Leveraging Deep Learning for Physical Model Bias of Global Air Quality Estimates

2025-08-06

AI Summary

Global modeling of surface ozone—particularly at urban scales—remains challenged by substantial biases in physics-based models and poorly understood driving mechanisms, limiting policy relevance. This paper proposes a physics-informed deep learning correction framework: a 2D convolutional neural network (CNN) is employed to model residuals of a chemical transport model (CTM), and, for the first time, high-resolution satellite-derived land-use data are systematically integrated as critical covariates to enable synergistic optimization of multi-source remote sensing observations and mechanistic modeling. Validation across North America and Europe demonstrates that the method significantly outperforms conventional machine learning correction approaches, reducing bias by up to 32%. Furthermore, it identifies urban surface heterogeneity, local emissions, and meteorological feedbacks as dominant contributors to model bias. This work establishes a novel paradigm for high-accuracy ozone mapping and targeted pollution mitigation.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Air pollution is the world's largest environmental risk factor for human disease and premature death, resulting in more than 6 million permature deaths in 2019. Currently, there is still a challenge to model one of the most important air pollutants, surface ozone, particularly at scales relevant for human health impacts, with the drivers of global ozone trends at these scales largely unknown, limiting the practical use of physics-based models. We employ a 2D Convolutional Neural Network based architecture that estimate surface ozone MOMO-Chem model residuals, referred to as model bias. We demonstrate the potential of this technique in North America and Europe, highlighting its ability better to capture physical model residuals compared to a traditional machine learning method. We assess the impact of incorporating land use information from high-resolution satellite imagery to improve model estimates. Importantly, we discuss how our results can improve our scientific understanding of the factors impacting ozone bias at urban scales that can be used to improve environmental policy.

Problem

• Estimating surface ozone model bias using deep learning
• Improving air quality models for urban-scale health impacts
• Enhancing ozone trend understanding for better environmental policy

Innovation

• 2D CNN for ozone model bias estimation
• High-resolution satellite land use data
• Improved urban-scale ozone bias understanding