Evaluating Regional Agricultural Resilience Using Dynamic CoCoSo with Hybrid Weights: A Case Study of Sichuan, China

Abstract

Regions with insufficient resilience in their agriculture industry can usually be exposed to threats of unstable supply of food and agricultural products. Therefore, agricultural resilience is important for regional development and welfare. To support the development of agricultural resilience, proper policies and incentives need to be implemented. To achieve this, the first step is to appropriately evaluate the regional agricultural resilience levels. In this study, a novel agricultural resilience evaluation method was developed based on hybrid weighting approaches and dynamic CoCoSo (i.e., Combined Compromise Solution). The method can capture the temporal change in resilience levels, integrate richer information, and provide more robust output. To confirm its effectiveness, the method was applied to the evaluation of regional agricultural resilience in 21 cities of Sichuan province in China across five years. Over a recent five-year period, the annual average levels of agricultural resilience in Sichuan have increased, although this trend became less significant in more recent years. Also, the resilience levels among cities are diverse, and some cities have experienced significant changes of resilience across years. When considering temporal effects integrating five years, Liangshanzhou city ranks the first and Bazhong city ranks the last in terms of their resilience levels, but such results can depend on CoCoSo parameters and time weight parameters, with the latter having more significant influence. This study can contribute to the existing literature by providing new methodological tools for agricultural resilience research and regional management studies. Also, this study can help identify cities with different agricultural resilience levels and dynamics, informing practitioners’ new perspectives for agricultural policy evaluation as well as business strategy planning.

1. Introduction

Agriculture is vitally important for economic growth and social welfare, and agricultural and food security have now become a key focus worldwide [1]. The robustness and stability of agricultural product supply can not only enable people’s quality of life but also support the realization of sustainable development goals (SDGs) proposed by the United Nations, such as SDG 2 (i.e., Zero Hunger).

In China, due to its high population, agriculture is of great importance to social and economic development. Each year, multiple policies have been implemented to support and enhance the security of the agricultural industry. For example, China unveiled the 2025 “No. 1 central document” recently. As one of the most important policy documents in China each year, the document in 2025 has pointed out the necessity of enhancing the sustainable supply capability of food and other key agricultural products [2]. The “plan to accelerate building up the strength in agriculture for the period from 2024 to 2035” was recently unveiled by the Communist Party of China Central Committee and the State Council, also explicitly emphasizing the importance of the stable and secure supply of food and agricultural products [3].

Sichuan is one of the provinces with the largest population in China. The province is located in southwestern China and has 21 cities. According to the China Statistical Yearbook 2024 [4], the population of Sichuan province has reached around 83.68 million, ranking in the top five provinces in terms of population size in China. Therefore, the regional agricultural security in Sichuan significantly determines regional development and welfare.

To enhance regional agricultural security, one of the most effective approaches is to strengthen agricultural resilience. Agricultural resilience, according to the existing literature, can be defined as the ability of agricultural systems to resist and recover from shocks and damage [5,6]. Specifically, the shocks and damage here, adapting the definition of resilience developed by the Food and Agricultural Organization (FAO) of the United Nations, can be considered as the threats that impact agriculture, nutrition, food security and food safety [7]. In recent years, due to the occurrence of extreme weather and natural disasters, as well as varied uncertain factors, the importance of agricultural resilience has attracted more attention in both academic and industrial communities. To enhance agricultural resilience, the first step is to properly evaluate the resilience level of agricultural systems, as it drives the policy and incentives to be better implemented to develop its resilience.

In the existing literature, multiple explorations have been conducted for agricultural resilience. Different topics have been covered, such as identifying the factors influencing agricultural resilience [8], developing evaluation approaches to measure the resilience levels [5], and comparing and ranking regions in terms of their agricultural resilience [9]. However, the published studies revealed research gaps that need to be addressed. First, the evaluation of agricultural resilience normally adopted a static perspective, with little focus on dynamic evaluation. This can potentially fail to catch the impact of temporal effects on resilience evaluation, delivering inappropriate or even inaccurate messages to policy makers. Second, the evaluation approaches were normally developed based on widely used methods (e.g., spatial regression, machine learning, etc.), but their robustness may not be fully examined and secured, potentially undermining the validity of study findings. Finally, the research scope of the studies on agricultural resilience normally adopted a national level perspective, with few applying a province-wide or region-wide scope. As agricultural industry in different provinces can present diverse structures and properties, examining agricultural resilience only from a national level may hinder the capture of unique features of regional agricultural industry.

Therefore, to properly address the above research gaps and inform practitioners and policy makers, this study aims to investigate the following research questions:

(1)

How can the agricultural resilience of cities in Sichuan province be evaluated dynamically?

(2)

What are the patterns of agricultural resilience of cities in Sichuan province in the past five years?

To address the research questions, this paper developed a hybrid weighting dynamic CoCoSo approach to examine the agricultural resilience levels across the 21 cities of Sichuan province. Through developing an agricultural resilience evaluation index system, the resilience of each city was examined and compared based on the data from Sichuan statistical yearbooks 2020 to 2024, and the policy and managerial implications were generated based on the evaluation results. By doing this, we believe this paper can contribute to the academic and practical implications:

• Academically speaking, this study introduced a novel and robust dynamic evaluation method which can capture rich information for agricultural resilience evaluation. This can enrich the analytical tools for agricultural risk and resilience management. Also, the findings can provide a theoretical basis for future studies of Sichuan province and other close regions in relevant topics (e.g., regional food safety study, regional economy, or regional risk prevention research).
• Practically speaking, this study can enable the policy makers to longitudinally evaluate the outcomes of agricultural policy, further supporting policy makers to design and implement more effective agricultural policy and incentives. In addition, local agricultural companies can benefit from the proposed method to upgrade their decision-making software to make more strategically appropriate plans.

This study has six sections. After the Introduction, Section 2 reviews the relevant literature, and Section 3 presents the methods of this study. After that, the results are presented in Section 4, followed by discussion and conclusion in Section 5 and Section 6, respectively.

2. Literature Review

This study concerns two streams of literature review, namely agricultural resilience development and resilience level evaluation.

2.1. Agricultural Resilience Development

Existing literature has examined resilience development under multiple factors. The first category is the resilience development under natural factors, such as extreme weather, climate change, natural disasters, or pandemics. For example, Rezvi et al. [8] indicated that climate change can be a key factor that poses negative effects to crop production and agricultural food security. To mitigate such effects and enhance agricultural resilience, the authors proposed the solution of developing tolerant and biofortified rice varieties. Zeng et al. [10] revealed that extreme weather events can do harm to the stability and robustness of the agricultural economy. They also mentioned that agricultural insurance can be an effective tool to mitigate negative consequences of these events. By using agricultural insurance, the ability to resist extreme weather consequences in agriculture can be enhanced, increasing its resilience. Lioutas et al. [11] discussed how agriculture can improve its ability to cope with major disasters based on the experience from the COVID-19 pandemic. They discussed the practices drawn by Canadian and American governments, and proposed that to promote agricultural resilience, policies should support the development of crisis management plans and enhance farmers’ ability to resist external disturbance.

