Multiple Pathways of Rural Digital Intelligence Driving Agricultural Eco-Efficiency: A Dynamic QCA Analysis

Abstract

The shift toward sustainable and efficient agricultural production has become a global imperative. Rural digital intelligence, which integrates advanced technologies into agricultural practices, emerges as a pivotal driver for advancing green transformation. Based on the technology–organization–environment (TOE) framework, this study explores how rural digital intelligence drives agricultural eco-efficiency. Drawing on panel data from 30 Chinese provinces (2013–2023), this study applies dynamic qualitative comparative analysis (QCA) to unravel the complex causal pathways influencing agricultural eco-efficiency. Key findings demonstrate that (1) no single element of rural digital intelligence suffices to improve agricultural eco-efficiency; the combination of various factors can affect agricultural eco-efficiency. (2) Four distinct pathways achieve high agricultural eco-efficiency, categorized into three archetypes: application-driven pathway, synergy-robust pathway, and policy-driven pathway. (3) Temporal analysis indicates time-dependent effects in these pathways, influenced by fragmented policy implementation and technological constraints. (4) Spatial heterogeneity is pronounced; western China primarily follows the application-driven pathway, while eastern China and central China align with the synergy-robust pathway. This research explores configurational pathways through which rural digital intelligence enhances agricultural eco-efficiency, offering theoretical and empirical foundations for regionally tailored sustainable agricultural policies.

1. Introduction

Global climate warming and continuous population growth pose severe challenges to agricultural sustainability. Despite rapid agricultural economic development in China, its growth predominantly relies on a high-input, high-consumption, high-pollution model [1]. Empirical studies indicate that agricultural activities contribute 19–29% of global greenhouse gas emissions [2] and account for 49.77% of chemical oxygen demand, 46.52% of total nitrogen, and 66.58% of total phosphorus discharges [3]. This extensive paradigm not only threatens long-term agricultural viability and exacerbates environmental stress but also undermines ecosystem functionality. As a critical metric balancing resource utilization and environmental impact in agricultural production, agricultural eco-efficiency serves as a vital indicator for assessing regional sustainable development capacity [4]. Consequently, enhancing agricultural eco-efficiency while ensuring food security and supply has emerged as a central research and practical imperative. Rural digital intelligence, which integrates digital and intelligent technologies into rural areas, is profoundly reshaping traditional agricultural practices and rural economies [5]. This provides a new path for the realization of green and sustainable agricultural development. Against this backdrop, investigating the mechanisms and pathways through which rural digital intelligence fosters agricultural sustainability holds significant theoretical and practical value.

The merging of the digital economy with agricultural advancement has garnered significant scholarly attention. Current research primarily focuses on unidirectional effects of digital technologies on eco-efficiency. Quantitative analyses demonstrate that digital solutions effectively reduce greenhouse gas emissions [6,7], while rural digital economies enhance crop net carbon efficiency [8,9], contributing to climate regulation and environmental protection. Furthermore, a symbiotic and mutually reinforcing relationship between the digital economy and agricultural green development has been identified [10]. Nevertheless, despite these findings substantiating the pivotal role of digitalization in boosting agricultural productivity and accelerating green transitions, several limitations persist in this research domain: Most studies examine digital economies or technologies in isolation, with scant attention to integrated digital intelligent frameworks for enhancing agricultural eco-efficiency. While conventional regression and spatial econometric models effectively capture average effects and spatial spillovers, they fail to unravel nonlinear, configurational causalities where multiple antecedents interactively drive outcomes.

This study aims primarily to systematically identify the configurational pathways and dynamic mechanisms through which rural digital intelligence enhances agricultural eco-efficiency. What sets this study apart is its innovative situating of rural digital intelligence and agricultural eco-efficiency within the technology–organization–environment (TOE) framework to explore multifaceted connections between rural digital intelligence and agricultural eco-efficiency. Using panel data from 30 Chinese provinces (2013–2023), we measure rural digital intelligence levels via the projection pursuit model and agricultural eco-efficiency via the slacks-based measure (SBM) model. Dynamic qualitative comparative analysis (QCA) is utilized to identify multifactorial pathways driving high eco-efficiency and their temporal evolution. Our research specifically answers the following questions: (1) are there necessary conditions for enhancing agricultural eco-efficiency? (2) Through which configurational pathways does rural digital intelligence elevate eco-efficiency? (3) How do regional heterogeneities (e.g., resource endowments, geographic conditions) shape pathway selection? This study not only offers a new perspective on understanding the complex interaction mechanism between rural digital intelligence and agricultural ecological efficiency but also provides a practical foundation for spatially differentiated green agricultural policies.

2. Literature Review and Theoretical Framework

2.1. Literature Review

The existing literature has established a clear foundation for understanding both rural digital intelligence and agricultural eco-efficiency. The concept of rural digital intelligence evolves from “digital villages,” defined as deep integration of digital technologies in agricultural production, rural industries, governance, and livelihoods [5]. With rapid AI advancements, rural areas are progressively transitioning toward intelligence [11,12]. Although lacking a universally standardized definition, rural digital intelligence fundamentally constitutes the application of the internet, big data, cloud computing, AI, and IoT technologies in rural areas [11,13], spanning agricultural production [14], supply chains [15], and governance [5]. Accurate measurement is a prerequisite for assessing its impact on agricultural eco-efficiency. Existing studies predominantly employ the entropy weight method [7,16] and the entropy-weighted TOPSIS method [17] to measure the digital economy index and rural digitalization index.

