Advancing operational PM2.5 forecasting with dual deep neural networks (D-DNet)

Abstract: PM2.5 forecasting is crucial for public health, air quality management, and policy development. Traditional physics-based models are computationally demanding and slow to adapt to real-time conditions. Deep learning models show potential in efficiency but still suffer from accuracy loss over time due to error accumulation. To address these challenges, we propose a dual deep neural network (D-DNet) prediction and data assimilation system that efficiently integrates real-time observations, ensuring reliable operational forecasting. It demonstrates notably higher efficiency than the Copernicus Atmosphere Monitoring Service (CAMS) 4D-Var operational forecasting system while maintaining comparable accuracy. Main Text: Accurately forecasting PM2.5 concentration (Particulate Matter with a diameter of 2.5 micrometers or smaller) is of paramount importance, given its significant impact on air quality and public health (1, 2). In general, atmospheric forecasting is a challenging task due to the complex and chaotic nature of the atmosphere system (3-5). In the context of PM2.5 forecasting, the complexity is compounded by the intricate interactions between various atmospheric processes, emissions, and depositions (6-8). Traditional forecasting methods, such as the atmospheric models in Copernicus Atmosphere Monitoring Service (CAMS), are based on the fundamental physical and chemical principles governing the emission, transformation, and transport of pollutants. The need to accurately represent numerous complex processes leads to significant computational demands (9, 10). Running these models in a timely manner, which is crucial for forecasting, requires access to high-performance computing resources (11, 12). To address these challenges, the scientific community has explored using advanced neural networks as complementary to physics-based models (13-15).