Geospatial artificial intelligence for estimating daytime and nighttime nitrogen dioxide concentration variations in...

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Geospatial Analysis

Background:

• Nitrogen dioxide (NO2) is a significant air pollutant from traffic and industry.
• Existing models often underestimate NO2 exposure due to unaddressed daily patterns.

Purpose of the Study:

• To develop high-resolution spatial models for NO2 concentrations.
• To investigate the diurnal and seasonal variations of NO2.
• To assess NO2 exposure changes during COVID-19 restrictions.

Main Methods:

• Ensemble mixed spatial models (EMSM) integrating kriging, land use regression, and machine learning.
• Geo-artificial intelligence (Geo-AI) applied to 26 years of NO2 data, meteorological, and geospatial variables.
• Separate models developed for daytime and nighttime NO2 variations.

Main Results:

• High model reliability with adjusted R² of 0.92 (daytime) and 0.93 (nighttime).
• Significantly higher mean NO2 concentrations at nighttime (9.60 ppb) versus daytime (5.61 ppb).
• Winter season showed the highest NO2 levels.

Conclusions:

• EMSMs provide accurate, high-resolution NO2 mapping, revealing critical nighttime peaks.
• Findings enhance understanding of NO2 exposure dynamics for public health and policy.
• Generated maps illustrate NO2 levels before and during COVID-19 restrictions in Taiwan.