Evolving patterns of agricultural production space in China: A network-based approach

Yang, Shuhui; Li, Zhongkai; Zhou, Jianlin; Gao, Yancheng; Cui, Xuefeng Article — Published Version Evolving patterns of agricultural production space in China: A network-based approach Geography and Sustainability Provided in Cooperation with: Leibniz Institute of Agricultural Development in Transition Economies (IAMO), Halle (Saale) Suggested Citation: Yang, Shuhui; Li, Zhongkai; Zhou, Jianlin; Gao, Yancheng; Cui, Xuefeng (2024) : Evolving patterns of agricultural production space in China: A network-based approach, Geography and Sustainability, ISSN 2666-6839, Elsevier, Amsterdam, Vol. 5, Iss. 1, pp. 121-134, https://doi.org/10.1016/j.geosus.2023.11.007 , https://www.sciencedirect.com/science/article/pii/S2666683923000755 This Version is available at: https://hdl.handle.net/10419/281442 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/ Geography and Sustainability 5 (2024) 121–134 Contents lists available at ScienceDirect Geography and Sustainability journal homepage: www.elsevier.com/locate/geosus Research Article Evolving patterns of agricultural production space in China: A network-based approach Shuhui Yang a , d , Zhongkai Li b , c , Jianlin Zhou a , Yancheng Gao a , Xuefeng Cui a , ∗ a School of Systems Science, Beijing Normal University, Beijing 100875, China b State Key Laboratory of Earth Surface Processes and Resource Ecology, Faculty of Geographical Science, Beijing Normal University, Beijing 100875, China c School of Natural Resources, Faculty of Geographical Science, Beijing Normal University, Beijing 100875, China d Leibniz Institute for Agricultural Development in Transition Economies (IAMO), Halle (Saale) 06120, Germany h i g h l i g h t s g r a p h i c a l a b s t r a c t •Agricultural production space is characterized by bipartite network. •Product and province spaces exhibit well core-periphery and community structures. •The community changes show relatively stable production patterns. •The study offers a holistic view to understand agricultural production system. a r t i c l e i n f o Article history: Received 10 September 2023 Received in revised form 9 November 2023 Accepted 27 November 2023 Available online 18 December 2023 Keywords: Agricultural system Complex network Agricultural production space Proximity matrix Production capability a b s t r a c t The agricultural production space, as where and how much each agricultural product grows, plays a vital role in meeting the increasing and diverse food demands. Previous studies on agricultural production patterns have predominantly centered on individual or specific crop types, using methods such as remote sensing or statistical metrological analysis. In this study, we characterize the agricultural production space (APS) by bipartite network connecting agricultural products and provinces, to reveal the relatedness between diverse agricultural products and the spatiotemporal characteristic of provincial production capabilities in China. The results show that core products are cereal, pork, melon, and pome fruit; meanwhile the milk, grape, and fiber crop show an upward trend in centrality, which is in line with diet structure changes in China over the past decades. The little changes in community components and structures of agricultural products and provinces reveal that agricultural production patterns in China are relatively stable. Additionally, identified provincial communities closely resemble China’s agricultural natural zones. Furthermore, the observed growth in production capabilities in North and Northeast China implies their potential focus areas for future agricultural production. Despite the superior production capabilities of southern provinces, recent years have witnessed a notable decline, warranting special attentions. The findings provide a comprehensive perspective for understanding the complex relationship of agricultural products’ relatedness, production capabilities and production patterns, which serve as a reference for the agricultural spatial optimization and agricultural sustainable development. 1. Introduction As part of the United Nations Sustainable Development Goals (SDGs), the Chinese Government has given priority to ending hunger and ensuring sufficient food supply, achieving food system sustainability ( Agnolucci et al., 2020 ). Driven by population growth and economic ∗ Corresponding author at: No. 19 Xinjiekouwai St, Haidian District, Beijing, China. E-mail address: [email protected] (X. Cui) . development, dietary structures have changed with people consuming numerous and diverse food than ever before ( Sun et al., 2019 ). Simultaneously, there is limited arable cropland in China, with over 40% of it classified as “moderately ”or “severely ” degraded, due to issue such as soil erosion, water pollution, or water scarcity ( Sommer et al., 