Multilayer Vertical Interactions of Air Pollution and Meteorological Drivers Revealed by a Unified Data-Driven Model

Area of Science:

• Atmospheric Science
• Environmental Science
• Artificial Intelligence

Background:

• Accurate vertical distribution of atmospheric pollutants is crucial for effective air quality management.
• Traditional numerical models and observational methods struggle to capture detailed vertical pollutant structures.

Purpose of the Study:

• To develop a unified machine learning (ML) framework for multipollutant vertical profiling and forecasting.
• To investigate the influence of meteorological factors and interlayer dependencies on pollutant vertical distribution.

Main Methods:

• Integration of long-term vertical observations from a tower building and lidar measurements.
• Development of a multilayer interactive machine learning framework.
• Analysis of meteorological factors, boundary layer height (BLH), and atmospheric thermal structure.

Main Results:

• Meteorological factors significantly improved 72-hour ozone (O3) forecast performance (correlation coefficient increase up to 0.29).
• Boundary layer height (BLH) and integrated temperature were key factors in upper atmospheric layers.
• Distinct pollutant dynamics were observed in lower (0-0.7 km) and upper (above 1.6 km) layers, governed by different meteorological parameters.
• The ML framework revealed asymmetric interlayer dependencies influencing forecast accuracy across layers.

Conclusions:

• The study provides novel insights into meteorological regulation and interlayer coupling of pollutant vertical structures.
• The developed ML framework offers a robust approach for three-dimensional air quality forecasting.
• Findings support the advancement of environment-oriented AI applications for air quality management.