Scalable penalized spatiotemporal land-use regression for ground-level nitrogen dioxide

Area of Science:

• Environmental Science
• Public Health
• Statistical Modeling

Background:

• Nitrogen dioxide (NO2) is a major air pollutant from traffic with known health risks.
• Accurate spatiotemporal NO2 data is crucial for assessing exposure and health impacts.
• Existing land-use regression (LUR) models have limitations in handling complex spatial correlations.

Purpose of the Study:

• To develop a scalable and accurate method for estimating daily, nationwide NO2 concentrations.
• To improve variable selection and estimation in LUR models for air pollution.
• To provide reliable data for epidemiological and risk assessment studies.

Main Methods:

• Developed a scalable approach combining Gaussian-process approximation with penalized LUR coefficients.
• Implemented simultaneous variable selection and estimation for spatiotemporally correlated errors.
• Validated the method using simulated data and applied it to daily US-wide NO2 data.

Main Results:

• The new approach demonstrated superior model selection (specificity and sensitivity) and prediction accuracy (calibration and sharpness) compared to existing methods.
• Applied to US NO2 data, the method yielded more accurate, sparser, and interpretable models.
• Generated daily NO2 predictions revealing significant spatial variations within and between US cities.

Conclusions:

• The developed method offers a significant advancement in modeling air pollution spatiotemporal distribution.
• The accurate and interpretable NO2 predictions are valuable for public health research and risk assessment.
• This approach facilitates national-scale, daily exposure assessments for epidemiological studies.