A Hybrid Model Driven by Remotely-Sensed Data for Simulating the Impact of High Spatiotemporal Variations in Fertilization and Atmospheric N2O on N2O Emissions in Paddy Fields

Paddy fields with excessive fertilizer application are potential N2O emission hotspots, which profoundly affect the greenhouse effect. Unlike prior regional models assuming uniform fertilization and static atmospheric N2O concentration, this study focuses on simulating the effects of fertilization heterogeneity and dynamic atmospheric N2O concentration on daily field-scale N2O emissions. Accordingly, this study presents a hybrid model (NAU-RSP-N2O) that combines multi-source remotely-sensed data, ML algorithms, and the water-air gas exchange model to predict daily field-scale N2O emissions in paddies. Firstly, we built Bayesian-optimized ML models using critical water quality parameters to predict dissolved N2O and incorporated SHAP analysis for interpretability. Secondly, the remotely-sensed data (fertilization information, LST) were used to drive models of water quality parameters, capturing their complex spatiotemporal changes. Finally, the water-air interface gas model considering the dynamic atmospheric N2O concentration simulated daily field-scale N2O emissions in paddy fields. The model demonstrated an average R2 of 0.72, with MAE of 1.56 mg center dot m-2 center dot d-1 and RMSE of 1.65 mg center dot m-2 center dot d-1. The NAU-RSP-N2O model effectively simulated daily field-scale N2O emissions and spatiotemporal patterns, highlighting the critical role of nitrogen management and atmospheric N2O levels in controlling emissions. Our findings present a novel approach for the large-scale prediction of N2O emissions from paddy fields, applicable across diverse rice-growing regions in China.