The concept of eco-efficiency, first proposed by the World Business Council for Sustainable Development (WBCSD) in 1990, quantifies the ability to create economic value while minimizing environmental impacts. Agricultural eco-efficiency emphasizes maximizing economic benefits while reducing ecological damage and enhancing resource utilization efficiency [4]. The measurement of agricultural eco-efficiency encompasses inputs, desirable outputs, and undesirable outputs. Agricultural production inputs encompass not only traditional factors (labor, land, capital) but also natural resources (water, energy) [18]. Outputs include desirable products (crop yields, gross agricultural output) and undesirable byproducts (agricultural non-point source pollution, carbon emissions) [19]. As a critical indicator for assessing sustainable development, accurate measurement of agricultural eco-efficiency is essential for evaluating greening progress in agricultural systems. The slack-based measure (SBM) model has emerged as the predominant methodology, overcoming the slack variable limitations inherent in radial and oriented traditional DEA approaches [20]. Identified determinants include technological progress [21,22], land-use strategies [23], soil health conditions [24], air pollution levels [25], and urbanization dynamics [26].

The relationship between digital technology, the digital economy, and green agricultural development has attracted widespread scholarly attention. Multiple studies employ conventional econometric models to confirm the significant positive impact of the digital economy on eco-efficiency, identifying key mediators such as agricultural research and development investment [27], human capital and technological progress [8], and labor transfer with industrial agglomeration [28]. Spatial econometric approaches further reveal regional heterogeneity in how digital economies influence green total factor productivity in agriculture, attributable to cross-regional interactions in digital infrastructure, technology diffusion, and environmental policies [28]. These studies effectively elucidate the impact of the digital economy and digital technologies on agricultural eco-efficiency from a quantitative analytical perspective. Although these methods are effective in quantifying net effects, transmission mechanisms, and spatial spillovers, their dependence on “net-effect thinking” restricts the exploration of how conjunctural antecedents collectively enhance eco-efficiency.

2.2. Theoretical Framework Analysis

The technology–organization–environment (TOE) framework examines technology adoption and diffusion within organizations across three dimensions: technological, organizational, and environmental [29]. Widely applied in development pathway studies, it has analyzed digital transformation-driven green innovation [30] and identified green transition pathways in industrial parks [31]. Notably, the TOE framework demonstrates strong methodological compatibility, integrating with structural equation modeling to identify drivers of outcome variables [32] or combining with fuzzy-set qualitative comparative analysis (fsQCA) to unravel complex causalities [33]. Grounded in the TOE framework, we have developed the theoretical model illustrated in Figure 1, which offers a comprehensive perspective for understanding how rural digital intelligence influences agricultural eco-efficiency.

In terms of technology dimensions, core elements comprise digital intelligent infrastructure and innovation outcomes. Rural digital infrastructure includes broadband networks, IoT facilities, and logistics systems. These demonstrate empirically verified positive impacts on agricultural eco-efficiency when adequately developed [34]. Digital innovation outcomes reflect achievements in intelligent agriculture technology, such as big-data-driven farming solutions and AI-assisted production management, which constitute key drivers for continuous eco-efficiency improvement [35]. These technological conditions provide the necessary hardware and software support for improving agricultural production efficiency and reducing resource waste. In terms of organizational dimensions, critical factors include digital intelligence industry level and livelihood applications. Digital intelligence industry level includes digital finance, e-commerce platforms, and digitized agricultural bases, facilitating rural tourism and leisure agriculture, thereby upgrading industrial structures and reducing reliance on polluting farming practices [36]. Concurrently, digital financial services provide farmers with convenient and reliable financial services [15]. Life applications refer to the adoption of digital tools in daily agricultural practices, which enhance farmers’ environmental awareness, further contributing to ecological sustainability [36]. In terms of environmental dimensions, factors such as policy support, funding availability, and human capital constitute the external enabling conditions. Empirical evidence confirms significant contributions from skilled talent pools [6] and innovation policies [36] to eco-efficiency. The supply of funds provides a necessary guarantee for agricultural technology research and development and equipment renewal, promotes the development of agricultural production in a green and efficient direction, and then improves agricultural ecological efficiency. The interplay among these three dimensions determines whether high or non-high agricultural eco-efficiency is achieved.

In conclusion, digital intelligence technologies, organizations, and environments do not operate in isolation to enhance agricultural ecological efficiency. Instead, they interact in a nonlinear and interconnected manner to collectively influence agricultural eco-efficiency; this configurational perspective aligns with the analytical approach of qualitative comparative analysis (QCA).

3. Materials and Methods

3.1. Research Methods

3.1.1. Accelerated Genetic Algorithm-Optimized Projection Pursuit Modeling

Projection pursuit modeling identifies optimal projection vectors that reveal inherent data structures, providing an alternative approach for analyzing high-dimensional datasets [37]. The method projects data onto low-dimensional subspaces, critically enhancing the capture of essential data features [38,39]. This study employs accelerated genetic algorithm-optimized projection pursuit modeling to quantify rural digital intelligence levels, as it objectively extracts intrinsic structures from complex indicators. Computational procedures are as follows.

First, one must normalize indicator values. Given a sample set x*(i, j) {i = 1, 2, …, n; j = 1, 2, …, p}, n and p are the sample size and the number of indicators, respectively. The normalized processing of positive indicators and negative indicators is carried out by Equations (1) and (2), respectively:

$$x(i,j)={\displaystyle \frac{x^{*}(i,j)-x_{min}(j)}{x_{max}(j)-x_{min}(j)}}$$

(1)

$$x(i,j)={\displaystyle \frac{x_{\max }(j)-x^{*}(i,j)}{x_{max}(j)-x_{min}(j)}}$$

(2)

Second, one must construct the projection index function. The projection tracking model integrates p-dimensional data x (i, j) {j = 1, 2, …, p} into a one-dimensional projection value z(i) with projection directions a = {a(1), a(2),…, a(n)}:

$$z(i)={\displaystyle \sum _{j=1}^{p}a}(j)x(i,j),(i=1,2,\dots ,n)$$

(3)

The projection index function Q(a) can be expressed as Equation (4):

$$Q\left(a\right)=S_z{D}_z$$

(4)

$$S_z=\sqrt{{\displaystyle \frac{{\displaystyle \sum _{i=1}^n(z(}i)-E(z){)}^2}{n-1}}}$$

(5)

$$D_z={\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n(R-r(}}i,j){)}^{*}u(R-r(i,j))$$

(6)

Here, a represents the unit-length vector. S_z denotes the standard deviation of the projection value z(i), D_z is the local density of the projection value z(i), E(z) is the average of the sequence z(i){i = 1, 2,…, n}, R is the window radius of the local density, r(i, j) indicates the distance between samples (r (i, j) = |z(i) − z(j)|), and u(t) is a unitary step function that equals 1 when t ≥ 0 and 0 when t < 0.

