Reevaluating the Drivers of Fertilizer-Induced N<sub>2</sub>O Emission: Insights from Interpretable Machine Learning
Xiaodong Ge, Danni Xie, Jan Mulder, Lei Duan
Abstract
Direct nitrous oxide (N 2 O) emissions from fertilizer application are the largest anthropogenic source of global N 2 O, but the factors influencing these emissions remain debated. Here, we compile 1134 observations of fertilizer-induced N 2 O emission factor (EF) from 229 publications, covering various regions and crops globally. We then employ an interpretable machine learning model to investigate the driving factors of fertilizer-induced N 2 O emissions. Our results reveal that pH, soil organic carbon, precipitation, and temperature are the most influential factors, overweighing the impacts of management practices. Nitrogen application rate has a positive impact on the EF, but the effect diminishes as nitrogen application rate increases, which has been overestimated in previous studies. Soil pH has three-stage influence on EF: positive when 7.3 ≤ pH ≤ 8.7, significantly negative between 6.8 and 7.3, and insignificant at lower pH levels (4.7 ≤ pH ≤ 6.8). Moreover, we confirm the nonlinear contributions of temperature and precipitation to EF, which may cause an unexpected increase in N 2 O emission under climate change. Our research provides crucial insights for global N 2 O modeling and mitigation strategies.