2023 ). The combination of these factors puts certain pressure on agrihttps://doi.org/10.1016/j.geosus.2023.11.007 2666-6839/© 2023 The Authors. Published by Elsevier B.V. and Beijing Normal University Press (Group) Co., LTD. on behalf of Beijing Normal University. This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/4.0/ ) S. Yang, Z. Li, J. Zhou et al. Geography and Sustainability 5 (2024) 121–134 cultural production in China ( Ye et al., 2022 ). The agricultural production space, as where and how much each agricultural product grows, plays an important role in satisfying the increasing and diversified food demands. China’s agricultural production capability stands as a critical determinant in maintaining nation’s future food supply. It is affected by multiple factors, including land and water resources ( Duan et al., 2021 ), climate ( Arora, 2019 ), households’ land-use behaviors ( Liu et al., 2020 ), fertilizer and pesticide inputs ( Seghezzo et al., 2020 ; Yu et al., 2021 ), management practices ( Cui et al., 2018 ), as well as national policies and strategies ( Lu et al., 2022 ). These factors are prior notions of what are the important factors of agricultural production. However, it is not easy to embrace all factors and to quantify all factors or components of agricultural production capability ( van Ittersum et al., 2008 ). Instead, according to the agnostic approach, an outcomes-based approach offers an alternative way to measure production capability. This approach has been used in the international trade field, where a country’s export goods could reveal its domestic production capabilities and specialization patterns ( Sciarra et al., 2020 ). In agricultural production, the output of agricultural products can serve as a reflection of the production capability of a particular region. The fact that the production of identical agricultural products across different regions may suggest a shared production capability to cultivate these products. Agricultural products are seen as a consequence of a region’s or province’s endowments or capabilities, possibly covering land, water resource, labor, technology, knowledge and capital ( Antle et al., 2004 ; Gumbau Albert, 2017 ). Analyzing agricultural production system has been proved to be a complex task due to its interrelated and heterogeneous characteristic ( Jones et al., 2017 ). Previous studies such as crop distribution and its spatial optimization ( Hu et al., 2022 ; Q. Liu et al., 2022 ), crop yields ( Wang et al., 2022 ), crop production conditions ( Pickson et al., 2022 ) and so on have been carried out. There are also numerous researches about the structure and space of agricultural production in China, such as spatiotemporal characteristics of agricultural production efficiency ( Guo et al., 2020 ), spatial feasibility and cost-effectiveness of agricultural land consolidation ( Duan et al., 2021 ), productivity and environmental costs of different-scale farms ( Qi et al., 2018 ), the pattern of product diversification ( Emran and Shilpi, 2012 ). Statistical metrology ( Baldoni and Esposti, 2020 ) and remote sensing with GIS ( Medennikov et al., 2020 ) are commonly used in these fields. However, these studies mainly focus on single agricultural product or some specific products ( Geng et al., 2017 ). Interrelation among diverse agricultural products in the whole nation is rarely explored. Furthermore, currently spatial mismatch continues to exist between major producing area, agricultural production structure and farmland resource in China ( Li et al., 2017 ). Analyzing single agricultural product alone poses challenges in effectively integrating various agricultural activities and production capabilities, thereby hindering the optimization of agricultural layout ( Lu et al., 2013 ). Recently, complex network (CN) has been introduced into agricultural filed. For example, Branco et al. (2021) used network approach to optimize the spatial distribution of new soybean and corn in Brazil. Ma et al. (2019) developed a network with 502 farms in the UK to compare resilience of genetically modified herbicide-tolerant management and conventional weed management. Z. Chen et al. (2022) applied a spatial network framework to analyze the cross-regional collaborative mechanisms for agricultural green development in China. Moreover, complex network approach also has been used to measure countries’ specialization patterns and analyze how these patterns affect food supply and food security in global food system ( Campi et al., 2020 , 2021 ). Combing complex network with global agricultural products is used to explore why and how a country produces goods, proposing that a country’s ability to produce new goods is constrained by the combination of its existing capabilities and the acquisition of novel capabilities ( Hidalgo and Hausmann, 2009 ). The global agricultural product space