The Role of New Quality Productivity in Enhancing Agricultural Product Supply Chain Resilience: A Predictive and Configurational Analysis

Abstract

Currently, factors such as geopolitical conflicts, frequent extreme weather events, and power struggles among major countries are threatening the stability of the global supply chain. Building a more resilient supply chain has received international consensus. Today, new quality productivity (NQP), spawned by disruptive innovation, is an important way for China to enhance its agricultural product supply chain resilience (SCR). However, studies often overlook the “time lag” problem of the panel data adopted, and their empowering paths require further investigation. Therefore, this study firstly constructs NQP and agricultural product SCR indicators. Based on the panel data produced by 31 Chinese provinces from 2011 to 2022, we solved the “time lag” problem by integrating a Backpropagation Neural Network (BPNN) with an Autoregressive Integrated Moving Average (ARIMA) model to predict the NQP level. Subsequently, the empowering paths through NQP-enhancing agricultural product SCR were explored via entropy weight TOPSIS and Fuzzy-Set Qualitative Comparative Analysis (fsQCA) method. Foundations: China’s agricultural product SCR exhibits a spatial differentiation characteristic of “prominent in the central region and weak in the western region”. A single factor is not a necessary condition for high resilience, and its improvement depends on the synergy of multiple factors. Three differentiated driving paths have been identified: “autonomous endogenous driving type”, “environment-enabled driving type”, and “system architecture driving type”. NQMP has become the bottleneck for improving agricultural product SCR, and the threshold of each factor has increased significantly as the resilience target is raised. High resilience stems from the synergy and functional compensation of core factors, while low resilience is mostly caused by the concurrent absence of key conditions or structural mismatch, showing distinct “multiple concurrencies” and “causal asymmetry” characteristics.

1. Introduction

Today, the world is facing complex situations such as geopolitical tensions, frequent extreme weather, and intensified competition among major powers, which have led to severe shocks to the global supply chain. As a major economy deeply embedded in the global network, China’s own industrial and supply chain risks are impacted by global challenges. Among them, the inherent vulnerability of the agricultural product supply chain, due to its fundamental position, is further highlighted in this process. Unblocking the upstream and downstream relationships of the supply chain and enhancing its resilience by “solidifying the chain”, “filling the chain”, and “strengthening the chain” are important strategies proposed by China in recent years for responding to complex international situations. The President of the People’s Republic of China and many vital Chinese documents have emphasized that “China will develop NQP according to local conditions and improve SCR”. This means that, currently, NQP is a key instrument for improving SCR in China.

Existing studies have shown that the development of NQP can enhance agricultural product SCR. For example, NQP is driven by technological innovation. Developing smart agriculture and agricultural facilities accelerates the innovation and iteration of agricultural product varieties, thus reducing large fluctuations in agricultural product supply [1]. The development of NQP can expand the boundaries of labor objects to break through resource constraints and help resolve the problem of regional development imbalance [2]. This shows that NQP has inherent advantages in solving problems such as poor supply stability, regional development imbalance, information asymmetry, and insufficient scientific and technological innovation and application in China’s agricultural product supply chain [3,4].

However, studies on the empowering path through NQP, enhancing agricultural product SCR, are insufficient. Existing studies mainly focus on the following three aspects:

(1)

Studies related to NQP. NQP is a new form of productivity driven by technological innovation (especially disruptive technology). Its emergence was rooted in China’s pursuit of high-quality “economic growth” [5]. Current studies on NQP are developing incrementally and can be roughly divided into three main areas: NQP measurement indicators, level measurement, and influencing factors.

(a)

Regarding the construction of indicators for NQP, from a systems theory perspective, NQP is a “factor-structure-function” system. Based on a deep understanding of NQP connotations, scholars have mostly built their measurement indicators on the three traditional factors of production (means of labor, labor objects, and laborers) [6,7,8,9,10,11]. While effective in reflecting NQP, the evaluation index system has limitations due to its lack of integration of new factors. Some scholars have also established NQP indicator systems from other dimensions. Lu and Wang (2024) [12] constructed an assessment system covering three dimensions: technological productivity, green productivity, and digital productivity. Han et al. (2024) [13] constructed an evaluation system for NQP from the two dimensions of importance and penetration. These studies have laid a theoretical foundation for measuring NQP level.

(b)

Regarding the measurement of the development level of NQP, it was noted that China’s development level is on an upward trend, but there are regional differences [14,15]. Currently, many studies have used the entropy method and TOPSIS method based on panel data from before 2022 to measure the development level of NQP [16,17,18]. Some scholars have achieved innovations in research methods. For example, Hua et al. (2024) [19] constructed the IDOCRIW-PROBID model to study the development level of NQP in the new energy vehicle (NEV) industry. Zhong et al. (2025) [20] used text analysis and machine learning to extract information from massive amounts of unstructured data to build more timely and fine-grained enterprise-level NQP indicators to measure the level of NQP.

It can be observed that panel data used to measure NQP development levels suffer from statistical lag due to data availability. Analysis based on such retrospective data can only reveal historical correlations and is insufficient to support timely and forward-looking strategy formulation. Predicting NQP levels using large-scale models is an effective way to address this limitation; however, relevant research is still insufficient.

(c)

A study on factors influencing the development of NQP is the one by Ruan Zhou and Guo (2025) [21], which emphasized that strengthening intellectual property protection could effectively promote the development of NQP. The two factors have significant synergistic effects. Over time, NQP and intellectual property protection showed strong complementarity. Some scholars have focused on the enterprise level, pointing out that public data openness and artificial intelligence have had a significant role in promoting the NQP of enterprises [20,21,22].

Therefore, the panel data measuring the NQP development level is time-lagged. Predicting the NQP level through some AI models is a good method. However, few studies have done this, especially those that combine ARIMA and BPNN models. In addition, the NQP indicator system overlooks the importance of the development environment and the development concept.

(2)

There have been studies on agricultural product SCR. Resilience was first applied in the field of ecology [23] and later introduced into the field of supply chains. Early studies focused on the connotation and characteristics of SCR. The complex structure of SCR encompasses multiple dimensions and levels. For example, SCR is conceptualized as the ability to recover, adapt, and continue after encountering risks and being damaged [24,25]. Delving into the agricultural product supply chain, Christopher and Peck (2004) [26] defined agricultural product SCR as the ability of a supply chain to recover to its original state or to a better state after a disruption. Later studies focused on the path to improving agricultural product SCR. For example, Aungkulanon et al. (2024) [27] proposed that iterative strategies can enhance agricultural product SCR. Yuan et al. (2024) [28] emphasized that under the influence of climate change, resilience can be enhanced by optimizing supply chain structure, integrating digital technologies, and strengthening collaborative mechanisms and policy guarantees. In recent years, research has focused more on the impact of new technologies and green development on SCR. The application of digital technologies such as blockchain, the Internet of Things, and artificial intelligence can significantly enhance efficiency, transparency, and agricultural product SCR [29,30]. Pu et al. (2025) [31] stressed that the digital transformation of agricultural product supply chains is crucial for improving efficiency, enhancing SCR, and promoting sustainable development. Coopmans et al. (2021) [32] found that deploying digital technology could optimize supply chain management and enhance the ability to respond to changes in market demand, thereby improving agricultural product SCR.

Constructing an evaluation index system for agricultural product SCR is crucial to exploring the influencing factors of agricultural product SCR [33]. However, scant empirical research explicitly addresses the development of standardized resilience metrics for agricultural product SCR. For example, Ren and Hao (2025) [34] built an evaluation system that includes indicators such as resilience, recovery, reorganization, and renewal. Sargani et al. (2025) [35] quantified agricultural product SCR using three observation indicators: resilience, recovery speed, and continuity. Zhao et al. (2024) [36] divided the resilience phases, covering the entire process of “preparation-response-recovery-adaptation”. Wang et al. (2023) [37] incorporated predictive power and sustainability into common metrics. Most scholars have directly focused on a supply chain perspective and built an indicator system covering absorptive capacity, adaptive capacity, and resilience [38].

From the above analyses, most studies suggest that the resilience, resistance, and transformative power of the supply chain can be used to measure the level of agricultural product SCR. Significantly, the integration of emerging technologies—such as IoT, blockchain, and predictive analytics—demonstrably enhances SCR robustness.

(3)

There are studies about NQP empowering SCR. Most of the related studies remain at the stage of qualitative analysis. Cui and Du (2025) [39] believed that NQP could significantly improve SCR, and in areas with high adaptability, the effect of NQP on improving SCR was more significant. At the same time, Zhu et al. (2025) [40] pointed out that NQP-improved SCR is more significant in the eastern region and high-resilience regions. Specifically, NQP can enhance the resilience of the industrial and supply chain by reducing costs, reducing dependence on traditional production factors, and optimizing resource allocation [41]. It can also enhance the competitiveness and resilience of the industrial chain by deepening the industrial division of labor [42]. Technological innovation (especially disruptive technology) is the core of NQP. It can significantly improve SCR [43,44,45].

Existing studies have the following shortcomings: (1) The construction of NQP indicators mostly focuses on three aspects, New Quality Laborers (NQL), New Quality Labor Objects (NQLO), and New Quality Means of Product (NQMP), and overlooks the importance of the New Quality Development Environment (NQDE) and the New Quality Development Concept (NQDC). (2) The measurement of the level of NQP mainly relies on panel data. However, there exists a “time lag” problem, and studies often overlook this issue. This oversight results in a lack of a forward-looking nature in current research findings. (3) The causal pathways through which NQP enhances SCR remain inadequately explored, particularly given the temporal disalignment inherent in conventional panel datasets.

