity of agricultural systems. Agricultural industry agglomeration is positively associated with agricultural economic resilience, with socialized services and production efficiency identified as potential channels [14,15]. Farming system research emphasizes robustness, recovery, and adaptation as important system capacities [16]. Infrastructure and material and information flows also shape how agricultural systems respond to disturbances [17]. Digital village development is associated with stronger agricultural resilience [18]. Related digital technologies facilitate information and resource flows [5] and support greener allocation of agricultural water resources [19].

Despite these advances, three questions remain. First, agricultural industry agglomeration is still commonly treated as the horizontal spatial concentration of agricultural production. Limited attention has been given to coordination between horizontal production agglomeration and vertical agglomeration. Existing measures therefore do not fully capture the organizational relationships among production, processing, and agricultural services. Second, studies of agricultural industry agglomeration, agricultural industry resilience, and agricultural carbon emissions usually estimate separate relationships. Agricultural industry resilience has rarely been tested as a transmission pathway operating at the system level between agglomeration coordination and environmental performance. The process linking the two remains insufficiently explained. Third, existing evidence connects digital village development with agricultural resilience. However, there is limited direct evidence on whether digital village development changes the strength of the relationship between agricultural industry agglomeration and agricultural industry resilience. These gaps leave an incomplete systems explanation of how industrial coordination relates to low-carbon performance through adaptive capacity and how digital conditions shape this relationship.

To address these gaps, this study conceptualizes provincial agriculture as an industrial system of interdependent production, processing, and agricultural service activities embedded in a broader digital environment. Agricultural industry agglomeration, agricultural industry resilience, digital village development, and agricultural carbon emission intensity represent system organization, adaptive capacity, digital conditions, and environmental performance, respectively. The analysis examines the association between coordination within the industrial system and carbon emission intensity, together with the possible intermediary role of resilience. It also tests whether the association between agglomeration and resilience varies with digital conditions. The empirical analysis uses a balanced panel of 390 observations covering 30 provincial regions in China from 2011 to 2023. All models include province and year fixed effects. Mediation and moderation analyses are used to examine the proposed pathway and moderating relationship. Robustness and endogeneity checks assess the stability and temporal ordering of the estimates, with the endogeneity check implemented as a temporal falsification test. Comparisons across grain functional zones and unconditional quantile regression examine their empirical scope.

This study makes three contributions. First, it extends conventional concentration measures by operationalizing AGG as the coupling coordination between horizontal production agglomeration and vertical agglomeration across production, processing, and agricultural services. Second, it connects system organization, adaptive capacity, digital conditions, and carbon performance within one testable framework. AR is specified as a possible intermediary pathway, while DV is examined as a condition shaping the relationship between AGG and AR. Third, the study provides provincial evidence and assesses the scope of the estimated relationship between AGG and ACE across grain functional zones and the unconditional distribution of ACE.

2. Theoretical Framework and Hypothesis Development

From a systems perspective, provincial agriculture comprises interdependent production, processing, and service activities linked through flows of products, production factors, information, and technology. Within this framework, Agricultural Industry Agglomeration (AGG) captures the coordination between horizontal agglomeration in agricultural production and vertical agglomeration across these activities, Agricultural Industry Resilience (AR) represents adaptive capacity, and Agricultural Carbon Emission Intensity (ACE) indicates carbon performance. Digital village development (DV) reflects the digital conditions affecting information transmission and resource circulation and is examined as a moderator of the association between AGG and AR. These roles provide the basis for the hypotheses developed below.

2.1. The Effect of Agricultural Industry Agglomeration on Agricultural Carbon Emission Intensity

Prior research establishes the spatial concentration of agricultural production and shows that its evolution reflects natural conditions, institutional change, public policy, and neighboring land-use patterns [20,21]. Evidence from agro-industrial clusters further indicates that vertical coordination across production and processing matters alongside spatial concentration [22]. Together, these studies identify spatial concentration and coordination across activities as distinct dimensions of agricultural organization. To capture this additional dimension, this study follows Wang [23] in distinguishing horizontal agglomeration within agricultural production from vertical agglomeration across production, processing, and agricultural services. AGG is measured as the coupling coordination degree between these two subsystems and therefore represents synergistic agglomeration rather than the scale of a single segment. In this study, ACE is operationalized as carbon emissions estimated from six specified agricultural production sources per unit of the real gross output value of agriculture, forestry, animal husbandry, and fishery at 2011 constant prices [4,10,24]. It captures agricultural carbon performance within this defined accounting boundary.

Externality theory provides the primary theoretical basis for explaining the relationship between AGG and ACE. It holds that one actor’s activities can alter the costs or benefits faced by others, while spatial concentration magnifies these interdependencies. Marshall identifies specialized suppliers, labor pooling, and knowledge spillovers as important sources of localized positive externalities [25]. Empirical evidence from U.S. manufacturing supports these mechanisms and identifies input–output linkages as a particularly important driver of coagglomeration [26]. Increasing returns can further reinforce spatial concentration [27]. However, agglomeration may also intensify congestion and competition for local resources [28]. AGG can therefore generate positive and negative environmental externalities at the same time. Previous findings across related environmental outcomes suggest that the efficiency gains and environmental costs of agglomeration may coexist and vary with its degree and organizational form [7,8].

Against this background, the negative externality pathway arises when excessive agglomeration intensifies competition for production factors, exceeds the capacity of local infrastructure, and encourages inefficient resource use [7]. The positive externality pathway, by contrast, operates through three mechanisms that may reduce resource and energy use per unit of real agricultural output. First, spatial proximity facilitates specialization and the shared use of machinery, infrastructure, logistics, and agricultural services, which can reduce duplicated inputs and improve production efficiency [29]. Second, proximity promotes information exchange and knowledge diffusion and may accelerate the adoption of more resource-efficient practices [12,27]. Third, coordination across production, processing, and agricultural services can improve factor matching and support structural upgrading [10]. Through these mechanisms, AGG can lower the carbon burden per unit of real agricultural output when the resulting efficiency gains outweigh the associated resource pressures. Because AGG captures coordination between horizontal and vertical agglomeration rather than production concentration alone, it is more closely aligned with the sharing of resources, the matching of production factors, and specialization across production stages. Over the observed sample range, these positive externalities are expected to generate efficiency gains that outweigh the associated congestion and resource pressures. Accordingly, the following hypothesis is proposed:

H1. 

A higher level of agricultural industry agglomeration is associated with lower agricultural carbon emission intensity.

2.2. The Potential Mediating Role of Agricultural Industry Resilience

Agricultural industry resilience (AR) denotes the capacity of an agricultural industrial system to withstand external shocks, restore production and resource flows, and reorganize its internal structure in response to changing conditions. In this study, AR is measured using a composite index constructed with entropy weights and based on ten indicators of robustness, recovery capacity, and reorganization capacity. Robustness refers to the ability to maintain essential agricultural functions during a disturbance. Recovery capacity captures the restoration of production after disruption, whereas reorganization capacity reflects structural adjustment in response to new environmental and market conditions.