Third, one must optimize the projection indicator function. Given a sample set, the projection indicator function Q(a) only varies with the projection direction a. Therefore, the optimal projection direction can be estimated by maximizing the projection indicator function. The expression of the optimal objective function and its constraints is:

$$maxQ(a)=S_z{D}_z,{\displaystyle \sum _{j=1}^p{a}^2}(j)=1$$

(7)

Fourth, one must calculate index indicators. Upon incorporating the identified optimal projection vector a* into Equation (3), the projection value z*(i) for each sample is derived, representing the computed index.

3.1.2. Super-Efficiency SBM Model Based on Non-Expected Output

Conventional DEA models fail to incorporate undesirable outputs and overlook input–output slack [40]. The slacks-based measure (SBM) model proposed by Tone Kaoru overcomes these limitations by integrating undesirable outputs into a non-radial distance function that incorporates slack variables within the objective function, enabling precise discrimination among efficient decision-making units (DMUs) [41]. Consequently, this study adopts the undesirable-output super-efficiency SBM model to measure agricultural eco-efficiency. The formalization is provided by Equations (8) and (9):

$$min\rho ={\displaystyle \frac{1+\frac{1}{m}{\displaystyle \sum _{i=1}^m\frac{s_i^{-}}{x_{ik}}}}{1-\frac{1}{c_1+c_2}\left[{\displaystyle \sum _{r=1}^{c_1}\frac{s_r^{+}}{y_{rk}}}+{\displaystyle \sum _{t=1}^{c_2}\frac{s_t^{b-}}{b_{tk}}}\right]}}$$

(8)

$$s.t.\left\{\begin{array}{l}{\displaystyle \sum _{j=1,j\ne k}^n{x}_{ij}}{\lambda }_j-s_i^{-}⩽x_{ik}\hfill \\ {\displaystyle \sum _{j=1,j\ne k}^n{y}_{rj}}{\lambda }_j+s_r^{+}⩾y_{rk}\hfill \\ {\displaystyle \sum _{j=1,j\ne k}^n{b}_{tj}}{\lambda }_j-s_t^{b-}⩾b_{tk}\hfill \\ 1-{\displaystyle \frac{1}{c_1+c_2}}\left({\displaystyle \sum _{r=1}^{c_1}\frac{s_r^{+}}{y_{rk}}}+{\displaystyle \sum _{t=1}^{c_2}\frac{s_t^{-}}{b_{tk}}}\right)>0\hfill \end{array}\right.$$

(9)

Here, $\mathsf{\rho }$ denotes agricultural eco-efficiency; m, c₁, and c₂ represent the number of inputs, desirable outputs, and undesirable outputs, respectively. x_ik, y_rk, and b_tk denote the actual required input variables, desirable output variables, and undesirable output variables for DMU, respectively. x_ij, y_rj, and b_tj denote the estimated required input variables, desirable output variables, and undesirable output variables for DMU, respectively. $s_i^{-}$ $s_r^{+}$ and $s_t^{b-}$ denote slack variables for inputs, desirable outputs, and undesirable outputs. λ_j defines weight constraints. DMUs are efficient when $\mathsf{\rho }$ ≥ 1 and inefficient when $\mathsf{\rho }$ < 1.

3.1.3. Dynamic QCA Method

Conventional QCA approaches, typically applied to cross-sectional data, struggle to capture dynamic causal processes and the temporal evolution of variable relationships. In contrast, the dynamic QCA method proposed by Castro and Ariño explicitly incorporates time into the causal interpretation, allowing for the identification of how sufficient configurations may strengthen, weaken, or emerge over different periods [42]. Dynamic QCA methodology comprises three key steps: calibration of condition and outcome variables, necessity analysis of individual conditions, and sufficiency analysis of condition configurations.

This study employs dynamic QCA for the following reasons: first, the research aims to examine the joint effects of complex interactions among multiple antecedent conditions on agricultural eco-efficiency, rather than testing the net effects of individual variables—a task for which conventional regression models exhibit inherent limitations. In contrast, QCA is particularly suited to addressing issues of multiple conjunctural causality and causal asymmetry, making it well aligned with the objectives of this study. Second, given the longitudinal nature of provincial panel data, the dynamic QCA method facilitates comparative analysis of configurations over time and across provinces, allowing us to evaluate the temporal validity of a given pathway while also discerning distinct routes by which diverse regions attain high eco-efficiency.

3.2. Variable Measurement and Calibration

3.2.1. Outcome Variables

Focusing on crop farming (narrowly defined agriculture) as the research subject, this study employs an undesirable-output-inclusive super-SBM model to measure agricultural eco-efficiency, utilizing three variable categories—inputs, desirable outputs, and undesirable outputs (

Table 1. Measurement of outcome variables.

First-Level Indicators	Secondary Indicators	Measurement Indicator
Input	Labor input	Number of persons engaged in farming, forestry, animal husbandry, and fishery * weight coefficient N (104 persons)
	Land input	Total sown area of crops (thousand hectares)
	Agricultural machinery input	Total power of agricultural machinery (104 kW)
	Chemical fertilizer input	Chemical fertilizer application (104 tons)
	Pesticide input	Pesticide application (104 tons)
	Agricultural film input	Agricultural film usage (104 tons)
	Irrigation inputs	Agricultural effective irrigation area (104 tons)
Expected output	Agricultural output	Total agricultural output value at constant prices with 2012 as the base period (104 tons)
	Agricultural carbon sink	Carbon sequestration by crops (104 tons)
Non-expected output	Agricultural carbon emission intensity	Total agricultural carbon emissions (104 tons)
	Agricultural non-point source pollution	Total agricultural non-point source pollution load (104 tons)

Note: the weighting coefficient N = total agricultural output value/total output value of agriculture, forestry, animal husbandry, and fishery. * represents the multiplication sign.