Therefore, considering the importance of NQDE and NQDC, and combining Marx’s theory of productive forces, an indicator system for new-quality productive forces was constructed from five dimensions: NQMP, NQLO, NQL, NQDC, and NQDE. At the same time, it selects resilience, resistance, and transformation power to measure the agricultural product SCR and constructs a corresponding indicator system. Then, based on panel data produced by 31 Chinese provinces from 2011 to 2022, we solved the “time lag” problem by integrating a BPNN with ARIMA modeling to predict the NQP level. Subsequently, the empowering paths through NQP that enhanced agricultural product SCR were explored via entropy weight TOPSIS and the fsQCA method. Finally, we put forward targeted policy recommendations for different provinces (the framework diagram is shown in Figure 1).

The innovations of this study: (1) Delving into the driving forces of organizational operations—the innovation of productivity—and integrating environmental and development concepts, a comprehensive indicator system for NQP can be constructed. (2) A comprehensive research method based on entropy weight TOPSIS-ARIMA-BPNN and the fsQCA method is proposed to empirically explore the specific path of NQP, empowering agricultural product SCR.

The research significance of this paper: Academic value: (1) An NQP index system was constructed, which enriched the NQP theory. (2) A new comprehensive research method was proposed to explore the specific path of NQP to enhance agricultural product SCR. This method enriched SCR management theory. Application value: (1) The constructed NQP indicator system can provide an indicator reference for Chinese government departments to measure the development level of NQP. (2) The proposed comprehensive research method based on entropy weight TOPSIS-ARIMA-BPNN and fsQCA can provide a theoretical reference for Chinese government departments to formulate NQP development policies around improving agricultural product SCR.

2. Theoretical Analysis and Research Hypothesis

Impacts of NQP on the Agricultural Product SCR

NQP can directly enhance agricultural product SCR, mainly by improving the resistance, resilience, and transformation power (as shown in Figure 2).

First, NQP can improve the resistance of the agricultural product supply chain. NQP uses technologies such as the Internet of Things, big data, and AI to monitor and accurately control the growth environment of crops in real time, improve the precision and intelligence of agricultural production, and increase output and quality [46]. On the other hand, NQP promotes the digitalization of agricultural product SCR, and blockchain technology realizes full traceability, enhancing transparency and collaborative efficiency [47]. It enhances rapid risk identification and response within the supply chain while enabling precise interlinkage across its stages, significantly strengthening agricultural product SCR.

Second, NQP can improve the resilience of the agricultural product supply chain. Through platform-based and modular production organization, NQP enables the supply chain to respond more flexibly to changes in market demand [48], promotes the flexible development of agricultural product SCR, enhances their ability to enable dynamic reconfiguration and restructuring, and improves economic resilience. In addition, NQP also promotes the green transformation of the agricultural product supply chain. For instance, it can apply technologies (such as clean energy and circular economy) to reduce the resource consumption and environmental impact of the production and circulation of agricultural products [49], and then greatly improve the ecological sustainability of the agricultural product supply chain.

Third, NQP can improve the transformation power of the agricultural product supply chain. NQP promotes platform development in the agricultural product supply chain. New models such as “agriculture + Internet” based on Internet platforms break the traditional supply chain pattern and realize efficient resource allocation and multi-party value creation, and their open sharing model greatly improves the flexibility and innovation of the supply chain.

Hypothesis 1.

NQP has a positive impact on product SCR.

3. Study Design

3.1. Research Methods

The research methods used in this study are shown in Figure 3.

(1)

Entropy-weighted TOPSIS method: In this paper, the entropy-weighted method is first used to determine the weights of the NQP and the agricultural product SCR indicators, and then the TOPSIS method is used to rank the indicators. The entropy-weighted TOPSIS method is a multi-indicator comprehensive evaluation method that combines entropy weighting (objective weighting) with the TOPSIS method (sorting by approaching ideal solutions). This method objectively assigns weights based on the degree of variation in the data itself to avoid subjective arbitrariness. NQP and agricultural product SCR indicators cover many aspects and have differences in magnitude and dimension. Before weighting, they need to be normalized for comparability. The indicators and their weights are weighted to make the comprehensive evaluation score more objective.

(2)

ARIMA-BPNN hybrid model: To solve the problem of the panel data related to NQP, which “lags” the panel data related to agricultural product SCR, we proposed the ARIMA-BPNN hybrid model to predict the NQP indicator data.

(a)

The ARIMA model, also known as the autoregressive moving average model, combines the autoregressive (AR) and moving average (MA) models. It is a highly accurate linear time series forecasting method that utilizes historical trends and recent fluctuations in data for predictions. Its expression is as follows:

$$X_t={\phi }_1{X}_{t-1}+{\phi }_2{X}_{t-2}+\dots +{\phi }_p{X}_{t-p}+{\epsilon }_t-{\theta }_1{\epsilon }_{t-1}-{\theta }_2{\epsilon }_{t-2}-\dots -{\theta }_q{\epsilon }_{t-q}$$

(1)

where ${\phi }_1,{\phi }_2,\dots ,{\phi }_p$ is the autoregressive coefficient, $p$ is the autoregressive order, ${\theta }_1,{\theta }_2,\dots ,{\theta }_q$ is the moving average coefficient, $q$ is the moving average order, and $\left\{{\epsilon }_{\mathrm{t}}\right\}$ is the white noise sequence. Usually, this model can be represented as ARIMA ($p,d,q$), where $d$ is the difference order.

The modeling process begins with a stationarity test, using the ADF unit root test to determine the difference order d (see

Table A1. Results of ADF test.

Variable	ADF Statistic	Critical Values			p-Value	Test Results	d
		1%	5%	10%
X1	−2.2289	−4.6652	−3.3672	−2.8030	0.1959	non-stationary	1
D(X1)	−5.1716	−6.0451	−3.9293	−2.9868	0.0000	stationary
X2	−2.8570	−5.3543	−3.6462	−2.9012	0.0506	non-stationary	1
D(X2)	−4.5819	−6.0451	−3.9293	−2.9868	0.0001	stationary
X3	−2.6069	0.0916	−4.9387	−3.4776	−2.8439	non-stationary	1
D(X3)	−6.9387	−6.0451	−3.9293	−2.9868	0.0000	stationary
X4	−4.6821	0.0000	−4.9387	−3.4776	−2.8439	stationary	0
X5	−2.8118	0.0566	−4.6652	−3.3672	−2.8030	non-stationary	1
D(X5)	−4.6430	−5.3543	−3.6462	−2.9012	0.0001	stationary
X6	−2.1248	0.2347	−4.6652	−3.3672	−2.8030	non-stationary	1
D(X6)	−5.3653	−6.0451	−3.9293	−2.9868	0.0000	stationary
X7	−9.4559	0.0000	−5.3543	−3.6462	−2.9012	stationary	0
X8	−3.4246	0.0101	−5.3543	−3.6462	−2.9012	stationary	0
X9	−6.9904	0.0000	−5.3543	−3.6462	−2.9012	stationary	0
X10	−0.4096	−5.3543	−3.6462	−2.9012	0.9085	non-stationary	2
D(X10)	−2.3573	−6.0451	−3.9293	−2.9868	0.1541	non-stationary
D2(X10)	−3.6914	−6.0451	−3.9293	−2.9868	0.0042	stationary
X11	−3.4225	−4.9387	−3.4776	−2.8439	0.0102	stationary	0
X12	−3.0326	−4.9387	−3.4776	−2.8439	0.0320	stationary	0
X13	−8.1334	−5.3543	−3.6462	−2.9012	0.0000	stationary	0
X14	0.0790	−5.3543	−3.6462	−2.9012	0.9646	non-stationary	2
D(X14)	−1.7338	−6.0451	−3.9293	−2.9868	0.4138	non-stationary
D2(X14)	−4.8876	−6.0451	−3.9293	−2.9868	0.0000	stationary
X15	−2.0860	−5.3543	−3.6462	−2.9012	0.2502	non-stationary	1
D(X15)	−5.1563	−6.0451	−3.9293	−2.9868	0.0000	stationary
X16	0.2598	−5.3543	−3.6462	−2.9012	0.9754	non-stationary	2
D(X16)	−1.7844	−6.0451	−3.9293	−2.9868	0.3882	non-stationary
D2(X16)	−5.0538	−6.0451	−3.9293	−2.9868	0.0000	stationary
X17	−5.0359	−4.6652	−3.3672	−2.8030	0.0000	stationary	0
X18	−3.2459	−5.3543	−3.6462	−2.9012	0.0175	stationary	0
X19	−3.1309	−4.9387	−3.4776	−2.8439	0.0243	stationary	0
X20	−3.8003	−4.9387	−3.4776	−2.8439	0.0029	stationary	0
X21	−2.4302	−4.9387	−3.4776	−2.8439	0.1334	non-stationary	1
D(X21)	−4.4498	−6.0451	−3.9293	−2.9868	0.0002	stationary
X22	−2.3112	−4.9387	−3.4776	−2.8439	0.1684	non-stationary	1
D(X22)	−4.0734	−6.0451	−3.9293	−2.9868	0.0011	stationary
X23	−3.6331	−4.9387	−3.4776	−2.8439	0.0052	stationary	0
X24	−3.3483	−5.3543	−3.6462	−2.9012	0.0129	stationary	0
X25	−0.6510	−5.3543	−3.6462	−2.9012	0.8590	non-stationary	1
D(X25)	−8.6741	−6.0451	−3.9293	−2.9868	0.0000	stationary
X26	0.2517	−5.3543	−3.6462	−2.9012	0.9750	non-stationary	2
D(X26)	−2.2938	−6.0451	−3.9293	−2.9868	0.1740	non-stationary
D2(X26)	−5.6542	−6.0451	−3.9293	−2.9868	0.0000	stationary
X27	−49.5688	−5.3543	−3.6462	−2.9012	0.0000	stationary	0
X28	−0.3150	−4.6652	−3.3672	−2.8030	0.9234	non-stationary	2
D(X28)	−1.8533	−5.3543	−3.6462	−2.9012	0.3543	non-stationary
D2(X28)	−5.1894	−6.0451	−3.9293	−2.9868	0.0000	stationary
X29	−3.2994	−4.6652	−3.3672	−2.8030	0.0149	stationary	0

Note: 1. The null hypothesis of the ADF test is “the time series has a unit root (non-stationary)”, while the alternative hypothesis is “the time series has no unit root (stationary)”. 2. The core criterion for stationarity judgment is the p-value: if p-value ≤ 0.05, the null hypothesis is rejected (the series is stationary). If p-value > 0.05, the null hypothesis cannot be rejected (the series is non-stationary). 3. Critical values provide auxiliary verification: when the ADF statistic is less than the critical value at the corresponding significance level, it also supports rejecting the null hypothesis (the series is stationary). The results in the table are consistent with the judgment based on p-values.

in Appendix A). Subsequently, the autoregressive order p and the moving average order q are preliminarily determined through autocorrelation function (ACF) and partial autocorrelation function (PACF) analysis, and parameter optimization is performed using Bayesian information criterion (BIC) minimization (Further details on model diagnostics are provided in Supplementary Materials). During model fitting, outliers are detected through residual analysis and influence functions, and identified outliers are Winsorized by adjusting extreme values to values at specified quantiles to mitigate their adverse effects on parameter estimation. After model establishment, a residual white noise test is performed to ensure effective extraction of linear information. Finally, dynamic prediction is achieved through a sliding window mechanism, with the window size set according to the periodic characteristics of the sequence.

