Background & Summary

Recently, economic conflicts such as the US-China trade war have intensified a transition from the hyper-globalization era toward greater regionalization, digitalization, and strategic nationalism. While cross-border flows of data, services, and capital remain robust, trade in goods and physical capital is fragmenting, giving rise to a more multipolar, risk-aware, and locally oriented global economy. Industry, supply chains, and value chains are facing challenges to resilience and restructuring, with trends toward localization, regionalization and diversification1,2. Against this backdrop, a dual structure covering economic flows within and among nations at both the subnational and international levels, such as provinces in China, states in the U.S. and countries in the EU, is becoming increasingly important, and with substantial implications for both the economy and the environment3. The motivation for our work stems from the observed discrepancies between coupled multi-regional input–output (MRIO) tables and separately subnational and international MRIOs4. The coupling MRIO provides higher accuracy, which can reveal key issues and uncover findings that separate tables cannot reflect. As a result, the subnational and international coupling (SIC) database offers critical implications for the study of supply chains, global value chains (GVC), and embodied carbon emissions, facilitating more robust policy analysis and targeted strategies.

MRIO tables at international and subnational levels differ substantially in their sector coverage, regional scope and time span, posing challenges for integration5. As quantitative tools for examining interregional economic structures, MRIOs are widely used in regional economics. International MRIO databases such as the global trade analysis project (GTAP), the organization for economic co-operation and development (OECD), EXIOBASE, the world input-output database (WIOD), The EMERGING database (EMERGING), and the EORA are prominent. GTAP covers 160 countries; EXIOBASE includes 163 sectors; OECD offers long time-series data (1995–2018); WIOD emphasizes value chains and environmental extensions; EMRGING contains 135 sectors across 245 economies; EORA provides high spatial and sectoral resolution (Table 1). MRIO tables play an important role in capturing economic flows6 and their environmental effects7 in the dual structure. The advantage of using international MRIO data lies in its global perspective, which enables macro-level analysis of ecological systems8, climate systems9, technological supply chains10, etc. This approach describes and analyses international interrelationships between the consumption and production sectors11. However, the lack of coupling with subnational MRIO leads to a loss of important details of subnational regions. In contrast to international MRIO tables, subnational MRIO data provide detailed insights into structural economic changes and environmental policy impacts within countries, including China[12](https://www.nature.com/articles/s41597-025-06040-2#ref-CR12 “Yang, Y. et al. Mapping global carbon footprint in China. Nature Communications 11, 2237, https://doi.org/10.1038/s41467-020-15883-9

(2020).“), the EU[13](https://www.nature.com/articles/s41597-025-06040-2#ref-CR13 “Bruckner, B. et al. Ecologically unequal exchanges driven by EU consumption. Nature Sustainability 6, 587–598, https://doi.org/10.1038/s41893-022-01055-8

(2023).“), Japan[14](https://www.nature.com/articles/s41597-025-06040-2#ref-CR14 “Long, Y. et al. Comparison of city-level carbon footprint evaluation by applying single-and multi-regional input-output tables. Journal of Environmental Management 260, 110108, https://doi.org/10.1016/j.jenvman.2020.110108

(2020).“), etc. Some studies have extended their scope to inter-city input-output relationships among a few adjacent cities[15](https://www.nature.com/articles/s41597-025-06040-2#ref-CR15 “Wang, S. & Chen, B. Energy–water nexus of urban agglomeration based on multiregional input–output tables and ecological network analysis: A case study of the Beijing–Tianjin–Hebei region. Applied Energy 178, 773–783, https://doi.org/10.1016/j.apenergy.2016.06.112

(2016).“), and even across entire national cities[16](https://www.nature.com/articles/s41597-025-06040-2#ref-CR16 “Wang, Y. et al. An inter-city input-output database distinguishing firm ownership in the Greater China area during 2002–2017. Scientific Data 12, 723, https://doi.org/10.1038/s41597-025-04996-9