).

Agricultural production inputs are characterized by seven variables: crop cultivation labor force, total sown area, total agricultural machinery power, fertilizer application, pesticide usage, agricultural plastic film coverage, and irrigated area [34]. Desirable outputs comprise gross agricultural output value (economic benefit) and carbon sequestration during crop growth periods (ecological benefit) [34]. Carbon absorption during the crop growth period was estimated according to the economic coefficient and carbon absorption rate of different crops [43]. Outputs include agricultural carbon emission intensity and agricultural non-point source pollution [28]. Total carbon emissions of agricultural production are obtained by multiplying the carbon emission coefficients of chemical fertilizers, pesticides, agricultural films, diesel oil, ploughing, and irrigation with their corresponding values and then summing them up [44]. Total amount of agricultural production non-point source pollution was calculated by multiplying the use of chemical fertilizer, pesticidem, and agricultural film by their non-point source pollution coefficient, respectively [45]. Carbon emission coefficients were specified as fertilizer (0.8956 kg/kg), pesticide (4.9341 kg/kg), plastic film (5.18 kg/kg), diesel (0.5927 kg/kg), sowing (312.6 kg/hm²), and irrigation (20.47 kg/hm²) [44], and residue coefficients were specified as pesticide loss rate (0.5), fertilizer residue rate (0.75), and agricultural film residue rate (0.1) [45].

3.2.2. Condition Variables

In qualitative comparative analysis (QCA), conditional variables refer to potential causes or influencing factors hypothesized to affect an outcome. A key feature of conditional variables is multiple conjunctural causation—meaning different combinations of conditions may lead to the same outcome, and the same outcome may be achieved through different configurational paths. Based on the previous analysis, this paper selects five condition variables from three dimensions of digital technology, digital organization, and digital environment based on the TOE framework, drawing lessons from previous studies [9,34,35]. We built a digital and intelligent indicator system for rural areas (

Table 2. Measurement of condition variables.

First-Level Indicators	Secondary Indicators	Measures of the Indicator	Index Attribute	Weight
Digital intelligence infrastructure	Rural logistics facilities	Length of rural delivery routes (km)	+	0.304
	Optical cable development	Length of long-distance optical cable lines (104 km)	+	0.131
	Internet infrastructure	Number of rural broadband subscribers (104 households)	+	0.398
	Agro-meteorological observation stations	Number of agro-meteorological observation stations (units)	+	0.078
Digital intelligence innovation outcomes	AI industry patents	Number of patent applications in the AI industry (items)	+	0.349
	Level of agricultural science and technology innovation	Number of agricultural scientific and technological achievements (items)	+	0.348
	Blockchain industry patents	Number of patent applications in the blockchain industry (items)	+	0.233
Digital intelligence life application	IT service level	Rural total revenue from telecommunication services (CNY 108)	+	0.157
	Rural smartphone penetration	Mobile phones per 100 rural households at the end of the year (units)	+	0.147
	Information service consumption	Transportation and communication spending per capita for rural inhabitants (CNY)	+	0.163
Digital intelligence industry level	Rural digital transactions	Rural e-commerce sales volume (CNY 108)	+	0.228
	Rural digital finance development	Rural digital inclusive finance index	+	0.152
	Rural digital bases	Number of Taobao villages (villages)	+	0.284
Digital intelligence development environment	Digital talent level	Rural proportion of information technology employees (%)	+	0.123
	Digital and intelligent capital investment	Investment in information transmission, software, and IT services (CNY 108)	+	0.311
	Rural capital supply	Local fiscal expenditure on urban and rural community affairs (CNY 108)	+	0.296

).

3.2.3. Variable Calibration

Before dynamic QCA analysis, data underwent calibration to ensure attributes were appropriately represented through membership scores. Using the direct calibration method, we assigned the 95th, 50th, and 5th percentiles of data distributions as thresholds for full membership, crossover points, and full non-membership, respectively [46]. Calibrated results and descriptive statistics are shown in

Table 3. Results of variable calibration and descriptive statistics.

Variable Type		Calibrate the Anchor			Descriptive Statistics
		Full Membership	Crossover Point	Full Non-Membership	Mean	SD	Min	Max
Outcome variable	Agricultural eco-efficiency	1.021	0.673	0.411	0.705	0.215	0.273	1.043
Condition variables	Digital intelligence infrastructure	0.627	0.259	0.035	0.286	0.179	0.020	0.744
	Digital intelligence innovation outcomes	0.404	0.064	0.009	0.106	0.132	0.000	0.880
	Digital intelligence life application	0.220	0.146	0.054	0.142	0.052	0.011	0.320
	Digital intelligence industry level	0.304	0.086	0.017	0.106	0.087	0.000	0.529
	Digital intelligence development environment	0.361	0.124	0.030	0.153	0.105	0.015	0.494

.

3.3. Data Sources

This study selected panel data from 30 Chinese provinces and spanning the years 2013 to 2023 as case study samples. Tibet, Hong Kong, Macao, and Taiwan were excluded due to substantial data gaps. Data for calculating the rural digital intelligence index and agricultural eco-efficiency were obtained from multiple official statistical yearbooks and public databases, including the China Statistical Yearbook, China Rural Statistical Yearbook, China Environmental Statistical Yearbook, and China Science and Technology Statistical Yearbook. These sources provide comprehensive provincial-level datasets covering rural areas across the 30 studied provinces in China. Data on digital intelligence innovation in rural areas were sourced from the Qiyan Social Science Big Data Platform. Given the absence of direct local statistics, provincial-level innovation metrics were utilized as a proxy. Minor missing data were imputed using linear interpolation to ensure temporal continuity and cross-regional comparability.