can also reveal the relatedness between diverse products and find different structures in different network locations ( Hidalgo et al., 2007 ). Complex network is extensively applied in varieties of real-world systems, such as agriculture system ( Albert and Barabási, 2002 ; Mariani et al., 2019 ), which can help to simulate interconnections among diverse agricultural elements and capture dynamic characteristics of agricultural system, providing a new perspective and an effective tool to study complex systems ( Lambiotte et al., 2019 ). In this study, we use complex network approach to characterize China’s agricultural production space, and identify core-periphery structure and community structure of agricultural product space and province space, revealing the relatedness of multiple agricultural products and the spatial characteristic of provincial production capabilities. This work provides a systematic and holistic perspective to investigate agricultural production space and structure, offers us a better understanding of the complex relationship of agricultural production space and production capability, and the relatedness of diverse agricultural products. It is significant reference to upgrade and diversify agricultural product bundles in China to achieve the spatial optimization of agricultural production. We will address the following questions in this study: (a) What characterizes the “province-product ” agricultural production space in China from 1995 to 2019, and what are the structures and patterns of agricultural product space and province space? (b) What is the relatedness of multiple agricultural products and the spatial characteristic of provincial production capabilities under current production structure? 2. Materials and methods 2.1. Theoretical background of complex network Complex network serves as a type of graph that can abstract activities or event into elements and their relations ( Lacasa et al., 2008 ). Within the network framework, elements are assigned as nodes, meanwhile connection or relationship between pairs of elements are regarded as links. In this study, we use complex network connecting provinces and agricultural products to build a series of agricultural production space in China. The fact that different provinces producing the same products might indicate that they share possession of the required essential capabilities for producing these goods ( Fig. 1 (a)). When two agricultural products share similarities in terms of essential requirements such as climate, soil, water, technology, and other pertinent factors, they tend to be cultivated together in the same regions. Conversely, dissimilar agricultural products exhibit a lower likelihood of co-production ( Fig. 1 (b)). Measuring the similarity of capabilities required for products and the similarity of endowment of provinces are called “proximity ”in the complex network approach. Proximity matrix connects pairs of agricultural products yielding the “product space ”( Hidalgo et al., 2007 ). Equally, proximity matrix can also link pair of locations, giving rise to “province space ”( Bahar et al., 2014 ). The proximity of product space and province space is built as proximity network to reveal distinct agricultural network characteristics in China ( Börner et al., 2012 ; Balland and Rigby, 2017 ). Despite their specific structures, proximity network serves as a foundation for measuring the relatedness among agricultural products ( Hidalgo et al., 2007 ). In this study, agricultural product network is projected into agricultural province space and product space by proximity matrix ( Fig. 1 (c)). 2.2. Data Data on agricultural production from 1995 to 2019 for all 31 provinces are sourced from the National Bureau of Statistics in China 122 S. Yang, Z. Li, J. Zhou et al. Geography and Sustainability 5 (2024) 121–134 Fig. 1. Theoretical map of the complex network in agricultural production. Table 1 Categories of agricultural products at Levels 1, 2 and 3. Level 1 Level 2 Level 3 Cereal Wheat Maize Rice Sorghum Barley Millet Vegetable and melon Melon Watermelon Muskmelon Fruit and nut Tropical and subtropical fruit Banana Pineapple Citrus Grape Berry Strawberry Pome fruit Apple Pear Jujube Persimmon Oil crop Soybean Peanut Temporary oilseed crop Flaxseed Rapeseed Sesame Sunflower seed Root crop Potato Irritant and spice crop Irritant crop Tea Bean Mung Adzuki Sugar crop Sugar beet Sugarcane Livestock product Beef Mutton Pork Milk Poultry egg Honey Aquatic product Seawater product Freshwater product Others Fiber crop Cotton Sisal Jute Flax Ramie Tobacco Sources : Bureau of Statistics in China ( https://data.stats.gov.cn/easyquery.htm?cn = E0103 ); the World Agriculture Census Program in 2020 of the FAO ( https://www.fao.org/world-census-agriculture/wcarounds/wca2020/en/ ). (NBSC). Product classification is conducted following the Revised Indicative Crop