Resilience theory provides the complementary theoretical basis for the mediating mechanism. Holling defines resilience as the capacity of a system to absorb disturbance while retaining its essential functions [30]. Subsequent research extends this idea to resistance, recovery, adaptation, and structural renewal in regional systems [31,32]. Research on coupled social and ecological systems further emphasizes adaptability and transformability under changing conditions [33]. Studies of farming systems apply these capacities to the maintenance, adjustment, and transformation of agricultural functions [16,34]. The three dimensions used in this study are empirical expressions of these general resilience capacities, but they do not correspond exactly to the dimensions adopted in every existing framework.

The proposed mediating pathway comprises two relationships. The first concerns the association between AGG and AR. The positive externalities generated by AGG expand the network resources available to agricultural actors during disturbances. Closer linkages among production, processing, and agricultural services facilitate the circulation of information and resources across the value chain [23,35]. Agglomeration also expands access to agricultural socialized services and improves production efficiency [14]. Agricultural technical efficiency provides an additional channel through which agglomeration enhances agricultural economic resilience [15]. Coordination across stages of the agricultural value chain provides agricultural actors with alternative suppliers, services, and market connections when one link is disrupted [23,35]. These mechanisms strengthen robustness, accelerate recovery, and support reorganization following disturbances.

The second relationship concerns the association between AR and ACE. Greater resilience does not necessarily imply lower carbon emissions because buffers and redundant capacity may require additional resources [16,34]. Its environmental effect depends on how the three capacities represented by AR influence the relationship between emissions and output. Robustness limits production losses and reduces imbalances between resource inputs and agricultural output during disturbances. Recovery capacity shortens periods in which resource use remains high relative to agricultural output. Reorganization capacity facilitates the replacement of inefficient production arrangements with alternatives that are more adaptive and use resources more efficiently. When the efficiency gains from avoiding losses caused by disturbances exceed the costs of maintaining resilience, stronger AR is expected to be associated with lower carbon emissions per unit of real agricultural output. Taken together, these capacities may reduce efficiency losses arising from disturbances and facilitate more resource-efficient adjustments in agricultural production.

These arguments suggest that higher AGG is expected to be associated with stronger AR, while stronger AR is, in turn, expected to be associated with lower ACE. AR is therefore examined as one potential intermediary pathway between AGG and ACE. Accordingly, the following hypothesis is proposed:

H2. 

A higher level of agricultural industry agglomeration is associated with stronger agricultural industry resilience, which is, in turn, associated with lower agricultural carbon emission intensity.

2.3. The Moderating Role of Digital Village Development

Digital village development (DV) refers to the integration of digital infrastructure, digital technologies, data resources, and digital services into rural production and livelihoods [36]. In this study, DV is measured using a composite index constructed with the entropy weight method and covering digital infrastructure, industrial digitalization, and digital livelihoods. DV is treated as an external development condition rather than a component of AGG or AR.

Externality theory provides the primary entry point for understanding the moderating role of DV. The positive externalities of AGG depend on how efficiently information, knowledge, services, and production factors circulate among agglomerated actors. Weak digital conditions can restrict these flows even when firms and service providers are geographically concentrated. Digital development can reduce these frictions and enable existing agglomeration networks to generate stronger information spillovers, benefits from shared services, and coordination gains [18,37]. Resilience theory explains why these stronger external effects matter for AR. Evidence from organizational settings indicates that digitalization does not automatically generate benefits; its effects depend on whether actors can update existing knowledge and reconfigure resources in response to new information [38]. Faster information transmission and more flexible resource reconfiguration improve the capacity to maintain functions, recover from disruption, and reorganize after shocks. Evidence of a positive association between DV and agricultural resilience supports this connection [18]. This direct association does not by itself establish moderation, which requires the association between AGG and AR to vary with the level of DV.

The positive association between AGG and AR may become stronger as DV advances for three reasons. First, digital platforms improve the timeliness of production, market, weather, and disaster information [18,37]. Information and knowledge spillovers generated by AGG can therefore reach connected actors more quickly. Second, data sharing reduces information asymmetry and improves coordination among farmers, processors, service providers, financial institutions, and public agencies [18,37]. This allows the shared-service and factor-matching benefits of AGG to be used more effectively. Third, digital connectivity reinforces links across production stages and provides alternative channels for resource mobilization during disturbances. Recent evidence based on firm data indicates that integrating digital technologies with real economic activities improves resource allocation and coordination within supply chains, especially in regions with more advanced digital infrastructure [39]. Recent studies generally associate rural digitalization with stronger resilience in agriculture and agricultural supply chains through improved information transmission and resource coordination [18,40,41,42,43]. However, these studies mainly examine direct relationships or individual mechanisms, leaving limited evidence on whether the association between AGG and AR varies with the level of DV. Taken together, these arguments suggest that the positive association between AGG and AR is stronger where DV is more advanced. Accordingly, the following hypothesis is proposed:

H3. 

The positive association between agricultural industry agglomeration and agricultural industry resilience is stronger in regions with higher levels of digital village development.

Figure 1 summarizes the proposed systems framework and the three hypotheses.

Figure 1. Systems framework and hypothesized relationships among system organization, adaptive capacity, digital conditions, and environmental performance. The dashed outlines group the proposed coordination channels for AGG and the components of AR and DV

3. Research Design

3.1. Data Sources and Sample Selection

This study constructs a balanced panel dataset for 30 provincial regions in China from 2011 to 2023, excluding Hong Kong, Macao, Taiwan, and Tibet. The study period was selected for two reasons. First, the underlying data required to construct ACE and AGG are consistently available and sufficiently comparable across provinces during this period. Second, the period captures an important stage in China’s agricultural transformation, beginning with the policy emphasis on sustainable agriculture under the Twelfth Five-Year Plan and extending to the acceleration of the green transition following the announcement of the carbon peaking and carbon neutrality goals in 2020 [44].

Primary data are drawn from the China Statistical Yearbook, China Rural Statistical Yearbook, and provincial statistical yearbooks. Prior to estimation, a standard data preprocessing procedure was applied. Cultivated land area data for 2018 were missing for all provinces and were therefore imputed using linear interpolation. The gross output value of agriculture, forestry, animal husbandry, and fishery used to construct ACE was converted to 2011 constant prices using the corresponding gross output value index, thereby removing the effect of price changes from the denominator of ACE. The disaster variable is transformed using ln(x + 1), while trade openness is transformed using the inverse hyperbolic sine function. The remaining variables enter the regressions in levels.

3.2. Variable Description and Measurement

To ensure the scientific rigor and replicability of the empirical analysis, this section details the definitions, measurement methodologies, and data sources for the primary variables. Table 1 summarizes the definitions and measurement procedures for the variables used in the analysis.

Table 1. Definitions of the variables.