4. Results

4.1. Necessity Analysis of a Single Condition

4.1.1. QCA Result

Before undertaking the configurational analysis, we must first determine if any individual condition is necessary for the outcome. A condition is deemed necessary if its consistency score exceeds 0.9 and its coverage surpasses 0.5 [47]. As shown in

Table 4. Results of single-factor necessity analysis.

Condition Variables	High Agricultural Eco-Efficiency				Non-High Agricultural Eco-Efficiency
	Aggregate Consistency	Aggregate Coverage	BECONS Adjusted Distance	WICONS Adjusted Distance	Aggregate Consistency	Aggregate Coverage	BECONS Adjusted Distance	WICONS Adjusted Distance
Digital intelligence infrastructure (X1)	0.589	0.609	0.136	0.529	0.627	0.636	0.245	0.460
~Digital intelligence infrastructure (~X1)	0.647	0.639	0.237	0.495	0.614	0.595	0.056	0.466
Digital intelligence innovation outcomes (X2)	0.605	0.709	0.215	0.449	0.520	0.598	0.358	0.500
~Digital intelligence innovation outcomes (~X2)	0.658	0.583	0.252	0.391	0.748	0.650	0.132	0.339
Digital intelligence life application (X3)	0.73	0.737	0.373	0.270	0.527	0.521	0.542	0.351
~Digital intelligence life application (~X3)	0.526	0.531	0.512	0.443	0.734	0.728	0.324	0.259
Digital intelligence industry level (X4)	0.699	0.79	0.331	0.316	0.479	0.531	0.557	0.397
~Digital intelligence industry level (~X4)	0.584	0.533	0.399	0.391	0.810	0.726	0.177	0.207
Digital intelligence development environment (X5)	0.669	0.709	0.177	0.431	0.538	0.560	0.331	0.483
~Digital intelligence development environment (~X5)	0.585	0.563	0.267	0.506	0.721	0.681	0.102	0.339

Note: ~ represents non-high

, the consistency values of all individual condition variables are lower than 0.9, which allows for the initial inference that there is no single condition indispensable to achieving high agricultural eco-efficiency. However, an adjusted distance exceeding 0.2 indicates significant temporal and case-specific effects, necessitating further analysis of case consistency and coverage across both between-group and within-group dimensions to assess result stability.

Regarding the between-group adjusted consistency distance, values exceeding 0.2 are observed for several condition variables, warranting validation using cross-sectional data for specific causal configurations. To visualize the results clearly, this study employed Origin 2024 software to generate plots illustrating the between-group consistency and coverage across the study period. Figure 2 reveals that for three specific configurations across years (non-high rural digital intelligence life application paired with non-high agricultural eco-efficiency in both 2013 and 2014, and high rural digital intelligence life application paired with non-high agricultural eco-efficiency in 2019), the between-group consistency exceeds 0.9 while coverage exceeds 0.5.

Consequently, for these three causal configurations, X-Y scatter plots depicting the relationship between the relevant condition variables and the outcome variable for their respective years were generated for further necessary condition testing. Figure 3 demonstrates that the configurations for non-high rural digital intelligence life application and non-high agricultural eco-efficiency in 2013 and 2014 failed the necessary condition test, as nearly one-third of the case points plot above the diagonal [48]. Similarly, the 2019 configuration (high rural digital intelligence life application and non-high agricultural eco-efficiency) also failed the test, with the majority of case points distributed along the right side of the Y-axis [48].

Regarding the within-group consistency adjusted distance, all five condition variables (digital intelligence infrastructure, digital intelligence innovation outcomes, digital intelligence life application, digital intelligence industry level, and digital intelligence development context) exceeded the threshold of 0.2. This phenomenon may be associated with our sample selection encompassing 30 Chinese provinces, which exhibit substantial heterogeneity in resource endowments, locational conditions, and digital infrastructure development.

4.1.2. NCA Result

Nevertheless, the QCA method assesses the necessity of condition variables from a qualitative perspective. To overcome this shortcoming, this study employed the NCA package in R programming language (R 4.5.1) to quantitatively analyze whether digital intelligence infrastructure, digital intelligence innovation outcomes, digital intelligence life application, digital intelligence industry level, and digital intelligence development context constitute necessary conditions for enhancing agricultural eco-efficiency and to what extent. NCA necessity analysis primarily utilizes two methods: ceiling regression (CR) and ceiling envelopment (CE) [49]. Ceiling envelopment is mainly suited for dichotomous variables or discrete variables with fewer than five categories, whereas ceiling regression is appropriate for continuous variables and discrete variables with more than five categories. Given that both the outcome variable and condition variables in this study are continuous, the ceiling regression method was applied.

As shown in

Table 5. Analysis of the necessary condition of NCA.

Condition Variables	Method	Ceiling Zone	Effect Size	Accuracy (%)	p Value
Digital intelligence infrastructure	CR	0.002	0.002	97.9	0.779
Digital intelligence innovation outcomes	CR	0.023	0.025	91.8	0.339
Digital intelligence life application	CR	0.002	0.002	100	0.866
Digital intelligence industry level	CR	0.043	0.047	93.6	0.205
Digital intelligence development environment	CR	0.012	0.013	93.9	0.502

, the effect size for five single variables is less than 0.1, and the p-values lack statistical significance (p > 0.05), signifying that none of the variables can be regarded as a necessary condition for the outcome variable [49]. This finding corroborates the results of the necessary condition analysis derived from the dynamic QCA, reaffirming that no single factor of rural digital intelligence alone constitutes a necessary condition for enhancing agricultural eco-efficiency. Furthermore, bottleneck-level analysis results for each condition variable under the ceiling regression (CR) method, generated via RStudio, are presented in Figure 4. The results show that for low-to-medium levels of agricultural eco-efficiency, all conditions are NN (not necessary), implying that no specific conditions are required to achieve these efficiency levels. However, to attain the target efficiency of 100%, at least 52.6% of digital intelligence infrastructure, 26.4% of digital intelligence innovation outcomes, 45.7% of digital intelligence life application, 51.5% of digital intelligence industry level, and 17.7% of digital intelligence development context are required. This again validates that achieving high agricultural eco-efficiency necessitates the complementary synergy of multiple factors rather than reliance on any single determinant.

