A Bayesian Multisource Fusion Model for Spatiotemporal PM2.5 and NO2 Concentrations: For Exposure Health Assessment in an Urban Setting
Conference
Format: IPS Abstract
Keywords: air pollution, bayesian, data fusion, methodology, spatiotemporal
Tuesday 3 December 9:30 a.m. - 11 a.m. (Australia/Adelaide)
Abstract
Exposure to high air pollution concentrations is a rising concern globally, with known risk factors for cardiovascular and respiratory health outcomes. Epidemiological studies are working to monitor, model, and predict the mortality and morbidity of high exposures worldwide, especially in large urban areas. These studies benefit from accurate and precise exposure estimates available at fine spatial resolutions across time. This study develops a Bayesian spatiotemporal fusion model for monthly fine particulate matter (PM2.5) and nitrogen dioxide (NO2) concentrations over Greater London from 2010 to 2019 at a 1km x 1km spatial resolution.
The base of the fusion model is validated ground-measured air pollution concentrations. Then, additional proxy data sources and predictive covariates are used to model concentrations across the spatial and temporal domains. Satellite-derived aerosol optical depth and tropospheric column NO2 are shown to support prediction across unmeasured locations and are available globally. Additionally, the fixed-effect covariates include temperature, humidity, and population, as well as indicators for urban/rural areas and proximity to large roads and industrial areas.
The model is based on the Integrated Nested Laplacian Approach (INLA) framework, implementing a hierarchical Bayesian spatiotemporal model coupled with a Stochastic Partial Differential Equation (SPDE) spatiotemporal process term. We then increase the model's flexibility by relaxing the stationary assumption, allowing spatially-varying covariate effects for meteorological and satellite variables. Models are assessed and compared through spatial and temporal cross-validation methods using both spatial subsetting and temporal forecasting.
The output of this work will be used to study the effect of air pollution exposure on children's mental health and wellbeing. Using a two-stage Bayesian approach, we use exposure estimates at the children's residence and school, along with exposure uncertainty, within an exposure-health model for various mental health outcomes.