Another category is resilience development under economic and technological factors, such as industrial and market structures or technological advancement and digitalization. For example, Zhou et al. [12] analyzed agricultural resilience from an economic perspective. Using panel data of provinces in China, they found that the integration of rural industries can significantly contribute to agricultural economic resilience, and such effect can vary in different regions. Xue et al. [13] examined how a digital economy can impact agricultural industry. Their findings claimed that a digital economy can positively contribute to the agricultural development resilience.

The third category is resilience development under social factors, such as population and labor structures. For example, Zhang et al. [14] applied the spatial autoregressive model and the generalized moment method to measure the effect of an aging rural workforce on agricultural economic resilience. Using the Chinese agricultural industry as the research context, they found that an aging rural workforce significantly reduced agricultural economic resilience. Han et al. [15] analyzed the panel data of China to identify the effect of population from the agricultural economy perspective. They also found negative impacts posed by population aging in risk resilience of the agricultural economy.

The above literature indicated that there have been rich explorations in agricultural resilience development, and many of them were conducted in China. However, the explorations seem to prefer to adopt national level investigation, with few focusing on provincial level and regionally specific properties in agriculture industry.

2.2. Agricultural Resilience Evaluation

The second stream of the literature is the agricultural resilience evaluation, and three topics can be identified in this stream. First, the existing literature of this stream explored the construction of an agricultural resilience index system, where multiple indicators have been considered to obtain an inclusive evaluation. For example, Lin et al. [6] developed an analysis index system of agricultural resilience by considering the resistance recovery capacity, adaptive regulatory capacity, and innovation and transformation capacity. Using this system, the authors examined the evolutionary pattern of Chinese agricultural resilience. Qiao et al. [5] developed an agricultural resilience index system from both resistance and reconstructive power perspectives. The system considered five dimensions of agricultural resilience, including production resilience, ecological resilience, economic resilience, adaptability, and transformation and innovation capabilities. Luo et al. [16] examined the cropland resilience evaluation, and they developed an index system incorporating resistance, adaptation, and innovation dimensions with 18 sub-factors.

The second direction is the methods proposed for agricultural resilience evaluation. For example, Yang et al. [17] adopted machine learning techniques such as random forest to measure the agri-economy resilience and explore its influential factors. Chao et al. [18] used the entropy method to calculate the agri-economy resilience index and analyze the effect of agricultural production agglomeration by applying the Spatial Durbin model, kernel density estimation, and threshold regression. In the literature, the MCDM methods are also applied to the agricultural resilience evaluation. For example, Xue et al. [13] applied entropy-TOPSIS (i.e., Technique for Order Preference by Similarity to Ideal Solution) method to measure the levels of agricultural development resilience and the digital economy, further exploring their relationship. Also, Kertolli et al. [19] measured the resilience of the water–energy–food–ecosystem nexus of the Fès–Meknès region in Morocco to help farmers adapt to climate change, and the SAW (i.e., simple additive weighting) method was used to prioritize innovative strategies for adaption.

Finally, the third direction is the practical applications of agricultural resilience evaluation, in which the evaluation results are used to support decision-making (e.g., policy design and business strategy planning). For example, Robert et al. [20] evaluated the economic vulnerability and resilience of agriculture in England and Wales based on individual farm business data, supporting the decision-making and strategy planning for sustainable agricultural promotion. Qiao et al. [5] measured the agricultural resilience levels of each province in China, finding that the overall levels of agricultural resilience in China are still relatively low and presenting regional difference. The findings therefore can inform the necessity of customized policies for different regions. Chen et al. [21] used the fuzzy analytic hierarchy process and entropy methods to evaluate the resilience to drought of Huaibei plain, and their results supported the design of proper policy and emergency plans in agriculture.

The above literature revealed that multiple studies have been conducted in agricultural resilience evaluation, and different methods were adopted. However, compared with the widely used methods (i.e., statistical inference/regression or machine learning models), the MCDM methods seem less popular, with dynamic MCDM attracting even fewer academic attentions in agricultural resilience evaluation.

2.3. Research Gap Identification

Based on the above literature review, the following research gaps can be identified. On the one hand, although rich studies have been conducted to explore the factors and evaluation of agricultural resilience, the majority of them adopted a national level and were lacking in more specific investigation (i.e., provincial level). As suggested by the above literature [5] that significant difference can exist among different regions, lacking in such a view may hinder the way of appropriate and tailored policy design for specific provinces. On the other hand, although multiple methods have been developed to evaluate the agricultural resilience level, the MCDM method, especially its dynamic variations, has rarely been used. This makes the current research fail to benefit from the robustness of MCDMs as well as their ability to capture dynamic information in alternative evaluations.

Therefore, to address the above research gaps, this paper aims to develop a dynamic hybrid weighting CoCoSo method and apply it to evaluating agricultural resilience in Sichuan province. Through this, this paper is expected to contribute to the methodological innovation in agricultural resilience studies, as well as to the support of policy design and strategy planning for regional policy-makers and agricultural companies.

3. Materials and Methods

Having identified the research gaps from the literature, this study attempts to address them by following the methodological flows in Figure 1. The proposed method is based on the CoCoSo model which is a widely used and robust algorithm in MCDM [22]. The extension of this algorithm made by this paper enables its objective evaluation using dynamic data. The R (version 4.4.2) was used for programming. Specifically, there are four main steps to achieve this extension, namely MCDM formation, hybrid weight, dynamic CoCoSo calculation and sensitivity analysis.

3.1. MCDM Formation

The first step of this study is to form the MCDM problem. To properly construct the MCDM problem, the alternatives should be decided. To obtain a thorough evaluation of agricultural resilience in Sichuan, all cities in the province were taken into account. Specifically,

Table 1. Cities in Sichuan province.

City Name	Chengdu	Zigong	Panzhihua	Luzhou	Deyang	Mianyang	Guangyuan
Abbreviation	CD	ZG	PZH	LZ	DY	MY	GY
City Name	Suining	Neijiang	Leshan	Nanchong	Meishan	Yibin	Guangan
Abbreviation	SN	NJ	LS	NC	MS	YB	GA
City Name	Dazhou	Yaan	Bazhong	Ziyang	Aba	Ganzi	Liangshan
Abbreviation	DZ	YA	BZ	ZY	ABZ	GZZ	LSZ

summarizes the 21 cities in Sichuan which are the alternatives in the model.

After constructing the alternatives, the evaluation index system should be developed. Multiple indicators should be included in the evaluation index to obtain a comprehensive and accurate evaluation of agricultural resilience. After thoroughly reviewing the previous literature related to agricultural resilience (i.e., [5,16,21,23,24,25]) as well as carefully considering the data accessibility (e.g., statistical data of cities in Sichuan province), the index system with ten indicators are developed in

Table 2. Index system for agricultural resilience.