4.2. Condition Configuration Sufficiency Analysis

The preceding analysis demonstrates that no single element of rural digital intelligence can drive the enhancement of agricultural eco-efficiency; instead, it is the combination of multiple factors through configurational pathways that exerts an influence. Therefore, building upon existing research, a frequency threshold of 4, a raw consistency threshold of 0.8, and a PRI (proportional reduction in inconsistency) consistency threshold of 0.6 were set [50], collectively covering 309 case observations. This study utilized the intermediate solution to specify the conditions included in each configuration and determined the core conditions based on the nested relationship between the intermediate and parsimonious solutions. The resulting configurational pathways through which rural digital intelligence empowers agricultural eco-efficiency are presented in

Table 6. Results of configuration analysis.

Condition Variables	High Agricultural Eco-Efficiency				Non-High Agricultural Eco-Efficiency
	Configuration H1	Configuration N1	Configuration M1	Configuration N2	Configuration G1	Configuration L1
Digital-intelligence infrastructure	⊗	●	$\otimes$		⬤	⬤
Digital-intelligence innovation outcomes			⊗	●	⬤	$\otimes$
Digital-intelligence life application	⬤	⬤	$\otimes$	⬤	⊗	⊗
Digital-intelligence industry level	⬤	⬤	⬤	⬤	⊗	$\otimes$
Digital-intelligence development environment	$\otimes$	⬤	⬤	⬤	●	⊗
Consistency	0.860	0.801	0.864	0.823	0.856	0.833
PRI	0.677	0.635	0.622	0.688	0.624	0.677
Coverage	0.349	0.462	0.279	0.521	0.293	0.386
Unique coverage	0.072	0.016	0.019	0.042	0.051	0.144
BECONS adjusted distance	0.113	0.177	0.105	0.158	0.188	0.211
WICONS adjusted distance	0.242	0.230	0.213	0.224	0.262	0.313
Aggregate consistency	0.813				0.835
Aggregate PRI	0.691				0.693
Aggregate coverage	0.643				0.437

Note: ⬤ and ⊗ represent core presence and absence, respectively; ● and $\otimes$ represent marginal presence and absence, respectively; blank space represents present or absence.

.

4.2.1. Aggregated Results

As presented in Table 6, the overall solution consistency for achieving high agricultural eco-efficiency is 0.813, the PRI consistency is 0.691, and the overall coverage is 0.643. This indicates that the four identified configurational pathways collectively explain over 64.3% of the cases achieving high agricultural eco-efficiency and can be considered sufficient condition combinations for this outcome. Analyzing the characteristics of these pathways enhances understanding of the underlying mechanisms through which different factor combinations improve agricultural eco-efficiency. Based on their core conditions, the four pathways are categorized into three types. The following section presents representative provincial cases of each pathway to illustrate specific manifestations of these pathways.

(1)

Application-driven pathway (H1). Configuration H1 is characterized by the presence of a high digital intelligence industry level and high digital intelligence life application, coupled with the absence of high digital intelligence infrastructure. This suggests that even in regions with less developed digital infrastructure, promoting internet platforms and digital industries can effectively enhance agricultural eco-efficiency. This pathway has a raw coverage of 34.9%. A representative case is Chongqing Municipality; characterized by mountainous terrain, it has historically faced challenges in digital intelligence infrastructure development. However, through implementing the “Five Upgrades and One Gap-Filling” initiative for digital village network development, Chongqing achieved comprehensive mobile coverage among natural villages. By 2021, the overall agricultural and rural informatization development level reached 43.3%, significantly promoting agricultural green and low-carbon development.

(2)

Synergy-robust pathway (N1–N2). Both N1 and N2 represent pathways where multiple digital intelligence conditions interact synergistically. Configuration N1 is driven by high digital intelligence industry level, life application, and development context, supported by high digital intelligence infrastructure. It has a raw coverage of 46.2%. Configuration N2 is characterized by high digital intelligence industry level, life application, and development context, supplemented by high digital intelligence innovation outcomes. It has a raw coverage of 52.1%. A representative case is Jiangsu Province, leveraging its robust policy environment and digital platform to promote agricultural informatization. By 2021, Jiangsu had established 12 national and 412 provincial-level model bases for agricultural and rural informatization, achieving a provincial digital development level of 65.4%. The province also adopted innovative ecological models such as circular agriculture and fishery–photovoltaic complementarity, further advancing agricultural sustainability.

(3)

Policy-led pathway (M1). Configuration M1 features high digital intelligence industry level and development context as core present conditions, with low digital intelligence innovation outcomes as a core absent condition. This indicates that, even in regions with limited innovation outcomes, fostering a supportive policy environment and applying digital technologies to traditional agricultural practices can enhance eco-efficiency. This pathway has a raw coverage of 27.9%. A representative case is Henan Province, where the government issued the “Implementation Opinions on Accelerating Agricultural Informatization and Digital Village Development” in 2020 to vigorously support the establishment of IoT technology application demonstration bases for major crops like wheat and corn. Emphasis was placed on smart management, digital seedling cultivation, and integrated water–fertilizer systems, significantly contributing to green agricultural development.