Variable Name	Variable	Definition	Unit
Agricultural carbon emission intensity	ACE	Ratio of aggregate carbon emissions from six agricultural production sources to the real gross output value of agriculture, forestry, animal husbandry, and fishery at 2011 constant prices	t C per 104 CNY at 2011 constant prices
Agricultural industry agglomeration	AGG	Coupling coordination degree between horizontal agricultural production agglomeration and the entropy-weighted vertical agglomeration index of production, processing, and agricultural services	Dimensionless index
Agricultural Industry Resilience	AR	Entropy-weighted composite index based on 10 indicators reflecting robustness, recovery capacity, and reorganization capacity	Dimensionless index
Digital Village Development	DV	Entropy-weighted composite index covering digital infrastructure, industrial digitalization, and digital livelihoods	Dimensionless index
Per Capita GDP	PGDP	GDP/permanent resident population, measured in CNY per person	CNY per person
Urbanization Rate	Urban	Urban population/Total population	Proportion, 0–1
Agricultural Fiscal Support Level	Fiscal	Agriculture, forestry, and water expenditure as a percentage of general public budget expenditure	%
Agricultural Disaster Area	Disaster	ln (Affected agricultural area + 1)	Log-transformed; original unit: 103 ha
Rural Human Capital	HC	Average years of schooling among rural residents aged six and above	Years
Effective Irrigated Area	IRR	Effective irrigated area, measured in thousand hectares	103 ha
Trade Openness	Open	IHS (Total imports and exports GDP)	Dimensionless
Industrial Structure Upgrading	Indus	Tertiary industry value added/GDP	Proportion, 0–1

3.2.1. Agricultural Carbon Emission Intensity (ACE)

Following Li et al. [24] and subsequent provincial studies, aggregate production-related carbon emissions are estimated from six sources: chemical fertilizer, pesticides, agricultural plastic film, agricultural diesel, irrigation, and tillage [4,10,45,46]. Following the specification of Feng et al. [4], ACE is calculated as the ratio of these emissions to the real gross output value of agriculture, forestry, animal husbandry, and fishery at 2011 constant prices. The indicator therefore measures the carbon burden associated with six specified production sources per unit of real provincial agricultural output.

$$C_{\mathit{it}}=\sum _{k=1}^6{C}_{k_{\mathit{it}}}=\sum _{k=1}^6{\delta }_k{T_k}_{\mathit{it}}$$

(1)

In Equation (1), C_it denotes aggregate carbon emissions from the six agricultural production sources in province i and year t; $C_{k_{\mathit{it}}}$ denotes the emissions generated by carbon source k; ${T_k}_{\mathit{it}}$ represents the corresponding activity level; δ_k is the emission coefficient for carbon source k. The emission coefficients are reported in Table 2.

Table 2. Carbon emission coefficients of major agricultural sources.

Carbon Source	Coefficient	Unit	Reference
Fertilizer	0.8956	kg C/kg	[24]
Pesticide	4.9341	kg C/kg	[24]
Agricultural films	5.1800	kg C/kg	[24]
Diesel	0.5927	kg C/kg	[24]
Tillage	312.60	kg C/km2	[24,47]
Irrigation	20.476	kg C/hm2	[24,48]

Note: All emission coefficients are expressed in units of carbon, and kg C denotes kilograms of carbon. Dubey and Lal [48] report a baseline irrigation coefficient of 25 kg C/hm². The adjusted coefficient used in this study is 20.476 kg C/hm², obtained by applying the adjustment factor for thermal power generation reported by Li et al. [24]. Before aggregation, each activity variable is converted to the unit required by its corresponding emission coefficient.

Based on the total emissions obtained from Equation (1), ACE is calculated as follows:

$$\mathit{AC}{E}_{\mathit{it}}={\displaystyle \frac{C_{\mathit{it}}}{A{V}_{\mathit{it}}{,}_{2011}}}$$

(2)

In Equation (2), ACE_it denotes agricultural carbon emission intensity for province i and year t. C_it represents aggregate agricultural carbon emissions from the six specified sources, as calculated using Equation (1). AV_it,₂₀₁₁ denotes the real gross output value of agriculture, forestry, animal husbandry, and fishery for province i and year t, expressed at 2011 constant prices. The nominal gross output value is converted to 2011 constant prices using a price index for which 2011 is the base year and the index is set to 100. Equation (2) expresses the emissions estimated from the six specified sources relative to the real gross output value of agriculture, forestry, animal husbandry, and fishery. This treatment reduces the influence of production scale and facilitates comparisons across provinces and years.

3.2.2. Agricultural Industry Agglomeration (AGG)

International studies commonly use the location quotient to measure the relative spatial concentration or specialization of agricultural activities because it facilitates comparisons across regions of different economic sizes [12,20]. At the same time, a provincial agricultural industry system extends beyond primary production and includes vertically connected processing and service activities [5]. The conceptual distinction between horizontal and vertical agglomeration follows Kalathil et al. [13], while the operational coupling procedure is adapted from Wang [23]. Accordingly, this study measures Agricultural Industry Agglomeration (AGG) from two complementary dimensions: horizontal concentration in agricultural production and vertical agglomeration across production, processing, and agricultural services.

The location quotient is calculated separately for the three components of the vertical subsystem. The production location quotient compares the share of agricultural output in provincial GDP with the corresponding share at the national level. The processing location quotient compares revenue from agricultural product processing as a share of provincial industrial revenue with the corresponding national share. The service location quotient compares the output share of specialized and auxiliary activities in agriculture, forestry, animal husbandry, and fishery with the corresponding national share. The inclusion of the service component is also consistent with the use of location quotients to identify the regional concentration of producer services [49]. The three components are combined using weights determined by the entropy method to construct the Vertical Agglomeration Index (VAI), as reported in Table 3.

Table 3. Measurement Framework for Vertical Agglomeration Indicators (VAI).

Dimension	Indicator	Calculation Method	Weight
Production	LQ1	(Regional Ag-Output/National Ag-Output)/(Regional GDP/National GDP)	0.1745
Processing	LQ2	(Regional Ag-Processing Revenue/National Ag-Processing Revenue)/(Regional Total Industrial Revenue/National Total Industrial Revenue)	0.5909
Services	LQ3	(Regional Ag-Service Output/National Ag-Service Output)/(Regional GDP/National GDP)	0.2346

Horizontal agglomeration (HAI) is represented by the production location quotient. The production component therefore enters the two subsystems for different analytical purposes. In HAI, it independently measures the relative spatial concentration of primary agricultural production. In VAI, it represents the upstream production component required to evaluate the coordinated concentration of production, processing, and agricultural services. AGG is subsequently calculated as the coupling coordination degree between HAI and VAI using Equations (4)–(6). A higher AGG value indicates stronger coordination between the horizontal concentration of agricultural production and the vertical concentration of activities along the agricultural industry chain.

VAI_i = w₁ × LQ_1i + w₂ × LQ_2i + w₃ × LQ_3i

(3)

where w_m represents the entropy weight for dimension m. To capture the coordination between horizontal concentration and vertical agglomeration, a coupling coordination model comprising two subsystems is constructed. We define horizontal agglomeration (HAI) using the production location quotient (HAI_i = LQ_1i), which represents the foundational spatial clustering of agricultural activities. The production location quotient serves different analytical roles in the two subsystems. In HAI, it measures the spatial concentration of primary agricultural production, whereas in VAI, it represents the upstream production component required to evaluate coordination across production, processing, and agricultural services. The comprehensive agglomeration index (AGG) is then computed as the coupling coordination degree (D) between the HAI and VAI subsystems:

$$D_i=\sqrt{C_i\times T_i}$$

(4)

The coupling degree C measures the mutual reinforcement intensity between the two subsystems:

$$C_i=2\times \sqrt{{\displaystyle \frac{\mathit{HA}{I}_i\times \mathit{VA}{I}_i}{{(\mathit{HA}{I}_i+\mathit{VA}{I}_i)}^2}}}$$

(5)

Finally, the comprehensive coordination index T represents the overall development level of both dimensions, with equal contributions assumed:

T_i = α ⋅ HAI_i + β ⋅ VAI_i, α = β = 0.5

(6)

A higher AGG value indicates a higher level of coordinated development between production-based spatial concentration and the concentration of production, processing, and agricultural service activities.