(2025).“). However, subnational MRIO databases typically include only a single column for exports, offering limited information on international trade. This makes it difficult to evaluate the role and contribution of subnational regions or cities in the global economy, as represented by the division of labor in GVC. Some databases provide both international and subnational MRIO, such as the carbon emission accounts and datasets (CEADs). The linkage between subnational and international MRIO tables has been advanced through the development of region-extended Inter-Country Input-Output (ICIO) tables17, in which linear programming is applied to harmonize macroeconomic data from customs and the Broad Economic Categories (BEC)18. However, the advent of frontier technologies, including machine learning, deep learning, large language models (LLMs), etc., has enabled the exploration of advanced methods such as model coupling in this field19,20.

Regarding the measurement of the dual structure, existing research is generally divided into macro-level and micro-level approaches. Macro-level studies often assess the dual structure under the assumption that subnational and national trade structures are identical. This approach, however, would overlook subnational heterogeneity and compromise the accuracy of the data. For example, existing research has disaggregated the Chinese portion of the GTAP database using provincial shares from China’s provincial MRIO tables21,22, or integrated the Asian International Input-Output (IOT) database with China’s MRIOs under the assumption of uniform regional import structures23. Due to the unavailability of subnational import and export data, existing studies commonly assume that subnational trade patterns resemble those at the national level, thus failing to capture the provincial distinction of economic development and participation in the initernatonal division. Micro-level studies typically rely on firm-level trade data to measure either domestic or foreign trade in isolation. For instance, foreign trade studies have used multinational enterprises’ investment data24, value-added firm evidence25, etc., and domestic trade studies have integrated datasets such as capital and tax information26,[27](https://www.nature.com/articles/s41597-025-06040-2#ref-CR27 “Chen, Q. et al. An interprovincial input–output database distinguishing firm ownership in China from 1997 to 2017. Scientific Data 10, 293, https://doi.org/10.1038/s41597-023-02183-2

(2023).“). This approach fails to capture the structural linkages between subnational and international trade within the dual structure.

While artificial intelligence techniques such as machine learning, deep learning and large language models (LLMs) have been widely applied in data-intensive fields[28](https://www.nature.com/articles/s41597-025-06040-2#ref-CR28 “Zhang, G. et al. China’s environmental policy intensity for 1978–2019. Scientific Data 9, 75, https://doi.org/10.1038/s41597-022-01286-6

(2022).“),[29](https://www.nature.com/articles/s41597-025-06040-2#ref-CR29 “Huang, C. et al. ORLM: A Customizable Framework in Training Large Models for Automated Optimization Modeling. Operations Research https://doi.org/10.1287/opre.2024.1233

(2025).“), their application to input-output estimation remains limited. Recent methodological advances in input-output table estimation have introduced innovative computational approaches that substantially improve upon traditional techniques. Hybrid algorithms have substantially improved the accuracy of interregional input-output table estimation. These include combinations of the RAS method with a real-coded genetic algorithm30 and a boundary tightening algorithm31. In addition, advanced balancing algorithms that can handle unequal net row and column sums32 have also shown strong performance. Similarly, deep neural network architectures have shown superior performance in forecasting input-output tables. They perform well in both short-term and long-term scenarios. This approach has substantially reduced prediction error rates33. More recent evidence suggests that intelligent methods make a difference in the future inference of the IO table; however, the potential of artificial intelligence in MRIO table compilation remains largely untapped.

This study compiles a subnational and international coupling MRIO (SIC-MRIO) covering 30 provinces in China, 66 countries & economies worldwide, and 30 sectors, thereby linking international MRIO with subnational MRIO. The main contributions of this study are as follows. First, our methodology incorporates LLMs into the compilation of MRIO tables, to compensate for missing critical information and frameworks. Second, in addition to the applications of mathematical and planning methods, we use a large amount of real data to fill in the microstructure between the international and subnational sectors. Third, we take advantages of both the macro-level and the micro-level research perspectives, our model integrates macro-level aggregate data (e.g., national customs import–export data and service trade data) with micro-level structural data (e.g., firm-level import and export logistics data).

