Identify Ghost Cities Using Random Forest Algorithm in Hohhot City, China

Keywords: Ghost City, China study, landuse policy, urban study
Session Type: Virtual Poster Abstract
Day: Friday
Session Start / End Time: 2/25/2022 03:40 PM (Eastern Time (US & Canada)) - 2/25/2022 05:00 PM (Eastern Time (US & Canada))
Room: Virtual 39

Abstract

Over the past two decades, China has urbanized rapidly from coastal to inland cities, resulting in a remarkable expansion of the urban landscape. A ghost city is a newly developed residential area with extremely high vacancy. Our goal is to identify ghost cities in Hohhot city using machine learning. Existing works rarely conduct a microscopic study on individual cities and decipher the influencing factors. In this paper, we propose a practical framework to identify ghost cities using random forest (RF) regression based on the Google Earth Engine (GEE) web API. First, we identify recent residential development by classifying satellite images of three time periods (2008, 2013, and 2018.) Then, we use point of interest (POI) data representing residential buildings and grided population counts based on mobile phone signals to identify ghost cities among the recently developed residential POIs. The results show that in 2008-18, built-up areas almost doubled. We identified 228 "ghost" residential areas, and 1478 “potential ghost” residential, in total8878 high and newl built-up residential areas. Thus, our study provides spatially explicit insight into the "ghost city" phenomenon, which is an advanced method for extracting urban structure. The multi-variable-based analysis may help local governments in sustainable urban planning and provide a prognosis for future research on the causes of ghost city formation and the prediction of ghost cities.