One country, two “urban” systems: focusing on bimodality in China’s city-size distribution

Abstract

In this study, we demonstrate the existence of bimodality in China’s city-size distribution and develop an urban-growth forecast model that incorporates this bimodality. Main data for our analysis are $0.{25}^{\circ }\times 0.{25}^{\circ }$ population density grids for the past 32 years, created from China’s official census data and county-level statistics. Our results show that the mixture of two Gaussian distributions outperforms unimodal distributions in explaining China’s historic urban-growth patterns, suggesting that the conventional unitary urban-hierarchy assumption lacks ground in China’s context. We also find that the higher-density mixture component increasingly dominates the entire distribution, and this gradual transition toward a unimodal city-size distribution is partly related to increased domestic population mobility.

Appendix

1.1 Grid conversion of conventional population data

A brief summary of the method for converting census data into population grids is as follows. First, we compute population density for each census tract from the available census data. Then we assign the population density to overlapped grids. If there is a grid where multiple census tracts intersect, then we assign the area-weighted average density to the grid cell. For example, the population density for grid cell $i (d_i)$, displayed in Fig. 8, is ($d_\mathrm{A} w_\mathrm{A} +d_\mathrm{B} w_\mathrm{B} +d_\mathrm{C} w_\mathrm{C})$. As the land area of grid cell i ($l_i)$ is already known from the SEDAC dataset, the population size for the same grid cell ($p_i)$ can easily be computed by multiplying $d_i$ and $l_i$, if $p_i$, instead of $d_i$, is used to estimate China’s city-size distribution.

Fig. 8

Overlay of census data with the $0.25{^{\circ }}\times 0.25{^{\circ }}$ grid layer, Beijing and its vicinity