The Evolution of Global Surface Ammonia Concentrations during 2001–2019: Magnitudes, Patterns, and Drivers

Abstract

Ammonia (NH₃) is the most prevalent alkaline gas in the atmosphere, with its elevated concentrations posing significant adverse impacts on air quality, ecosystems, and human health across diverse spatial and temporal scales. Given the ongoing global change and intensified anthropogenic NH₃ emissions, it is projected that the global surface NH₃ concentration will escalate further. Here, based on ground observations, gridded data of organic and inorganic nitrogen fertilizer applications, meteorological data, and ancillary information, we estimated changes in global monthly surface NH₃ concentration during 2001–2019 at a 0.1°× 0.1° resolution. A novel scale-adaptive approach, essentially an Ensemble Random Forest Model built upon Rotated Quadtree Partitioning and Box-Cox Transformation, was developed. The model well reproduced the spatial and temporal patterns of surface NH₃ observations, particularly capturing peak and valley values (R² = 0.91 and slope = 0.82 for the whole; R² = 0.79 and slope = 0.70 for testing). The results indicate a global increase in surface NH₃ concentration over 2001–2019, from 1.44 μg m^–3 yr^–1 in 2001 to 1.51 μg m^–3 yr^–1 in 2019. Notably, hotspots of elevated NH₃ concentrations were located in northern South Asia, northern China, the Sahel area, southeast South America, and central United States. Decreased SO₂ emissions and increased fertilizer applications dominated the increase of surface NH₃ concentrations in China, while in South Asia, the increase was primarily driven by organic and inorganic nitrogen inputs. Temperature changes were identified to play an important role in affecting surface NH₃ concentrations in most regions, particularly in Africa, South America, and Oceania. These findings have the potential to facilitate research on global nitrogen cycle and its environmental footprints and inform the development of locally or regionally tailored nitrogen management strategies. Furthermore, the proposed modeling algorithm showcases its capability in capturing intricate patterns and relationships within highly spatially heterogeneous data, thereby addressing up-scaling challenges associated with multimodal site observations.