Causal attribution and external validity of the non-monotonic urban density–spatial carbon performance relation in Chinese megacities

For Beijing and Shanghai (1999–2019), Yang et al. (2024) examined the relations between urban density (population, building, and economic density per unit of built-up area) and spatial carbon performance, defined as CO2 emissions per unit of built-up area ($CO_2/L$). The study's non-monotonic polynomial fits - including an N-shaped cubic for population density with turning points near 12,600 and 16,400 people/km2 and an M-shaped building-density relation - were estimated for Beijing alone, while Shanghai's curve shapes were taken from an earlier Shanghai study cited in the paper: population- and economic-density shapes consistent with Beijing, but an inverted-U building-density relation inconsistent with it. The authors nonetheless asserted that the findings apply to the same type of megacities worldwide. The study itself flags what remains unresolved: technical (efficiency) factors were excluded by construction, the effects of policy interventions such as the relocation of high-emission manufacturing out of inner urban districts were never quantified, and built-up-area measurement coarseness leaves the density–performance relation undemonstrated at finer spatial granularity; in addition, the fits are correlational and rest on built-up-area aggregates, and the worldwide-applicability claim is untested beyond the two case cities. The open problem is to produce an externally validated account of how urban density drives megacity spatial carbon performance: re-estimating the density–$CO_2/L$ relation across additional cities and at finer spatial resolution to test whether the non-monotonic shapes and thresholds replicate, and decomposing - with causal identification where feasible - the observed $CO_2/L$ trajectories into the contributions of densification, technological change, and policy-driven industrial relocation.

Background

Cities dominate anthropogenic carbon emissions, and whether increasing urban density helps or hurts mitigation is a long-standing question in the compact-city debate: higher density can cut per-capita transport energy and building energy use through shared walls and shorter trips, but excessive density can induce congestion and rising per-area energy use. The source study approaches this with a spatial accounting lens. The built-up area $L$ is the administratively recognized, continuously developed urban land of a Chinese city. Spatial carbon performance is then the ratio $CO_2/L$ - total carbon dioxide emissions divided by built-up area - i.e., an emission intensity per unit of urban space in which a decline is treated as an improvement. Urban density is decomposed into population density $P/L$, economic density $G/L$ (gross urban product per unit built-up area), and building density $B/L$ (floor area per unit built-up area, measured with data at built-up-area granularity, which the authors themselves flag as a possible source of inaccuracy). Using annual data for Beijing and Shanghai from 1999 to 2019, the study tracks the temporal evolution of densities and $CO_2/L$ in both cities, computes Pearson correlations, and maps industrial agglomeration with Point-of-Interest (POI) data - georeferenced records of individual facilities such as factories, shops, and offices - smoothed by kernel density estimation, a standard method for converting point locations into a continuous surface of local intensity; and, for Beijing, it fits cubic and quadratic log-polynomials of $CO_2/L$ on each density dimension and runs a multiple log-linear regression of $CO_2/L$ on all three densities. The headline finding is non-monotonicity: for Beijing, population density relates to $CO_2/L$ through an N-shaped cubic ($R^2$ = 0.732) with two turning points at roughly 12,600 and 16,400 people/km2 - called "inflection points" in the source, though a cubic has a single true inflection point and these are the two thresholds at which the fitted slope changes sign - so that rising density worsens per-area emissions below the first and above the second threshold but improves the emissions outcome between them; Beijing's building density shows an M-shaped relation; the economic-density relation changes sign across sub-periods and is mediated by industrial structure. For Shanghai the paper does not fit these polynomials itself but reports, from a cited earlier Shanghai study (Chen et al.), N-shaped and inverted-N-shaped relations for population and economic density consistent with Beijing, and an inverted-U building-density relation inconsistent with Beijing's M-shape. The comprehensive regression deliberately excludes technical factors (energy-supply and process efficiency improvements) to simplify the analysis, and the authors note that both city governments, especially Beijing's, pushed high-emitting and high-polluting manufacturing out of urban districts to suburbs or adjacent regions - a policy channel whose effect on $CO_2/L$ they did not study. Three limitations are stated explicitly by the source: unquantified policy effects; built-up-area measurement coarseness and the absence of finer-grained (parcel-level floor area, building energy, sub-municipal emissions) data; and the omission of technical factors, leaving open how much of the observed decline in $CO_2/L$ reflects densification rather than technological change. Beyond the paper's own list, its curve-fitting evidence is correlational and Beijing-only, and the claim of worldwide applicability to similar megacities is asserted but untested - even as the source's own Shanghai comparison already records one non-replicating building-density shape. Adjacent multi-city research on density, compactness, and urban form - using different outcome variables such as total or per-capita city emissions and carbon emission intensity - consistently finds heterogeneous, scale- and city-type-dependent associations, which sharpens but does not answer the generalizability question for the $CO_2/L$ framework. A neighboring-subfield reader should also know the convention that in this Chinese urban-planning literature the fitted non-monotonic curves echo the environmental Kuznets curve tradition, in which an emission outcome is modeled as a polynomial function of a development-related variable so that the relation changes sign across development stages.