Indicator	Name	Explanation	Effect
C1	Effective irrigation rate	Total irrigated area of cultivated land/total sown area (ratio)	+
C2	Agricultural machinery power rate	Total power of agricultural machinery/total sown area (10 kw/hectare)	+
C3	Proportion of rural population	Regional rural population/regional total population (ratio)	+
C4	Grain crops per capita	Total grain crops yield/regional total population (tons/person)	+
C5	Fertilizer application rate	Total consumption of chemical fertilizers/total sown area (100,000 tons/hectare)	-
C6	Agricultural economy size	Gross output value of farming, forestry, animal husbandry and fishery (100 million yuan)	+
C7	Average value-added in primary industry	Value-added in primary industry/number of employed persons in primary industry (10,000 yuan/person)	+
C8	Rural disposable income per capita	Per capita disposable income of rural households (yuan)	+
C9	Rural expenditure consumption per capita	Per capita expenditures for consumption (yuan)	+
C10	General public budget expenditure	General public budget expenditure for agriculture, forestry and water conservancy (10,000 yuan)	+

.

Finally, the data for MCDM were collected. To capture the changes of the agricultural resilience across years, this paper adopted a novel dynamic perspective for resilience evaluation. This means the indicator data needed to be collected by multiple years, which is different from the classical static MCDM models. To achieve a thorough evaluation, consistent with the previous literature [26], we collected the data of five years for each indicator by cities from the Sichuan Statistical Yearbooks published from 2020 to 2024 [27] to ensure the dynamic CoCoSo could be conducted to comprehensively evaluate the cities’ agricultural resilience levels. Due to the time delay, what should be noted is that the published year means the data covering the last year. For example, statistical yearbook 2022 represents the data covering from the beginning of 2021 to the end of 2021 (i.e., the beginning of 2022). Also, the Sichuan Statistical Yearbook 2020–2024 is the most recent 5-year available data that we could collect when conducting this study.

Based on the data collected, the decision array can be developed as ${\mathit{H}}_{\mathit{T}\times \mathit{N}\times \mathit{M}}$, where $T$ is the time dimension which is five (i.e., 2020 to 2024), while $N$ and $M$ are the number of alternatives (i.e., 21) and indicators (i.e., 10). Specifically, it is formulated as follows, and $h_{tij}$ is the element of ${\mathit{H}}_{\mathit{T}\times \mathit{N}\times \mathit{M}}$, with $1\le t\le 5$, $1\le i\le 21$, and $1\le j\le 10$.

$${\mathit{H}}_{\mathit{T}\times \mathit{N}\times \mathit{M}}=\begin{array}{c}\left[\begin{array}{c}\begin{array}{ccc}{h}_{111}& h_{112}\dots & h_{11m}\\ h_{121}& h_{122}\dots & h_{12m}\\ \dots & \dots & \dots \end{array}\\ \begin{array}{ccc}{h}_{1n1}& h_{1n2}\dots & h_{1nm}\end{array}\end{array}\right]\\ \left[\begin{array}{c}\begin{array}{ccc}{h}_{211}& h_{212}\dots & h_{21m}\\ h_{221}& h_{222}\dots & h_{22m}\\ \dots & \dots & \dots \end{array}\\ \begin{array}{ccc}{h}_{1n1}& h_{1n2}\dots & h_{1nm}\end{array}\end{array}\right]\\ \begin{array}{c}\dots \\ \left[\begin{array}{c}\begin{array}{ccc}{h}_{511}& h_{512}\dots & h_{51m}\\ h_{521}& h_{522}\dots & h_{52m}\\ \dots & \dots & \dots \end{array}\\ \begin{array}{ccc}{h}_{5n1}& h_{5n2}\dots & h_{5nm}\end{array}\end{array}\right]\end{array}\end{array}$$

(1)

3.2. Hybrid Weights

After constructing the MCDM problem, the second step is to determine the relative importance of each indicator in the index system. This needs to weigh indicators. To obtain appropriate weights, this study conducted a hybrid weighting approach integrating entropy, MEREC (i.e., method based on the removal effects of criteria), and LOPCOW (i.e., logarithmic percentage change-driven objective weighting) weights. The justification of adopting hybrid weights is that three weighting approaches can extract indicator information from different perspectives (see following mathematical formulae for details), expecting to contribute to a more appropriate decision-making in greater possibility.

3.2.1. Entropy Method

The first way of weighting is the Entropy method, which is one of the most frequently used weighting method (e.g., [28]). Compared with the previous literature, as this paper considers the dynamics of agricultural resilience, the effects of multi-period need to be considered in the weight generation. Based on the entropy method, the weight of the $j^{\mathrm{t}\mathrm{h}}$ indicator ($1\le j\le 10$) is derived as $e{w}_j$. The specific steps of entropy method are presented in Appendix A.1.1.

3.2.2. MEREC Method

The second way of weighting is the MEREC method. It was developed in [29]. To fit the dynamic evaluation in the modelling, the approach is also adapted to enable temporal effects. Based on the MEREC method, the weight of $j^{\mathrm{t}\mathrm{h}}$ indicator ($1\le j\le 10$) is derived as $d{w}_j$. The specific steps of MEREC method are presented in Appendix A.1.2.

3.2.3. LOPCOW Method

The third way of weighting is the LOPCOW method, which was developed in [30]. To fit the dynamic evaluation in the modelling, the approach is also adapted to enable temporal effects. Based on the LOPCOW method, the weight of $j^{\mathrm{t}\mathrm{h}}$ indicator ($1\le j\le 10$) is derived as $p{w}_j$. The specific steps of LOPCOW method are presented in Appendix A.1.3.

3.2.4. Hybrid Weighting Process

After obtaining the weights of indicators by using entropy, MEREC, and LOPCOW methods individually (i.e., $e{w}_j$, $d{w}_j$, and $p{w}_j$), the hybrid indicator weights are constructed to achieve a holistic indicator weighting. In this study, the geometric mean formula is used to obtain the hybrid weights, with a similar means of observation in the previous literature (e.g., [31]):

$$w_j={\displaystyle \frac{\sqrt[3]{e{w}_j\times d{w}_j\times p{w}_j}}{{\sum }_{j=1}^{10}\sqrt[3]{e{w}_j\times d{w}_j\times p{w}_j}}}$$

(2)

According to the previous literature, the hybrid weights can reduce the possible bias induced from single weighting methods [32]. Also, using hybrid weights can contribute to obtaining diverse information from different features [33]. Therefore, in this study, the hybrid weights can integrate the information captured by the individual methods, expectedly resulting in more informative and robust weights of dynamic agricultural resilience for each indicator.

The formulation of hybrid weights reflects the economic implications for indicators: the larger weights represent the bigger impacts on the agricultural resilience from certain indicators. In other words, policy makers should pay more attention to the indicators with larger weights when designing and implementing policies (e.g., subsidies or public budget planning).

3.3. Dynamic CoCoSo

Having derived the weights from the above methods, we next apply the dynamic CoCoSo to comprehensively evaluate the regional agricultural resilience. Specifically, based on the weights derived from the hybrid processes, the CoCoSo is used to calculate the agricultural resilience levels for the cities. CoCoSo was developed in [22], and the justification of using CoCoSo rather than other methods such as TOPSIS, ARAS (i.e., additive ratio assessment) or WASPAS (i.e., weighted aggregated sum product assessment) is because of its superiority in robustness in evaluating alternatives. Specifically, the robustness and reliability of CoCoSo are secured from its unique combination of three compromised score functions [34]. Benefiting from such a mathematical structure, the regional agricultural resilience considering temporal effects can be better examined under dynamic changes.