3.2.3. Explanation of Other Variables

(1) Agricultural Industry Resilience (AR). Agricultural Industry Resilience (AR) is measured using a framework of system capacities derived from research on agricultural and regional resilience. Agricultural resilience studies distinguish the capacity to withstand disturbances from the capacities to adapt operations and transform system structures [16,34]. Regional economic resilience research further emphasizes resistance, recovery, and the reorientation of economic structures following shocks [31]. Adapting these perspectives to the provincial agricultural industry system, this study evaluates AR through three dimensions: robustness, recovery, and reorganization.

Robustness refers to the capacity to maintain core agricultural production functions during disturbances. Recovery captures the capacity to restore production and resource flows after a disruption. Reorganization reflects the capacity to reallocate resources and adjust internal structures in response to changing environmental and market conditions. Consistent with empirical studies published in English that construct multidimensional agricultural resilience indices for Chinese provinces [35,50], the ten indicators reported in Appendix Table A1 are combined using the entropy weight method. The index comprises four indicators of robustness, four indicators of recovery, and two indicators of reorganization. The entropy weight method determines indicator weights from their observed variation, thereby reducing reliance on subjectively assigned weights. Wang [23] is retained as a supplementary reference for adapting the specific indicators to the availability and statistical definitions of Chinese provincial data.

(2) Digital Village Development (DV). Digital Village Development (DV) is measured using a composite index that captures the integration of digital technologies across multiple dimensions of rural systems. Drawing on the framework of Wang and Guo [40], we evaluate DV across three dimensions: digital infrastructure, industrial digitalization, and digital livelihoods. This design aligns with the broader transition toward Digital Agriculture 4.0 [37] and adopts representative metrics tailored to China’s rural context [36]. To mitigate subjective bias in weighting, the composite index is subsequently constructed via the entropy weight method, following Lei et al. [51]. The complete measurement framework for DV is detailed in Appendix Table A2.

(3) Control Variables. To mitigate potential omitted variable bias, this study incorporates a set of control variables grounded in established literature, including Per Capita GDP(PGDP), Urbanization Rate (Urban), Agricultural Fiscal Support Level (Fiscal), Agricultural Disaster Area (Disaster), Rural Human Capital (HC), Effective Irrigation Area (IRR), Trade Openness (Open) and Industrial Structure Upgrading (Indus). These variables capture important economic, structural, environmental, and institutional factors that may simultaneously influence ACE and AGG. Detailed variable definitions, measurement approaches, and expected directional effects are summarized in Table 1. Additionally, HC is calculated as (the number of rural residents with primary education × 6 + the number of rural residents with junior secondary education × 9 + the number of rural residents with senior secondary or technical secondary education × 12 + the number of rural residents with a college degree or above × 16) divided by the total rural population aged six and above.

3.3. Model Specification

3.3.1. Baseline Model Specification

To empirically examine the association between AGG and ACE, this study specifies the following baseline regression model (7):

ACE_it = β₀ + β₁AGG_it + β₂Cont_it + μ_i + λ_t + ϵ_it

(7)

where i and t index provinces and years, respectively. ACE_it denotes agricultural carbon emission intensity for provinces i in year t. AGG_it denotes agricultural industry agglomeration and serves as the core explanatory variable. Cont_it denotes a vector of control variables. μ_i denotes province fixed effects, which account for unobserved provincial characteristics that do not vary over time. λ_t denotes year fixed effects, which absorb macroeconomic shocks and other temporal changes common to all provinces. ϵ_it denotes the idiosyncratic error term.

The parameter of primary interest is β₁, which estimates the conditional association between AGG and ACE. If β₁ < 0 and is statistically significant, it indicates that a higher level of AGG is negatively associated with ACE. Finally, standard errors are clustered at the provincial level to account for potential heteroskedasticity and serial correlation within each province.

3.3.2. Mediation-Effect Model

To assess whether the estimated relationships are consistent with a mediating role for AR, this study follows the sequential regression procedure discussed by Jiang [52]. Model (8) estimates the association between AGG and AR. Model (9) then estimates the association between AR and ACE after accounting for AGG.

AR_it = α₀ + α₁AGG_it + α₂Cont_it + μ_i + λ_t + ϵ_it

(8)

ACE_it = γ₀ + γ₁AGG_it + γ₂AR_it + γ₃Cont_it + μ_i + λ_t + ϵ_it

(9)

where AR_it denotes the index of agricultural industry resilience, which represents the dynamic capacity of the agricultural system to withstand, recover from, and adapt to external shocks. The indirect association is calculated as α₁γ₂, where α₁ represents the estimated association between AGG and AR and γ₂ represents the estimated association between AR and ACE after accounting for AGG. It is evaluated using a province-level cluster bootstrap with 1000 replications. Provinces are resampled with replacement, and all annual observations within each selected province are retained. A bias-corrected 95% confidence interval that excludes zero is interpreted as evidence consistent with an intermediary role for AR.

3.3.3. Moderating-Effect Model

Model (10) is constructed to examine whether the association between AGG and AR varies with the level of DV.

AR_it = η₀ + η₁AGG_it + η₂DV_it + η₃(AGG_it × DV_it) + η₄Cont_it + μ_i + λ_t + ϵ_it

(10)

where DV_it denotes the level of digital village development. A positive and statistically significant η₃ indicates that the association between AGG and AR becomes stronger as DV increases. To interpret this interaction, the conditional marginal association between AGG and AR is estimated at the mean of DV and at values one standard deviation below and above the mean. The corresponding 95% confidence intervals are calculated using the delta method and the province-clustered covariance matrix from Model (10).

4. Empirical Results and Analysis

4.1. Descriptive Statistics

Table 4 reports the descriptive statistics for all variables. The dependent variable (ACE) exhibits substantial cross-provincial heterogeneity (mean = 0.113). The key independent variable (AGG) averages 0.446, ranging from 0.029 to 0.774, reflecting substantial differences in AGG across provinces. DV shows a relatively low mean (0.207) but considerable dispersion, underscoring the uneven geographic distribution of rural digitalization. Control variables generally display reasonable distributions without extreme outliers. The descriptive statistics indicate sufficient variation across provinces and years, providing empirical support for subsequent panel regression analysis.

Table 4. Descriptive statistics for each variable.

Variable	N	Mean	Median	Std. Dev.	Min	Max
AGG	390	0.446	0.473	0.149	0.029	0.774
ACE	390	0.113	0.089	0.130	0.007	0.940
AR	390	0.270	0.242	0.110	0.092	0.631
DV	390	0.207	0.177	0.144	0.017	0.752
PGDP	390	61,004.385	52,394.000	32,262.530	16,024.000	200,278.000
Urban	390	0.602	0.591	0.121	0.344	0.942
Fiscal	390	11.306	11.376	3.371	3.870	20.384
HC	390	9.329	9.240	0.913	7.470	12.780
IRR	390	2225.161	1685.480	1714.779	109.240	6666.390
Indus	390	0.484	0.479	0.097	0.297	0.849
Disaster	390	5.776	6.206	1.633	0.000	8.349
Open	390	0.250	0.144	0.237	0.008	1.153

Table 5 reports the pairwise correlation coefficients for the study variables. The unconditional correlation between AGG and ACE is negligible and statistically insignificant. AGG and AR exhibit a weak negative correlation, whereas AR is negatively correlated with ACE. These bivariate relationships do not account for observed covariates or unobserved provincial characteristics and therefore should not be interpreted as evidence of conditional or causal relationships. Several covariates exhibit relatively high pairwise correlations. However, the VIF diagnostics for the baseline specification yield a maximum value of 6.22 for Urban and a mean value of 4.02. All VIF values remain below the conventional threshold of 10, suggesting that severe multicollinearity is unlikely to be a dominant concern. Because bivariate correlations do not account for observed covariates, temporal dynamics, or unobserved time-invariant provincial characteristics, the subsequent analysis employs panel fixed-effects models.

