Our study pioneers a PyTorch-based atmospheric assimilation system that leverages automatic differentiation and Deep Computing Unit acceleration to achieve order-of-magnitude speedups while establishing a direct pathway for future integration with deep learning.

The Singular Value Decomposition-three Dimensional Ensemble Variational data assimilation scheme is applied for the first time in the Tropical Regional Atmospheric Model System. With optimized three-dimensional perturbation generation and parallel strategies, computational costs were greatly reduced. Results indicate that the optimized scheme maintains reasonable accuracy while achieving much higher efficiency, suggesting good potential for practical forecasting use.

Our research presents a novel deep learning approach called "TemDeep" for downscaling atmospheric variables at arbitrary time resolutions based on temporal coherence. Results show that our method can accurately recover evolution details superior to other methods, reaching 53.7 % in the restoration rate. Our findings are important for advancing weather forecasting models and enabling more precise and reliable predictions to support disaster preparedness, agriculture, and sustainable development.

This study uses WRF-Chem to assess how meteorological conditions and emission reductions affected fine particulate matter (PM_2.5) in the North China Plain (NCP). It highlights regional disparities: in the northern NCP, adverse weather negated emission reduction effects. In contrast, the southern NCP featured a PM_2.5 decrease due to favorable weather and emission reductions. The research highlighted the interaction between emissions, meteorology, and PM_2.5.

This study developed a four-dimensional variational assimilation (4DVAR) system based on WRF–Chem to optimise SO₂ emissions. The 4DVAR system was applied to obtain the SO₂ emissions during the early period of the COVID-19 pandemic over China. The results showed that the 4DVAR system effectively optimised emissions to describe the actual changes in SO₂ emissions related to the COVID lockdown, and it can thus be used to improve the accuracy of forecasts.