As the weights have considered temporal effects, the model is essentially a dynamic CoCoSo model. Specifically, by extending previous literature, the dynamic CoCoSo value can be derived as $R_{ti}$ ($1\le t\le 5$, $1\le i\le 21$). To interpret the meaning of $R_{ti}$, it represents the agricultural resilience level of each city by year. For example, $R_{13}$ means the resilience level of the third city in 1st year. More specifically, it means the resilience level of PZH (see Table 1) in 2020 (as 2020 is the first year the data were collected). Also, the property of mathematical models determines higher $R_{ti}$ implies a higher agricultural resilience level.

First, the ${\mathit{H}}_{\mathit{T}\times \mathit{N}\times \mathit{M}}$ is normalized to ${\mathit{A}}_{\mathit{T}\times \mathit{N}\times \mathit{M}}$ based on the following formula with $a_{tij}$ as the element of ${\mathit{A}}_{\mathit{T}\times \mathit{N}\times \mathit{M}}$:

$$a_{tij}=\left\{\begin{array}{c}{\displaystyle \frac{a_{tij}-\underset{t,i}{\mathit{min}}{(a}_{tij})}{\underset{t,i}{\mathit{max}}{(a}_{tij})-\underset{t,i\mathrm{ }}{\mathit{min}}(a_{tij})}}, {\mathrm{C}}_{\mathrm{j}} \mathrm{has}+\mathrm{effect}\\ {\displaystyle \frac{\underset{t,i}{\mathit{max}}(a_{tij})-a_{tij}}{\underset{t,i}{\mathit{max}}{(a}_{tij})-\underset{t,i}{\mathit{min}}(a_{tij})}}, {\mathrm{C}}_{\mathrm{j}} \mathrm{has}-\mathrm{effect}\end{array}\right.$$

(3)

Second, the weighted sum and the power weighted sum of each alternative under indicators are calculated as follows [22]:

$${\Phi }_{ti}={\sum }_{j=1}^{10}{w}_j\times a_{tij}$$

(4)

$${\Theta }_{ti}={\sum }_{j=1}^{10}{\left(a_{tij}\right)}^{w_j}$$

(5)

Third, the three appraisal scores are derived as follows [35]:

$${\beta }_{ti}={\displaystyle \frac{{\Phi }_{ti}+{\Theta }_{ti}}{{\sum }_{t=1}^5{\sum }_{i=1}^{21}({\Phi }_{ti}+{\Theta }_{ti})}}$$

(6)

$${\epsilon }_{ti}={\displaystyle \frac{{\Phi }_{ti}}{\underset{t,i}{\mathit{min}}{\Phi }_{ti}}}+{\displaystyle \frac{{\Theta }_{ti}}{\underset{t,i}{\mathit{min}}{\Theta }_{ti}}}$$

(7)

$${\rho }_{ti}={\displaystyle \frac{\tau \times {\Phi }_{ti}+(1-\tau )\times {\Theta }_{ti}}{\tau \times \underset{t,i}{\mathit{max}}{\Phi }_{ti}+(1-\tau )\times \underset{t,i}{max}{\Theta }_{ti}}}$$

(8)

The $\tau$ here indicates the relative importance between ${\Phi }_{ti}$ and ${\Theta }_{ti}$. Supported by the previous literature [22], $\tau$ is set to 0.5. The rationale to adopt this baseline value is to assume a neutral decision maker who has equal preference on both values. In the sensitivity analysis below, the effect of different values of $\tau$ will be analyzed.

Fourth, the CoCoSo values for each city by year can be derived as follows [35]

$$R_{ti}=\sqrt[3]{{\beta }_{ti}{\times \epsilon }_{ti}\times {\rho }_{ti}}+{\displaystyle \frac{{\beta }_{ti}{+\epsilon }_{ti}+{\rho }_{ti}}{3}}$$

(9)

Finally, as the existence of the temporal effect, the levels of agricultural resilience in different years may have varied importance. This leads to the consideration of time weights for resilience levels each year. Notating ${\phi }_t$ as the time weights with $1\le t\le 5$, the dynamic CoCoSo values for comprehensive evaluation of the city’s agricultural resilience level can be derived as follows [36]:

$$R_i={\sum }_{t=1}^5{\phi }_t\times R_{ti}$$

(10)

To interpret the meaning of $R_i$, it represents the comprehensive measurement of the agricultural resilience level of city $i$ ($1\le i\le 21$) across five years. For example, $R_2$ means the agricultural resilience level of ZG (see Table 1) integrating the temporal effect of data of five years.

According to the previous literature (e.g., [36,37]), to appropriately derive the ${\phi }_t$, the widely used nonlinear optimization approach is applied where ${\phi }_t$ satisfies the following relationship:

$$\left\{\begin{array}{c}\mathit{max}\left(-{\sum }_{t=1}^5{\phi }_t\times \mathit{ln}\left({\phi }_t\right)\right)\\ s.t. \lambda ={\sum }_{t=1}^5\left({\displaystyle \frac{5-t}{5-1}}\right){\phi }_t\\ {\sum }_{t=1}^5{\phi }_t=1\end{array}\right.$$

(11)

The value of $\lambda$ ranges from 0 to 1, and it indicates the relative importance attached to the more recent data [36]. The lower the $\lambda$ is, the more relative importance of recent data is considered by the decision maker. For example, if $\lambda$ is 0.1, it means the extreme importance attached to the recent data. On the contrary, if $\lambda$ is 0.9, the recent data is of low importance. The mathematical property of the above nonlinear programming also indicates that if $\lambda$ is 0.5, the data across different time periods will have equal importance. Therefore, it can be justified that the adoption of this way of determining time weights can enable the decision-makers to flexibly adjust the relative importance of recent data, better fitting the practices and informing policy making. For more information about this way of determining time weights, we refer interested readers to [36] for details. Supported by the previous literature of dynamic MCDM [36,37], the $\lambda$ is set to 0.2 in this study to give higher weights to the more recent data. The rationale is that the old data close to 2019 may be affected more by the COVID-19 pandemic. Considering the effects of the COVID-19 pandemic have gradually diminished, less consideration should be given to the old data to achieve a fairer evaluation. However, the robustness of results subject to different $\lambda$ will be tested in the sensitivity analysis below.

3.4. Sensitivity Analysis

When calculating the results of dynamic CoCoSo, the results of the evaluation of agricultural resilience of cities in Sichuan essentially depend on two parameters, i.e., $\tau$ for CoCoSo calculation, and $\lambda$ for time weight calculation. Therefore, it is necessary to alter the parameters and check result sensitivity, which can better inform the policy makers and local agricultural enterprises in their decision-making to enhance the regional agricultural resilience.

3.5. Method Summary

The above Section 3.1, Section 3.2, Section 3.3 and Section 3.4 have depicted the method flows of this study. To sum up, this study first forms the MCDM problem (identifying alternatives and criteria), then calculates criterion weights (using entropy, MEREC, and LOPCOW) and derives hybrid weights (using geometric mean) and calculates dynamic CoCoSo values (by integrating CoCoSo and time weights) and finally conducts sensitivity analysis (by altering $\tau$ and $\lambda$). In the following section, the results of each step are presented.