Table 5. Correlation Matrix.

	AGG	ACE	AR	DV	PGDP	Urban
AGG	1
ACE	0.006	1
AR	−0.136 *	−0.314 *	1
DV	−0.341 *	−0.176 *	0.700 *	1
PGDP	−0.708 *	−0.260 *	0.423 *	0.701 *	1
Urban	−0.695 *	−0.210 *	0.235 *	0.479 *	0.819 *	1
Fiscal	0.751 *	0.202 *	−0.347 *	−0.421 *	−0.608 *	−0.627 *
HC	−0.610 *	−0.356 *	0.209 *	0.354 *	0.776 *	0.873 *
IRR	0.482 *	−0.186 *	0.535 *	0.095	−0.217 *	−0.275 *
Indus	−0.517 *	−0.225 *	0.212 *	0.492 *	0.750 *	0.733 *
Disaster	0.573 *	0.014	−0.044	−0.355 *	−0.658 *	−0.648 *
Open	−0.749 *	−0.246 *	0.321 *	0.321 *	0.640 *	0.722 *
	Fiscal	HC	IRR	Indus	Disaster	Open
Fiscal	1
HC	−0.562 *	1
IRR	0.286 *	−0.200 *	1
Indus	−0.370 *	0.752 *	−0.312 *	1
Disaster	0.500 *	−0.563 *	0.533 *	−0.662 *	1
Open	−0.756 *	0.659 *	−0.301 *	0.575 *	−0.583 *	1
	Fiscal	HC	IRR	Indus	Disaster	Open

Note. Pairwise correlation coefficients are reported. * p < 0.10.

4.2. Baseline Regression Results

Table 6 reports four specifications. Column (1) presents pooled ordinary least squares estimates. This specification stacks all province-year observations and estimates a common relationship without controlling for unobserved province or year effects. Columns (2) and (3) add year and province fixed effects separately, while Column (4) includes both sets of fixed effects and serves as the preferred specification. The coefficient on AGG is negative across all four specifications.

Table 6. Baseline Regression Results.

Variable	(1)	(2)	(3)	(4)
AGG	−0.549 * (−1.91)	−0.546 * (−1.96)	−0.401 *** (−6.69)	−0.365 *** (−6.98)
PGDP	−0.000 ** (−2.54)	−0.000 ** (−2.43)	−0.000 (−1.31)	−0.000 (−0.34)
Urban	0.447 (1.29)	0.563 (1.33)	−0.072 (−0.94)	−0.046 (−0.37)
Fiscal	0.009 (0.95)	0.008 (0.91)	−0.002 (−1.13)	−0.002 (−0.98)
HC	−0.080 (−1.39)	−0.096 (−1.39)	−0.029 (−1.59)	−0.024 (−1.57)
IRR	0.000 (0.00)	0.000 (0.29)	0.000 ** (2.29)	0.000 ** (2.40)
Indus	−0.019 (−0.10)	0.171 (0.70)	0.006 (0.11)	−0.007 (−0.09)
Disaster	−0.022 (−1.24)	−0.021 (−1.21)	−0.000 (−0.15)	0.000 (0.18)
Open	−0.204 (−1.68)	−0.266 (−1.53)	0.019 (0.79)	0.045 (0.88)
Constant	1.024 (1.68)	1.017 (1.69)	0.601 *** (3.41)	0.508 *** (2.75)
Year FE	No	Yes	No	Yes
Province FE	No	No	Yes	Yes
N	390	390	390	390
R2	0.352	0.365	0.965	0.967
Adj R2	0.336	0.329	0.962	0.962
F	2.16	1.34	17.41	14.58

Notes. T-statistics are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10.

4.3. Robustness and Endogeneity Checks

Table 7 presents five robustness tests. Column (1) winsorizes all continuous variables at the 1st and 99th percentiles. Column (2) replaces current AGG with its value from the preceding period. Column (3) excludes observations from 2020 through 2022. The estimate changes little, suggesting that the baseline result is not driven solely by disruptions associated with COVID-19. Column (4) adds a separate linear trend for each province to the model with province and year fixed effects. Column (5) uses an alternative AGG measure. Under this construction, VAI includes only processing and service agglomeration, with equal weights assigned to the two components before VAI is coupled with HAI. The coefficient is −0.276 and remains statistically significant at the 5% level. Overall, the estimates are stable across alternative data treatments, sample periods, model specifications, and AGG constructions.

Table 7. Robustness test results.

Variable	(1)	(2)	(3)	(4)	(5)
	Winsor	Lagged AGG	Sub-Sample	Province Trend	Alternative AGG
AGG	−0.350 *** (−6.05)	−0.193 *** (−3.38)	−0.358 *** (−6.97)	−0.402 *** (−6.14)	−0.276 ** (−2.74)
Controls	YES	YES	YES	YES	YES
Constant	0.420 *** (2.99)	0.384 *** (2.67)	0.510 *** (2.90)	0.424 *** (2.87)	0.430 ** (2.59)
Year FE	Yes	Yes	Yes	Yes	YES
Province FE	Yes	Yes	Yes	Yes	YES
N	390	360	300	390	390
R2	0.971	0.964	0.979	0.991	0.965
Adj R2	0.967	0.959	0.975	0.989	0.960
F	10.48	4.34	8.55	11.56	6.14

Notes. T-statistics are reported in parentheses. *** p < 0.01, ** p < 0.05.

A lead-based falsification test is further conducted to examine potential reverse temporal ordering. The model includes both current AGG and its one-period lead while retaining the baseline control variables, province fixed effects, and year fixed effects. If future AGG is significantly associated with current ACE after current AGG is taken into account, the estimated association may reflect reverse temporal ordering. By contrast, an insignificant coefficient on the lead term would be consistent with the temporal ordering specified in the empirical framework. Table 8 reports the results.

Table 8. Endogeneity Check Using the One-Period Lead of AGG.

Variable	ACE
Current AGG	−0.349 *** (−6.18)
One-period lead of AGG	−0.032 (−0.94)
Control variables	Yes
Year FE	Yes
Province FE	Yes
N	360
R2	0.968
Adj R2	0.963
F	13.98
p-value for one-period lead of AGG	0.355

Notes. The F tests the joint significance of the explanatory variables. The p-value for the one-period lead of AGG corresponds to the null hypothesis that its coefficient equals zero. T-statistics are reported in parentheses. *** p < 0.01.

Table 8 shows that current AGG remains negatively associated with ACE at the 1% level, whereas its one-period lead is statistically insignificant, with a p-value of 0.355. This pattern is consistent with the temporal ordering specified in the empirical framework and complements the lagged AGG evidence reported in Table 7.

