Revisiting the dynamics of gaseous ammonia and ammonium aerosols during the COVID-19 lockdown in urban Beijing using machine learning models.

The concentration of atmospheric ammonia (NH) in urban Beijing substantially decreased during the COVID-19 lockdown (24 January to 3 March 2020), likely due to the reduced human activities. However, quantifying the impact of anthropogenic interventions on NH dynamics is challenging, as both meteorology and chemistry mask the real changes in observed NH concentrations. Here, we applied machine learning techniques based on random forest models to decouple the impacts of meteorology and emission changes on the gaseous NH and ammonium aerosol (NH) concentrations in Beijing during the lockdown. Our results showed that the meteorological conditions were unfavorable during the lockdown and tended to cause an increase of 8.4 % in the NH concentration. In addition, significant reductions in NO and SO emissions could also elevate NH concentrations by favoring NH gas-phase partitioning. However, the observed NH concentration significantly decreased by 35.9 % during the lockdown, indicating a significant reduction in emissions or enhanced chemical sinks. Rapid gas-to-particle conversion was indeed found during the lockdown. Thus, the observed reduced NH concentrations could be partially explained by the enhanced transformation into NH. Therefore, the sum of NH and NH (collectively, NH) is a more reliable tracer than NH or NH alone to estimate the changes in NH emissions. Compared to that under the scenario without lockdowns, the NH concentration decreased by 26.4 %. We considered that this decrease represents the real decrease in NH emissions in Beijing due to the lockdown measures, which was less of a decrease than that based on NH only (35.9 %). This study highlights the importance of considering chemical sinks in the atmosphere when applying machine learning techniques to link the concentrations of reactive species with their emissions.