4. Results

Based on the methods in Section 3, the agricultural resilience levels of each city are reported in this section.

4.1. Results for Agricultural Resilience Evaluation

First, to evaluate the resilience levels, the raw data of the index system developed in Table 2 were collected from the Sichuan statistical yearbook published in 2020 to 2024. The data are presented in Appendix A,

Table A1. Raw data for regional agricultural resilience indicators.

City by Year	C1	C2	C3	C4	C5	C6	C7	C8	C9	C10
CD2020	0.506939	0.563115	0.373827	0.150373	0.023161	1003.343	6.058585	24,357.09	17,572.1	1,125,457
ZG2020	0.274302	0.312839	0.573883	0.433405	0.023878	324.2333	3.324244	17,276.54	14,271.19	219,824
PZH2020	0.57906	0.979105	0.47786	0.236687	0.035461	140.1198	4.869357	18,351.87	13,383.66	141,786
LZ2020	0.287312	0.432391	0.582776	0.450514	0.018996	357.1778	2.231948	16,531.47	12,774.62	542,006
DY2020	0.328465	0.4072	0.671782	0.50801	0.03567	406.3268	3.363584	18,248.73	14,538.17	284,063
MY2020	0.331584	0.514284	0.642078	0.433694	0.029672	504.4739	3.559609	17,734.57	13,991.71	512,035
GY2020	0.187649	0.577042	0.746903	0.527817	0.019339	272.044	2.263275	13,127.36	11,020.6	545,244
SN2020	0.340974	0.330459	0.714325	0.393101	0.032009	309.9816	2.921912	16,358.1	13,789.54	359,919
NJ2020	0.280088	0.490257	0.7244	0.4183	0.022552	377.323	4.663952	16,450.23	13,007.62	300,627
LS2020	0.413446	0.787047	0.620453	0.350097	0.023392	385.9616	3.309696	16,727.95	13,724.19	548,329
NC2020	0.258116	0.337956	0.708737	0.425469	0.022955	656.0281	3.519983	15,027.21	12,022.54	706,477
MS2020	0.538572	0.692802	0.650891	0.365058	0.035907	333.1927	2.475084	18,177.29	14,800.64	327,988
YB2020	0.317056	0.427591	0.621328	0.458869	0.012695	450.8414	2.095788	16,999.23	13,418.44	655,813
GA2020	0.259927	0.607983	0.745423	0.392234	0.024659	331.6753	2.086064	16,444.98	12,522.87	486,481
DZ2020	0.226432	0.341582	0.662618	0.482547	0.025251	551.4275	2.194959	15,504.2	11,535.51	764,290
YA2020	0.461797	1.41545	0.556209	0.234494	0.041097	193.4038	2.748444	14,585.56	12,189.13	142,393
BZ2020	0.180312	0.372042	0.72046	0.522394	0.022708	214.3556	1.738295	13,231.54	11,089.67	480,083
ZY2020	0.221612	0.352914	0.835476	0.485456	0.009664	261.792	1.627004	17,592.5	13,323.81	310,414
ABZ2020	0.360051	1.006597	0.718889	0.175471	0.013919	115.0571	2.387051	14,252.14	12,861.52	654,424
GZZ2020	0.432077	1.091248	0.828624	0.206609	0.003174	100.3635	1.409919	12,807.81	9388.539	991,955
LSZ2020	0.249621	0.479897	0.789226	0.461927	0.017218	600.2257	2.241951	13,906.78	10,327.37	2,470,088
CD2021	0.508431	0.56925	0.33171	0.149937	0.021777	1071.945	4.089632	26,431.74	18,500.95	1,201,524
ZG2021	0.272072	0.317916	0.617999	0.443043	0.021748	374.8134	4.844221	18,787.89	14,742.05	268,816
PZH2021	0.615686	0.986346	0.467593	0.239352	0.029263	148.848	4.924289	19,938.25	14,293.19	185,823
LZ2021	0.284239	0.429889	0.601772	0.455897	0.017903	430.2896	3.049584	18,035.25	13,630.58	582,466
DY2021	0.324619	0.40663	0.649752	0.513714	0.034256	460.3089	4.17913	19,790.32	14,761.57	324,608
MY2021	0.328392	0.518523	0.634059	0.437358	0.027953	631.148	4.83916	19,303.43	15,038.29	576,083
GY2021	0.184883	0.575049	0.726263	0.536767	0.017861	336.5695	3.623666	14,367.11	12,083.37	605,437
SN2021	0.339312	0.335086	0.694962	0.401552	0.03098	363.1067	3.494956	17,814.5	14,771.72	420,378
NJ2021	0.277819	0.503303	0.719359	0.424551	0.020449	422.9692	5.462147	17,918.42	14,076.29	336,686
LS2021	0.404897	0.771356	0.60977	0.354986	0.022188	436.6401	4.205388	18,174.69	14,836.65	506,026
NC2021	0.2542	0.335242	0.705825	0.433257	0.021084	763.9239	4.200365	16,431.4	13,075.13	832,648
MS2021	0.534059	0.693657	0.608836	0.36826	0.034135	380.4231	3.195125	19,730.13	15,314.05	428,390
YB2021	0.314411	0.422004	0.616878	0.46341	0.01215	563.7472	3.225707	18,568.89	14,605.64	752,123
GA2021	0.258373	0.611898	0.75417	0.39749	0.023125	384.4388	3.2675	17,867.27	13,539.06	469,848
DZ2021	0.236509	0.339766	0.638634	0.489237	0.022108	636.0389	3.177444	16,876.15	12,495.93	803,732
YA2021	0.458933	1.415356	0.536697	0.237012	0.038155	237.6164	4.598304	15,889.94	13,211.62	181,304
BZ2021	0.174846	0.364475	0.708597	0.529453	0.020842	291.1581	2.551561	14,428.71	12,023.28	621,154
ZY2021	0.221942	0.354502	0.809088	0.490142	0.00949	318.2045	2.808968	19,075.99	13,705.01	307,181
ABZ2021	0.393027	0.99161	0.717949	0.17836	0.013126	142.2087	3.260628	15,539.04	12,162.18	737,790
GZZ2021	0.457231	1.08584	0.82965	0.212396	0.003025	123.2873	1.785746	13,966.64	9867.617	890,742
LSZ2021	0.251612	0.480367	0.772463	0.463255	0.016144	698.7156	2.765006	15,232.37	11,288.79	2,024,651
CD2022	0.408366	0.572858	0.321167	0.148163	0.021242	942.6124	3.66971	29,126.33	20,459.94	1,227,997
ZG2022	0.291736	0.318794	0.640985	0.451677	0.020998	392.9342	5.222533	20,693.86	16,266.32	277,968
PZH2022	0.607537	0.961184	0.476279	0.244592	0.028416	159.5456	5.281489	21,979.14	15,688.46	177,928
LZ2022	0.282083	0.429629	0.602131	0.464906	0.017174	444.853	3.240191	20,008.24	15,225.35	616,560
DY2022	0.319175	0.403308	0.629096	0.522803	0.033424	475.2248	4.418722	21,858.32	16,390.64	336,286
MY2022	0.32759	0.527846	0.633397	0.44627	0.027065	641.2822	5.088874	21,339.53	16,623.55	574,983
GY2022	0.18384	0.57591	0.722561	0.546486	0.016968	358.7886	3.975185	15,925.12	13,309.14	592,615
SN2022	0.333598	0.340261	0.698546	0.410634	0.029429	367.1668	3.653375	19,726.62	16,238.53	437,085
