2020/05/19 by Kyle P. Messier, Messier, Kyle P, Matthias Katzfuß +1
Environmental Science · Social Sciences · #Air Quality and Health Impacts #Applications (stat.AP) #FOS: Computer and information sciences #Health, Environment, Cognitive Aging #Urban Transport and Accessibility
paper · pdf · doi:10.48550/arxiv.2005.09210
openalex publication_date 2020/05/19 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Nitrogen dioxide (NO2) is a primary constituent of traffic-related air\npollution and has well established harmful environmental and human-health\nimpacts. Knowledge of the spatiotemporal distribution of NO2 is critical for\nexposure and risk assessment. A common approach for assessing air pollution\nexposure is linear regression involving spatially referenced covariates, known\nas land-use regression (LUR). We develop a scalable approach for simultaneous\nvariable selection and estimation of LUR models with spatiotemporally\ncorrelated errors, by combining a general-Vecchia Gaussian process\napproximation with a penalty on the LUR coefficients. In comparisons to\nexisting methods using simulated data, our approach resulted in higher\nmodel-selection specificity and sensitivity and in better prediction in terms\nof calibration and sharpness, for a wide range of relevant settings. In our\nspatiotemporal analysis of daily, US-wide, ground-level NO2 data, our\napproach was more accurate, and produced a sparser and more interpretable\nmodel. Our daily predictions elucidate spatiotemporal patterns of NO2\nconcentrations across the United States, including significant variations\nbetween cities and intra-urban variation. Thus, our predictions will be useful\nfor epidemiological and risk-assessment studies seeking daily, national-scale\npredictions, and they can be used in acute-outcome health-risk assessments.\n