We developed a near-real-time approach that consistently estimates the daily emissions of air pollutants. Compared to previous emission inventory, the new emission estimates better supported air quality simulation and efficiently detected short-term emission change due to unexpected events at the provincial level. By combining machine learning, moreover, the major sources of temporal variability of air quality were identified for effective policy making of air pollution controls.

Accurate maps of nitrogen dioxide pollution at fine scales are essential for assessing air quality and protecting public health. We developed a machine learning model that produces daily high-resolution 3D nitrogen dioxide fields across Western Europe by combining large-scale atmospheric simulations with ground-based measurements. This approach outperforms traditional methods, especially over cities and complex terrain, and can enhance satellite-based air quality monitoring.

High-resolution weather and climate simulations produce massive amounts of data, creating major storage challenges. This study explores a method that reduces unnecessary numerical detail by keeping only a limited number of significant digits. The results show that substantial data reduction can be achieved while preserving key physical features. 

We implement a new 12-km global nested simulation capability in GEOS-Chem, an open-source global 3-D model of atmospheric chemistry. Compared with the standard 25-km simulation, the 12-km simulation features stronger vertical transport due to better resolved horizontal convergence, along with improved representation of urban NO₂ and ozone titration. Application to methane emission inversion yields higher information content and resolves finer spatial structure in emission sectors.

This study explores the neglected role of atmospheric deposition in δ¹⁵N_4a-3s (δ¹⁵N difference between NH₄⁺ and source NH₃), a key parameter in atmospheric NH₄⁺ source tracing. We develop an integrated model coupling these two processes and find that the difference in δ¹⁵N_4a-3s between the combined effect and the pure fractionation scenario increases with f (NH₄⁺/NH_x molar ratio). This work provides a new idea for precise NH₄⁺ source partitioning.

We studied how environmental factors influence the release of mercury from seawater to the atmosphere. We applied a novel approach in which prior knowledge about cause-and-effect is captured as graphical models and then used to estimate effect sizes. Results showed that sunlight affects mercury both directly and indirectly, with about 32 % of the effect explained by temperature increase. Thus, causal models can improve our understanding of pollution processes and the effect policies.

Aviation impacts climate by forming contrails that trap heat and can persist for hours at cruising altitudes. Forecasting these humid regions is difficult, as satellites lack accuracy, aircraft data are limited, and ERA5 reanalysis has random errors. This study presents a hybrid machine learning method that corrects ERA5 with aircraft data, using decision trees in dry air and neural networks in humid air. It improves relative humidity predictions, especially in the lower stratosphere.

Gas-phase chemical mechanisms are key components of air quality models used by regulatory agencies for air quality and public health planning. We use modeled ozone concentrations and Ozone Production Efficiency (OPE) to compare four chemical mechanisms and find that OPE is a viable comparison metric under atmospheric conditions where nitrogen oxides are limited. Using OPE to predict how ozone responds to emissions reductions, however, is an oversimplification that can overstate ozone reductions.