Problem Statement

The source question, kept in its full generality: for Chinese megacities (and the megacities worldwide of the same type as Beijing and Shanghai - large, high-density urban agglomerations - to which the source claims its findings apply), establish how urban density shapes spatial carbon performance, defined as $CO_2$ emissions per unit of built-up area ($CO_2/L$, lower being better). Concretely, this requires an empirical account that: (1) re-estimates the relation between each density dimension ($P/L$, $B/L$, $G/L$) and $CO_2/L$ - taking as the baseline the Beijing-fitted non-monotonic (N-shaped, M-shaped) forms and their slope sign-change thresholds, such as the population-density turning points near 12,600 and 16,400 people/km2 - and tests whether they replicate across additional cities and with finer-grained (sub-city, parcel-level) data rather than built-up-area aggregates alone; and (2) quantifies how much of the observed decline in megacity $CO_2/L$ over recent decades is attributable to densification itself, to technological and efficiency change in energy supply and industrial processes, and to policy interventions such as the explicit relocation or exclusion of high-emission manufacturing from inner urban districts, with causal identification of these channels where data and quasi-experimental variation permit.

References

1. Yang Yang, Jianmin Zhang, Jingyuan Huang, Yongchao Zhao, Qian Zhao (2024), "Urban density and spatial carbon emission performance of megacities in China", Scientific Reports 14:23862. DOI: 10.1038/s41598-024-75163-0. https://doi.org/10.1038/s41598-024-75163-0
2. Jianmin Zhang, Yang Yang, Jingyuan Huang, Xiaoxuan Kao (2024), "Can urbanization improve carbon performance?", Frontiers in Environmental Science 12:1431324. DOI: 10.3389/fenvs.2024.1431324. https://doi.org/10.3389/fenvs.2024.1431324
3. Fei Chen, Yonghe Li, Xiaoqing Ma (2023), "Is the city low-carbon because of its compactness? An empirical study in Shanghai, China", Urban Climate 52:101690. DOI: 10.1016/j.uclim.2023.101690. https://doi.org/10.1016/j.uclim.2023.101690
4. Yanchun Yi, Yajun Wang, Yaqin Li, Ji Qi (2021), "Impact of urban density on carbon emissions in China", Applied Economics 53(53):6153–6165. DOI: 10.1080/00036846.2021.1937491. https://doi.org/10.1080/00036846.2021.1937491
5. Kai Zhu, Manya Tu, Yingcheng Li (2022), "Did Polycentric and Compact Structure Reduce Carbon Emissions? A Spatial Panel Data Analysis of 286 Chinese Cities from 2002 to 2019", Land 11(2):185. DOI: 10.3390/land11020185. https://doi.org/10.3390/land11020185
6. Muhan Li, Minghao Zuo, Saiyi Chen, Shuang Tang, Tian Chen, Jia Liu (2024), "Impact of Urban Spatial Compactness on Carbon Emissions: Heterogeneity at the County Level in the Beijing–Tianjin–Hebei Area, China", Land 13(12):2104. DOI: 10.3390/land13122104. https://doi.org/10.3390/land13122104
7. Tian Feng, Bo Zhou (2023), "Impact of urban spatial structure elements on carbon emissions efficiency in growing megacities: the case of Chengdu", Scientific Reports 13:9939. DOI: 10.1038/s41598-023-36575-6. https://doi.org/10.1038/s41598-023-36575-6
8. Tiankui Li, Yi Liu, Chaoran Wang, Gustaf Olsson, Zishu Wang, Hao Wang (2019), "Decentralization of the non-capital functions of Beijing: Industrial relocation and its environmental effects", Journal of Cleaner Production 224:545–556. DOI: 10.1016/j.jclepro.2019.03.247. https://doi.org/10.1016/j.jclepro.2019.03.247
9. Ke-Liang Wang, Ru-Yu Xu, Wei Jiang, Yuan Liu (2023), "Impact of National Industrial Relocation Demonstration Zones (NIRDZs) policy on urban carbon emissions in China", Environmental Impact Assessment Review 102:107165. DOI: 10.1016/j.eiar.2023.107165. https://doi.org/10.1016/j.eiar.2023.107165
10. Fei Fan, Shangze Dai, Bo Yang, Haiqian Ke (2023), "Urban density, directed technological change, and carbon intensity: An empirical study based on Chinese cities", Technology in Society 72:102151. DOI: 10.1016/j.techsoc.2022.102151. https://doi.org/10.1016/j.techsoc.2022.102151
11. Xuelian Zhu, Jianan Che, Xiaogeng Niu, Nannan Cao, Meiyu Liu (2025), "Study on the effect of technological innovation on carbon emission intensity in 278 prefecture-level cities in China", Scientific Reports 15:14917. DOI: 10.1038/s41598-025-99370-5. https://doi.org/10.1038/s41598-025-99370-5
12. Zichen Zhao, Yixuan Zou, Cheng Shen, Hua Chen, Renlu Qiao, Zhiqiang Wu, Xin Li (2026), "Spatiotemporal Impacts of Vertical Urban Development Patterns on Carbon Emission Intensity: A GeoAI Framework Using Multisource Remote Sensing Data", Remote Sensing 18(11):1775. DOI: 10.3390/rs18111775. https://doi.org/10.3390/rs18111775
13. Heli Lu, Siqi Lu, Huan Li, Liang Cao, Zongran Han, Fang Liu, Chuanrong Zhang, Changhong Miao, Xiaoye Zhang (2025), "Estimation of high-resolution emissions using functionally-filtered nighttime lights in monocentric and polycentric cities: Fusion of remotely sensed human activities and social media data for carbon neutral settlements", Applied Energy 401:126562. DOI: 10.1016/j.apenergy.2025.126562. https://doi.org/10.1016/j.apenergy.2025.126562
14. Xiao Zhou, Weihao Chen, Zhou Huang, Yanling Lu, Guoqing Zhou (2026), "High-resolution assessment of embodied carbon emissions in the urban built environment: Multi-scale patterns and socio-economic drivers across 35 Chinese cities", Sustainable Cities and Society 148:107701. DOI: 10.1016/j.scs.2026.107701. https://doi.org/10.1016/j.scs.2026.107701