Multivariate Spatial Prediction of Air Pollutant Concentrations with INLA

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Accurate daily air pollution estimates are crucial for public health research.
• Existing models often analyze pollutants in isolation, limiting comprehensive assessment.
• There's a need for integrated models that leverage pollutant relationships.

Purpose of the Study:

• To develop a spatial multipollutant data fusion model for improved air quality prediction.
• To create a comprehensive daily air pollution dataset for the contiguous United States.
• To provide model-based prediction error estimates.

Main Methods:

• Developed a spatial multipollutant data fusion model combining monitoring and chemical transport model data.
• Leveraged inter-pollutant dependencies to enhance spatial prediction accuracy.
• Utilized integrated nested Laplace approximation (INLA) for Bayesian inference.

Main Results:

• Generated a daily dataset for 12 pollutants (CO, NOx, NO2, SO2, O3, PM10, and PM2.5 species EC, OC, NO3, NH4, SO4) from 2005-2014.
• Achieved strong out-of-sample prediction performance, particularly for PM2.5 species (e.g., R2=0.84 for NH4).
• The model provides reliable prediction error estimates.

Conclusions:

• The developed model successfully integrates diverse data sources for accurate air pollution forecasting.
• This novel data product offers unprecedented spatial and temporal resolution for PM2.5 species and gases.
• The publicly available dataset will significantly aid epidemiological studies and health impact assessments.