Predicting Nitrous Oxide Emissions from China’s Upland Fields Under Climate Change Scenarios with Machine Learning

Abstract

Upland fields are a significant source of N2O emissions. Thus, an accurate estimation of these emissions is essential. This study employed four classical modeling approaches—the Stepwise Regression Model, Decision Tree Regression, Support Vector Machine, and Random Forest (RF)—to simulate soil N2O emissions from Chinese upland fields. The upland crops considered in this study covered food crops, oil crops, cash crops, sugar crops, fruits, and vegetables, excluding flooded rice. Comparative analysis revealed that the RF algorithm performed the best, with the highest R2 at 0.66 and the lowest root mean square error at 0.008 kg N2O ha−1 day−1. The application rate of mineral nitrogen fertilizers, mean temperature during the growing season, and soil organic carbon content were the key driving factors in the N2O emission model. Utilizing the RF model, total N2O emissions from Chinese upland fields in 2020 were estimated at 183 Gg. Future projections under Representative Concentration Pathway (RCP) scenarios indicated a 2.80–5.92% increase in national N2O emissions by 2050 compared to 2020. The scenario analysis demonstrated that the proposed nitrogen reduction strategies fail to counteract climate-driven emission amplification. Under the combined scenarios of RCP8.5 and nitrogen reduction strategies, a net 4% increase in national N2O emissions was projected, highlighting the complex interplay between anthropogenic interventions and climate feedback mechanisms. This study proposes that future attention should be paid to the development of nitrogen optimization strategies under the impact of climate change.