Spatial Bayesian Fusing Models for Sparse Networks and Health Risk

• Activity Number: 57 - Developments in Bayesian Spatial and Spatio-Temporal Modeling of Small Area Health Data
• Type: Topic Contributed
• Date/Time: Sunday, July 29, 2018 : 4:00 PM to 5:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Andrew B Lawson* and Raymond Boaz
• Companies: Medical University of South Carolina and Medical University of South Carolina
• Keywords: Bayesian; spatial; health; fusing; downscale; CMAQ

Abstract

Multivariate estimation of air pollution fields is well developed. Assimilation of different sources of data on a large scale has also been a focus. However, in many situations there is a need to be able to handle estimation of sparsely sampled components where the multivariate mixture can vary in its completeness across different sampling densities. In addition, the balance between downscaled fusion product use (e.g. CMAQ) and monitor sites adds another dimension to the estimation problem. In this talk we will demonstrate a novel sparse interpolation and prediction approach to situations where variable densities of sites and variable numbers of pollutants are found at sites. The approach is fully Bayesian and utilises dense/sparse monitors and complete and incomplete pollutant sets. Some novel spatio-temporal models are proposed to handle sparsity and imputation. A Case study of South Carolina multi -pollutant estimation is demonstrated.