classification from the World Agriculture Census Program in 2020. This classification categorized all agricultural products into three main types, as outlined in Table 1 . The Level 1, consists of 11 groups, while the more detailed Level 2 includes 31 specific products. For the purposes of this study, Level 2 products are used due to their increased specificity. In case where NBSC does not directly provide data on certain products, we calculate them from data in Level 3. As shown in Table 1 , the products are melon, tropical and subtropical fruit, berry, pome fruit, temporary oilseed crop, irritant crop, and fiber crop. All data of the selected agricultural products are completely recorded in 31 provinces from 1995 to 2019. During this period, the recorded data of each product remains stable, with rare missing values. The proportion of each agricultural product is also demonstrated, which is in line with China’s food production situation (Fig. S1, Fig. S2). 2.3. Methods 2.3.1. Agricultural production space To explore how province grows agricultural production, we adopt the complex network to build agricultural production space, as introduced by Campi et al. (2020) . This approach enables the representation of the relationship between provinces and agricultural products. The specific type of complex network in this study is bipartite network which is described as a bipartite matrix M . Within this framework, bipartite network, designated as a triplet G = ( C, P, X ), consists of three components: C , representing the set of top nodes corresponding to provinces; P , signifying the set of bottom nodes corresponding to agricultural products; X , denoting the set of links. The specific element 𝑋𝑡 𝑖𝑘 indicates that the province 𝑖 produces product k in year t . We constructed the APS with agricultural products at Level 2 from 1995 to 2019 over 31 provinces. 123 S. Yang, Z. Li, J. Zhou et al. Geography and Sustainability 5 (2024) 121–134 2.3.2. Link weight The link between two nodes within the APS indicates that the specific province produces the specific agricultural products. Moreover, the weight assigned to each link serves as an indicator of the province’s capabilities or abilities concerning that particular product. With the theory of an outcomes-based method, we use the data of final agricultural products to calculate capabilities or performance of this province in producing a particular agricultural product. It is identified as follows: 𝑀𝑡 𝑖,𝑘 =𝑥( 𝑖, 𝑘 ) ∑𝑘 𝑥( 𝑖, 𝑘 ) ×∑𝑖 𝑥( 𝑖, 𝑘 ) ∑𝑖,𝑘 𝑥( 𝑖, 𝑘 ) (1) where, x is production; 𝑥 (𝑖, 𝑘 ) is the production of product 𝑘 in province 𝑖 in year 𝑡 . The first fraction is the ratio of the production of product k in that of all products in province 𝑖 , and the second fraction is the proportion of a given product k among all products in China. 2.3.3. Proximity matrix In the network-based approach, relatedness captures the interactions among elements that are similar, which quantifies these relatedness or similarity call ‘proximity’ ( Hidalgo et al., 2007 ). Proximity matrix can evaluate a province’s similar capabilities in producing an agricultural product related with its other products. In agricultural production space, if two agricultural products are related because they require similar climate, soil, water, technology, or some combination thereof, they are likely to be produced together in the same provinces. Metrics of proximity connect pairs of agricultural products or pairs of provinces. The proximity between the pairs of products k and  𝑘 is the minimum probabilities of a province producing a specific product given that it produces another ( Hidalgo, 2021 ). To analyze the evolution of productive structure, the APS is projected into “product space ”and “province space ”by defining their similarity with proximity matrix. Within the agricultural product space, each pair of nodes is linked with their degree of similarity. According to the Hidalgo (2021) , the similarity 𝑁𝑘 𝑘 is defined as follows: 𝑀𝑖 =∑ 𝑘 𝑀𝑖𝑘 (2) 𝑀𝑘 =∑ 𝑖 𝑀𝑖𝑘 (3) 𝑁𝑘 𝑘 =∑ 𝑖 𝑀𝑖𝑘 𝑀𝑖 𝑘 𝑀𝑖 𝑀𝑘 (4) where, 𝑀𝑖 and 𝑀𝑘 represent the number of products produced by that province, and the number of provinces producing a given products, respectively. 𝑁𝑘 𝑘 is the similarity of the pairs of products k and  𝑘 . Following the same strategy, an agricultural province space defined which nodes are provinces, and links between provinces 𝑖 and  𝑖 are weighted by 𝑁𝑖 𝑖 : 𝑁𝑖 𝑖 =∑ 𝑘 𝑀𝑖𝑘 𝑀 𝑖 𝑘 𝑀𝑖 𝑀𝑘 (5) where 𝑁𝑖 𝑖 represent the similarity of the pairs of provinces 𝑖 and  𝑖 . 