(b)

Backpropagation Neural Network (BPNN) is a typical multi-layer feedforward neural network structure. This network typically consists of an input layer, an output layer, and one or more hidden layers, each layer composed of several neurons. In terms of connectivity, neurons in adjacent layers are fully connected through adjustable weights, while nodes within the same layer are not interconnected. A typical BP neural network is a three-layer structure with one hidden layer, as shown in Figure 4.

Suppose a neural network has $d$ input neurons, $q$ output neurons, and $l$ hidden neurons. Then, the output of each neuron is as follows:

$$H_j=f_h\left({\displaystyle \sum _{i=1}^d{w}_{ji}^{(h)}}{x}_i+b_j^{(h)}\right)$$

(2)

The output of the output layer neurons is as follows:

$$O_k=f_o\left({\displaystyle \sum _{j=1}^l{w}_{jk}^{(o)}}{H}_j+b_k^{(o)}\right)$$

(3)

where $w$ and $b$ are the connection weights and bias terms, respectively, $f_h$ and $f_o$ are the activation functions of the hidden layer and the output layer, respectively.

The core of the network training process is the backpropagation algorithm. This process first obtains the network output through forward computation, and then calculates the output error $E={\displaystyle \frac{1}{2}}{\displaystyle \sum _{k=1}^q{(T_k-O_k)}^2}$ ($T_k$ is the target input). During the crucial “backpropagation” phase, this error signal propagates backward along the network, and the gradient of the error function with respect to the weights and biases of each layer is calculated according to the chain rule ${\displaystyle \frac{\partial E}{\partial w}}$ and ${\displaystyle \frac{\partial E}{\partial b}}$. Finally, the network uses this gradient information to update all weights and bias parameters through optimization algorithms such as the Levenberg–Marquardt algorithm, i.e., $w\leftarrow w-\eta {\displaystyle \frac{\partial E}{\partial w}}$ ($\eta$ is the effective learning rate), thereby iteratively minimizing the error between the network output and the true value.

In this study, the BPNN employs a single hidden layer structure with 5 nodes in the input layer, and the number of hidden layer nodes is optimized between 3 and 8 using a grid search. The hidden layer uses the hyperbolic tangent activation function, while the output layer uses a linear activation function. The network is trained using the Levenberg–Marquardt backpropagation algorithm, with a maximum training iteration count of 1000 and a target error of 0.001. To enhance the model’s robustness to outliers, L2 regularization is employed during training, and outlier detection is used to weight the training samples, reducing the impact of outliers on network parameter updates.

(c)

The combined model is determined. Most data sequences related to NQP levels exhibit composite characteristics, including both linear and nonlinear time series components. Simply using BPNN or ARIMA models may lead to excessive errors. Considering the nonlinear representation capabilities of BPNN and the statistical modeling advantages of ARIMA, as well as their different sensitivities to outliers, this paper proposes a novel inverse error weighting strategy based on backtesting performance. This method calculates the prediction errors of each model on the validation set and assigns weights according to the inverse of the error, giving models with smaller errors a larger share in the combination, thus adaptively highlighting the predictive power of the better-performing model. Weights are assigned based on the reciprocal of the error:

$$w_{\mathrm{A}}={\displaystyle \frac{{\mathrm{MAPE}}_{\mathrm{BP}}^{-1}}{{\mathrm{MAPE}}_{\mathrm{ARIMA}}^{-1}+{\mathrm{MAPE}}_{\mathrm{BP}}^{-1}}},\hspace{0.33em}\hspace{0.33em}\hspace{0.33em} w_{\mathrm{B}}=1-w_{\mathrm{A}}$$

(4)

$W_A$ and $W_B$ represent the weights of the ARIMA and BPNN models, respectively, and MAPE represents the mean absolute percentage error of each model on the validation set. The final combined prediction value is as follows:

$${\widehat{Y}}_t=w_{\mathrm{A}}\cdot {\widehat{Y}}_t^{(\mathrm{ARIMA})}+w_{\mathrm{B}}\cdot {\widehat{Y}}_t^{(\mathrm{BP})}$$

(5)

(3)

fsQCA: This method is applied to analyze the path of NQP, empowering the agricultural product SCR. fsQCA is a methodology between qualitative and quantitative research, which is used to analyze causal complexity and identify multiple combinations of conditions (variables) that lead to specific results.

(4)

Necessary condition analysis (NCA) is an emerging method for analyzing causal relationships. It supplements traditional statistical methods and quantitative comparative analysis (QCA) by identifying the critical level that antecedent conditions must reach for an outcome to occur. To accurately test whether a single antecedent condition is necessary for high agricultural product SCR, this study introduced two estimation methods from NCA—upper bound regression (CR) and upper bound envelope (CE)—to calculate the effect size d and p-value before configuration analysis. A condition is considered necessary when both α ≥ 0.1 and α < 0.05 are simultaneously satisfied.

3.2. Sample Selection and Data Sources

Based on panel data of 31 provinces (autonomous regions and municipalities) in China from 2011 to 2022, this paper first uses the BPNN and ARIMA models to predict the NQP level separately in 2022. However, the R² value of some indicators is less than 0. This phenomenon shows that the fitting effect of the corresponding model on the sample data is significantly worse than that of the mean model, and it is difficult to meet the data prediction needs. Then, we use the hybrid model of ARIMA-BPNN to predict the NQP level in 2022. The comparison with the actual data from 2022 is within an acceptable error. Therefore, in this paper, we will use the hybrid model of ARIMA-BPNN to predict the NQP level in 2023. Then, we will combine the data of the agricultural product SCR in 2023 to study the path through which NQP empowers agricultural product SCR.

The original data comes from the official website of the China National Bureau of Statistics, the China Statistical Yearbook, the China Rural Statistical Yearbook, the Ministry of Industry and Information Technology of the People’s Republic of China, and the statistical yearbooks of various provinces (autonomous regions and municipalities) from 2012 to 2024. A small amount of missing data is supplemented by interpolation. At the same time, considering the lag of the actual situation, when evaluating the role of NQP on agricultural product SCR, the method of Zhang et al. (2020) [50] is used to match the conditional variables of the previous period with the result variables of agricultural product SCR, lagged by one year.

3.3. Construction of the Indicator System

(1)

Construction of NQP indicators: Combining the existing literature, a comprehensive evaluation indicator system of NQP consisting of five levels, seven Tier-1 Indicators, fifteen Tier-2 Indicators, and twenty-nine Tier-3 Indicators was constructed (see

Table A2. Comprehensive evaluation indicator system of NQP.

Dimension	Tier-1 Indicators	Tier-2 Indicators	Tier-3 Indicators	Attributes
NQMP	Level of Agricultural Science and Technology	Science and Technology Innovation	Agricultural Science and Technology Employees [9,16]	Positive
			Agricultural S&T Input [7,16]	Positive
			R&D Expenditure [8]	Positive
			R&D Personnel [8,11]	Positive
		Level of Agricultural Mechanization	Total Power of Agricultural Machinery and Machinery [60]	Positive
		Grain Production Capacity	Total Grain Production [7,9]	Positive
	Agricultural Information Development	Information Infrastructure Development	Number of Agrometeorological Station Observations [6,7,18]	Positive
			Number of Rural Broadband Access Users [7,9]	Positive
			Length of Fiber Optic Cable Lines Per Square Meter [6,7,9,12]	Positive
NQLO	The Scale of a Technological Organization (Organizational Perceived Capacity)	Above-Scale Agriculture-Related Enterprises	Number of Agricultural Production Enterprises [7,8,13]	Positive
			Number of Agricultural IT Enterprises [7,8,13]	Positive
			Agricultural Breeding Enterprises [7,8,13]	Positive
			Number of Pesticide And Fertilizer Enterprises [7,8,13]	Positive
			Number of Agricultural Logistics Enterprises [7,8,13]	Positive
NQL	Labor Force Development Level	Educational Attainment	Years of Education Per Rural Laborer [6,7,8,10]	Positive
		Per Capita Output Value of Primary Industry	Output Value of Primary Industry/Number of Employees in Primary Industry [7,9]	Positive
		Per Capita Income of Rural Residents	Per Capita Disposable Income of Rural Residents [7,9,10]	Positive
NQDC	Organizational Development Concept (Organizational Sustainability)	Resource Consumption	Water Consumption in Agriculture [7]	Negative
			Agricultural Electricity Consumption [16]	Negative
		Environmental Pollution	Fertilizer Application [8,16]	Negative
			Pesticide Application [8,16]	Negative
		Environmental Protection	Soil Erosion Control Area [6]	Positive
			Forest Cover [10]	Positive
NQDE	Market Transaction Environment	Transaction Scale	Gross Agricultural Output [9]	Positive
	Natural Environment	Water Resources	Total Water Resources [10]	Positive
	Industry Development Environment	Value Added to the Industry	GDP Output Value-Added of Primary Industry [60]	Positive
			Value Added of Tertiary Industry [60]	Positive
		External Environment	Water Resources Irrigation Level [8]	Positive
			Total Retail Sales of Consumer Goods [61]	Positive

in Appendix A).