Methods

Data collection

The SIC is constructed based on a comprehensive data framework, integrating extensive trade information to capture subnational and international economic linkages. The SIC is compiled using a partial-survey method. It describes the main industrial input-output relationships through more than 2 million subnational and international trade data points. This study focuses on 30 Chinese provinces and 66 other countries & economies using 2017 as the base year. The trade data include more than 5,200 commodities, which are classified into 30 economic sectors following the Harmonized System (HS) Codes. Table 2 provides an overview of the raw data used in SIC construction. In this table, we use web crawlers on the customs website to obtain “Chinese customs import and export data”. We collect “Firm-level import and export logistics data” and “American firm-level import and export logistics data” from the Alayun and ImportYeti websites respectively, and “China MRIO 2017” is based on our previously published work in Environmental Science & Technology (EST). And we account all the MRIO tables to reflect values at producers’ prices.

The basic structure of SIC

The MRIO tables link multiple domestic input-output tables to improve data consistency within and across economies. The standard MRIO structure consists of three main components: I intermediate matrix (Z) section, II final demand (FD) section, III value added (VA) section. The subnational MRIO and international MRIO share similar structures, as shown in Fig. 1. In the figure, Ctry denotes a country; Prov, a province; S-MRIO and I-MRIO represent the subnational and international MRIOs, respectively. SIC-MRIO_ZExport refers to China’s exports in the intermediate matrix section of the SIC, while SIC-MRIO_ZImport represents China’s imports in the same section. SIC-MRIO_FDExport and SIC-MRIO_FDImport denote China’s exports and imports in the final demand section, respectively. Figure 1a,b illustrate their layouts. However, integrating them is challenging. Figure 1c shows the subnational-international dual structure (grey squares). It connects international trade relations (green squares) with subnational trade relations (red squares). As we have mentioned in the first section, there are severe data gaps in the coupling process. To address this, we construct two key datasets: the subnational-international import table (SIC_Import) and the subnational-international export table (SIC_Export). In addition, the MRIO_Z section adopts EST_MRIO developed by our research team, while the ICIO_Z section uses OECD_ICIO.

Fig. 1

Main structure of MRIO tables. (a) International MRIO; (b) Subnational MRIO; (c) SIC MRIO.

The main work is divided into macro-level and micro-level components. We compile the SIC_Import and the SIC_Export using both macro and micro frameworks, as shown in Fig. 2. We calculate SIC_Import and SIC_Export by applying the transnational sector–sector trade ratio matrix, derived from a micro perspective, to the province–country total trade values estimated from a macro perspective. The macro dataset includes over 740,000 records from the General Administration of Customs, while the micro dataset contains more than 1,050,000 records from the Alayun Platform and 500,000 records from the ImportYeti Platform.

Fig. 2

Overview of the framework for building the SIC.

Data processing from a macro perspective

We calculate the province-country total value (goods) and province-country total value (services) through the steps shown in Fig. 2.

The provincial goods trade import and export data provided by General Administration of Customs of the People’s Republic of China, include the following indicators: “HS code”, “product name”, “registration location code”, “registration location name”, “trade partner code”, “trade partner name”, and “transaction value in RMB”[34](https://www.nature.com/articles/s41597-025-06040-2#ref-CR34 “General Administration of Customs of the People’s Republic of China. http://stats.customs.gov.cn/indexEn

(2017).“). Provinces are identified using registration location names, and trade partner countries are identified using the trade partner name. It is worth noting that the trade partner category for imports includes China itself, indicating the presence of domestic (China-to-China) trade flows. Supplementary Information 2 provides further details. The HS code, administered by the United Nations Statistics Division, serves as the global standard for product classification in international trade. This HS code employs a six-digit numerical framework to systematically categorize traded commodities, enabling uniform identification and description of products. We compile the goods sectors and HS code comparison table by extending the classification from industry to goods sectors based on the industry and HS code comparison table35. In addition, we use HS code to differentiate economic sectors and transform customs trade data into subnational and international sectoral input-output relationships. This method enables the calculation of the province-country total value (goods).

