More urbanization, more polarization: evidence from two decades of urban expansion in China

The study investigates whether China’s rapid urbanization over the past two decades has produced a more balanced or increasingly polarized city size distribution. Urban polarization—growing disparities between larger and smaller cities—can exacerbate socioeconomic inequalities and instability. The literature is divided on whether polarization is increasing or whether city-size distributions remain stable over time. China is a critical case given the unprecedented scale of its urbanization, the growth of mega-cities, and state-led efforts—such as the 2014 New Urbanization Plan—to promote coordinated and sustainable development across urban agglomerations. Traditional population-based measures of city size are challenged by institutional features (e.g., hukou) and migration patterns that obscure true urban dynamics. The authors therefore examine physical infrastructure growth via built-up area (BUA) to assess city size distributions and polarization within and across 23 officially designated urban agglomerations, asking whether polarization is occurring within agglomerations, across agglomerations, or both.

City size distributions have been studied through frameworks such as the law of the primate city, rank-size rule (Pareto principle), Zipf’s law, and scaling laws. Deviations from these patterns are often attributed to local socioeconomic and political conditions. China is known to diverge from commonly observed development patterns (e.g., rank-size conformity), potentially due to its agglomeration strategies and policy interventions. Prior studies offer mixed evidence on polarization trends, with some suggesting increased polarization and others indicating stability. Population-based analyses are complicated by hukou restrictions, temporary migration, and measurement uncertainties across administrative scales. Advances in remote sensing and nighttime lights, combined with machine learning, enable fine-grained assessments of urban infrastructure expansion, offering advantages over population counts for tracking city size dynamics. Existing research often focuses on select coastal agglomerations; comprehensive comparative analyses across all designated agglomerations have been limited, motivating this study.

Study area: 23 urban agglomerations comprising 168 cities designated in China’s 11th Five-Year Plan (2006–2010), grouped along three axes: coastal, northern inland (Longhai/Lanxin railways), and southern inland (Yangtze River drainage). The three largest by GDP are the Yangtze River Delta (YRD), Beijing–Tianjin–Hebei (BTH), and Pearl River Delta (PRD). Hong Kong and Macao are not included in the PRD delineation per official designation. Data and urban classification: Built-up area (BUA) maps for 1995, 2000, 2005, 2010, 2015, and 2018 were produced on Google Earth Engine using Landsat imagery and DMSP/OLS and VIIRS nighttime light imagery. Landsat preprocessing selected pixels with low cloud scores and computed per-band medians. Six spectral indices (NDWI, NDBI, UI, NDSV, NDVI, EVI) plus all bands were inputs to a Random Forest classifier (20 trees) labeling pixels as built-up (>50% built) vs non-built (≥50% natural). Nighttime light thresholds generated labeled training data. Inter-annual correction improved accuracy, and 60,000 hand-labeled validation points (10,000/year) from Google Earth provided accuracy assessment using overall accuracy, TPR, TNR, balanced accuracy, and user’s accuracy. Growth measures: Annual growth rate (AGR) of BUA was computed as AGR = (Uj − Ui)/(j − i)/Ui × 100%, where Ui and Uj are BUA at years i and j. Polarization across agglomerations is inferred when larger agglomerations have higher AGR than smaller ones. Within-agglomeration distribution: Primacy index defined as the ratio of BUA of the largest city to the second-largest city within each agglomeration. Values near 1 indicate dual-core; 2–4 indicate moderate primacy; >4 indicate high primacy. Increasing primacy over time suggests rising within-agglomeration polarization. Rank-size distribution: Pareto exponent (α) estimated for YRD, BTH, and PRD (each with ≥9 cities) using the Gabaix–Ibragimov rank − 1/2 correction: ln(Ri/2) − 1/2 = ln A − α ln Si + u, where Ri is rank, Si is city BUA, A is largest city size. Larger α implies more even distribution. The mean value of rank (MVR) across cities in each agglomeration was also computed to track changes in relative standing within the national system. Data sources: Administrative boundaries from China’s 11th Five-Year Plan and National Earth System Science Data Center. Landsat from USGS; nighttime lights from NOAA DMSP/OLS and VIIRS/NPP.

