Inferring Surface NO2Over Western Europe: A Machine Learning Approach With Uncertainty Quantification

Nitrogen oxides (NOx= NO + NO2) are of great concern due to their impact on human health and the environment. In recent years, machine learning (ML) techniques have been widely used for surface NO2estimation with rapid developments in computational power and big data. However, the uncertainties inherent to such retrievals are rarely studied. In this study, a novel ML framework has been developed, enhanced with uncertainty quantification techniques, to estimate surface NO2and provide corresponding data‐induced uncertainty. We apply the Boosting Ensemble Conformal Quantile Estimator (BEnCQE) model to infer surface NO2concentrations over Western Europe at the daily scale and 1 km spatial resolution from May 2018 to December 2021. High NO2mainly appears in urban areas, industrial areas, and roads. The space‐based cross‐validation shows that our model achieves accurate point estimates (r= 0.8, R2= 0.64, root mean square error = 8.08 μg/m3) and reliable prediction intervals (coverage probability, PI‐50%: 51.0%, PI‐90%: 90.5%). Also, the model result agrees with the Copernicus Atmosphere Monitoring Service (CAMS) model. The quantile regression in our model enables us to understand the importance of predictors for different NO2level estimations. Additionally, the uncertainty information reveals the extra potential exceedance of the World Health Organization (WHO) 2021 limit in some locations, which is undetectable by only point estimates. Meanwhile, the uncertainty quantification allows assessment of the model's robustness outside existing in‐situ station measurements. It reveals challenges of NO2estimation over urban and mountainous areas where NO2is highly variable and heterogeneously distributed. Inferring surface NO2concentrations is an effective way to monitor and mitigate NOxpollution which is of great concern due to its impact on human health and the environment. Machine learning (ML) techniques have been widely used for surface NO2estimation with rapid developments in computational power and big data. However, such estimations can be uncertain due to inherent errors in the data, and this uncertainty is rarely studied. We develop a novel ML framework to estimate surface NO2concentrations and provide corresponding uncertainty information. We infer surface NO2levels over Western Europe at the daily scale and 1 km spatial resolution from May 2018 to December 2021. Our model's performance is reliable as verified by in‐situ station measurements and an independent physics‐based model. We observe NO2hotspots over urban areas, industrial areas, and major roads. The uncertainty quantification (UQ) techniques allow us to analyze the influence of different input data on estimating different NO2levels. The UQ also helps to identify potential NO2exceedances of the WHO 2021 limit, which have not been observed in previous research. Additionally, we assess the model's robustness outside of in‐situ stations and witness the challenge of NO2estimation over urban and mountainous areas. A novel and reliable machine learning model with uncertainty quantification is applied to infer surface NO2levels over Western EuropeOur work uncovers how predictors impact model inference of various surface NO2levels differentlyOur approach identifies areas of high uncertainty in surface NO2mapping and potential environmental risks from overlooking uncertainty A novel and reliable machine learning model with uncertainty quantification is applied to infer surface NO2levels over Western Europe Our work uncovers how predictors impact model inference of various surface NO2levels differently Our approach identifies areas of high uncertainty in surface NO2mapping and potential environmental risks from overlooking uncertainty