Quantifying Spatiotemporal Heterogeneities in PM2.5-Related Health and Associated Determinants Using Geospatial Big Data: A Case Study in Beijing

Abstract

Air pollution has brought about serious challenges to public health. With the limitations of available data, previous studies overlooked spatiotemporal heterogeneities in PM2.5-related health (PM2.5-RH) and multiple associated factors at the subdistrict scale. In this research, social media Weibo data was employed to extract PM2.5-RH based on the Bidirectional Encoder Representations from Transformers (BERT) model, in Beijing, China. Then, the relationship between PM2.5-RH and eight associated factors was qualified based on multi-source geospatial big data using Geographically Weighted Regression (GWR) models. The results indicate that the PM2.5-RH in the study area showed a spatial pattern of agglomeration to the city center and seasonal variation in the spatially non-stationary effects. The impacts of varied factors on PM2.5-RH were also spatiotemporally heterogeneous. Specifically, nighttime light (NTL), population density (PD) and the normalized difference built-up index (NDBI) had outstanding effects on PM2.5-RH in the four seasons, but with spatial disparities. The impact of the normalized difference vegetation index (NDVI) on PM2.5-RH was significant in summer, especially in the central urban areas, while in winter, the contribution of the air quality index (AQI) was increased. This research further demonstrates the feasibility of using social media data to indicate the effect of air pollution on public health and provides new insights into the seasonal impacts of associated driving factors on the health effects of air pollution.