(2)

Construction of agricultural product SCR indicators: Starting from the three dimensions of resistance, resilience, and transformation power, an agricultural product SCR indicator system consisting of three Tier-1 Indicators, seven Tier-2 Indicators, and sixteen Tier-3 Indicators was constructed (see

Table A3. Agricultural product SCR indicator system.

Tier-1 Indicators	Tier-2 Indicators	Tier-3 Indicators	Attributes
Resistance [62]	Stability of Agricultural Production	Total Sown Area [63,64]	Positive
		Effective Irrigation Area [63,64,65,66]	Positive
		Total Power of Agricultural Machinery [65]	Positive
	Agricultural Product Supply Level	Total Grain Production [63,64]	Positive
		Number of People Employed in One Industry [63,65]	Positive
Resilience [62]	Economic Recoverability	Industry Value Added [64,65]	Positive
		The Proportion of Primary Industry-Added Value in Regional GDP [63,64]	Positive
		Agricultural Producer Price Index [63]	Positive
		Total Agricultural Output [63]	Positive
	Ecological Sustainability	Amount of Agricultural Fertilizers Used [66]	Negative
		Agricultural Plastic Film Usage [63,65]	Negative
		The Area of Crops Affected [63,65]	Negative
Transformative Power [62]	Technological Advancement	Agricultural Electricity Consumption [66]	Negative
		Agricultural Diesel Consumption [63,65]	Negative
	Diversity and Coordination	Rural Per Capita Disposable Income [66]	Positive
	Innovation	Agricultural Technology Investment [66]	Positive

in the Appendix A). Resistance comprises agricultural product production stability and supply level, and resilience encompasses economic recoverability and ecological sustainability. Transformative power integrates scientific and technological progress, diversity coordination, and innovation.

3.4. NQP Level Prediction

The hybrid ARIMA-BPNN model performs better on evaluation indicators such as mean square error (MSE) and mean absolute error (MAE), showing good prediction accuracy (see Figure 5 and

Table A4. Prediction and evaluation of NQP level.

Tier-3 Indicators	MSE	RMSE	MAE	R2
Agricultural Science and Technology Employees	126,717,585.1212	11,256.8906	9787.1783	0.9317
Agricultural S&T Input	4570.8838	67.6083	48.9924	0.7294
R&D Expenditure	26,248.0240	162.0124	126.4613	0.8501
R&D Personnel	1,027,515,575.5921	32,054.8838	27,488.1832	0.9457
Total Power of Agricultural Machinery and Machinery	1114.9808	33.3913	21.3362	0.9872
Total Grain Production	806.1365	28.3925	24.7510	0.9639
Number of Agrometeorological Station Observations	0.1694	0.4116	0.3432	0.9847
Number of Rural Broadband Access Users	0.0050	0.0710	0.0568	0.8861
Length of Fiber Optic Cable Lines Per Square Meter	4140.2197	64.3445	49.1519	0.6977
Number of Agricultural Production Enterprises	52.3230	7.2335	5.8294	0.7072
Number of Agricultural IT Enterprises	66.6693	8.1651	6.7185	0.9163
Agricultural Breeding Enterprises	0.1894	0.4352	0.3313	0.9488
Number of Pesticide and Fertilizer Enterprises	0.4689	0.6848	0.6131	0.9616
Number of Agricultural Logistics Enterprises	1.6913	1.3005	1.0034	0.7915
Water Consumption in Agriculture	8.2931	2.8798	2.2239	0.9865
Agricultural Electricity Consumption	48,247.7969	219.6538	176.4976	0.9329
Fertilizer Application	31.8832	5.6465	4.5023	0.9539
Pesticide Application	0.2006	0.4479	0.3931	0.8005
Soil Erosion Control Area	0.4961	0.7043	0.4638	0.9877
Forest Cover	2767.5808	52.6078	45.2128	0.9476
Years of Education Per Rural Laborer	0.0127	0.1126	0.0743	0.9864
Output Value of Primary Industry/Number of Employees in Primary Industry	2 × 10−9	4 × 10−5	3 × 10−5	0.6857
Per Capita Disposable Income of Rural Residents	438,454.1293	662.1587	541.8965	0.9245
Gross Agricultural Output	31,358.0325	177.0820	139.5979	0.8973
Total Water Resources	10,570.8289	102.8145	89.4070	0.7926
GDP Output Value-Added Of Primary Industry	8,500,737.3359	2915.6024	2489.5789	0.9537
Value Added of Tertiary Industry	9,286,204.5357	3047.3274	2568.6276	0.9548
Water Resources Irrigation Level	2558.7567	50.5842	39.7776	0.9404
Total Retail Sales of Consumer Goods	1,457,832.1501	1207.4072	897.6685	0.9431

in Appendix A, taking Guangdong Province data as an example).

Therefore, we integrated the two deep learning models BPNN and ARIMA, used the inverse variance method to achieve model integration, and constructed a combined forecasting model to achieve the forecast of the NQP data for 2023 (see Figure 6).

Before model training, the original data are normalized to the minimum and maximum values, and the original data are scaled to the target interval using the maximum and minimum values in the dataset to eliminate the dimensional differences between different indicators. The sliding window method is then used, and the time step is set to 3. Specifically, the data of three consecutive time points are used as input features each time to predict the value of the next time point. Then, a three-layer feedforward backpropagation neural network is constructed to minimize the error between the predicted and the actual values. To enhance the stability and interpretability of the prediction results, ARIMA is introduced, and on this basis, a weighted fusion strategy based on the inverse error method is further proposed. The prediction results of the BPNN and the ARIMA model are integrated, and the mean square error, root mean square error, mean absolute error, and determination coefficient are introduced to comprehensively evaluate the model’s predictive accuracy to avoid misjudging performance by a single indicator.

3.5. Variable Measurement and Calibration

(1)

Measurement of outcome variables. The outcome variable is the agricultural product SCR, which is comprehensively reflected by the three dimensions of resistance, resilience, and transformation power. We employed the entropy weight TOPSIS method to objectively determine indicator weights for generating a comprehensive evaluation score. Resistance constitutes the capacity of the agricultural product SCR to mitigate or hedge against risk exposures during disruptive events. It encompasses data such as agricultural product production stability and supply level. Resilience refers to the ability of the supply chain to maintain its function and structure in the face of major risks and to quickly recover to a normal or better state after being affected. It is divided into economic and ecological resilience. The transformation power emphasizes the ability of the supply chain to self-adjust, optimize, and quickly respond to changes when external factors change. It is represented by indicators such as scientific and technological progress, diversity and coordination, and innovation.

(2)

Measurement of conditional variables. The conditional variable is the development level of NQP, which is divided into five dimensions—NQMP, NQLO, NQL, NQDC, and NQDE—based on the inherent structure of NQP. The entropy-weighted TOPSIS method is used to objectively weight the indicators at each level and calculate a comprehensive score to characterize the development level of NQP at each level. Specifically, the NQMP dimension focuses on technology application and infrastructure, covering indicators such as agricultural technology input, mechanization level, and information infrastructure. The NQLO dimension mainly reflects the carriers and scale of technology application, measured by the number of agriculture-related enterprises. The NQL dimension covers labor quality and income levels, reflected by indicators such as education level and per capita output. The NQDC dimension emphasizes the organization’s sustainable perception and development orientation. The NQDE dimension includes the natural environment, industrial base, and market conditions, comprehensively evaluated through indicators such as resource endowment, industrial added value, and consumption scale.

(3)

Variable calibration. Drawing on the research approach of Fiss (2011) [51] and the standard practices of the fsQCA methodology [52], this study employs a direct calibration method to convert conditional variables into fuzzy sets. The specific anchor points are set as follows: the 95th percentile of the sample is used as the complete membership point (0.95), the 50th percentile (median) as the crossover point (0.5), and the 5th percentile as the complete non-membership point (0.05). This scheme follows the calibration principle of “based on substantial knowledge”, using sample quantiles as a robust alternative in the absence of external industry benchmarks. Choosing the 95th and 5th quantiles (rather than extreme values) effectively suppresses the interference of outliers, ensuring that only cases with extremely excellent or insufficient conditional performance are assigned extremely high (>0.95) or extremely low (<0.05) membership. Setting the median as the crossover point best reflects the membership state of the set with “maximum fuzziness”, thus theoretically conforming to the definition of intermediate states in fuzzy sets. The result variables and five conditional variables were calibrated using fsQCA software (v 4.1)at different time intervals (see

Table 1. Calibration anchor points for conditional and outcome variables.

Variable Classification	Variable Name	Fully Affiliated	Intersection	Not Affiliated at All
Outcome Variable	Agricultural Product SCR	0.5450	0.1980	0.0319
Condition Variables	NQMP	0.7571	0.2796	0.0075
	NQL	0.6388	0.2864	0.1492
	NQLO	0.4403	0.2226	0.1283
	NQDC	0.6102	0.3130	0.0677
	NQDE	0.6428	0.3460	0.1650

).

4. Analysis of Empirical Results

4.1. Measurement of Agricultural Product SCR

Based on China’s provincial-level administrative divisions, agricultural product SCR in 31 provinces in 2023 was measured using the entropy weight TOPSIS method. The full-sample quartile method was used to classify agricultural product SCR level into five levels: low (0–0.2197), low-medium (0.2198–0.2935), medium (0.2936–0.3783), medium-high (0.3784–0.4660), and high (0.4661–0.7043). The differences in resilience levels among regions were presented in a visual form, as shown in Figure 7.