The primary data source for the service sectors is the China Commerce Yearbook, specifically the provincial STIE (see Table 3 for comparison details). Additional data are collected from provincial bureaus of statistics, provincial commerce bureaus and other government websites to supplement the analysis (see Supplementary Information 3). For tourism-related service trade, we use data from the World Travel & Tourism Council (WTTC) to distribute tourist expenditures across sectors[36](https://www.nature.com/articles/s41597-025-06040-2#ref-CR36 “WTTC. https://researchhub.wttc.org/product/global-retail-tourism-trends-insights-full-report

(2023).“) including accommodation and catering, transportation, warehousing and postal services, culture, sports and entertainment, as well as other service sectors. However, STIE provides only aggregate provincial-level data without detailing trade flows between provinces and countries. To address this, we use the sectoral distribution of China’s service exports from the OECD_ICIO database. By integrating these data sources, we derive the province-country total value (services).

Data processing from a micro perspective

From a micro perspective, we employ large language models to classify firms by sector and convert firm-level product logistics relationships into inter-industry trade linkages. We use data from the Alayun Platform, which provides province-level firm-level import and export logistics data. These data include key indicators such as “HS code”, “purchaser country”, “supplier country”, “detailed product name”, “quantity”, “purchaser name”, “supplier name”, “transaction date”, “product description”, “purchase amount”, and “gross weight”. To extract firm-level trade relationships, we focus on the “purchaser name” and “supplier name” indicators. Using the GPT4 API, we retrieve firm business scope, core product information according to the purchaser and supplier name from e-commerce platforms and firm websites, summarizing each firm’s primary business activities. Specifically, we use 500,000 trade records of U.S. companies from the ImportYeti Platform in 2017 to assist in identifying their primary business activities. To ensure the accuracy, we conduct a manual verification, which is provided in the Technical Validation section.

We employ LLM techniques to extract and refine sector classification descriptions from China’s input-output sector classification explanation and code, constructing a sector classification corpus. We fine-tune the Llama 3 model37 to train a firm information-to-sector classification model based on the sector classification corpus. This model achieves an accuracy of 0.96. We apply this firm-to-sector model to categorize the extracted firm business information into corresponding sectors. By integrating these sector classifications with purchase amount, country of origin and other trade-related attributes, we derive the transnational sector-sector trade ratio for China’s international trade.

Owing to the limited number of trade partner countries available on the Alayun Platform, we classify 66 countries & economies from the OECD_ICIO database into three income groups according to World Bank standards: high income, upper middle income and lower middle income (see Supplementary Information 1 for details). For high-income countries, we use Russia and the US as reference points in our analysis. We incorporate geographical and economic characteristics to analyze the correlation between each country’s trade structure and that of the US and Russia. This allows us to assess which country’s trade pattern a high-income country more closely follows.

Due to severe data deficiencies for certain provinces on the Alayun Platform, we apply a regional or GDP-based adjacency principle. This method imputes missing values for provinces with substantial missing data, based on the three scenarios, i.e. reference provinces for data-missing provinces in Export, Import_US and Import_Others, outlined in Table 4 and Supplementary Information 8. As a result, we compute 1980 transnational sector-sector trade ratios, covering the trade flows between 30 Chinese provinces and 66 countries & economies.

Final demand section

Based on the aforementioned micro- and macro-level data, we proceed with the table compilation.