2.3.4. Core-periphery structure detection The detection of core and periphery within the network can reveal localized network structures ( Malliaros et al., 2020 ; Gallagher et al., 2021 ). Various methods and algorithms have been employed to extract network’s core and periphery, such as statistical inference ( Kojaku and Masuda, 2017 ; Peixoto, 2019 ), spectral decomposition ( Cucuringu et al., 2016 ; Tudisco and Higham, 2019 ), diffusion mapping ( Rossa et al., 2013 ) to motif counting ( Ma et al., 2018 ), geodesic tracing ( Cucuringu et al., 2016 ), and rich clubs ( Ma and Mondragón, 2015 ). There is also a new approach to detect the core-periphery structure of weighted and undirected network. Unlike other methods determining whether a node belongs to core with “yes ”or “no ”, this method uses a scoring system ranging between 0 and 1 to measure the degree of core. A higher score indicates closer proximity to the core. Notably, a node is more likely in the network’s core position which not only densely connects among other core nodes and has high strength, but also has links with peripheral nodes ( Rombach et al., 2017 ). Moreover, coreperiphery structure can also be nested with networks’ community structure ( Leskovec et al., 2009 ; Yang and Leskovec, 2012 ). 2.3.5. Community detection Community detection may be a helpful tool for analyzing the differences of products group and provinces group. The traditional strategy of community detection focuses on optimizing modularity ( Esfahlani et al., 2021 ), with example including Guimera’s modularity ( Guimera et al., 2007 ) and Barber’s modularity ( Barber, 2007 ). QuanBiMo is the first algorithm to maximize weighted modularity in bipartite network. However, its sensitivity to specific input variables that might not be accessible can lead to deviations from expected outcomes ( Dormann and Strauss, 2014 ). Furthermore, DIRTLPAwb + , an extension of QuanBiMo, performs well on small networks, which can search the optimal modularity in whole possible space ( Beckett, 2016 ). In this study, the DIRTLPAwb + algorithm was used to maximize modularity scores in the networks. DIRTLPAwb + has more meaningful input parameters and exhibits enhanced performance. Its implementation is also more stable than that of QuanBiMo on the test ( Beckett, 2016 ). 3. Results 3.1. Overview of the “province-product ” agricultural production space The “province-product ” agricultural production space has been built to exhibit how provinces grow agricultural products from 1995 to 2019. Fig. 2 shows the case of 2019 and it clearly demonstrate that the network’s central nodes include maize, wheat, rice, pork, sugarcane, pome fruit, and melon, indicating the critical role of these products in the food production system. Additionally, this network effectively depicts the geographical distribution of agricultural products across China. Maize predominantly links with provinces in the norther China, such as Hebei, Henan, Heilongjiang, Liaoning, Jilin, Shanxi, and the Inner Mongolia Autonomous Region. Meanwhile, rice exhibits strong connections with southern provinces like Anhui, Jiangsu, Jiangxi, Hunan, Hubei, Sichuan, and Guangdong. Similarly, wheat primarily connects with Henan, Shandong, Hebei, Shaanxi, and Anhui. These findings are well consistent with previous researches conducted by Ye et al. (2020) , Bai et al. (2021) , L. Zhang et al. (2022) . Sugarcane, a typical tropical and subtropical product, is logically planted in Hainan, Guangdong, Guangxi, and Yunnan. Blue provincial nodes represent traditional agricultural production areas that have diverse products and higher output, which include Shandong, Henan, Hebei, Heilongjiang, Shanxi, Jilin, and Liaoning. In contrast, provinces like Qinghai, Hainan and Xizang Autonomous Region occupy a more peripheral position, in keeping with the actual state of agricultural production ( Guo et al., 2020 ). The darker-colored links connecting maize, rice, wheat, melon, and pork to their provinces signify that these products have considerable importance in their provinces. Products located in the outer ring exhibit sparse and weaker links, likely due to their requirement for specialized production conditions, often connect to specific provinces within a limited geographic area. Our subsequent investigation will delve into the core-periphery structure and community structure within both the agricultural product space and the province space. 3.2. Analysis of the core-periphery structure 3.2.1. Core-periphery structure of agricultural product space Fig. 3 (a) shows the bubble matrix diagram that illustrates the core score ranking of each product from 1995 to 2019. Among all products, maize, wheat and rice consistently maintain the highest position, with maize gradually surpassing wheat and rice, which there is no doubt that 124 S. Yang, Z. Li, J. Zhou et al. Geography and Sustainability 5 (2024) 121–134 Fig. 2. Agricultural production space in 2019. Note: Nodes are two disjoint sets that represent products and provinces. Product nodes are orange, and province nodes are blue. The size of the nodes represents their importance in the network. Links are color coded with their RCA value. Links are always between the node of the product set and a node of the province