4.1.1. Overall Analysis

Spatially, agricultural product SCR fluctuates across the east, center, and west, with an overall pattern of high resilience in the east and center and low resilience in the west. This pattern exhibits significant regional imbalance, with resilience “highlands” emerging in certain areas (such as those centered in and surrounding Hubei). A total of 90% of highly resilient regions are located in plains (in the Northeast, the Huanghuaihai region, and the middle reaches of the Yangtze River), while 80% of low-resilience regions are concentrated in mountainous areas and coastal areas (in the Southeast, Southwest, and Northwest).

4.1.2. Analysis of High Resilience Regions

Highly resilient regions are not scattered but rather clustered around agricultural infrastructure, logistics hubs, or foreign trade locations. The Northeast Plain Commercial Grain Base (Heilongjiang) is strengthening the upstream SCR through a combination of “scale + policy + logistics”. With its vast expanse of black soil, per capita arable land is three times the national average, a mechanization rate exceeding 90%, and corn and soybean production accounting for a quarter of the national total. The base maintains efficient supply chain circulation and enhanced regulatory oversight by leveraging national grain reserves and specialized transportation networks. Combined with policies such as subsidies for commercial grain bases and black soil protection, the base’s disaster resilience has been enhanced. During Typhoon Dusurui in 2023, the yield reduction was only one-third of the national average.

The Huanghuaihai agricultural cluster (Shandong and Henan) is leveraging “industrial collaboration + market networks” to revitalize midstream supply chains. Over 2000 vegetable processing companies in Shouguang, Shandong, form a comprehensive supply chain, and Henan’s wheat processing capacity accounts for one-third of the national total. Leveraging the Beijing-Tianjin market, Shandong Xinfadi radiates into North China, while Henan Wanbang covers Central China. Shouguang vegetables reach 200 cities 24 h a day. Shandong, known as the “Silicon Valley of Vegetable Seeds”, and Henan’s leading wheat breeding technology have reduced Henan’s risk of yield reduction by 40%.

The Yangtze River Economic Belt logistics hubs (Hubei, Chongqing, and Sichuan) are improving SCR through “water-land intermodal transport + product synergy”. Leveraging the Yangtze River waterway and the China–Europe Railway Express, logistics costs are 30% lower than those in western China, and transportation time for Sichuan citrus has been reduced by 50%. Collaborative efforts are underway to fill gaps in multiple product categories, including rice and pigs. Chongqing’s pig slaughtering operations and Hubei’s crayfish processing (with an annual output value exceeding CNY 50 billion) have enhanced resilience, ensuring a six-month supply of frozen meat during the pandemic.

South China’s export-oriented agricultural regions (Guangdong and Guangxi) are leveraging a “foreign trade location + cold chain e-commerce” strategy to enhance downstream resilience. Under the RCEP framework, agricultural product trade between Guangdong and ASEAN has grown by 15% annually. Guangxi’s Pingxiang port accounts for 70% of China’s durian imports. Guangdong’s lychee cold chain, with a four-hour closed loop, has reduced the loss rate from 25% to 5%. Guangxi’s cross-border cold chain shelf life has been extended to 15 days. These policy measures are strengthening resilience to market fluctuations and helping to cushion market shocks.

4.1.3. Analysis of Low Resilience Regions

The resilience of agricultural supply chains in low-resilience regions exhibits significant fragility. The low resilience of agricultural supply chains in Tibet, Ningxia, and Qinghai stems from a combination of natural constraints, industrial shortcomings, and lagging infrastructure. Naturally, Tibet’s high altitude results in scarce arable land and low yields; Qinghai’s high altitude and drought conditions weaken its agricultural and animal husbandry sector, and Ningxia’s reliance on the Yellow River for water resources and severe desertification all hinder production stability. Industrially, these three regions have a low agricultural share, primarily focusing on primary products, with weak deep processing capabilities (for example, the processing rate for Tibetan barley and Ningxia wolfberry is less than 20%), and a lack of industrial chains to withstand market fluctuations. Regarding infrastructure, transportation accessibility is poor (some towns in Tibet close their mountains in winter, and Yushu in Qinghai requires transit through Xining). Cold chain coverage is less than 10%, and logistics costs account for over 30% of the sales price, exacerbating vulnerability due to losses in distribution.

Beijing has only 2% of its arable land, and its grain and vegetable self-sufficiency rate is low (less than 5% of grain). A total of 95% of grain and 70% of vegetables are imported. The “functional alienation” of agriculture has exacerbated this lack of resilience. Beijing’s current agricultural sector is primarily focused on “high-quality products and tourism” rather than ensuring basic supply, severely weakening its ability to scale production and mitigate risks. This “transfer of industrial status” and “functional trade-offs” have made it difficult for Beijing, despite its strong economic and technological strength, to transform its agricultural supply chain into risk-resistant capabilities. This has ultimately led to a unique mismatch between urban development and agricultural resilience.

To address the identified vulnerabilities, the following tiered policy interventions can be considered:

• At the level of natural constraints, differentiated measures should be adopted based on the unique natural conditions of each region: In Tibet, the focus should be on promoting cold-resistant and barren-tolerant crop varieties and greenhouse agriculture; in Qinghai, developing plateau-specific water-saving agriculture and animal husbandry; and in Ningxia, promoting a production model combining water-saving irrigation with sand-fixing crop cultivation. For Beijing, the focus should be on developing vertical agriculture and facility agriculture, compensating for insufficient arable land resources through technological innovation.
• At the level of industrial weaknesses, a precise industrial upgrading strategy should be implemented. Addressing the insufficient processing rate of specialty agricultural products in Tibet, Qinghai, and Ningxia, a special fund for deep processing should be established to promote the transformation from primary products to high-value-added products. For Beijing, its technological and market advantages should be fully utilized to extend agriculture towards high-value-added segments such as R&D, services, and branding, while maintaining a moderate scale of production capacity to balance the multi-functional positioning of agriculture.
• At the level of infrastructure, a logistics system adapted to regional characteristics should be constructed. In western regions, the focus should be on improving transportation accessibility and developing logistics solutions adapted to seasonal demands. In megacities like Beijing, an efficient agricultural product distribution network should be established, reducing dependence on external sources through digital supply chain management and regional collaboration.

Specifically addressing the “functional alienation” issue emerging in megacities like Beijing, innovative policy tools are needed: establishing an integrated urban–rural food system plan, balancing the ecological, living, and production functions of agriculture through multiple pathways such as urban agriculture, regional cooperation, and strategic reserves. Simultaneously, leveraging the advantages of the digital economy to build a smart supply chain system will enhance the ability to respond to unforeseen risks.

4.1.4. Analysis of Unconventional Areas

(1)

As the main grain-producing area in Northeast China and a hub in the Bohai Rim Economic Zone, Liaoning boasts superior agricultural resource endowments (ranking among the top ten in the country in corn and rice production and eighth in pig production) and industrial location advantages. However, due to “triple structural imbalances”, the resilience of the agricultural product supply chain and resource conditions are seriously inverted. Industrial resource mismatch: Industry is prioritized, and agricultural input is diluted. Liaoning invests 60% of its fiscal resources in industry (such as the Shenyang Intelligent Connected Vehicle Industrial Park), while agricultural expenditure accounts for only 50% of Heilongjiang’s total expenditures. High-standard farmland accounts for 60% (90% in Heilongjiang), and the agricultural mechanization rate is 65% (90% in Heilongjiang). The “production end” of the supply chain has weak disaster resistance (in 2023, typhoons caused a 15% reduction in corn production, while in Heilongjiang it was only 5%). Industrial chain fault: weak processing and loss of added value. Liaoning’s corn processing is mainly based on primary products such as starch and alcohol (accounting for 70%), lacking deep processing (such as corn germ oil and protein feed in Heilongjiang). The agricultural product processing conversion rate is 40% (65% in Heilongjiang). When corn prices plummet in 2023, processing companies will have insufficient capacity to absorb the impact (Heilongjiang already absorbs 30% of excess capacity through deep processing).

The core issue: Industrial advantages have not effectively fed back into agriculture, and resource endowments have become systematically ineffective under the “heavy industry, light agriculture” strategy.

(2)

As a major agricultural province, Jiangxi Province should theoretically have high SCR due to its natural resource endowment and industrial foundation, but in reality, it has medium-low resilience. Possible reasons are as follows. The single industrial structure: Jiangxi’s agriculture has long relied on the “four dominant” products of rice, pigs, citrus, and conventional aquatic products. In 2022, grain production accounted for more than 60% of the total agricultural output value, while cash crops and high-value-added industries accounted for less. Jiangxi’s agricultural product processing rate is only 61%, lower than the national average of 65%. Market dependence is high. Exports are concentrated in Southeast Asian and Middle Eastern markets. The ecological carrying capacity is limited. Jiangxi’s mountainous and hilly areas account for more than 60%, and cultivated land is severely fragmented. The transformation of low-yield fields, such as cold slurry fields, is difficult. In 2024, the transformation of cold slurry fields in Yuanzhou District only covered 10% of the affected area. Jiangsu has saved 20% of its irrigation water through “unmanned farms” and precision irrigation technology, achieving a balance between ecology and production.

4.2. Necessity Analysis

4.2.1. Necessity Analysis of Individual Conditions

In the fsQCA, consistency is an important criterion for measuring necessary conditions. It reflects the reliability of the causal relationship between the conditional variable and the outcome variable. It is generally believed that when a single variable consistency value is greater than 0.9, it indicates a necessary condition, and it is also a condition that must exist for the outcome to occur. Coverage is used to evaluate the extent to which the subset covers the target set, that is, how much the conditional variable can explain the occurrence of the outcome variable. As shown in

Table 2. Necessity analysis of single condition.