4.4. Mechanism Analysis

4.4.1. Mediating Effect

Table 9 reports the results from the two mediation equations. In Column (1), the coefficient on AGG is positive and statistically significant. This result indicates that an increase in AGG within a province is associated with higher AR after accounting for the baseline controls, province fixed effects, and year fixed effects. In Column (2), the coefficient on AR is negative and statistically significant. The coefficient on AGG also remains negative and statistically significant, but its absolute magnitude declines from 0.365 in the baseline model to 0.312 after AR is included. These coefficient patterns provide the component relationships used to estimate the indirect association reported in Table 10.

Table 9. Mediation analysis results.

Variable	(1)	(2)
AGG	0.303 ** (2.42)	−0.312 *** (−7.74)
AR		−0.176 ** (−2.21)
Controls	YES	YES
Constant	0.028 (0.13)	0.513 *** (2.95)
Year FE	Yes	Yes
Province FE	Yes	Yes
R2	0.950	0.968
Adj R2	0.942	0.963
F	4.27	29.16
N	390	390

Notes. Column (1) reports the mediator equation with AR as the dependent variable. Column (2) reports the outcome equation with ACE as the dependent variable. T-statistics are reported in parentheses. *** p < 0.01, ** p < 0.05.

Table 10. Province-Level Cluster Bootstrap Results for the Estimated Indirect Association through AR.

Effect	Coefficient	Boot SE	95% Confidence Interval
Indirect association (AGG → AR → ACE)	−0.053	0.046	[−0.173, −0.010]
Direct effect	−0.312	0.087	[−0.523, −0.125]
Total effect	−0.365	0.102	[−0.652, −0.233]

Notes: Bootstrap standard errors and bias-corrected 95% confidence intervals are based on 1000 province-level cluster bootstrap replications. Provinces are resampled with replacement, with all observations within each selected province retained.

Table 10 reports the province-level cluster bootstrap results for the indirect association through AR. The estimated indirect association is −0.053, with a bias-corrected 95% confidence interval of [−0.173, −0.010]. Because this interval excludes zero, the result is consistent with AR serving as a partial intermediary pathway in the association between AGG and ACE. The direct association remains negative, indicating that AR accounts for part of the estimated association between AGG and ACE.

4.4.2. Moderating Effect

Table 11 reports the moderating effect of DV on the relationship between AGG and AR. In Column (3), the coefficient on the interaction between AGG and DV is 1.4923 and statistically significant at the 1% level. This estimate indicates that the conditional association between AGG and AR increases with DV.

Table 11. Moderating Effect of DV on the Relationship between AGG and AR.

Variable	(1)	(2)	(3)
AGG	0.303 ** (2.42)	0.305 ** (2.51)	0.268 ** (2.51)
DV		0.069 (0.69)	0.159 ** (2.41)
AGG × DV			1.492 *** (4.51)
Controls	YES	YES	YES
Constant	0.028 (0.13)	0.020 (0.10)	−0.032 (−0.16)
Year FE	Yes	Yes	Yes
Province FE	Yes	Yes	Yes
R2	0.950	0.950	0.965
Adj R2	0.942	0.943	0.960
F	4.40	4.27	12.23
N	390	390	390

Notes. In Column (3), AGG and DV are centered at their sample means before the interaction term is constructed. T-statistics are reported in parentheses. *** p < 0.01, ** p < 0.05.

The simple slope estimates clarify the interaction. At one standard deviation below the mean of DV, the estimated association between AGG and AR is 0.054 and statistically insignificant, with a 95% confidence interval from −0.174 to 0.281 and p = 0.645. At the mean and one standard deviation above it, the estimates increase to 0.268 and 0.483, with corresponding 95% confidence intervals from 0.059 to 0.477 and from 0.252 to 0.713; p = 0.012 and p < 0.001, respectively. Figure 2 presents these estimates and confidence intervals. The stronger association at higher levels of DV is consistent with H3.

Figure 2. Conditional Association between AGG and AR at Selected Levels of DV. Notes: Low, mean, and high DV correspond to values of 0.063, 0.207, and 0.351, respectively, representing one standard deviation below the sample mean, the sample mean, and one standard deviation above it. The 95% confidence intervals are calculated using the delta method and the covariance matrix clustered at the province level in Model (10).

4.5. Heterogeneity Analysis

Following Du [53], provinces are classified according to their grain production and consumption functions into major grain production areas, major grain consumption areas, and grain production and consumption balance areas. Columns (1) to (3) of Table 12 report the corresponding subsample estimates. The coefficient on AGG is negative and statistically significant in the producing and consuming areas, but negative and insignificant in the balance area, with a larger absolute point estimate in consuming areas. Formal coefficient equality tests show that only the difference between the major grain production areas and the balance areas is marginally significant at the 10% level. This difference becomes statistically insignificant when wild cluster bootstrap inference is applied, while the remaining pairwise comparisons and the joint test are also insignificant. Overall, the tests do not establish a stable pattern of heterogeneity across functional zones. The negative relationship between AGG and ACE is therefore not confined to a particular grain production and consumption category, although its magnitude should not be assumed to be identical across regions.

Table 12. Heterogeneity Analysis across Grain Functional Zones.

Variable	Balance Areas (1)	Producing Areas (2)	Consuming Areas (3)
AGG	−0.3079 (−0.73)	−0.3650 *** (−7.08)	−0.8745 *** (−6.10)
Controls	YES	YES	YES
Constant	0.5993 ** (2.30)	0.4512 *** (3.71)	0.5782 *** (5.62)
Year FE	YES	YES	YES
Province FE	YES	YES	YES
N	130	169	91
R2	0.951	0.984	0.829

Note: t-statistics are reported in parentheses. *** p < 0.01, ** p < 0.05.

Consistent with Ye et al. [54], unconditional quantile regression is employed to examine whether the association between AGG and ACE varies across the unconditional distribution of ACE. Columns (1) to (5) of Table 13 report the estimates at the 10th, 25th, 50th, 75th, and 90th percentiles, denoted by Q10, Q25, Q50, Q75, and Q90, respectively. The estimated AGG coefficients tend to be larger in absolute magnitude at the lower percentiles of ACE. However, the coefficient differences across the quantile pairs for which bootstrap inference is available are not statistically significant. The results therefore do not provide evidence of systematic heterogeneity in the effect of AGG across the ACE distribution. Comparisons involving Q90 should be interpreted cautiously because stable bootstrap standard errors could not be obtained.

Table 13. Unconditional Quantile Regression Results across the ACE Distribution.

Variable	Q10 (1)	Q25 (2)	Q50 (3)	Q75 (4)	Q90 (5)
AGG	−0.5500 *** (−3.09)	−0.5693 *** (−3.51)	−0.3115 * (−1.87)	−0.1828 * (−1.70)	−0.3740 (−1.29)
Controls	YES	YES	YES	YES	YES
Constant	0.5177 * (1.70)	0.3204 (1.55)	0.1676 (0.81)	0.5636 ** (2.09)	0.6669 * (1.81)
Year FE	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes
N	390	390	390	390	390
R2	0.2779	0.3918	0.3863	0.2458	0.1027

Notes. Q10, Q25, Q50, Q75, and Q90 denote the 10th, 25th, 50th, 75th, and 90th percentiles of the unconditional distribution of ACE, respectively. t-statistics are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10.