In the MRIO, the total output of an economy can be calculated as:

$$X={AX}+{FD}$$

(1)

where (X) is the total output, (A) is the direct requirements matrix, ({AX}) or Z is the intermediate matrix, and ({FD}) is the final demand. To compile the SIC_Z and SIC_FD, we calculate the proportion of ({FD}) in goods sectors using the EST_MRIO, and the proportion of ({FD}) in service sectors using the OECD_ICIO database. Since the ({FD}) proportion of imports in China is not directly available in the OECD_ICIO database, we apply the ({FD}) proportion of imports from China’s non-competitive input-output table instead[38](https://www.nature.com/articles/s41597-025-06040-2#ref-CR38 “State Statistics Bureau. https://data.stats.gov.cn/files/html/quickSearch/trcc/trcc0801.html

(2025).“).

Intermediate matrix section

Given the quantitative relationship between Z and FD components established in Eq. (1), the residual term (X – FD) mathematically corresponds to Z (Fig. 3). We take a province-country input-output matrix from SIC_ZExport as an example, focusing on export flows from province r to country m (Fig. 3a). This matrix contains 30 economic sectors, categorized into 19 goods and 11 services sectors. The interaction of these categories generates four distinct submatrices: goods-goods, goods-services, services-goods and services-services (as shown in Fig. 3b). First, the detailed calculation procedure for goods-goods is illustrated in Fig. 3c. It is calculated as the product of the Province-country total value (from Data processing from a macro perspective section) and the Transnational sector-sector trade ratio (from Data processing from a micro perspective section). The calculations for goods-services, services-goods and services-services are computed using a similar method, except that they rely on the Transnational sector-sector trade ratio from the OECD_ICIO database for China’s trade with each country.

Fig. 3

Calculating methods of SIC_Z.

Value added section

In the MRIO, the total input of an economy can also be calculated as:

$$X=Z+{VA}$$

(2)

where X represents the total input, Z denotes the Intermediate matrix, and VA refers to the value added. The VA values of each province in China are sourced from our EST_MRIO table.

Table balancing

The table balancing approaches include RAS, generalised RAS (GRAS), entropy, constrained optimization, linear programming approaches, Monte Carlo, hybrid algorithms, etc., and they have different scopes of applicability and advantages. With imputed microdata, the chosen GRAS method can achieve convergence at the 89th iteration. The GRAS method is an extension of the RAS approach, which aims to preserve the original matrix structure as much as possible while making minimal adjustments to align with new row and column totals, accommodating both positive and negative values in the matrix. Using the GRAS method, we ensure the input-output table satisfies the row-column relationships specified in Eqs. (1, 2).

$$\left{\begin{array}{c}{\sum }_{j}{r}_{i}{Z}_{{ij}}{s}_{j}={u}_{i}(\forall ,{i})\ {\sum }_{i}{r}_{i}{Z}_{{ij}}{s}_{j}={v}_{j}(\forall ,{j})\end{array}\right.$$

(3)

$$\mathop{\min }\limits_{R,S}\sum _{i,j}\left|{Z}_{{ij}}^{{\prime} }-{Z}_{{ij}}\right|$$

(4)

Here, r**i denotes the row scaling multiplier for the i-th row, which may assume either positive or negative values; ({s}_{j}) represents the column scaling multiplier for the j-th column, also permitting both positive and negative values. The original intermediate matrix is denoted as ({Z}_{{ij}}), while ({Z}_{{ij}}^{{\prime} }) refers to the adjusted intermediate matrix. Furthermore, ({u}_{i}) and ({v}_{j}) correspond to the row and column totals of the adjusted matrix for the i-th row and j-th column, respectively.

Data Records

The SIC dataset is available at the figshare website[39](https://www.nature.com/articles/s41597-025-06040-2#ref-CR39 “Zheng, J. & Mi, Z. A multi-regional input-output database linking Chinese subnational regions and global economies. figshare. Dataset. https://doi.org/10.6084/m9.figshare.28973396.v3

(2025).“). We provide 6 processing folders containing source data and code: 1. Custom data processing, 2. Service trade data processing, 3. Micro data processing, 4. Table compiling_Z, 5. Table compiling_FD, 6. Table balancing and one final table of results: “SIC-MRIO_2017”. This table contains an intermediate matrix (2820*2820), a final demand vector (2820*282) and a value added vector (2820*1). The SIC dataset can be freely downloaded from https://doi.org/10.6084/m9.figshare.28973396.v3.

