资源环境科技发展态势分析平台

Despite research into the response of ammonia (NH₃) volatilization in farmland to various meteorological factors, the potential impact of future climate change on NH₃ volatilization is not fully understood. Based on a database consisting of 1063 observations across China, nonlinear NH₃ models considering crop type, meteorological, soil and management variables were established via four machine learning methods, including support vector machine, multi-layer perceptron, gradient boosting machine and random forest (RF). The RF model had the highest R² of 0.76 and the lowest RMSE of 0.82 kg NH₃-N ha⁻¹, showing the best simulation capability. Results of model importance indicated that NH₃ volatilization was mainly controlled by total input of N fertilizer, followed by meteorological factors, human managements and soil characteristics. The NH₃ emissions of China's cereal production (paddy rice, wheat and maize) in 2018 was estimated to be 3.3 Mt NH₃-N. By 2050, NH₃ volatilization will increase by 23.1−32.0% under different climate change scenarios (Representative Concentration Pathways, RCPs), and climate change will have the greatest impact on NH₃ volatilization in the Yangtze river agro-region of China due to high warming effects. However, the potential increase in NH₃ volatilization under future climate change can be mitigated by 26.1−47.5% through various N fertilizer management optimization options.