Improved surface NO₂ Retrieval: Double-layer machine learning model construction and spatio-temporal characterization...

Area of Science:

• Atmospheric Chemistry and Physics
• Environmental Monitoring
• Machine Learning Applications

Background:

• Surface nitrogen dioxide (SNO2) is a critical air pollutant with significant health and environmental impacts.
• Existing SNO2 retrieval models inadequately account for the vertical structure of NO2, leading to inaccuracies.
• Conventional machine learning models struggle with the complex spatiotemporal dynamics of SNO2 and tropospheric NO2 columns (XNO2).

Purpose of the Study:

• To enhance the accuracy of SNO2 level inversion by integrating NO2 vertical stratification characteristics and spatiotemporal variation mechanisms.
• To develop and validate a novel Double-Layer Machine Learning (DLML) framework for estimating SNO2 levels across China.
• To analyze the temporal and spatial variation patterns of SNO2 levels in China from 2018 to 2023.

Main Methods:

• Development of a Double-Layer Machine Learning (DLML) framework utilizing Light Gradient Boosting Machine (LGBM) and Extremely Randomized Forests (ERF).
• Incorporation of NO2 vertical stratification and spatiotemporal variation mechanisms into the DLML model.
• Estimation of SNO2 levels across China for the period 2018-2023.

Main Results:

• The DLML model demonstrated superior performance with an R² of 0.87 in spatiotemporal cross-validation, a 10% improvement over traditional models.
• Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) were reduced to approximately 4.24 μg/m³ and 5.79 μg/m³, respectively.
• SNO2 levels exhibited a decreasing trend from central/eastern coastal areas to surrounding regions, aligning with WHO air quality guidelines; a U-shaped temporal variation was observed, peaking in winter (January, December) and reaching a minimum in summer (June-August).

Conclusions:

• The DLML framework effectively integrates NO2 vertical structure and spatiotemporal dynamics for accurate SNO2 estimation.
• Winter meteorological conditions significantly influence SNO2 variations, with regional economic and industrial activities contributing to elevated levels in areas like Wuhan and the Yangtze River Delta.
• The study provides a robust method for SNO2 monitoring, crucial for understanding and mitigating air pollution impacts.