The overall solution consistency for non-high agricultural eco-efficiency is 0.835, with a PRI consistency of 0.693 and an overall coverage of 0.437. These results indicate that both identified configurations (G1 and L1) represent sufficient conditions for producing non-high agricultural eco-efficiency. Analyzing the characteristics of these configurations associated with non-high efficiency aids in identifying key bottlenecks that hinder improvements in eco-efficiency. As detailed in Table 6, configuration G1 is characterized by the core absence of both a high digital intelligence industry level and high digital intelligence life application despite the presence of high digital intelligence infrastructure, high digital intelligence innovation outcomes, and high digital intelligence development context as supplementary conditions. Configuration L1 is defined primarily by the core presence of high digital intelligence infrastructure coupled with the core absence of digital intelligence life application and development context along with the peripheral absence of digital intelligence industry level and innovation outcomes. Together, these pathways demonstrate that even well-developed digital infrastructure (high digital intelligence infrastructure) is insufficient to achieve high eco-efficiency when accompanied by a weak digital industrial base (low digital intelligence industry level) and an under-supportive policy environment (low digital intelligence development context).

4.2.2. Between-Group Results

Between-group consistency is used to assess whether different configurational pathways act as sufficient conditions for the outcome variable across various time points within the study period. To visualize the temporal trends of the four pathways achieving high agricultural eco-efficiency, between-group consistency level plots were generated using Origin 2024 software. As shown in Figure 5, configuration H1 exhibits temporal effects during 2015–2016; configurations N1 and N2 show temporal effects spanning 2014–2019, whereas configuration M1 maintains a consistency level consistently above 0.75 throughout the study period, indicating no significant temporal effects and robust explanatory power for the outcome variable. Overall, the consistency levels of all four pathways collectively declined from 2013 to 2016, which was likely constrained by the fragmented implementation of early-stage digital village pilot policies in China and associated technological adaptation costs. Specifically, although China’s implementation plan for the “Internet Plus Modern Agriculture Initiative” was issued in 2015, policy implementation lag delayed its full rollout until 2017. Post-2017, the consistency levels of all four pathways exhibited an upward trend, facilitated by synergistic improvements in institutional systems and breakthroughs in technological costs. Key drivers included the joint launch of China’s “Digital Village Pilot” policy in 2017 by seven ministries, including the Cyberspace Administration of China, and the continuous enhancement of provincial digital infrastructure. From 2020 to 2023, the consistency levels of all configurations stabilized above 0.75, signifying the absence of temporal effects during this period and strong explanatory power. This trend underscores the significant efficacy of China’s 14th Five-Year Plan (2021–2025) strategies, demonstrating how rural digital intelligence synergistically empowers the enhancement of agricultural eco-efficiency.

4.2.3. Within-Group Results

Within-group consistency assesses whether distinct configurations of conditions constitute sufficient conditions for the outcome variable across different sample types during the study period. To visualize disparities among the 30 Chinese provinces regarding the four configurations achieving high agricultural eco-efficiency, a within-group-consistency-level map was generated using Origin 2024 software. As illustrated in Figure 6, the within-group consistency exceeds 0.7 for most provinces, indicating that these four configurations possess strong explanatory power as sufficient conditions for the outcome variable across the vast majority of regions. Additionally, Table 6 shows that the adjusted coverage distance of consistency for four configurations exceeds 0.2, indicating that their case distributions are influenced by differences in resource endowments and locational conditions across various regions of China. This study analyzes the case distribution characteristics of the four configuration pathways by dividing China into three regions: eastern, central, and western.

As shown in

Table 7. Regional case coverage of each configuration.

Region	Application-Driven Pathway	Synergy-Robust Pathway		Policy-Driven Pathway
	H1	N1	N2	M1
Eastern China	0.307	0.559	0.670	0.268
Central China	0.502	0.724	0.658	0.490
Western China	0.546	0.472	0.498	0.348

, cases explained by the application-driven pathway are predominantly located in central and western China. This regional trend suggests that the dual-core model driven by both digitalization and intellectualization aligns well with the developmental foundations of these areas. Representative cases include Chongqing, Yunnan, and Ningxia in the west, which leverage digital intelligent technologies to achieve leapfrog development in their agricultural green transition, offsetting disadvantages in geography and infrastructure.

Cases explained by the synergy-robust pathway are concentrated in eastern and central China. The coverage of the N1 pathway is higher in central China, as illustrated by major grain-producing provinces such as Henan and Hubei, which utilize intelligent agricultural machinery and IoT platforms to enhance precision and reduce pollution. In contrast, the N2 pathway shows higher coverage in eastern China, with provinces like Jiangsu and Zhejiang accelerating the transformation and application of digital intelligent technologies through integrated industry–academia–research mechanisms.

Cases explained by the policy-driven pathway are mainly found in central and western China. In western regions like Yunnan and Sichuan, constrained by limitations in technological innovation, improving agricultural green and low-carbon development levels relies primarily on financial and policy support. Central regions such as Hunan, benefiting from their favorable geographical positions, are better positioned to absorb policy dividends emanating from surrounding areas.

In accordance with the methodological principles of qualitative comparative analysis, this study conducted robustness tests on the sufficiency of conditional configurations by adjusting two parameters: setting the case frequency threshold to 5 and modifying the original consistency threshold to 0.75. The results demonstrated that these adjustments neither altered the original configuration pathways nor substantially changed the core conditions or overall consistency. Consequently, the conditional configuration pathway linking rural digitalization development levels to agricultural ecological efficiency remains robust.

5. Discussion

In this study, we identify multiple conjunctural pathways leading to high agricultural eco-efficiency and delve into their temporal and spatial characteristics. Our findings corroborate and extend the existing literature on the complex, nonlinear relationships between digital transformation and agricultural sustainability.

First, the impact of distinct elements within rural digital intelligence on agricultural eco-efficiency is intrinsically asymmetric and complex, consistent with the principles of configurational theory. No single element constitutes a necessary condition for improvement; instead, combinations of multiple factors jointly influence eco-efficiency. This absence of necessity for any single condition underscores that the development of rural digital intelligence is a systemic endeavor, where the deficiency in one dimension can be compensated by strengths in others within a specific configuration. This aligns with the findings of Liang et al. [51]: enhancing digital innovation investment, information sharing, and factor mobility accelerates sustainable agricultural development. More profoundly, it echoes the core tenet of the TOE framework applied in this study [30,33], emphasizing that technological, organizational, and environmental contexts must interact synergistically to drive meaningful change.

