Predicting gestational personal exposure to PM2.5 from satellite-driven ambient concentrations in Shanghai. (October 2019)

• Record Type: Journal Article
• Authors: Zhu, Qingyang; Xia, Bin; Zhao, Yingya; Dai, Haixia; Zhou, Yuhan; Wang, Ying; Yang, Qing; Zhao, Yan; Wang, Pengpeng; La, Xuena; Shi, Huijing; Liu, Yang; Zhang, Yunhui
• Is Part Of: Chemosphere. Volume 233(2019)
• Page Start: 452
• Page End: 461
• Publication Date: 2019-10
• Subjects: PM2.5 -- Personal exposure -- Machine learning -- Remote sensing -- MAIAC AOD; AOD aerosol optical depth -- MODIS Moderate Resolution Imaging Spectroradiometer -- MAIAC Multi-Angle Implementation of Atmospheric Correction -- Shanghai MCPC Shanghai Maternal-Child Pairs Cohort -- CF cloud fraction -- PBL planetary boundary layer -- T temperature -- RH relative humidity -- CV cross-validation -- OOB out-of-bag; Pollution -- Periodicals; Pollution -- Physiological effect -- Periodicals; Environmental sciences -- Periodicals; Atmospheric chemistry -- Periodicals; 551.511
• DOI: 10.1016/j.chemosphere.2019.05.251
• Languages: English
• ISSNs: 0045-6535

Abstract

Abstract: Background: It has been widely reported that gestational exposure to fine particulate matters (PM2.5 ) is associated with a series of adverse birth outcomes. However, the discrepancy between ambient PM2.5 concentrations and personal PM2.5 exposure would significantly affect the estimation of exposure-response relationship. Objective: Our study aimed to predict gestational personal exposure to PM2.5 from the satellite-driven ambient concentrations and analyze the influence of other potential determinants. Method: We collected 762 72-h personal exposure samples from a panel of 329 pregnant women in Shanghai, China as well as their time-activity patterns from Feb 2017 to Jun 2018. We established an ambient PM2.5 model based on MAIAC AOD at 1 km resolution, then used its output as a major predictor to develop a personal exposure model. Results: Our ambient PM2.5 model yielded a cross-validation R 2 of 0.96. Personal PM2.5 exposure levels were almost identical to the corresponding ambient concentrations. After adjusting for time-activity patterns and meteorological factors, our personal exposure has a CV R 2 of 0.76. Conclusion: We established a prediction model for gestational personal exposure to PM2.5 from satellite-based ambient concentrations and provided a methodological reference for further epidemiological studies. Highlights: Gestational personal exposure to PM2.5 is predictable with novel machine learning approaches. Satellite-driven ambient PM2.5 concentrations is the dominant predictor and major contributor of personal exposure in Shanghai. Meteorological factors, socio-demographic status and time-activity patterns are also useful in predicting personal PM2.5 exposure.