Spatiotemporal model fine for predicting fine particulate matter concentration in the Metropolitan Area of São Paulo

Author(s)

Aline Santos Damascena

• Document type: Doctoral Thesis
• Press: São Paulo.
• Institution: Universidade de São Paulo (USP). Faculdade de Medicina (FM/SBD)
• Defense date: 2020-11-26

Examining board members

Paulo Hilario Nascimento Saldiva; Glauber Guimarães Cirino da Silva; Julio da Motta Singer

• Advisor: Paulo Hilario Nascimento Saldiva

Abstract

The spatiotemporal pattern of ne particulate matter (PM2:5) concentrations is an important factor in predicting some health outcomes of inhabitants in urban areas. PM2:5 concentrations are measured in air quality monitoring networks, which, in general, are sparse and do not represent the exposure of an entire population in a given region. From an epidemiological point of view, it is necessary to develop methodologies that allow us to estimate PM2:5 concentrations with high spatial and temporal resolution in places without monitoring stations, so that it is possible to study the short-term (days/weeks) and long term (years) outcomes. In this thesis, we applied a methodology proposed by researchers from the Harvard School of Public Health to predict the concentrations of PM2:5 in the Metropolitan Area of São Paulo (MASP), in the 2012-2017 period. In addition, we proposed changes in that methodology, in order to consider in the prediction models the two sources of correlation (spatial and temporal) inherent to the data, since the methodology proposed by Harvard considers only the spatial correlation. We also presented the standard deviations of the prediction errors associated with the estimates obtained in those models. The Harvard methodology is based on the Linear Mixed Model (LMM) and on the use of Aerosol Optical Depth (AOD) data obtained from satellites. AOD data have been shown to be a predictor of surface PM2:5 and/or PM10 concentrations in dierent regions of the world. With LMM it is possible to incorporate in the predictions the daily variation of AOD-PM2:5 relationship, caused by the variation of meteorological conditions. For the rst time in the MASP, we used AOD data with high spatial resolution (1 km2) from version 6 of the Multiangle Implementation of Atmospheric Correction (MAIAC) algorithm. Our results showed that, unlike what was observed in other regions where Harvard methodology has already been applied, AOD data from MAIAC were not predictive of surface PM2:5 and PM10 concentrations over the MASP. We found that the aerosol model used by MAIAC in the AOD retrievals does not adequately represent the aerosol properties of that region. However, the proposed prediction models that did not consider AOD data were able to explain, respectively, about 60% and 70% of the variability of diunal and daily PM2:5 concentrations over the MASP. The performance of the prediction models for daily PM2:5 concentrations were quite satisfactory and comparable to the performance of the prediction models developed in Switzerland, Israel and Mexico City, which used AOD data from MAIAC. In addition, the prediction models that considered the two correlation sources had about 5% greater predictive power than the models that considered only the spatial correlation, and also generated more accurate estimates. The maps with the annual of the estimates of diurnal and daily PM2:5 concentrations over the MASP were made available in an open access repository. Those estimates can be used as an approximation for the exposure to air pollution of the inhabitants in the MASP, contributing to epidemiological studies that aim to understand their impacts on the population\'s health (AU)

• FAPESP's process: 16/09411-1 - Spatio-temporal model for predicting the concentration of fine particulate matter (PM 2.5) in metropolitan area of São Paulo
• Grantee: Aline Santos Damascena
• Support Opportunities: Scholarships in Brazil - Doctorate