Fusing physics and AI enhances accuracy and diagnostic capabilities in air quality forecasting.
11 affiliations
- 1State Key Laboratory of Atmospheric Environment and Extreme Meteorology, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029, China.
- 2Earth System Numerical Simulation Science Center, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029, China.
- 3College of Earth and Planetary Sciences, University of Chinese Academy of Sciences, Beijing 100049, China.
- 4College of Environmental Sciences and Engineering, Peking University, Beijing 100871, China.
- 5School of Mathematical Sciences and Center for Statistical Science, Peking University, Beijing 100871, China.
- 6Department of Geography, Hong Kong Baptist University, Hong Kong 999077, China.
- 7Gaoling School of Artificial Intelligence, Renmin University of China, Beijing 100872, China.
- 8China National Environmental Monitoring Center, Beijing 100012, China.
- 9Aerosol Chemistry Department, Max Planck Institute for Chemistry, Mainz 55128, Germany.
- 10Institute of Surface-Earth System Science, Tianjin University, Tianjin 300072, China.
- 11Center for Global and Regional Environmental Research, University of Iowa, Iowa City, IA 52242, USA.
Abstract
Accurate air quality forecasts with diagnostic value are pivotal for emission-control policies and public health, yet chemical transport models (CTMs) are hampered by uncertainties in physicochemical parameterizations, whereas AI-only approaches sacrifice interpretability for predictive power. Here, we introduce ProNet, a coupled forecasting framework that embeds observation-constrained neural correctors directly into the governing equations of CTMs. Rather than treating AI as an external postprocessor or a process-replacing surrogate, ProNet integrates trainable networks into physical simulations to systematically reduce process-level uncertainties, suppress error propagation and maintain physical constraints of mass transport. Applied to 2022-23 data across 338 Chinese cities, ProNet consistently outperforms physical CTMs and AI-only baselines, and better generalizes to extreme events beyond the training data. It also surpasses forecaster-adjusted operational forecasts, which often improve upon raw CTM guidance in severe episodes, improving accuracy by 6.1% for PM2.5 and 3.0% for O₃. Notably, ProNet provides error bookkeeping, allowing forecasters to quantitatively attribute simulation errors to specific space and time. In-depth analysis of these process errors reveals systematic biases in current CTM parameterizations, providing actionable insights to refine the physical and chemical modules of numerical models. These results highlight the promise of physics-AI coupling for advancing air quality forecasting and deepening our understanding of the complex Earth system.
Related
- Is Seismic Forecasting Possible with Physics-based AI?
- FAST-ML: A Hybrid Physics-Machine Learning Framework for Tropical Cyclone Intensity Forecasting
- Dual-constraint physics-informed LSTM-XAJ hybrid model for 10-minute flood forecasting in small-to-medium basins: A case of the Xiang River Basin, China
- PhyTGNet: A Physics‐Prior‐Guided Interpretable Forecasting Method for Thunderstorm Gusts in the Beijing–Tianjin–Hebei Region
- A physics guided hybrid modeling framework for rainfall runoff forecasting in data scarce regions
BibTeX
@article{wu2026fusing,
title = {{Fusing physics and AI enhances accuracy and diagnostic capabilities in air quality forecasting.}},
author = {Huangjian Wu and Zifa Wang and Yuanhang Zhang and Wei Lin and Xinwei Li and Meng Gao and Hao Sun and Dawei Zhang and Wei Wang and Xiao Tang and Jiang Zhu and Yu Song and Chaoqun Ma and Pingqing Fu and Hang Su and Yafang Cheng and Gregory R Carmichael},
journal = {National science review},
year = {2026},
month = sep,
doi = {10.1093/nsr/nwag570},
url = {https://europepmc.org/article/MED/42899117},
note = {CC BY},
}
RIS
TY - JOUR
TI - Fusing physics and AI enhances accuracy and diagnostic capabilities in air quality forecasting.
AU - Huangjian Wu
AU - Zifa Wang
AU - Yuanhang Zhang
AU - Wei Lin
AU - Xinwei Li
AU - Meng Gao
AU - Hao Sun
AU - Dawei Zhang
AU - Wei Wang
AU - Xiao Tang
AU - Jiang Zhu
AU - Yu Song
AU - Chaoqun Ma
AU - Pingqing Fu
AU - Hang Su
AU - Yafang Cheng
AU - Gregory R Carmichael
PY - 2026
DA - 2026/09/10
JO - National science review
DO - 10.1093/nsr/nwag570
UR - https://europepmc.org/article/MED/42899117
AB - Accurate air quality forecasts with diagnostic value are pivotal for emission-control policies and public health, yet chemical transport models (CTMs) are hampered by uncertainties in physicochemical parameterizations, whereas AI-only approaches sacrifice interpretability for predictive power. Here, we introduce ProNet, a coupled forecasting framework that embeds observation-constrained neural correctors directly into the governing equations of CTMs. Rather than treating AI as an external postprocessor or a process-replacing surrogate, ProNet integrates trainable networks into physical simulations to systematically reduce process-level uncertainties, suppress error propagation and maintain physical constraints of mass transport. Applied to 2022-23 data across 338 Chinese cities, ProNet consistently outperforms physical CTMs and AI-only baselines, and better generalizes to extreme events beyond the training data. It also surpasses forecaster-adjusted operational forecasts, which often improve upon raw CTM guidance in severe episodes, improving accuracy by 6.1% for PM2.5 and 3.0% for O₃. Notably, ProNet provides error bookkeeping, allowing forecasters to quantitatively attribute simulation errors to specific space and time. In-depth analysis of these process errors reveals systematic biases in current CTM parameterizations, providing actionable insights to refine the physical and chemical modules of numerical models. These results highlight the promise of physics-AI coupling for advancing air quality forecasting and deepening our understanding of the complex Earth system.
SN - 2095-5138
SN - 2053-714X
ER -