NJ2022	0.274314	0.506123	0.723573	0.433335	0.019032	434.9058	5.823245	19,819.39	15,544.5	354,612
LS2022	0.397352	0.767672	0.611175	0.361948	0.021311	437.8487	4.327208	20,042.99	16,557.72	510,632
NC2022	0.254255	0.336342	0.713206	0.44365	0.020206	787.4623	4.446577	18,246.56	14,655.62	842,710
MS2022	0.531848	0.691349	0.609692	0.37453	0.033477	391.8993	3.367526	21,771.45	16,959.66	407,445
YB2022	0.318398	0.422225	0.623978	0.471883	0.011611	583.4159	3.387747	20,591.14	16,302.12	763,828
GA2022	0.268681	0.607837	0.754413	0.405389	0.021964	397.0713	3.43698	19,751.6	15,003.66	474,982
DZ2022	0.232913	0.341658	0.633338	0.499495	0.021101	665.2611	3.391562	18,637.51	13,893.35	856,300
YA2022	0.452625	1.400642	0.537451	0.241422	0.03688	247.2839	4.900683	17,579.88	14,691.11	207,003
BZ2022	0.177725	0.363477	0.7099	0.541509	0.019795	314.8797	2.844336	15,962.13	13,346.19	630,112
ZY2022	0.221337	0.358497	0.82428	0.499094	0.011474	324.6032	3.017587	21,022.69	15,034.5	324,676
ABZ2022	0.398465	0.990348	0.729097	0.182743	0.012659	153.2017	3.728885	17,161.35	13,199.3	695,979
GZZ2022	0.453096	1.089342	0.802198	0.213455	0.00299	120.5172	1.835338	15,379.48	10,783.96	998,664
LSZ2022	0.254224	0.48271	0.771874	0.465921	0.015932	742.5621	2.961238	16,807.83	12,443.51	1,584,416
CD2023	0.375769	0.575358	0.311721	0.144431	0.020642	950.4171	3.614473	30,931.11	21,195.64	1,246,105
ZG2023	0.293587	0.316706	0.639783	0.447086	0.020122	408.8582	4.943816	21,976.1	17,081.75	278,331
PZH2023	0.579061	0.963366	0.478017	0.243445	0.027725	172.3233	5.050382	23,364.15	16,357.25	181,420
LZ2023	0.2862	0.424858	0.600635	0.456996	0.01643	464.1492	3.222455	21,348.46	15,851.83	634,320
DY2023	0.317022	0.397233	0.62642	0.517971	0.032274	498.1186	4.345364	23,192.32	17,415.89	352,976
MY2023	0.318942	0.526646	0.631349	0.439118	0.025986	648.3117	4.785598	22,726.18	17,345.82	583,873
GY2023	0.183446	0.573677	0.721597	0.539694	0.01632	385.3789	3.981957	16,881.23	14,058.88	577,700
SN2023	0.326776	0.34331	0.697799	0.408061	0.028353	367.1592	3.402094	20,986.27	17,100.78	463,771
NJ2023	0.26894	0.506617	0.720551	0.424599	0.018189	457.7679	5.560485	20,995.68	16,236.7	347,104
LS2023	0.389449	0.765411	0.61603	0.354748	0.020674	449.8108	4.178904	21,338.98	17,344.39	514,623
NC2023	0.259362	0.339549	0.711685	0.440837	0.019697	830.8139	4.470316	19,468.94	15,501.75	819,587
MS2023	0.521275	0.683155	0.608092	0.370438	0.032498	411.782	3.365315	23,098.87	17,731.25	443,823
YB2023	0.317534	0.426807	0.623268	0.458467	0.011167	646.2158	3.543478	21,845.55	17,205.58	840,985
GA2023	0.26484	0.605401	0.759511	0.399817	0.020894	401.874	3.252302	20,963.9	15,898.63	499,818
DZ2023	0.232201	0.340429	0.634445	0.495823	0.020265	698.0941	3.347971	19,905.78	14,646.42	856,301
YA2023	0.44433	1.379628	0.537294	0.239999	0.035582	264.6207	4.875467	18,793.73	15,797.9	239,944
BZ2023	0.174136	0.360049	0.709273	0.534386	0.019112	345.5546	2.858801	16,966.96	14,169.81	635,478
ZY2023	0.217586	0.355351	0.831032	0.486122	0.011263	359.4847	3.185459	22,326.26	15,701.84	315,232
ABZ2023	0.391618	0.942808	0.730726	0.182115	0.011937	159.2392	3.652916	18,260.52	13,886	686,401
GZZ2023	0.4185	1.028249	0.803294	0.211384	0.002731	127.4596	1.92464	16,362.69	11,349.83	887,000
LSZ2023	0.248278	0.48071	0.741621	0.455248	0.015469	812.3198	3.060438	17,949.8	13,066.11	1,565,648
CD2024	0.398042	0.612805	0.302653	0.145125	0.021417	957.762	3.868002	30,088.6	21,163.71	1,372,916
ZG2024	0.292645	0.325266	0.638168	0.458972	0.019768	417.1546	5.440209	20,573.53	15,900.74	314,018
PZH2024	0.604467	0.96652	0.477976	0.244829	0.027456	176.3121	5.39401	22,977.95	15,911.68	196,556
LZ2024	0.286826	0.438411	0.597969	0.468574	0.016208	470.2071	3.384366	20,015.54	14,845.91	649,168
DY2024	0.316436	0.398951	0.622617	0.534287	0.031796	508.4059	4.719236	21,898.72	16,281.38	382,598
MY2024	0.320361	0.539408	0.628234	0.448838	0.025713	656.2364	5.229981	21,620.81	16,438.27	662,394
GY2024	0.185513	0.572812	0.720467	0.555712	0.01591	386.017	4.270995	17,191.88	12,712.75	608,027
SN2024	0.340244	0.360678	0.697253	0.418919	0.028234	365.9757	3.606322	19,958.73	14,863.29	478,647
NJ2024	0.266122	0.507778	0.719829	0.439235	0.017682	462.9293	5.962911	20,098.79	15,624.98	404,343
LS2024	0.399536	0.802518	0.614671	0.363639	0.021055	458.8082	4.517216	20,366.58	15,200	532,724
NC2024	0.257194	0.347864	0.710306	0.454035	0.019309	835.829	4.62573	18,960	14,694	886,825
MS2024	0.516101	0.682368	0.606805	0.377752	0.031946	411.8193	3.576595	21,688.8	15,613	502,419
YB2024	0.31294	0.441074	0.6221	0.473996	0.010864	657.1534	3.741714	20,452.59	15,431.14	855,043
GA2024	0.260245	0.60488	0.757772	0.410788	0.020083	401.2889	3.417655	19,811.98	14,444.84	517,435
DZ2024	0.232557	0.348398	0.633636	0.51045	0.019872	707.752	3.519454	19,435.1	14,396.82	964,881
YA2024	0.441754	1.376257	0.536069	0.242513	0.035181	270.2676	5.155786	18,500.72	13,293.64	262,914
BZ2024	0.172136	0.35964	0.708918	0.550035	0.018676	350.376	3.002153	17,246.87	14,454.99	666,613
ZY2024	0.216072	0.363574	0.831067	0.507107	0.011084	362.6616	3.403843	20,923.6	14,565.83	339,636
ABZ2024	0.382947	0.876942	0.732441	0.18394	0.011518	172.0071	4.069421	18,140.91	13,530.16	713,236
GZZ2024	0.412654	0.999425	0.804189	0.212265	0.002607	130.8422	2.05196	16,917.81	12,568.04	940,013
LSZ2024	0.246432	0.484105	0.740201	0.460623	0.015272	817.9504	3.286181	17,685.77	13,086.62	1,570,320

