Both empirical models and process-based models of ammonia (NH3) volatilization simulation in paddy field have some disadvantages. This work describes a semi-empirical semi-process model, which combines the simplicity of empirical models with the precision of process-based models, and its combinatorial structure is sufficient for reproducing the characteristic course of NH3 volatilization in paddy fields under different irrigation modes and urea application regimes over time. A dataset consisting of seven rice seasons across three locations was used for model development. Performance indexes showed the model is good (R2 values ranged from 0.67 to 0.94). The influence of air temperature and wind speed on NH3 volatilization was not as great as our conventional understanding, and soil temperature was a more accurate indicator than air temperature for model development. Model performance was better where the less urea splits and the higher urea rate, but there was no significant difference between alternate wetting and drying (AWD) irrigation mode (average values of R2, IA, NSI, and RMSE were 0.82, 0.94, 0.96, and 0.53 kg ha-1 d-1) and flooding irrigation (FI) mode (average values of R2, IA, NSI, and RMSE were 0.86, 0.94, 0.96, and 0.58 kg ha-1 d-1). The ammonium nitrogen (NH4+-N) concentration and pH of surface water and soil water in paddy fields had the greatest impact on NH3 volatilization, and they were mainly related to the amount of urea applied. Splitting the urea application and delaying the first urea application are the fundamental countermeasures to reduce NH3 volatilization. The magnitude of developed model performance varied among rice cropping systems, irrigation modes, urea application regimes, and climatic regions, highlighting the need to understand why measured NH3 volatilization varies among locations. The new model may be a useful tool for predicting fertilizer efficiency of paddy fields applied urea, assessing NH3 volatilization factors, and reducing emission.