set, and there is not a link between two nodes in the same set. The darker color of links indicates the stronger capabilities of this province to produce a specific product. these three products dominate agricultural production in China. It is also worth noting that melon (watermelon and muskmelon), pork, and pome fruit (apple, pear, jujube, and persimmon) are in the fourth, fifth and sixth places, respectively. The core score of pork remained relatively stable from 1995 to 2018 but witnessed a sharp decline in 2019 due to the impact of the African swine fever and “Enhanced Pig Cycle ” (EPC). This event led to a significant reduction in domestic pork production, plummeting from 54.075 million tons to 42.553 million tons ( Song et al., 2022 ). Melon, as one of the most consumed food in China, experienced an expansion in its planting area from 1,013.52 ha in 1995 to 1,894.05 ha, representing approximating 40% of the global yield from China ( Luo et al., 2018 ). Pome fruits, including apple, pear, jujube, persimmon, serve as necessary food in our daily diet. Their planting areas in China are concentrated in the northern provinces, maintaining stable growth in their output ( Wang et al., 2020 ; Qi et al., 2023 ). For example, cultivation area of apple has expanded from semi humid (irrigated) to semi-arid (rain-fed) area ( Zhang et al., 2023 ). Fiber crops (flaxseed, rapeseed, sesame, and sunflower) and grape have experienced a consistent increase in their ranking since 1999, while milk began to rise in 2004. On the contrary, oil crops (temporary oilseed crop, peanut, and soybean) displayed a declining trend until 2019. Other products with invisible color such as barley and mutton are usually away from the core area in the network except some extreme years. For example, citrus achieved relatively higher rankings in 2015, 2018, and 2019 compared to other years, mainly owing to a rapid surge in citrus output during these years (from 245.25 million tons to 274.01 million tons). Fig. 3 (b) shows the box diagram presenting the core score for each product from 1995 to 2019, offering a clearer understanding of the overall differences among agricultural products. The mean core scores for maize, wheat, and rice are 0.9443, 0.9187, and 0.9015, respectively. Following these, melon (0.8428), pork (0.7944), and pome fruit (0.67769) hold relatively high average core scores. These products, as highlighted by Wang et al. (2019) , are fundamental foods in the Chinese diet and are cultivated in relatively high-producing regions, which dominate Chinese food system. In the second order, with slightly lower core score compared to the first order, are agricultural products such as potato, temporary oilseed crop, milk, peanut, fiber crop, grape, and soybean. These products are still substantially important in our daily diet or living ( Liu et al., 2021 ). Products limited to specific regions with specific conditions are positioned on the network’s periphery, with relatively lower scores. For instance, sugar beet, millet, adzuki, mung, and sorghum are mostly cultivated in northern China ( Lin et al., 2023 ). Citrus, irritant crop, freshwater product, seawater product, tropical and subtropical fruit are mainly in south China ( Duan et al., 2020 ; Yan et al., 2020 ). Tobacco has widespread distribution in southern hilly area ( Liu et al., 2018 ). Beef, mutton, and honey are concentrated to produce in northwest China ( H. Zhang et al., 2022 ). 3.2.2. Core-periphery structure of agricultural province space As depicted in Fig. 4 (a), provinces such as Shanxi, Jilin, Hebei, Inner Mongolia Autonomous Region, Liaoning, Heilongjiang, Henan, and Shandong occupy relatively higher rankings. These provinces exhibit high strength in main agricultural products, particularly in maize, wheat, rice and pome fruit. Additionally, these provinces are wellconnected to each other or even to some peripheral provinces. For example, Shanxi not only share connection with Henan, Hebei, Shandong, Inner Mongolia Autonomous Region, but also demonstrate strong links with Gansu, Shaanxi due to their mutual cultivation of specific agricultural products. The main agricultural products of high-ranking provinces have been outlined in Table S1. Liaoning, Tianjin and Heilongjiang have demonstrated a gradual upward trend in their rankings. Conversely, Shaanxi, Ningxia, and Gansu show a decreasing trend. Anhui and Xinjiang Uygur Autonomous Region exhibit noticeable fluctuations. While provinces like Xizang Autonomous Region and Qinghai, situated on the Qingzang Plateau with limited cultivated land and lower temperature, maintain lower rankings. Hainan, Guangdong, and 125 S. Yang, Z. Li, J. Zhou et al. Geography and Sustainability 5 (2024) 121–134 Fig. 3. Core-periphery structure in agricultural product space. (a) The bubble matrix diagram of core score rankings of each product; (b) Box diagram of core scores from 1995 to 2019 for each product. Note: Dots with darker colors and wider sizes have higher core scores. The line in the box represents the median value, and the solid dot indicates the mean value, which is linked by a black solid line. The asterisk denotes the outliers. The two ends of the box are the first and third quartiles. 