Technical Validation

Comparison between SIC and other MRIO datasets for China’s import & export total value

China’s import and export statistics show major discrepancies across key datasets, including those from the OECD and CEADs. These differences arise mainly from limited data availability and inconsistent source data. As evidenced in Table 5, our estimates show minimal deviations of merely 2.53% for imports and 5.09% for exports when benchmarked against official China Customs statistics. This represents a marked improvement in accuracy relative to alternative datasets, with EXIOBASE exhibiting deviations of 17.82% (imports) and 16.16% (exports), and the OECD showing variances of 11.82% (imports) and 11.12% (exports).

Comparison between SIC and other MRIO datasets for the structure of intermediate use and final use

In MRIO tables, the structure of intermediate use and final demand serves as another crucial indicator. Comparison between SIC and other major domestic and international MRIO databases—including OECD, EXIOBASE, CEADs and Li et al.—reveals generally consistent structural patterns. Among these, CEADs shows the highest proportion of intermediate use at 63.54%, while EXIOBASE records the lowest at 60.47%. When compared specifically with Chinese regional input-output tables, more pronounced discrepancies emerge in the provinces of Tianjin, Jilin and Heilongjiang, shown in Fig. 4. We conduct a detailed verification of the ratio of the structure of intermediate use and final use in Supplementary Information 5.

Fig. 4

Comparisons between SIC and other MRIO tables for the structure of intermediate use and final use. (a) Compares the structure of intermediate use and final demand in CHN; (b) Compares the provincial structure of intermediate use and final demand in CHN.

Manual annotations and checks

We undertake a systematic process of manual annotation and verification of the data from a micro perspective. First, the firm-level logistics data are cross-validated with China Customs data to ensure consistency and reliability, shown in Supplementary Information 6. Second, we assess the accuracy of company-level business information collected through the GPT4 API by conducting a secondary manual verification, with results presented in Table 6. While both GPT4 and manual methods maintain high precision, GPT4 shows slightly lower semantic accuracy. Specifically, GPT4 occasionally generates vague descriptors like “Industry & Manufacturing”, “Industrial” and “Technology development” for certain companies’ primary businesses - terms too ambiguous for proper industry classification. These non-specific terms (detailed in Supplementary Information 7) are subsequently excluded from our analysis.

Data availability

The SIC dataset is available in the figshare repository[39](https://www.nature.com/articles/s41597-025-06040-2#ref-CR39 “Zheng, J. & Mi, Z. A multi-regional input-output database linking Chinese subnational regions and global economies. figshare. Dataset. https://doi.org/10.6084/m9.figshare.28973396.v3

(2025).“) and is publicly accessible at https://doi.org/10.6084/m9.figshare.28973396. The shared output includes a final table of results titled “SIC-MRIO_2017”, and 6 processing folders containing the source data and the Python code.

Code availability

The program used in this study is based on Python. Code is freely available on the figshare website.

References

Zhang, Y., Tian, K., Li, X., Jiang, X. & Yang, C. From globalization to regionalization? Assessing its potential environmental and economic effects. Applied Energy 310, 118642 (2022).

Article CAS Google Scholar 1.

Zhao, R. et al. Cascade effects of global supply chains induced by regional conflicts increase global inequality. Resources, Conservation and Recycling 209, 107740 (2024).

Article Google Scholar 1.

Haas, W., Krausmann, F., Wiedenhofer, D. & Heinz, M. How circular is the global economy?: An assessment of material flows, waste production, and recycling in the European Union and the world in 2005. Journal of Industrial Ecology 19, 765–777 (2015).

Article Google Scholar 1.

Meng, B., Zhang, Y. & Inomata, S. Compilation and applications of IDE-JETRO’s international input–output tables. Economic Systems Research 25, 122–142 (2013).

