Authors: Min Tan*, Sun Yat-sen University, Lin Liu, University of Cincinnati，Sun Yat-Sen University, Kai Liu, Sun Yat-sen University, Yuanhui Zhu, Sun Yat-sen University, Dashan Wang, Sun Yat-sen University
Topics: Population Geography, Urban Geography, Geographic Information Science and Systems
Keywords: population distribution, the Pearl River Delta, random forest model
Session Type: Paper
Start / End Time: 2:00 PM / 3:40 PM
Room: Balcony M, Marriott, River Tower Elevators, 4th Floor
Presentation File: No File Uploaded
High resolution data on population distribution is crucial for disaster reduction, emergency response, environment protection, resources utilization and policy decision. However, the census population in China is only available at the urban street district (Jiedao) level, each having a population that may reach 100,000. Other grid-based population datasets, such as Worldpop (100m grid), Population Grids of China (1km grid) and GPW (30 arc-second grid), also have coarse resolutions. In this study, we integrate nighttime lights images, land cover, slope and DEM to the census data, to create a population grid with a resolution of 30m. The model used in this study is a random forest model, which is also capable of revealing the importance of individual variables. The estimated population was compared with three other open datasets. The model is calibrated and then validated using two different samples. The results suggest that nighttime lights, distance to water, distance to built-up area and density of roads contribute most to the model. The accuracy of the derived 30m population grid has reached at 83.32%, which is better than those of WorldPop and Population Grids of China, and similar to that of GPW.