126 S. Yang, Z. Li, J. Zhou et al. Geography and Sustainability 5 (2024) 121–134 Fig. 4. Core-periphery structure in agricultural province space. (a) The bubble matrix diagram of core score rankings of each province; (b) Box diagram of core scores from 1995 to 2019 for each province. Note: Dots with darker colors and wider sizes have higher core scores. The line in the box represents the median value, and the solid dot indicates the mean value, which is linked by a black solid line. The asterisk denotes the outliers. The two ends of the box are the first and third quartiles. 127 S. Yang, Z. Li, J. Zhou et al. Geography and Sustainability 5 (2024) 121–134 Fig. 5. Community presented in agricultural product space from 1995 to 2019. Guangxi which primarily focus on tropical and subtropical products, have fewer connections to products of other provinces. Therefore, they are at the lower rankings of the network. The box diagram in Fig. 4 (b) shows a clearer insight of each province’s core score from 1995 to 2019. Shanxi, Jilin, Hebei, and Inner Mongolia are in the high positions, with average scores over 0.8. Following closely are Liaoning, Heilongjiang, Henan, and Shandong, with scores fluctuating between 0.1 to 1. Conversely, other provinces have scores that are approximately below 0.3, and in some case, nearly 0. Despite these relatively lower scores, these provinces still play vital roles in some specific products. For example, Fujian, Guizhou, and Yunnan are significant for tea production; Qinghai, Xinjiang Uygur Autonomous Region and Xizang Autonomous Region are important for mutton production ( Yan et al., 2021 ). 3.3. Changes in the relatedness among agricultural products The evolving relatedness among different agricultural products are investigated by community detection and the outcomes are presented within the agricultural product space. These communities gather products that share similar production conditions or capabilities. As shown in Fig. 5 , the node size corresponds to the centrality of each product, determined by its core score. The modularity ranges from 0.4749 to 0.5957, which can support the possibility to present network substructures. A total of three to four communities are detected from 1995 to 2019. From 1995 to 2002 and from 2003 to 2011, the networks are divided into three communities. The purple community mainly includes maize, wheat, melon, pome fruit soybean, temporary oilseed crops and some peripheral products that are mainly suitable in north of China. The orange community is dominated by pork and rice, both of which require more heat and water resources for their production compared to the products in the purple community. The green community includes sugarcane and tropical and subtropical products. The community compositions in 2003 resemble those in 1995, with the exception of soybean, which has moved to the orange community with pork and rice. Since 2012, four communities are detected, and this configuration remains relatively stable until 2019. Notably, maize and soybean are observed within the same community after 2012, while other communities exhibit minimal changes over these years. The Sankey graph presented in Fig. 6 shows the dynamic changes in community components over time, with core products highlighted in bold. The core products within each community remain almost unchanged from 1995 to 2019, suggesting a relatively stable network structure in this period. This finding implies product structure in China’s agricultural production demonstrates a remarkable stability ( Huang and Tian, 2019 ; Guo et al., 2020 ), in line with the actual situation of agricultural production in China, because production conditions such as soil, cropland, climate, cultivation habits are long-term elements and infrequent changes. Additional noteworthy finding is the transition from three communities during 1995 to 2003 to four communities after 2012. More importantly, soybean is separated from the rice and pork groups to the community with maize, milk, pome fruit, grape, and fiber crops until 2019. Furthermore, it is also important to point out that the position of barley frequently changes, influenced by fluctuations in output within its primary production regions. Barley output declines from 0.805 million tons to 0.239 million tons in Jiangsu, 0.123 million tons to 0.002 million tons in Zhejiang, 0.369 million tons to 0.109 million tons in Henan, 0.125 million tons to 0.0254 million tons in Hubei, and 0.546 million tons to 0.133 million tons in Gansu. However, an increase 128