.

Based on the data, the entropy weights, MEREC weights, and LOPCOW weights are derived individually, and the hybrid weighs are calculated based on Equation (2). The weighting result is presented in

Table 3. Indicator weights by hybrid methods.

Methods	C1	C2	C3	C4	C5	C6	C7	C8	C9	C10
Entropy	0.130	0.250	0.036	0.079	0.054	0.114	0.063	0.075	0.049	0.149
MEREC	0.075	0.068	0.094	0.119	0.095	0.169	0.115	0.047	0.053	0.165
LOPCOW	0.075	0.045	0.158	0.113	0.124	0.081	0.115	0.101	0.127	0.062
Hybrid	0.098	0.100	0.088	0.112	0.094	0.127	0.103	0.077	0.075	0.125

Remark: The weights presented here are rounded to three decimal places.

.

By applying the above weights to CoCoSo model, the values of regional agricultural resilience levels can be derived by year. Figure 2 presents the results, and several insights can be identified. First, the average resilience level among all cities each year shows an increasing trend. This increasing trend of average resilience level is consistent with the previous studies of national agricultural resilience [38], as well as studies focusing on the resilience in other provinces [39]. More specifically, in Sichuan province, this increasing trend seems the most significant from 2020 to 2021 and less significant in the more recent years. Also, the relative ranks of cities in terms of their agricultural resilience levels keep dynamically changing. For example, from 2020 to 2021, CD ranked first but fell out of the top three after 2021. Another city, YB, did not show in the top three group until 2022, and remains in the group afterwards. What should be noticed is LSZ city that has always ranked in the top three since 2020 and has become the first place since 2022. This may suggest that LSZ is the city that possibly has the most sustained agricultural resilience. Finally, the city with the least resilience levels has been BZ since 2021, and SN city has ranked the second to the last since 2021. The information indicates that both cities are in low agricultural resilience and need more attention from the local government.

The above results indicate the changes of cities’ agricultural resilience by year. To comprehensively consider the resilience levels across all years, the time weights are considered to consolidate the levels. By solving the Formula (11) with $\lambda =0.2$, the weights (i.e., ${\phi }_t$) that can be found for the years from 2020 to 2024 are approximately {0.029, 0.060, 0.124, 0.256, 0.531}. Applying the weights to Equation (10), the comprehensive evaluation considering the time effects (i.e., $R_i$) was derived. Figure 3 visualizes the results. When considering all periods together, the city with the highest agricultural resilience level is LSZ, followed by MY city. The CD city, reaching the first places in 2020–2021, falls out of the top three place under the dynamic evaluation. This essentially confirms the observation in the above that LSZ is the city that has the most sustained and strongest agricultural resilience. Overall, the results suggest that development of agricultural resilience level may not be fully balanced among cities, consistent with the previous agricultural resilience literature focusing on other provinces [39]. Such regional diversity in agricultural resilience was also identified by studies focusing on places outside China like OECD countries [40]. Therefore, the finding here essentially advocates that diverse policies may need to be considered for improving agricultural resilience in different regions.

4.2. Sensitivity Analysis for Agricultural Resilience Evaluation

Having obtained the results from the dynamic CoCoSo, it is necessary to check the result robustness of dynamic CoCoSo results, and the following two sensitivity tests were conducted. First, consistent with the previous literature [41], the $\tau$ for CoCoSo calculation alters from 0.1 to 0.9 to check the robustness of the results. In practice, the change in $\tau$ reflects different decision makers’ preference of using a weighted sum model (i.e., ${\Phi }_{ti}$) or using a weighted product model (i.e., ${\Theta }_{ti}$) [22]. If τ = 0.9, it represents the decision maker prefers weighted sum model much more than weighted product model, with $\tau$ = 0.1 means the opposite. The special case lies at $\tau$ = 0.5 [35], representing both models are treated equally. As the change in $\tau$ might cause the change in the data scale and make the absolute CoCoSo values not compatible, the rank values under each value of $\tau$ are used in sensitivity analysis to expectedly lead to a fairer comparison [41]. Figure 4 presents the ranks of each city based on their agricultural resilience levels based on the comprehensive CoCoSo considering time effects (i.e., $R_i$). Although Figure 4 indicates some cities like CD, MS, and GA can have varied ranks under different $\tau$, the maximal rank difference is only two. The figure shows that the original results are generally robust, as the top three cities with highest resilience levels are always LSZ, MY, and YB, and the bottom three cities with lowest resilience levels are always YA, SN, and BZ. This confirms the results in Figure 3.

Apart from the parameter $\tau$, it is also necessary to consider the different relative importance of recent data. Therefore, the $\lambda$ for dynamic CoCoSo calculation alters as 0.1, 0.3, 0.5, 0.7, 0.9 to check the sensitivity of the results. The higher the $\lambda$ is, the higher importance is attached to the long-term data. In practice, if $\lambda$ = 0.1, it means the decision-makers or policy makers treat the recent data as much more important than the old data for decision-making (e.g., agricultural business strategy planning) or policy making (agricultural policy design). On the other hand, $\lambda$ = 0.9 means the opposite, representing that decision-makers or policy makers consider old data more strongly when designing and implementing the agricultural business strategies or policies. The special case lies at $\lambda$ = 0.5, and it represents all periods being treated equally. Consistent with the sensitivity analysis above, the city ranks under each value of $\lambda$ are used for a fairer comparison. Figure 5 visualizes the results of sensitivity analysis of $\lambda$. It can be seen that the results present diverse ranks under different $\lambda$ values. First, for cities ranking in the first three places, the LSZ is always in the party. MY and YB can rank in the top three together if $\lambda$ is relatively low (such as $\lambda$ = 0.1, 0.2 or 0.3), while CD and LS can enter the top three group if $\lambda$ is sufficiently high (such as $\lambda$ = 0.9). In addition, the cities located in the bottom three places also change depending on $\lambda$. Apart from BZ which almost always ranks in last place, the SN can be categorized in the last three if $\lambda$ is relatively low (such as $\lambda$ = 0.1, 0.2 or 0.3). In contrast, YA city can also fall into the last three group except $\lambda$ = 0.3. Some other significant cases whose ranks are largely influenced by $\lambda$ can be ZG, SN, and GZZ cities. Specifically, depending on $\lambda$, ZG’s rank can vary from 13 to 18 while SN’s rank can vary from 16 to 20. For GZZ, it can be the least one with high $\lambda$ (such as $\lambda$ = 0.9 or 0.7) but obtains higher rank if $\lambda$ gets lower. This essentially indicates that the different importance attached to the recent data can significantly vary the comprehensive evaluation of cities’ agricultural resilience levels, informing decision-makers carefully assigning the value of $\lambda$.

