Improved Daily Fertilizer-Induced Ammonia Emission Inventory for China by Combining Satellite Remote Sensing and Machine Learning

Abstract. Temporal variability of fine particulate matter (PM_2.5) pollution is strongly influenced by ammonia (NH₃) emissions, particularly their high-frequency variation driven by nitrogen fertilization. However, the unavailability of daily fertilization dates renders the spatiotemporal distribution of fertilizer-induced NH₃ emissions unclear, thereby limiting the accuracy of NH₃ emission estimates and PM_2.5 prediction. We developed a new method that identifies daily fertilization dates from multisource satellite observations and quantifies daily NH₃ emissions using machine learning. The identified fertilization dates correlated well with reported fertilization dates (mean correlation coefficient R = 0.96). Total NH₃ emissions from major crop fertilization were estimated at 3.01 Tg in 2021. The 10 % of days with the highest emissions accounted for 33.11 % of the annual total, with an average of 31.15 Gg d⁻¹ on these days. The new inventory substantially improved the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) simulation of PM_2.5, reducing the average root mean square error (RMSE) by 35.34 % during typical fertilization events. These daily datasets enable a shift from annual emission controls to temporally targeted mitigation strategies, and suggest that optimizing fertilization timing could mitigate severe PM_2.5 pollution.