Mapping Changing Population Distribution on the Qinghai–Tibet Plateau since 2000 with Multi-Temporal Remote Sensing and Point-of-Interest Data

• Main Authors: Lanhui Li, Yili Zhang, Linshan Liu, Zhaofeng Wang, Huamin Zhang, Shicheng Li, Mingjun Ding
• Format: Article
• Language: English
• Published: MDPI AG 2020-12-01
• Series: Remote Sensing
• Subjects: population mapping; remote sensing; point of interest data; random forest model; Qinghai–Tibet Plateau
• Online Access: https://www.mdpi.com/2072-4292/12/24/4059

Advanced developments have been achieved in urban human population estimation, however, there is still a considerable research gap for the mapping of remote rural populations. In this study, based on demographic data at the town-level, multi-temporal high-resolution remote sensing data, and local population-sensitive point-of-interest (POI) data, we tailored a random forest-based dasymetric approach to map population distribution on the Qinghai–Tibet Plateau (QTP) for 2000, 2010, and 2016 with a spatial resolution of 1000 m. We then analyzed the temporal and spatial change of this distribution. The results showed that the QTP has a sparse population distribution overall; in large areas of the northern QTP, the population density is zero, accounting for about 14% of the total area of the QTP. About half of the QTP showed a rapid increase in population density between 2000 and 2016, mainly located in the eastern and southern parts of Qinghai Province and the central-eastern parts of the Tibet Autonomous Region. Regarding the relative importance of variables in explaining population density, the variables “Distance to Temples” is the most important, followed by “Density of Villages” and “Elevation”. Furthermore, our new products exhibited higher accuracy compared with five recently released gridded population density datasets, namely WorldPop, Gridded Population of the World version 4, and three national gridded population datasets for China. Both the root-mean-square error (<i>RMSE</i>) and mean absolute error (<i>MAE</i>) for our products were about half of those of the compared products except for WorldPop. This study provides a reference for using fine-scale demographic count and local population-sensitive POIs to model changing population distribution in remote rural areas.