4.3. Results Summary

To better summarize the results of Section 4.1 and Section 4.2, this section reveals key findings:

• The average annual agricultural resilience levels of Sichuan province keep increasing, although such increasing trend has become less significant in more recent years.
• The agricultural resilience levels among cities are diverse, and the resilience of some cities have experienced significant change across years.
• When considering temporal effects integrating five years, LSZ is the city with the highest resilience level, and BZ is the city with the lowest resilience (when τ = 0.5 and λ = 0.2).
• The CoCoSo parameter (i.e., τ) has little effect on agricultural resilience measurements, while the time weight parameter (i.e., λ) can pose result variations when evaluating resilience levels.

5. Discussion

Based on the above results, several insights can be interpreted for supporting decision making for policy makers and local agricultural companies.

First, the results reflected from Figure 2 depict the longitudinal changes of agricultural resilience levels. Specifically, the annual average resilience level of Sichuan province keeps increasing in the five years, although this trend turns smaller in more recent years. This essentially suggests that the policies and actions implemented to enhance agricultural resilience in the five years in Sichuan worked generally well. However, it should also be noticed that this is possibly due to twofold reasons that make increasing effects become less significant. On the one hand, it could be the case that the currently implemented policies or actions have already exhausted the improvement potential of agricultural resilience in Sichuan province, leaving little space for future increase. If this is the case, it probably calls for new policies to implement and actions to take to obtain more significant increase for provincial agricultural resilience. For example, new risk prevention initiatives can be promoted in province wise to activate new directions for resilience increase, such as organic fertilizer applications and financial incentives for farmers [19]. Also, targeted irrigation improvements can be conducted, and digital agriculture technologies can be invested in, such as agricultural big data analytics, digital twins, and sensors.

On the other hand, it is also possible that there are new norms in the provincial agriculture industry in recent years, diminishing the effects of policies implemented for resilience enhancement. For example, the extreme weather becomes more frequent, which undermines the effectiveness of the agricultural system and brings higher risks and disruptions to agricultural production. Under such a case, additional actions may be needed to adapt to the new norm, so that a sustained increase can be maintained. For example, climate adaptation programs can be implemented like soil and water conservation [19], and the government public budget expenditures can be increased to adapt to new norms in the agriculture industry.

Apart from the provincial level, the city level results also give insights for agricultural resilience development. Results indicate that some cities always have relatively high resilience in their agricultural industry (e.g., LSZ), while some always have low resilience (e.g., BZ). Although such a rank can be related to the city’s own natural resources as well as the capacity and development levels of agriculture industry, it could also be possible that the policies and actions taken by the cities with higher resilience could be more effective in resilience enhancement. Therefore, it is worth comparing the policies implemented in cities with high and low resilience levels and promoting the suitable policies/actions to the agricultural industry of cities with low resilience levels.

Finally, the results also indicate the importance of decision-making processes in agricultural resilience evaluation, especially the decision-makers’ attitudes towards the more recent data. The results in Figure 5 indicate that the relative importance attached to the more recent data can pose influence of measuring levels of agricultural resilience among different cities. This directly leads to the evaluation of the performance and effectiveness of policies and actions taken by certain cities. Therefore, the processes in evaluation decision-making should be carefully designed and conducted. Suggested by the previous literature [36,37], to improve the efficiency and accuracy in this perspective, multiple decision-makers with relevant background in agriculture should form groups to design the evaluation processes together after thorough discussion.

6. Conclusions

This study evaluated the regional agricultural resilience levels. A novel dynamic CoCoSo model with hybrid weighting methods was proposed for evaluation purposes. Using 21 cities of Sichuan province in China as research objects, the agricultural resilience levels of the cities were comprehensively evaluated, and the dynamics of resilience across five years were examined. The results indicated several insights. First, the average agricultural resilience of Sichuan province has kept increasing since 2020, but the increase trend has been less significant in the more recent years. Also, among 21 cities, LSZ city seems to be the one with the strongest and most sustained agricultural resilience, while BZ city is the one with the lowest agricultural resilience. Finally, the evaluated agricultural resilience levels for some cities can change significantly under different relative importance attached to the more recent data, calling for the decision-makers to carefully design their evaluation protocols.

We believe this study has both theoretical and practical implications. Academically speaking, this study can contribute to the existing agricultural management research. To the best of our knowledge, this is the first paper applying dynamic CoCoSo with hybrid weights to agricultural resilience evaluation in Sichuan. In this study, the dynamic information of resilience captured as well as the comprehensive evaluation enabled by the method can therefore confirm its potential for agricultural management, providing new analytical tools for future agricultural resilience research. Also, the method offers flexible decision-making mechanisms to adapt different importance attached to the more recent data for agricultural resilience measurement, building a basis for future exploration on risk and disruption management in regional agriculture and relevant directions.

Practically, this study can inform policy makers and local agricultural companies for better decision-making. For example, the index system developed in this study, together with the method proposed, can become a useful tool for policy makers to evaluate the effectiveness of agricultural policies, supporting the opportunities for policy design or update. Also, the ways this study collects and analyzes data can inform practitioners (e.g., software engineers, information system managers, etc.) in agricultural companies to update their management and decision support systems. By utilizing these methods, agricultural practitioners can expectedly capture more improvement opportunities.

However, we also acknowledge there are limitations of this study, thus offering chances for future research. First, as this study only considers Sichuan province, the evaluation indicators, as well as the evaluation results, can be regionally specific. Due to data availability issues, some important indicators such as soil quality, water stress, land degradation, and biodiversity were not able to be included. In other words, the generalization of the index systems and results should be further validated in other regions [21,25]. Therefore, one of the interesting research directions in the future can be utilizing our method in other regions using various data to test its effectiveness and triangulate the findings. To achieve this, data adjustments and contextual considerations should be covered. For example, new indicators, such as soil quality, water stress, land degradation, and biodiversity can be added for examining resilience in other regions using our model once such data are available. Also, due to the properties of the data, the method and our analysis framework are limited to objective and quantitative data. However, as it is common that in agricultural MCDM research, some data are often fuzzy and subjective, and personal judgments can be applied in the decision-making process [42,43]. This leads future research to consider adding qualitative data into our framework. For example, future research can construct a fuzzy dynamic CoCoSo with hybrid weights, or even mixed data dynamic CoCoSo (e.g., integrating fuzzy and real numbers) to capture more information within the resilience evaluation. Finally, it could be interesting to extend the developed method to other industries, such as automotive or logistics, to check its effectiveness in resilience evaluation under different contexts.