Condition Variables	High Agricultural Product SCR		~High Agricultural Product SCR
	Consistency	Coverage	Consistency	Coverage
NQMP	0.8074	0.8153	0.4356	0.5116
~NQMP	0.5164	0.4403	0.8428	0.8358
NQLO	0.8046	0.7968	0.4493	0.5176
~NQLO	0.5129	0.4447	0.8236	0.8306
NQL	0.6204	0.5982	0.6168	0.6918
~NQL	0.6803	0.6042	0.6418	0.6629
NQDC	0.8311	0.7590	0.4920	0.5227
~NQDC	0.4774	0.4469	0.7732	0.8419
NQDE	0.8632	0.8352	0.4482	0.5044
~NQDE	0.4878	0.4318	0.8536	0.8789

Note: ~ indicates missing conditions, such as “~technical level” indicates missing technical level conditions.

, the consistency of the single antecedent variable of high SCR and non-high SCR is less than 0.9. It indicates that the single antecedent condition is not a necessary condition for high or non-high SCR, which fully reveals the complex causal mechanisms of the result.

4.2.2. NCA

Necessity analysis was performed using R 4.3.3 software, and the results of the NCA necessity analysis are shown in

Table 3. Necessary condition analysis results.

Variable	Method	d	p	C-Accuracy	Ceiling Zone	Scope
NQMP	CE	0.325	0.000	100%	0.290	0.89
	CR	0.334	0.000	80.6%	0.298
NQLO	CE	0.165	0.004	100%	0.144	0.87
	CR	0.192	0.031	90.3%	0.168
NQL	CE	0.097	0.141	100%	0.089	0.92
	CR	0.098	0.306	90.3%	0.090
NQDC	CE	0.291	0.000	100%	0.262	0.90
	CR	0.314	0.000	83.9%	0.283
NQDE	CE	0.297	0.000	100%	0.260	0.87
	CR	0.314	0.000	90.3%	0.275

Note: After calibration, a fuzzy set membership value of 0.0 ≤ d ≤ 0.1 indicates “low level”, 0.1 ≤ d ≤ 0.3 indicates “medium level”, and d > 0.3 indicates “high level”. In NCA, the permutation test uses repeated sampling = 10,000.

. The results show that the effect size of NQL is less than 0.1, and the p-value is not significant. The effect sizes of NQMP, NQLO, NQDC, and NQDE are all greater than 0.1 and the p-values are significant, but their precision is all less than 95%. Therefore, none of them can be considered necessary conditions [53], further verifying that a single NQP condition cannot constitute a necessary condition for high-level agricultural product SCR.

4.3. NCA Bottleneck Level Analysis Results

Using R 4.3.3 software, the bottleneck level analysis results for each antecedent condition under the upper bound regression method are further displayed, as shown in

Table 4. Results of bottleneck level analysis (%).

Agricultural Product SCR	NQMP	NQLO	NQL	NQDC	NQDE
0	NN	NN	NN	NN	NN
10	NN	NN	NN	NN	NN
20	4.8	NN	NN	NN	NN
30	14.0	NN	NN	0.4	9.6
40	23.2	3.3	NN	13.1	19.5
50	32.4	12.9	2.0	25.8	29.4
60	41.6	22.4	9.0	38.5	39.3
70	50.8	32.0	16.1	51.2	49.1
80	60.0	41.5	23.1	63.9	59.0
90	69.2	51.1	30.1	76.6	68.9
100	78.5	60.6	37.2	89.3	78.8

Note: NN signifies not necessary.

. The bottleneck level refers to the minimum critical value that must be reached for a specific condition variable to achieve the expected result. It identifies the quantitative threshold of the condition as a “necessary condition”.

It can be observed that NQMP initially becomes the bottleneck factor for improving agricultural product SCR. As the agricultural product SCR level increases, the bottleneck effect of each factor gradually intensifies. To reach a 20% agricultural product SCR, an NQMP of 4.8% is required. To reach a 50% agricultural product SCR, an NQMP of 32.4%, NQLO of 12.9%, NQL of 2.0%, NQDC of 25.8%, and NQDE of 29.4% are required. To reach a 100% agricultural product SCR, NQMP of 78.5%, NQLO of 60.6%, NQL of 37.2%, NQDC of 89.3%, and NQDE of 78.8% are required. As the agricultural product SCR level continues to increase from 20% to 100%, the bottleneck demand thresholds for the five antecedent conditions rise significantly. Moreover, the demand growth in each dimension shows a progressive characteristic of “the higher the resilience level, the more stringent the requirements”, which once again verifies that a single antecedent condition cannot constitute a necessary condition for a high agricultural product SCR.

4.4. Analysis of the Adequacy of Conditional Configuration

This paper contains 31 cases, which is a medium sample size. The frequency threshold is set to 1. Ragin (2008) [52] clearly points out that 0.8 is the minimum recommended standard for the consistency threshold of conditional configuration adequacy analysis. Drawing on this study, this paper sets the consistency threshold to 0.8 and the PRI consistency threshold to the minimum acceptable standard of 0.7. With this as the boundary, the truth table row with a value less than this value is assigned as 0, and the value greater than or equal to this value is assigned as 1.

By performing Boolean algebra operations on intermediate and simplified solutions, the configuration results for 2023 were obtained (

Table 5. Configuration analysis of agricultural product SCR.

Condition Variables	High Configuration Results			Low Configuration Results
	A	B	C	a	b	c	d
NQMP	●		●	⊗	⊗	●	●
NQLO	●	●			⊗	⊗	⊗
NQL	●	●	●		⊗	●	●
NQDC		●	●	⊗	⊗		⊗
NQDE	●	●	●	⊗		⊗	⊗
Consistency	0.9608	0.9803	0.9257	0.9470	0.9476	0.9483	0.9430
Original Coverage	0.3419	0.4856	0.5402	0.6543	0.4240	0.3078	0.2880
Unique Coverage	0.0621	0.0768	0.1677	0.1866	0.0091	0.0510	0.0275
Case Counts	2	3	8	10	6	2	2
Solution Consistency	0.8821			0.9440
Solution Coverage	0.8685			0.7600

Note: ● signifies the presence of core conditions. ⊗ signifies the lack of the core condition. Blank space signifies a “do not care” condition. Case counts only record cases with a membership greater than 0.5 in the term.

). These results show three high-configuration paths and four low-configuration paths, with the consistency levels of individual solutions across all seven paths exceeding the acceptable consistency level of 0.9. The overall consistency of high- and non-high-agricultural-product SCR configuration paths reached 0.8821 and 0.9440, respectively, both exceeding the standard value of 0.75. The overall coverage of both configurations exceeded 50%, meeting the established evaluation criteria. Based on the overall configuration analysis, multiple configuration paths were found to achieve high agricultural product SCR, and the unique coverage of different configuration paths varied. A higher unique coverage indicates a larger sample coverage and stronger explanatory power. All configurations leading to high agricultural product SCR included NQL and NQDE, demonstrating their necessity for agricultural product SCR.

4.4.1. Configuration Analysis of High SCR

We categorize the three paths under high configuration into three types, as shown in Figure 8.

(1)

Autonomous endogenously driven type (MP-LO-L-DC): Path H1 shows that the core combination of NQMP, NQLO, NQL, and NQDC constitutes an equivalent path for generating high agricultural product SCR. This path reveals that even in situations where external macroeconomic development environment support is relatively limited, regional agricultural systems can still build strong resilience by focusing on substantial upgrades of internal productivity factors and forward-looking guidance from development concepts. This reflects an “intrinsic” development logic, which does not rely on specific advantages of external location or policy dividends, but rather drives system evolution through synergistic innovation of internal technological foundation, human capital, industrial targets, and guiding ideologies. The original coverage of path H1 is 0.3419, indicating that this configuration can explain 34.19% of the high SCR samples. The unique coverage is 0.0621, indicating that 6.21% of the high SCR samples can only be explained by this configuration path. Typical regions include Guangxi and Shaanxi. Taking Guangxi as an example, although Guangxi is located in the west, in recent years it has focused on characteristic industries such as sugarcane and fruit, vigorously promoted small-scale agricultural machinery and water-saving irrigation technology in hilly and mountainous areas, implemented the “Local Expert” training program, promoted the development of primary and deep processing of agricultural products to optimize the value chain, and implemented the concept of “high-quality development of characteristic agriculture”. As a result, despite the unfavorable external circulation environment, it has formed a characteristic supply chain system with regional resilience.

The core of the autonomous endogenously driven type lies in the positive closed loop formed by the upgrading of internal elements and the guidance of concepts. Specifically, advanced labor resources, represented by intelligent equipment and digital platforms, improve the operational efficiency and process controllability of the supply chain. Highly skilled professional farmers and new agricultural business entities, along with high-value-added, branded specialty agricultural products, together constitute the core carriers and driving forces of supply chain value creation, while the development concepts of greening and high quality provide strategic focus and action guidelines for the entire chain. These three elements reinforce each other: technological equipment empowers laborers and products, and the upgrading of concepts guides technological investment and product selection. Ultimately, this enables the supply chain system to maintain stable output and rapid recovery capabilities even under external market fluctuations or natural constraints, achieving endogenous resilience through its solid internal capabilities.

