Identifying and Classifying Shrinking Cities Using Long-Term Continuous Night-Time Light Time Series

Abstract

Shrinking cities—cities suffering from population and economic decline—has become a pressing societal issue of worldwide concern. While night-time light (NTL) data have been applied as an important tool for the identification of shrinking cities, the current methods are constrained and biased by the lack of using long-term continuous NTL time series and the use of unidimensional indices. In this study, we proposed a novel method to identify and classify shrinking cities by long-term continuous NTL time series and population data, and applied the method in northeastern China (NEC) from 1996 to 2020. First, we established a long-term consistent NTL time series by applying a geographically weighted regression model to two distinct NTL datasets. Then, we generated NTL index (NI) and population index (PI) by random forest model and the slope of population data, respectively. Finally, we developed a shrinking city index (SCI), based on NI and PI to identify and classify city shrinkage. The results showed that the shrinkage pattern of NEC in 1996–2009 (stage 1) and 2010–2020 (stage 2) was quite different. From stage 1 to stage 2, the shrinkage situation worsened as the number of shrinking cities increased from 102 to 162, and the proportion of severe shrinkage increased from 9.2% to 30.3%. In stage 2, 85.4% of the cities exhibited population decline, and 15.7% of the cities displayed an NTL decrease, suggesting that the changes in NTL and population were not synchronized. Our proposed method provides a robust and long-term characterization of city shrinkage and is beneficial to provide valuable information for sustainable urban planning and decision-making.

Keywords

• city shrinkage
• night-time light
• geographically weighted regression
• random forest