Article Google Scholar 1.

Wang, Y., Geschke, A. & Lenzen, M. in Proceedings of the 22nd International Input-Output Conference, Lisbon, Portugal. 14–18. 1.

Zheng, J. et al. The slowdown in China’s carbon emissions growth in the new phase of economic development. One Earth 1, 240–253 (2019).

Article ADS Google Scholar 1.

Feng, K., Davis, S. J., Sun, L. & Hubacek, K. Drivers of the US CO2 emissions 1997–2013. Nature Communications 6, 7714 (2015).

Article ADS CAS PubMed Google Scholar 1.

Hamilton, H. A. et al. Trade and the role of non-food commodities for global eutrophication. Nature Sustainability 1, 314–321 (2018).

Article Google Scholar 1.

Shan, Y. et al. Impacts of COVID-19 and fiscal stimuli on global emissions and the Paris Agreement. Nature Climate Change 11, 200–206 (2021).

Article ADS CAS Google Scholar 1.

Han, X. et al. Comparative life cycle greenhouse gas emissions assessment of battery energy storage technologies for grid applications. Journal of Cleaner Production 392, 136251 (2023).

Article CAS Google Scholar 1.

Huo, J., Meng, J., Zheng, H., Parikh, P. & Guan, D. Achieving decent living standards in emerging economies challenges national mitigation goals for CO2 emissions. Nature Communications 14, 6342 (2023).

Article ADS CAS PubMed PubMed Central Google Scholar 1.

Yang, Y. et al. Mapping global carbon footprint in China. Nature Communications 11, 2237, https://doi.org/10.1038/s41467-020-15883-9 (2020).

Article ADS CAS PubMed PubMed Central Google Scholar 1.

Bruckner, B. et al. Ecologically unequal exchanges driven by EU consumption. Nature Sustainability 6, 587–598, https://doi.org/10.1038/s41893-022-01055-8 (2023).

Article Google Scholar 1.

Long, Y. et al. Comparison of city-level carbon footprint evaluation by applying single-and multi-regional input-output tables. Journal of Environmental Management 260, 110108, https://doi.org/10.1016/j.jenvman.2020.110108 (2020).

Article ADS CAS PubMed Google Scholar 1.

Wang, S. & Chen, B. Energy–water nexus of urban agglomeration based on multiregional input–output tables and ecological network analysis: A case study of the Beijing–Tianjin–Hebei region. Applied Energy 178, 773–783, https://doi.org/10.1016/j.apenergy.2016.06.112 (2016).

Article ADS Google Scholar 1.

Wang, Y. et al. An inter-city input-output database distinguishing firm ownership in the Greater China area during 2002–2017. Scientific Data 12, 723, https://doi.org/10.1038/s41597-025-04996-9 (2025).

Article PubMed PubMed Central Google Scholar 1.

Meng, B., Wang, Z. & Koopman, R. How are global value chains fragmented and extended in China’s domestic production networks? IDE Discussion Paper 424 (2013). 1.

Meng, B. & Yamano, N. Compilation of a regionally extended inter-country input–output table and its application to global value chain analyses. Journal of Economic Structures 6, 23 (2017).

Article Google Scholar 1.

Zheng, J., Hou, X., Duan, H. & Wang, S. Where we are and where we will go to achieve carbon neutrality: A review of quantitative socioeconomic modeling research. Fundamental Research 4, 1696–1709 (2024).

Article CAS PubMed PubMed Central Google Scholar 1.

Zheng, J. et al. The energy system transition pathway towards carbon reduction using a model-coupling approach. Global Environmental Change 89, 102945 (2024).

Article Google Scholar 1.

Mi, Z. F. et al. Chinese CO2 emission flows have reversed since the global financial crisis. Nature Communications 8 (2017). 1.

Ou, J. et al. Role of export industries on ozone pollution and its precursors in China. Nature Communications 11, 5492 (2020).

Article ADS CAS [PubMed](http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&dopt=Abstr