(2)

Environment-enabled driven type (LO-L-DE-DC). Path H2 shows that NQLO, NQL, NQDE, and NQDC are core conditions that can contribute to high agricultural product SCR. This path indicates that when a region possesses favorable external environments such as policy support, resource endowment, or market ecosystem, and has a clear development philosophy, even if its investment in traditional “hard” labor resources such as heavy mechanization and high-end intelligent facilities is not dominant, it can still achieve a leap in overall SCR through the deep empowerment of “people” (laborers) and “things” (labor objects) by the environment and philosophy. This highlights the decisive role of “soft environment” and “soft capital” under specific conditions. The original coverage of this configuration is 0.4856, which can explain 48.56% of the high agricultural product SCR sample; the unique coverage is 0.0768, indicating that 7.68% of the high agricultural product SCR sample can be explained only through this configuration path. Typical examples are Henan and Sichuan. Taking Henan as an example, Henan, relying on its strategic positioning as a core area of national grain production, enjoys stable policy expectations and a large-scale market, and firmly promotes the concepts of “storing grain in the land and storing grain in technology” and “digital transformation of agriculture”. Against this backdrop, Henan has continuously cultivated a large-scale, high-quality grain-growing workforce, developed high-quality specialty wheat and other crop varieties, and integrated resources through socialized service organizations to make up for the insufficient per capita machinery ownership, thus constructing a resilient generation path of “policy environment guidance-concept innovation drive-people and variety optimization”.

The core of the environment-enabled-driven type lies in the catalytic and synergistic effect of the “environment-concept” dual framework on micro-level elements. A superior development environment provides supply chain entities with a low-risk foundation of innovation, abundant access channels for factors of production, and a large-scale demand market, reducing transaction costs and uncertainties. Advanced development concepts act as the system’s “brain”, identifying opportunities and guiding resources towards human capital enhancement and the upgrading of labor objects. Empowered workers and objects become the direct units for achieving resilience; they can more effectively utilize the opportunities provided by the environment, respond to the direction guided by the concepts, and form a flexible and highly adaptive supply chain network. This empowerment mechanism, from the outside in and from macro to micro, allows the system to achieve resilience without bearing heavy fixed capital investments by activating and optimizing its most dynamic elements.

(3)

System architecture-driven type (MP-L-DE-DC). Path H3 indicates that the core conditions constituted by NQMP, NQL, NQDE, and NQDC can foster high agricultural product SCR. This path demonstrates that if a region can systematically integrate advanced material and technological foundations, unique resource endowments, a favorable macroeconomic environment, and top-level strategic concepts to construct a stable and efficient institutionalized operational paradigm, then even with relatively low immediate dependence on individual high-end human resources, it can ensure that its supply chain exhibits extremely high and stable resilience. This emphasizes a “systematic” or “institutionalized” logic for resilience generation. The original coverage of this configuration is 0.5402, which can explain 54.02% of the high agricultural product SCR sample. The unique coverage is 0.1677, indicating that 16.77% of the high agricultural product SCR sample can only be explained through this configuration path. Typical regions include Heilongjiang and Guangdong. Taking Heilongjiang Province as an example, leveraging its globally scarce black soil resources, Heilongjiang has established a leading large-scale agricultural machinery equipment system in China. Under the strategic positioning of Heilongjiang as a “ballast stone” for national food security, it has received strong and continuous policy and investment support, and has long adhered to the development concepts of “modernized large-scale agriculture” and “black soil protection”. This highly coordinated “resource-technology-policy-concept” system architecture ensures that SCR is rooted in the robustness of the entire system, rather than relying on scattered, individualized skill advantages.

The core of the system architecture-driven approach lies in solidifying key elements into a highly reliable system structure through top-level design. Its inherent operating mechanism is as follows: large-scale, standardized labor resources and objects constitute the standardized “hardware” modules and production processes of the supply chain, ensuring economies of scale and quality stability in output; a favorable development environment provides long-term, credible institutional commitments and resource guarantees, reducing external friction in system operation; and a clear development philosophy sets long-term goals and rule boundaries for system evolution. These three elements are interlocked, forming a resilient structure with low volatility and high predictability. Under this architecture, the operation of the supply chain relies more on pre-set rules, standard operating procedures, and stable resource inflows, reducing its immediate dependence on individual, random innovation. This allows it to exhibit strong structural stability and resilience in the face of shocks, achieving a kind of “system-inherent” resilience.

The relatively unique low coverage of paths A and B reveals their emerging characteristics and potential shortcomings, primarily reflecting the high resilience resulting from technological innovation or organizational scale within individual provinces. This result indicates significant case overlap in explaining high agricultural product SCR in these paths. Most high-resilience cases can be achieved through multiple paths, demonstrating a “multiple concurrent causal” characteristic. While this development model based on specific resource endowments is effective, its replicability is limited. This finding reveals a key implication for agricultural product SCR construction: paths A and B are effective solutions for regions with specific conditions, while achieving more general high resilience requires exploring diversified path choices. Therefore, the lower unique coverage not only does not diminish the value of these paths but also provides an important basis for differentiated policy formulation in regions with different endowments, deepening our understanding of the complexity of agricultural product SCR systems.

The core logic of the three highly configurable paths lies in the concurrent existence of core elements and the substitution and adaptation of non-core elements. That is, a deficiency in one dimension of new quality productivity can be compensated for by the synergistic effect of other core dimensions, without requiring all elements to reach their optimal levels, reflecting the flexibility and adaptability of “element substitution”. Element substitution is not a “direct replacement of a single element with another”, but rather a “systemic effect formed by the synergy of multiple core elements, compensating for the deficiency of a certain non-core element”. Its premise is the functional complementarity and synergistic adaptation of core elements, ultimately achieving the configuration logic of “multiple paths leading to the same result”.

4.4.2. Configuration Analysis of Non-High SCR

Non-high agricultural product SCR configuration results showed four paths. We categorize them into two types:

(1)

Comprehensive Shortcomings (L1 and L2). The core missing conditions for path L1 are NQMP, NQDC, and NQDE, while NQL and NQLO have no impact on the result. This means that when insufficient technical support, conservative development concepts, and lack of external environment coexist, regardless of the form of labor objects or the quality of laborers, a non-high agricultural product SCR will occur. The core missing conditions for path L2 are NQMP, NQDC, NQL, and NQLO, while NQDE has no impact on the result. This means that when the four core elements of technical equipment, product form, laborer quality, and development concepts are all insufficient, regardless of whether the external environment has supporting conditions, a non-high agricultural product SCR will occur.

(2)

Advantage Mismatch Type (L3, L4): The core missing conditions for path L3 are NQLO and NQDE, while the core existence conditions are NQMP and NQL. This is a typical path where “advantageous elements and key weaknesses coexist”. This means that even with advanced agricultural technology and highly skilled laborers, if agricultural products lack standardization, high-quality upgrading, and effective external support, the advantageous elements cannot be taken advantage of due to the lack of suitable scenarios, resulting in a non-high-value agricultural product SCR. The core missing conditions for path L3 are NQLO, NQDC, and NQDE, while the core existence conditions are NQMP and NQL. This is an “enhanced version of the weakness scenario” of path L3. This means that even with technological and human resource advantages, the three key conditions of product, concept, and environment are simultaneously missing, forming a triple constraint of “direction-value-platform”, which completely suppresses the role of advantageous elements, ultimately resulting in a non-high agricultural product SCR.

Path L1 has higher original coverage and unique coverage than the other three paths, indicating that path L1 is relatively more dominant in explaining the SCR of non-high-value agricultural product SCR. The core logic is that technology, concepts, and the environment are the “basic support triangle” for supply chain upgrading. The simultaneous absence of all three leads to a triple lock-in of “no technological efficiency improvement, no conceptual guidance, and no environmental empowerment” in the supply chain. Even if the labor objects or laborers have a certain potential, they cannot be transformed into a resilience advantage. The core logic of this non-high configuration result is the “concurrent absence” of core conditions rather than a single deficient factor. That is, the core elements of the key dimensions of NQP are missing simultaneously, or the key supporting conditions are not met, ultimately leading to a non-high agricultural product SCR. The results of the non-high configuration once again verify the “causal asymmetry” and “multiple concurrent absences” characteristics of NQP’s ability to enhance agricultural product SCR.

4.5. Robustness Test

To further confirm the robustness and reliability of the research results, a robustness test was carried out by adjusting the consistency threshold based on the established literature [54]. The original consistency threshold was adjusted from 0.8 to 0.85. The configuration results are shown in

Table 6. Configuration robustness test results after increasing the PRI consistency threshold.

Condition Variables	High Configuration Results			Low Configuration Results
	A	B	C	a	b	c	d
NQMP	●		●	⊗	⊗	●	●
NQLO	●	●			⊗	⊗	⊗
NQL	●	●	●		⊗	●	●
NQDC		●	●	⊗	⊗		⊗
NQDE	●	●	●	⊗		⊗	⊗
Consistency	0.9608	0.9803	0.9257	0.9470	0.9476	0.9483	0.9430
Original Coverage	0.3419	0.4856	0.5402	0.6543	0.4240	0.3078	0.2880
Unique Coverage	0.0621	0.0768	0.1677	0.1866	0.0091	0.0510	0.0275
Case Counts	2	3	8	10	6	2	2
Solution Consistency	0.9259			0.9440
Solution Coverage	0.7154			0.7600

Note: ● signifies the presence of core conditions. ⊗ signifies the lack of the core condition. Blank space signifies a “do not care” condition. Case counts only record cases with a membership greater than 0.5 in the term.

. It can be found that the high configuration results and non-high configuration results in Table 6 are consistent with those in Table 5. The original configuration path and type are still valid, and the consistency and coverage of the overall solution, as well as each configuration path solution, have no obvious changes, so the research conclusions of this paper are robust.

5. Discussion

Our research elucidates how the five dimensions of China’s NQP—NQMP, NQLO, NQL, NQDC, and NQDE—affect agricultural product SCR. Previous studies about NQP level have mainly focused on the three traditional factors, overlooking the importance of the NQDC and the NQDE. In addition, studies on the specific path through which NQP empowers SCR are insufficient, especially considering the time lag of the used panel data. Therefore, this study constructs NQP and agricultural product SCR indicators. Based on panel data produced by 31 Chinese provinces from 2011 to 2022, we solved the “time lag” problem by integrating a BPNN with ARIMA modeling to predict the NQP level. Subsequently, the empowering paths through NQP enhancing agricultural product SCR were explored via the entropy weight TOPSIS and fsQCA methods.

