Predicting Intra-Urban Variation in Air Pollution Concentrations with Complex Spatio-Temporal Dependencies

Area of Science:

• Environmental Science
• Biostatistics
• Epidemiology

Background:

• Chronic exposure to air pollution is linked to cardiovascular disease.
• Accurate estimation of long-term air pollution concentrations is crucial for epidemiological studies.
• Existing methods may not fully capture complex spatio-temporal correlations in air pollution data.

Purpose of the Study:

• To develop and validate a methodology for estimating individual long-term average air pollution concentrations.
• To account for complex spatio-temporal correlation structures and misaligned observations.
• To support the Multi-Ethnic Study of Atherosclerosis and Air Pollution (MESA Air) in assessing air pollution's impact on cardiovascular health.

Main Methods:

• A hierarchical spatio-temporal model was developed, decomposing the air pollution field into a covariate-dependent mean and spatially correlated residuals.
• The model characterizes temporal trends using basis functions and employs universal kriging for spatial fields of coefficients.
• A scalable single-stage estimation procedure was implemented, accommodating missing data at monitoring locations.

Main Results:

• The methodology was applied to predict oxides of nitrogen (NOx) concentrations in Los Angeles from 2005-2007.
• Cross-validation yielded an R-squared value of 0.67, indicating good predictive performance.
• A simulation study demonstrated that supplementary monitoring data can further enhance prediction accuracy.

Conclusions:

• The proposed methodology provides accurate individual estimates of long-term average air pollution concentrations.
• This approach effectively handles complex spatio-temporal dependencies and data limitations.
• The MESA Air study will benefit from improved air pollution exposure assessments for cardiovascular disease research.