An artificial intelligence (AI) technology has emerged that can fill the observation gap in ammonia concentrations that cause ultrafine dust.
A research team led by Professor Lim Jeong-ho of the Department of Earth, Environmental and Urban Construction Engineering at the Ulsan National Institute of Science and Technology (UNIST) said on the 15th that it developed an AI model that can accurately estimate daily ammonia concentrations in the atmosphere. The findings were published in July in the international environmental journal Journal of Hazardous Materials.
Ammonia is emitted as a gas from agricultural fertilizers, livestock manure and fire sites. It is harmless on its own, but when it meets acidic substances such as sulfuric acid or nitric acid in the atmosphere, it forms ultrafine dust, so precise monitoring is essential for air quality forecasts and environmental policy making.
However, ammonia has a short residence time in the atmosphere, resulting in large concentration fluctuations, and ground observation stations are rare, so observation data can be obtained every two weeks. There are climate models that estimate ammonia concentrations by calculation, but because they target broad areas, regional prediction errors have been large.
The team built an AI model based on deep neural networks that can reinforce the frequency and accuracy of ammonia observations. They trained the model using climate data from the European Centre for Medium-Range Weather Forecasts and satellite data as inputs, and data from the U.S. ground observation network as the ground truth.
This AI model recorded prediction errors up to 1.8 times lower than the European climate model. Although it was trained with U.S. data as the ground truth, it also detected the high-concentration event caused by a major fire in the Manchester area of the United Kingdom in 2019. This is experimental evidence that demonstrates the model's spatial scalability and field applicability.
The team said, "Existing climate models that estimate ground-level ammonia concentrations by calculation had limits in accuracy, and measurements through ground observation stations had long data provision cycles," adding, "This model can compensate for the shortcomings of existing monitoring methods."
Lim Jeong-ho said, "It can be directly used for nitrogen-based pollutant air quality forecasts and environmental management policy making," adding, "In particular, in Korea, ammonia concentration monitoring is conducted only at limited locations, but applying the developed technology could build a high-resolution monitoring system."
References
Journal of Hazardous Materials (2025), DOI: https://doi.org/10.1016/j.jhazmat.2025.139166