This study shares multiple consensuses with existing research on core understandings. First, all parties acknowledge the positive enabling role of NQP on SCR [39,40,41] and jointly emphasize the central role of technological innovation in this process—existing research indicates that technologies such as digitalization and the Internet of Things can significantly improve SCR [43,44,45], while this empirical study further confirms that NQP is a key core condition for constituting a highly resilient configuration. Second, at the regional impact level, this study’s findings on resilience differences between the eastern, central, and western regions echo existing research’s focus on regional differences in NQP effects [15,39,40] and spatial differentiation in high-quality agricultural development [55], jointly confirming the universal law of regional imbalance and extending the analytical perspective to the dynamic capability dimension of “SCR”. Furthermore, this study’s exploration of resilience “configuration” paths aligns with the research orientation of resilient agricultural supply network design [56]. The use of the NCA method to identify bottleneck factors shares methodological commonality with research on composite indicators for comprehensively assessing food security and environmental sustainability [57]. Both deepen the understanding of system resilience from the perspectives of “configuration synergy” and “constraint diagnosis”, respectively. Finally, the “factor synergy” and “system compensation” mechanisms emphasized in this study are highly relevant to cutting-edge discussions on innovation-driven supply chains and digital agriculture. They both point to the fact that digital technologies such as generative AI can be integrated into different driving paths as “empowering factors” and compensate for system shortcomings through “factor substitution” [58,59]. This provides a systems theory-level explanation for the “multiple concurrent effects” and “asymmetry” presented by NQP, thereby establishing an organic link between factor synergy and system resilience.

Compared with existing studies, this study has made breakthroughs in three aspects: First, in terms of indicator construction, the new development concept and environmental dimension are incorporated into the NQP evaluation system, which makes up for the previous focus on the three dimensions of “laborers-labor materials-labor objects” and is more in line with the complexity of agricultural ecosystems. Second, concerning research methods, the ARIMA-BPNN hybrid prediction model is combined with fsQCA for the first time to solve the “time lag” problem of panel data and overcome the problem that traditional regression analysis struggles to capture causal complexity. Finally, at the conclusion level, it clarifies the “configuration path” rather than the “linear relationship” of NQP empowering resilience, overcoming traditional methods’ reliance on “single factor determinism”, especially refining the substitution mechanism between technology and the environment in different regions and forming an empirical echo with the qualitative conclusion that “NQP has nonlinear spillover effects”.

This study has the following limitations: First, the contextual dependence and universality verification of the conclusions are insufficient. The findings, based on a Chinese context, require further examination regarding their cross-regional and cross-temporal external validity. Second, data constraints exist. Reliance on provincial aggregated data may mask micro-level differences, and some selected variables may be missing (e.g., favorable policies and government subsidies). Third, the predictive model has inherent uncertainties. The prediction results of ARIMA-BPNN contain certain errors, and the data used is from the past, which may not fully reflect the latest developments.

The impact of time lag on external validity is mainly reflected in three aspects: First, in the context of rapid technological development, predictions based on historical data struggle to accurately capture the actual impact of the latest technological breakthroughs on NQP. Second, in the process of dynamic adjustment of the policy environment, the analysis results may not reflect the effects of the latest policy interventions. Third, in the case of rapid evolution of market demand structure, the predictive model is not sensitive enough to emerging trends. All of these factors limit our ability to promote the research conclusions at the current point in time.

Future research can be deepened in three aspects: First, data enhancement: Collecting enterprise-level micro-data and high-frequency digital footprints. Systematic collection of enterprise/farm-level micro-data, conducting multi-case tracking and questionnaire surveys, and combining high-frequency digital trajectories such as e-commerce and remote sensing can build a more refined monitoring system to reveal behavioral differences and real-time response mechanisms among different entities in resilience configurations. Second, methodological expansion: Developing time-series configuration models and incorporating external shocks. Longitudinal qualitative comparative analysis or event history analysis can be used to establish panel datasets including data related to shocks such as pandemics and extreme climate events to test the robustness of configuration paths, and multi-dimensional dynamic models such as system dynamics can be constructed to simulate the effects of long-term policy interventions. Third, theoretical integration: exploring the coupling mechanism of “shock-configuration-resilience” in special situations. A key focus should be on examining how local resources and technologies adapt to form resilience paths in special regions such as the west and border areas and comparing the “compensation and substitution” mechanisms of various dimensions of new productivity under different shock types, thereby constructing a more general theoretical analysis framework.

6. Conclusions and Recommendations

6.1. Conclusions

This paper takes agricultural product SCR as the research object and integrates the entropy weight TOPSIS method, fsQCA method, BP neural network, and ARIMA prediction method to explore the driving factors and configuration path through which NQP empowers the improvement of agricultural product SCR. The study found the following:

(1)

China’s agricultural product SCR shows spatial differences with fluctuations in the east, central, and western regions. The central region (such as Henan) performs well due to resource integration and policy adaptability, while the western region (such as Tibet and Qinghai) has low resilience due to a weak foundation.

(2)

Both the fsQCA and NCA methods identified that a single factor does not constitute a necessary condition for high agricultural product SCR, but rather promotes the improvement of China’s agricultural product SCR by bolstering the synergy of factors. There are three driving paths for high agricultural product SCR: autonomous endogenously driven type (MP-LO-L-DC type), environment empowerment-driven type (LO-L-DE-DC type), and system architecture-driven type (MP-L-DE-DC type).

(3)

Bottleneck-level analysis results of NCA show that NQMP is the first to become a bottleneck factor for improving agricultural product SCR, and the bottleneck effect of each factor gradually increases with the improvement of the agricultural product SCR level.

(4)

The synergistic existence and functional complementarity of the core elements of high-resilience path dependence allow for system compensation through “element substitution”. Low resilience stems from the concurrent absence or structural mismatch of key conditions. Shortcomings in any core dimension may lead the system to a low-level equilibrium. NQP has multiple concurrent effects and asymmetric characteristics in improving agricultural product SCR.

6.2. Recommendations

Our results have the following implications for the development of China’s agricultural product SCR.

First, establish a cross-regional collaboration and dynamic adjustment mechanism. This includes the following: (1) Strengthening regional linkages, leveraging the radiating and driving role of eastern system architecture-driven regions and central environment-empowering regions, and assisting western regions with comprehensive shortcomings to quickly fill core gaps through technology transfer, talent exchange, and policy sharing; the central region should continue to consolidate its advantages in policy adaptation and resource integration, build a cross-regional supply chain collaboration hub, and promote complementary elements between the eastern, central, and western regions. (2) Establishing a dynamic monitoring and adjustment mechanism for resilience levels, tracking changes in the bottleneck effects of core elements in each region, and optimizing policy priorities for different evolutionary stages. (3) Strengthening the connection between top-level design and local practice, integrating the new quality development concept into regional agricultural policy formulation, avoiding structural mismatches caused by single-dimensional policies, and promoting the formation of a complementary synergy among technology, human resources, the environment, and other concepts through institutional guarantees, fundamentally breaking the “triple lock-in” of low-resilience regions and driving an overall leap in the national agricultural product SCR.

Second, all regions should base their development on their resource endowments and foundations, selecting appropriate resilience enhancement paths and strengthening the synergy of core elements. For autonomously driven regions such as Guangxi and Shaanxi, it is necessary to continuously focus on the closed-loop synergy of upgrading NQMP (such as agricultural machinery in hilly and mountainous areas and the promotion of water-saving technologies), cultivating NQL (such as the “local expert” program), and optimizing NQLO (such as the intensive processing of agricultural products), consolidating the internal capacity base under the guidance of development concepts and making up for the shortcomings of insufficient external environmental support. For environment-enabled driven regions such as Henan and Sichuan, the empowering effect of NQDE such as policy environment and market ecology should be further amplified, guided by clear development concepts, promoting the allocation of environmental resources towards human capital enhancement and the upgrading of characteristic agricultural products, and making up for the gap in insufficient investment in hard technology by integrating resources through socialized service organizations. For system architecture-driven regions such as Heilongjiang and Guangdong, it is necessary to maintain the systematic integration of NQMP, NQLO, NQDE, and NQDC, solidify the synergistic paradigm of “resources-technology-policy-concept”, and strengthen the robustness of the system architecture through standardized production processes and stable institutional guarantees. At the same time, all regions should flexibly utilize the “factor substitution” mechanism. For example, regions lacking heavy mechanization investment can achieve a resilient leap by optimizing the policy environment and cultivating a highly skilled workforce, thus avoiding the misconception of “total factor optimization”.

Third, we must focus on addressing core weaknesses and structural contradictions with precise policies. On the one hand, we must prioritize overcoming the key bottleneck of NQMP and increase research and development and promotion of intelligent agricultural equipment, digital platforms, and green production technologies, especially allocating resources to underdeveloped western regions such as Tibet and Qinghai to break the deadlock of “no technology to improve efficiency”. On the other hand, for regions with comprehensive weaknesses and low resilience, we must implement a three-pronged approach to address these weaknesses: technology, concepts, and environment. While improving the agricultural technology service system, we must update development concepts through policy dissemination and case demonstrations, and simultaneously optimize the external environment, such as county-level logistics networks and market access channels, to avoid the predicament of being unable to break the low-level equilibrium by addressing only one factor. For regions with mismatched advantages, we must focus on resolving the contradiction of “having advantages but no application scenarios”: for regions with advanced technologies but lacking suitable conditions, we must focus on promoting the standardization and branding of agricultural products, while building policy support platforms and market access channels to transform the advantages of labor resources and workers into actual resilience. For regions with outstanding technological and human resources advantages but lacking in development concepts and the environment, it is necessary to strengthen the strategic guiding role of development concepts, integrate green and high-quality orientations into industrial policies, attract resource agglomeration through policy dividends, and break the triple constraints of “direction-value-platform”.