Second, we identified four sufficient pathways to high agricultural eco-efficiency, which we categorized into three archetypes: application-driven pathway, synergy-robust pathway, and policy-driven pathway. Among these, the comprehensive synergistic-robust pathways (N1, N2) exhibit the highest raw coverage, indicating their greater universality in enhancing eco-efficiency. In contrast, the policy environment-dominant pathway (M1) shows lower coverage, suggesting its applicability may be confined to specific regions. This result aligns with findings from Stovba et al. [52], indicating that the introduction of digital innovation technologies in agriculture objectively requires both national policy support and incentive mechanisms that encourage agricultural digital innovation.

Third, the spatial heterogeneity of these pathways is striking and reflects China’s vast regional disparities in economic development, resource endowments, and institutional capacity. The digital application-dominant pathway predominantly covers the widest range of cases in western China. Conversely, the comprehensive synergistic-robust type prevails in the more developed eastern and central regions. This finding aligns with the results of Ren [28], indicating that the digital economy exhibits regional heterogeneity in its impact on agricultural green total factor productivity.

Fourth, our analysis of the configurations constraining high agricultural eco-efficiency delivers a crucial insight: robust digital intelligence infrastructure alone is insufficient. The core reason lies in the complex nature of digital intelligence-driven eco-efficiency improvement, which necessitates the joint effects of infrastructure development, industrial advancement, and a conducive policy environment. The absence of a solid digital industry base or a supportive policy environment can effectively negate the benefits of infrastructure, leading to suboptimal outcomes.

Our findings provide significant theoretical foundations and practical guidance for formulating region-specific policies to promote agricultural green development through rural digital intelligence. First, differentiated support policies should be implemented based on the distinct configurational pathways to high agricultural eco-efficiency. For the application-driven pathway, increased research and development investment and promotion efforts for digital intelligent technology applications in agriculture, encouraging collaboration between enterprises and research institutions, would be beneficial. For the synergy-robust pathway, we suggest strengthening of inter-departmental coordination to foster synergistic development across policy, technology, talent, and other dimensions. For the policy-driven pathway, we suggest further optimization of the policy environment to attract more resources towards enhancing agricultural eco-efficiency. Second, tailored development strategies should be adopted according to regional characteristics and strengths. In western China, promotion of the application-driven pathway should be prioritized, increasing investment in agricultural digital intelligence infrastructure to elevate the region’s digitalization level. In eastern and central China, we suggest consolidating and reinforcing the synergy-robust pathway by enhancing collaboration among regional stakeholders to drive overall improvement in agricultural eco-efficiency.

Our findings extend existing international research by showing that the association between digitalization and sustainability is often context-dependent rather than uniformly linear. The identified pathways, especially the policy-driven one, may offer valuable insights for countries with relatively weak agricultural foundations but strong central governance that aim to accelerate green transformation. Conversely, the synergy-driven pathway reflects patterns observed in more advanced agricultural systems such as those characterized by robust technological innovation capabilities and well-developed infrastructure.

However, this study acknowledges limitations. First, the measurement of agricultural eco-efficiency primarily focused on crop cultivation, excluding forestry and animal husbandry, which may constrain the generalizability of the findings. Future studies should refine the scope by separately investigating pathways for enhancing eco-efficiency in forestry and animal husbandry through rural digital intelligence alongside in-depth case studies of digital intelligent practices and their ecological effects in specific regions or agricultural subsectors. Second, while the dynamic QCA approach effectively analyzes the spatiotemporal impact of different configurational paths on the outcome, it emphasizes qualitative analysis and has relatively limited capability for quantitative analysis. Future studies should explore integrating QCA with quantitative analytical methods (e.g., structural equation modeling or in-depth econometric analysis of key mechanisms) to provide a more comprehensive investigation of the key antecedent conditions influencing agricultural eco-efficiency levels and their underlying mechanisms. Furthermore, future work could apply this configurational approach to cross-national comparative studies to disentangle universal patterns from nation-specific contexts, thereby further testing the generalizability of the pathways identified here.

6. Conclusions

Grounded in the TOE framework, this study unravels the complex mechanisms through which rural digital intelligence enhances agricultural eco-efficiency across 30 Chinese provinces (autonomous regions, municipalities) from 2013 to 2023, adopting a configurational perspective. This study captures the impact of different factor combinations in rural digital intelligence on agricultural eco-efficiency, thereby addressing a gap in the existing literature to some extent. We draw the following conclusions:

(1)

No individual element of rural digital intelligence can independently enhance agricultural ecological efficiency; instead, it is the combinatorial pathways of multiple factors that exert an impact on agricultural ecological efficiency.

(2)

Results from the conditional sufficiency analysis indicate that there are four configurational pathways to achieve high agricultural ecological efficiency and two configurational pathways that restrict it. The overall solution consistency for achieving high agricultural eco-efficiency is 0.813, and the PRI consistency is 0.691, indicating that the four identified configurational pathways can be considered sufficient condition combinations for achieving high agricultural eco-efficiency.

(3)

The configurational pathways (H1, N1, and N2) for achieving high agricultural ecological efficiency exhibit a certain time effect. The explanatory power of different configurational pathways for high agricultural eco-efficiency shows a “decline–rise–stabilization” trend over time, with policy synergy and technological advancement serving as key driving factors.

(4)

Among the 30 Chinese provinces, the within-group consistency exceeds 0.7 in most provinces, indicating that the four configurational pathways possess strong explanatory power as sufficient conditions for high agricultural eco-efficiency. However, all configurational pathways for achieving high agricultural eco-efficiency demonstrate spatial effects. The application-driven pathway is primarily concentrated in central China and western China, the synergy-stabilized pathway is mainly distributed in eastern China and central China, and the policy-driven pathway is predominantly located in central China and western China.