High-resolution mapping of demographic compositions based on remote sensing and social media data: a multimodal, multi-output modeling approach

Understanding fine-scale demographic compositions, such as sex and age structures, is fundamental to many fields, including public health, urban planning, and marketing. However, most existing fine-scale population products did not have spatial heterogeneities of demographic compositions within administrative units. This study, for the first time, developed a novel approach to estimate complete sex and age structures on 100 × 100 m grid cells, based on multimodal data including census, remote sensing imagery, and geotagged social media data. Key environmental variables derived from remote sensing imagery – such as artificial light intensity and vegetation index – and demographics-related variables extracted from social media texts were integrated using a multi-output random forest model, which, accounting for correlations among multiple demographic attributes, estimated the associations between the aforementioned variables and census-derived demographic compositions, including sex and age structures. The root mean square error (RMSE) of the final product was 6.18 for the sex ratio (the number of males per 100 females), and 1.89%, 3.82%, and 3.40% for the percentages over the total population of age groups 0–14, 15–59, and ≥60 years, respectively. This approach holds potential for detailed demographic modeling, which can be extended to other sociodemographic factors and benefit multiple disciplines.