Table 14 summarizes the core empirical relationships and their links to the three hypotheses.

Table 14. Summary of the Core Empirical Relationships.

Relationship Path	Type of Relationship	Main Empirical Evidence	Hypothesis
AGG → ACE	Direct negative association	The AGG coefficient is negative and statistically significant in the preferred specification and remains stable across the reported robustness checks. Formal comparison tests do not identify systematic differences across grain functional zones or ACE quantiles.	Consistent with H1
AGG → AR → ACE	Indirect association through AR	AGG is positively associated with AR, while AR is negatively associated with ACE after accounting for AGG. The bias-corrected 95% confidence interval from the province-level cluster bootstrap excludes zero.	Consistent with H2
AGG × DV → AR	Positive moderation	The interaction coefficient is positive and statistically significant. The simple slope estimates become larger as DV rises.	Consistent with H3

Notes. The reported relationships are interpreted as conditional associations.

5. Discussion and Policy Recommendations

5.1. Discussion

The baseline results show a significant negative association between AGG and ACE. This association remains stable across the alternative specifications, while the insignificant lead term is consistent with the temporal ordering specified in the empirical framework. Previous studies have reported different environmental effects of agricultural industrial agglomeration. Some studies find that agricultural industrial agglomeration improves agricultural green production efficiency [12], whereas others identify U-shaped or inverted U-shaped effects on agricultural green development or environmental efficiency [3,7]. Despite these different findings, the studies share a common measurement feature. They use location quotients to capture the regional specialization of agricultural production or the primary industry, thereby focusing mainly on spatial concentration within the production segment. This study extends that perspective by combining horizontal production concentration with vertical agglomeration across production, processing, and agricultural services and measuring AGG through their coupling coordination. The negative estimate therefore links lower ACE to stronger coordination across the horizontal and vertical agglomeration subsystems, extending the analytical focus beyond production concentration. From a systems perspective, AGG captures structural coordination within the agricultural industrial system. Shared infrastructure, resource circulation, and technological diffusion offer plausible system-level interpretations of this relationship.

The province-level cluster bootstrap results are consistent with an intermediary role for AR in the negative association between AGG and ACE. This finding accords with evidence that agricultural industrial agglomeration strengthens agricultural economic resilience through socialized services and improved production efficiency [14]. It is also consistent with farming systems research that defines resilience as the capacity to maintain essential functions and adapt under disturbances [16]. In the present analysis, higher AGG is associated with stronger AR, which is in turn associated with lower ACE. From a systems perspective, robustness may limit output losses during disturbances, recovery may shorten imbalances between resource use and output, and reorganization may facilitate adjustments that improve resource efficiency. Together, these capacities help interpret the estimated indirect association between AGG and ACE through AR from a systems perspective.

The interaction and simple slope results indicate that the positive association between AGG and AR is larger at higher levels of DV. At the low DV level, the estimated association is small and not statistically significant. At the mean and high DV levels, it is positive and statistically significant. This pattern is consistent with evidence that digital village development improves access to information and resource allocation [18] and that digital technologies strengthen linkages among agricultural production, processing, and services [5]. Digital infrastructure, information sharing, and platform connectivity can accelerate the transmission of production, market, and disaster information. These conditions may facilitate resource flows and coordinated adjustment among agricultural actors. From a systems perspective, DV can be interpreted as an enabling condition under which structural coordination is more strongly associated with adaptive capacity.

The AGG coefficients are negative in all three grain functional zones, although the estimate for the production and consumption balance areas is not statistically significant. Previous studies report regional differences in the nonlinear effects of agricultural industrial agglomeration [8], but the formal coefficient comparison tests and UQR results provide no stable evidence of heterogeneity. The negative association between AGG and ACE therefore does not appear to be confined to a particular type of region and shows a degree of generality within the study sample. However, insignificant differences across groups do not imply that the effect is identical in all regions. Differences in coefficient magnitude and estimation precision, together with the instability of the Q90 results, still require cautious interpretation. From a systems perspective, structural coordination may produce similar environmental benefits across different regional systems, although the strength of these benefits remains dependent on regional conditions.

Taken together, the findings describe an integrated pattern within provincial agricultural industrial systems. Higher AGG is associated with lower ACE partly through AR, and the positive association between AGG and AR is stronger at higher levels of DV. By linking system organization, adaptive capacity, digital conditions, and carbon performance, this framework extends agricultural agglomeration research beyond the environmental implications of production concentration.

5.2. Limitations and Future Research Directions

Several limitations define useful directions for further research. First, provincial data provide a consistent basis for nationwide comparisons, while finer-scale data could reveal variation within provinces. Second, ACE relates emissions from six specified agricultural production sources to the real gross output value of agriculture, forestry, animal husbandry, and fishery. This accounting boundary focuses the analysis on emissions associated with agricultural production inputs. Future research could incorporate rice cultivation, enteric fermentation, manure management, agricultural soils, and straw burning, and examine alternative output denominators [55,56,57]. Third, the fixed-effects specifications, lagged AGG model, and temporal falsification test support the stability and temporal ordering of the estimated association. Research designs drawing on external sources of variation could further strengthen causal identification. Finally, the findings reflect China’s agricultural and institutional context. Comparative studies could assess their applicability under different agricultural structures and institutional arrangements. The baseline framework treats each province as the analytical boundary of the agricultural industry system. This choice aligns the analysis with the available data and the provincial policy context, while leaving interprovincial spillovers for further examination. Spatial econometric models with alternative spatial weight matrices could assess whether the relationship between AGG and ACE extends across provincial boundaries and examine how such spillovers are transmitted. This extension would clarify the spatial scope of the estimated relationship.

Future research can build on the provincial DV index with farm or county data to examine how specific digital services relate to resource coordination, resilience, and carbon intensity [58,59]. It could also test whether these relationships vary by crop and climate. Research on highland barley illustrates why these conditions matter: adaptation measures expanded suitable growing areas on the Tibetan Plateau [60].

5.3. Policy Suggestions

Based on the empirical findings, three policy implications are proposed.

First, policies on agricultural industry agglomeration should focus on coordination among production, processing, and agricultural services. Shared logistics, machinery services, and common resource-use standards can reduce duplicated inputs and improve factor matching. The evaluation of agricultural industrial parks should consider the quality of cross-activity coordination alongside their physical scale. Local pilot evaluation is needed before these measures are replicated more broadly.

Second, resilience-oriented investment should focus on the capacities represented by AR, including maintaining production during disturbances, restoring resource flows, and reorganizing inefficient arrangements. Early-warning systems, resilient irrigation, and agricultural risk-management services are practical priorities. Their implementation should be evaluated through changes in recovery time, output stability, and input efficiency.

Third, digital village policies should improve the information and coordination conditions surrounding existing agglomerations. Priority should be given to interoperable data services, accessible market and disaster information, digital support for efficient agricultural input use, and practical digital training for farmers and service providers. Practice-oriented education has been shown to strengthen engagement with agricultural entrepreneurship, highlighting the value of applied learning in agricultural capacity building [61]. Capacity-building programs should be adapted to participants’ backgrounds and operational needs. These measures are particularly relevant where limited digital access constrains the conversion of agglomeration into resilience.

