China's population spatialization based on three machine learning models. (20th May 2020)

• Record Type: Journal Article
• Authors: Zhao, Song; Liu, Yanxu; Zhang, Rui; Fu, Bojie
• Is Part Of: Journal of cleaner production. Volume 256(2020)
• Publication Date: 2020-05-20
• Subjects: Population spatialization -- Deep learning -- Convolutional neural network -- Geolocated social media data -- Nighttime light; Factory and trade waste -- Management -- Periodicals; Manufactures -- Environmental aspects -- Periodicals; Déchets industriels -- Gestion -- Périodiques; Usines -- Aspect de l'environnement -- Périodiques; 628.5
• DOI: 10.1016/j.jclepro.2020.120644
• Languages: English
• ISSNs: 0959-6526

Abstract

Abstract: Spatial demographic data are one of the most common type of basic data for sustainability research on a regional scale. Accurate and effective spatial downscaling of demographic data is required, which can provide basic data support for coupling the analysis of natural resource and social factors and could be a fundamental indicator for the spatial consumption of various products. In this study, geolocated social media, nighttime light, land use and terrain data were selected as factors that affect the population distribution. Convolutional neural network, deep neural network, and random forest models were used to spatialize the 2015 statistical population data of mainland China to a 1 km grid, and the spatialization results were compared with the published Gridded Population of the World (GPW) dataset and the WorldPop dataset for accuracy verification. The results show that the population spatialization result of the convolutional neural network model has the highest accuracy, and the average relative error is 24.4%; the accuracy of the deep neural network model is slightly higher than that of the random forest model but lower than that of the GPW dataset. The spatialization results of all the models are better than those of the WorldPop dataset. Consequently, deep learning can acquire and learn multisource data better than shallow machine learning and can achieve a higher quality of population spatialization, which can be an effective tool for downscaling socioeconomic data and provide basic support for sustainability research.