A hybrid model based on machine learning and improved Jayaweera-Mikkelsen model to simulate NH3 volatilization in paddy field water

Abstract

NH₃ volatilization is one of the primary pathways of nitrogen loss in paddy field ecosystems, and accurately quantifying its flux remains a significant challenge, particularly under flooded conditions. To address this issue, we developed a hybrid model named NAU-_PNH₃. Based on the original Jayaweera-Mikkelsen (JM) model, this new model incorporates improvements in the NH_3(aq) concentration and key volatilization functions (K_gN and K_IN), accounting for the effects of solution activity coefficients, crop growth, and rainfall events. In addition, the model integrates machine learning (ML) algorithms to efficiently simulate the NH⁺₄-N concentrations (A_N) and pH in paddy field water, thereby enhancing overall model performance. We conducted a comprehensive evaluation of the original and improved models using field observations from four representative sites across major rice-producing regions in China. The results demonstrated that the NAU-_PNH₃ model outperforms the original JM model in simulating NH₃ volatilization fluxes from paddy fields, significantly reducing uncertainty and improving adaptability. Based on the NAU-_PNH₃ model, we developed the NAU-_PNH₃ Tool, an online platform for simulating daily NH₃ volatilization flux in paddy fields. This model offers a novel process-based approach for simulating NH₃ volatilization in paddy fields and provides an effective tool for understanding the mechanisms and dynamics of ammonia loss under varying environmental conditions.