- National and agglomeration-scale expansion: China’s total BUA increased from 51,238.66 km² (1995) to 198,602.14 km² (2018), a 287.60% rise, with classification accuracies averaging over 90%. Across the 23 agglomerations, BUA rose from 33,622.42 km² to 128,069.24 km² (+280.90%), accounting for about 65% of national BUA in 2018. - Size distribution of agglomerations: In 1995, only one agglomeration exceeded 5,000 km² BUA; by 2018, eight did. The smallest agglomeration (Jiuquan–Jiayuguan–Yumen) was >40× smaller than the largest (YRD) in 2018, highlighting pronounced size disparities. - Coastal vs inland divergence: Coastal agglomerations were larger and expanded more, with average BUA 10,479.91 km² vs 2,948.66 km² for inland agglomerations in 2018. The coastal–inland BUA gap widened over time. Several large agglomerations (YRD, BTH, Shandong Peninsula, Chengdu–Chongqing, PRD) each added >5,000 km² since 1995, exceeding the 1995 BUA of all but BTH. Smaller inland agglomerations (e.g., Jiuquan–Jiayuguan–Yumen, Central Guizhou, Changsha–Zhuzhou–Xiangtan, Nanning–Beihai–Qinzhou–Fangchenggang, Lanzhou–Baiyin–Xining) expanded by <1,400 km². Regional AGRs: southern inland 22.36%, coastal 11.59%, northern inland 9.44%. - Within-agglomeration primacy declined overall: Agglomerations with a primate city (primacy ≥2) fell from 10 (1995) to 7 (2018). Many agglomerations saw declining primacy indices, indicating more uniform internal size distributions. Nanning–Beihai–Qinzhou–Fangchenggang maintained high primacy; some shifted from high to moderate/no primacy. Three agglomerations (Yinchuan Plain, Changsha–Zhuzhou–Xiangtan, Central Plains) developed primacy by 2018. - Pareto exponent trends (three largest agglomerations): YRD α increased from 1.0411 (1995) to 1.3953 (2018); BTH α from 2.0036 to 2.3717—both indicating more even size distributions. PRD α decreased from 1.7151 to 1.2613, signaling increased polarization and a convergence toward Zipf-like scaling. - Changes in mean national rank: Mean city rank improved (decreased) in YRD (83rd to 45th) and BTH (43rd to 37th), but worsened (increased) in PRD (62nd to 84th). No spatial autocorrelation in city ranks was detected for the three agglomerations.

The analysis reveals a dual pattern: polarization among agglomerations increased—particularly between larger coastal and smaller inland systems—while within most agglomerations city sizes became more evenly distributed. This suggests that policy-driven coordination and agglomeration economies may be successfully diffusing growth within agglomerations, but not sufficiently bridging disparities across the national hierarchy. The findings align with broader dynamics including decoupling of population and infrastructure growth, evolving housing demand, and national planning interventions (e.g., green belts, regional revitalization strategies). YRD and BTH show decreasing internal polarization, likely reflecting economic dynamism, industrial transfers, transport connectivity, and supportive local policies that elevate smaller cities. Conversely, PRD exhibits rising internal polarization driven by the rapid growth of Shenzhen and Guangzhou relative to smaller neighbors and the distinct development context adjacent to Hong Kong and Macao. The coastal–inland divide mirrors the historic Huhuanyong line, underscoring structural geographic and policy factors. Methodological differences with prior nighttime-lights-only studies may explain contrasting results for PRD due to saturation and spillover effects and coarser resolution in earlier datasets.

Using a physical-infrastructure proxy (BUA) derived from satellite data and machine learning, the study shows that from 1995 to 2018 China’s urban agglomerations expanded dramatically. Polarization increased among agglomerations (coastal vs inland), while within-agglomeration size distributions generally became more uniform, with notable exceptions such as PRD. The results indicate partial success of policies aimed at coordinated development inside agglomerations but persistent national-scale imbalances. Future research could extend the analysis to include additional years, incorporate population/economic indicators alongside BUA to assess multi-dimensional polarization, integrate Hong Kong and Macao for PRD assessments, and evaluate policy impacts causally to determine the effectiveness of specific interventions in reducing inter-agglomeration disparities.

- Administrative delineations exclude Hong Kong and Macao from the PRD agglomeration, which may affect internal distribution assessments there. - The Pareto analysis was restricted to three agglomerations with ≥9 cities due to small city counts elsewhere, limiting generalizability of rank-size results across all agglomerations. - Primacy index considers only the top two cities within an agglomeration and may not capture full distributional dynamics across the entire city set. - Although classification accuracy exceeded 90% on average, remote sensing-based BUA estimates still entail uncertainties; nighttime lights can suffer from saturation and spillover, and data sources/resolutions can influence results. - BUA measures physical expansion and may be decoupled from population or economic growth, potentially leading to different interpretations versus population-based analyses.