Offline Meteorology-Pollution Coupling Global Air Pollution Forecasting Model with Bilinear Pooling

TL;DR

The paper tackles the challenge of efficiently forecasting global air pollutants by introducing a DL-based offline coupling framework that fuses meteorological fields with pollutant data using bilinear pooling. By predicting pollutant changes from inputs at times $t-1$, $t$, and meteorology at $t$, $t+1$, and leveraging a pollutant base, the model achieves competitive accuracy with only about $13\%$ of the parameters of online DL approaches. Compared to CAMS, the offline bilinear approach yields improvements across a majority of variables, especially for forecasts longer than $48$ hours, and demonstrates a 15% relative RMSE reduction on average. This work establishes a practical, scalable paradigm for real-time global atmospheric forecasting that reduces computational burden while maintaining high predictive fidelity, with implications for rapid pollution warnings and policy support.

Abstract

Air pollution has become a major threat to human health, making accurate forecasting crucial for pollution control. Traditional physics-based models forecast global air pollution by coupling meteorology and pollution processes, using either online or offline methods depending on whether fully integrated with meteorological models and run simultaneously. However, the high computational demands of both methods severely limit real-time prediction efficiency. Existing deep learning (DL) solutions employ online coupling strategies for global air pollution forecasting, which finetune pollution forecasting based on pretrained atmospheric models, requiring substantial training resources. This study pioneers a DL-based offline coupling framework that utilizes bilinear pooling to achieve offline coupling between meteorological fields and pollutants. The proposed model requires only 13% of the parameters of DL-based online coupling models while achieving competitive performance. Compared with the state-of-the-art global air pollution forecasting model CAMS, our approach demonstrates superiority in 63% variables across all forecast time steps and 85% variables in predictions exceeding 48 hours. This work pioneers experimental validation of the effectiveness of meteorological fields in DL-based global air pollution forecasting, demonstrating that offline coupling meteorological fields with pollutants can achieve a 15% relative reduction in RMSE across all pollution variables. The research establishes a new paradigm for real-time global air pollution warning systems and delivers critical technical support for developing more efficient and comprehensive AI-powered global atmospheric forecasting frameworks.

Figures (6)

• Figure 1: Framework of our global air pollution forecasting.
• Figure 2: Weighted root mean square error (RMSE) comparison of CAMS model v.s. ours on pollutant variables of total column (TC) and particulate matter (PM). Normalized RMSE values based on CAMS results are given below each RMSE curves. (Best viewed in color.)
• Figure 3: The normalized RMSE score matrix relative to the CAMS model for the pollutant variables at all pressure levels and all lead days. Red represents good results. (Best viewed in color.)
• Figure 4: The normalized RMSE values relative to the CAMS model at 3/5 days of lead time. A lower normalized value indicated better performance.
• Figure 5: Case study of PM10 in North Africa, we list the forecast results for the next 12, 24 and 36 hours. The columns are respectively the groundtruths, predictions of our model and forecast results of CAMS. (Best viewed in color.)