6. Conclusions

Using provincial panel data from China, this study examines the relationships among AGG, ACE, AR, and DV. Higher AGG is significantly associated with lower ACE. AGG is positively associated with AR, and AR is negatively associated with ACE. The bias-corrected bootstrap confidence interval for the indirect association excludes zero, which is consistent with a partial intermediary role for AR. The interaction and simple slope estimates indicate that the positive association between AGG and AR is larger at higher levels of DV. The negative association between AGG and ACE is broadly observed across grain functional zones and ACE quantiles, while formal comparison tests find no systematic differences across zones or quantiles. Taken together, these findings are consistent with the three proposed hypotheses. By combining a coordination measure of horizontal and vertical agglomeration with a unified systems framework and evidence across zones and quantiles, this study extends research on agricultural decarbonization from a systems perspective.

Appendix A.1. Agricultural Industry Resilience

Table A1 presents the ten indicators used to measure agricultural industry resilience across robustness, recovery capacity, and reorganization capacity.

Table A1. Measurement framework for Agricultural Industry Resilience.

Dimension	Indicator	Calculation	Unit	Direction
Robustness	Share of effective irrigated area	Effective irrigated area/Cultivated land area	%	+
	Agricultural mechanization level	Total agricultural machinery power/Cultivated land area	kW/ha	+
	Grain production capacity	Total grain output	104 t	+
	Agricultural economic share	Primary industry added value/Regional GDP	%	+
Recovery	Rural residents’ per capita disposable income	Rural residents’ per capita disposable income as reported in the statistical yearbook	CNY	+
	Agricultural land productivity	Gross agricultural output value/Sown area of crops	104 CNY/ha	+
	Rural electricity consumption	Rural residential electricity consumption	108 kWh	+
	Financial support for agriculture	Local fiscal expenditure on agriculture, forestry and water affairs	108 CNY	+
Reorganization	Agricultural production service level	Output of specialized and auxiliary activities in farming, forestry, animal husbandry and fishery/Primary industry added value	%	+
	Average education years of rural residents	the number of rural residents with primary education × 6 + the number of rural residents with junior secondary education × 9 + the number of rural residents with senior secondary or technical secondary education × 12 + the number of rural residents with a college degree or above × 16	Years	+

Note: All data are obtained from the China Rural Statistical Yearbook for 2011 to 2023. The composite index is constructed using the entropy weight method described in Section 3.2.3.

Table A1. Measurement framework for Agricultural Industry Resilience.

Dimension	Indicator	Calculation	Unit	Direction
Robustness	Share of effective irrigated area	Effective irrigated area/Cultivated land area	%	+
	Agricultural mechanization level	Total agricultural machinery power/Cultivated land area	kW/ha	+
	Grain production capacity	Total grain output	104 t	+
	Agricultural economic share	Primary industry added value/Regional GDP	%	+
Recovery	Rural residents’ per capita disposable income	Rural residents’ per capita disposable income as reported in the statistical yearbook	CNY	+
	Agricultural land productivity	Gross agricultural output value/Sown area of crops	104 CNY/ha	+
	Rural electricity consumption	Rural residential electricity consumption	108 kWh	+
	Financial support for agriculture	Local fiscal expenditure on agriculture, forestry and water affairs	108 CNY	+
Reorganization	Agricultural production service level	Output of specialized and auxiliary activities in farming, forestry, animal husbandry and fishery/Primary industry added value	%	+
	Average education years of rural residents	the number of rural residents with primary education × 6 + the number of rural residents with junior secondary education × 9 + the number of rural residents with senior secondary or technical secondary education × 12 + the number of rural residents with a college degree or above × 16	Years	+

Note: All data are obtained from the China Rural Statistical Yearbook for 2011 to 2023. The composite index is constructed using the entropy weight method described in Section 3.2.3.

Appendix A.2. Digital Village Development

Table A2 presents the ten indicators used to measure digital village development across digital infrastructure, industrial digitalization, and digital livelihoods.

Table A2. Measurement framework for Digital Village Development.

Dimension	Indicator	Calculation	Unit	Direction	Data Source
Digital Infrastructure	Mobile phones owned per 100 rural households	Number of mobile phones owned per 100 rural households	units per 100 households	+	China Statistical Yearbook
	Population coverage rate of rural radio and television programs	Population covered by rural radio and television programs/Total rural population × 100%	%	+	China Radio & Television Yearbook
	Number of rural postal service outlets	Total number of rural postal service outlets	outlets	+	China Postal Statistical Yearbook
	Rural Internet penetration	Administrative villages covered by 4G networks/Total administrative villages × 100%	%	+	China Communications Industry Statistical Yearbook
Industrial Digitalization	Rural online retail sales	Total value of rural online retail sales	108 CNY	+	China E-commerce Report
	Number of Taobao Villages	Total number of officially identified Taobao villages	Units	+	AliResearch Report on the Development of Taobao Villages in China
	Digital Financial Inclusion Index	Digital Financial Inclusion Index score	Index	+	Peking University Digital Finance Research Center
Digital Livelihoods	Per capita transportation and communications expenditure of rural residents	Total transportation and communications expenditure of rural residents/Rural population	%	+	China Rural Statistical Yearbook
	Share of transportation and communications expenditure in rural household consumption	Per capita transportation and communications expenditure/Per capita consumption expenditure × 100%	CNY per person	+	China Rural Statistical Yearbook
	Average population served per rural postal outlet	Rural population/Number of rural postal service outlets	104 persons per outlet	-	China Postal Statistical Yearbook

Note: Data sources are reported in the final column. The composite index is constructed using the entropy weight method described in Section 3.2.3.

Table A2. Measurement framework for Digital Village Development.

Dimension	Indicator	Calculation	Unit	Direction	Data Source
Digital Infrastructure	Mobile phones owned per 100 rural households	Number of mobile phones owned per 100 rural households	units per 100 households	+	China Statistical Yearbook
	Population coverage rate of rural radio and television programs	Population covered by rural radio and television programs/Total rural population × 100%	%	+	China Radio & Television Yearbook
	Number of rural postal service outlets	Total number of rural postal service outlets	outlets	+	China Postal Statistical Yearbook
	Rural Internet penetration	Administrative villages covered by 4G networks/Total administrative villages × 100%	%	+	China Communications Industry Statistical Yearbook
Industrial Digitalization	Rural online retail sales	Total value of rural online retail sales	108 CNY	+	China E-commerce Report
	Number of Taobao Villages	Total number of officially identified Taobao villages	Units	+	AliResearch Report on the Development of Taobao Villages in China
	Digital Financial Inclusion Index	Digital Financial Inclusion Index score	Index	+	Peking University Digital Finance Research Center
Digital Livelihoods	Per capita transportation and communications expenditure of rural residents	Total transportation and communications expenditure of rural residents/Rural population	%	+	China Rural Statistical Yearbook
	Share of transportation and communications expenditure in rural household consumption	Per capita transportation and communications expenditure/Per capita consumption expenditure × 100%	CNY per person	+	China Rural Statistical Yearbook
	Average population served per rural postal outlet	Rural population/Number of rural postal service outlets	104 persons per outlet	-	China Postal Statistical Yearbook

Note: Data sources are reported in the final column. The composite index is constructed using the entropy weight method described in Section 3.2.3.