Comparison of three models for simulating N2O emissions from paddy fields under water-saving irrigation

N2O emissions simulated by WNMM, DAYCENT and Crop-DNDC models were compared to the observed data sets from rice–wheat rotation systems under water-saving irrigation at Kunshan City, Jiangsu Province, China. On the basis of the correlation and paired t-test for evaluation, the simulation of N2O emission by Crop-DNDC and WNMM models provided better agreement with the observed data than by DAYCENT model. The daily time step, Crop-DNDC model was consistently the best for predicting daily N2O emissions (R2 = 0.892, n = 28, p > 0.05), and WNMM model performed better (R2 = 0.87, n = 28, p > 0.05). The Crop-DNDC model simulated the seasonal cumulative N2O emissions were the closest to the measured value of 1.07 kg N ha−1, and WNMM and DAYCENT models predicted 8.4% and 15.0% more N2O emissions than that in field experiments. The three models predicted well the seasonal cycle of soil temperature, soil moisture and could provide reliable estimations. The simulation of daily average soil temperature at 10 cm were consistently with the field observed data, which by Crop-DNDC (R2 = 0.92, n = 67, p > 0.05) and WNMM (R2 = 0.91, n = 67, p > 0.05). The comparison of observed to simulated results indicated that soil WFPS was simulated by Crop-DNDC (R2 = 0.52, n = 50, p > 0.05), WNMM (R2 = 0.56, n = 50, p > 0.05) and DAYCENT (R2 = 0.37, n = 50, p > 0.05). Accurate simulation of soil moisture, soil temperature and accurate partitioning of gaseous nitrogen loss into NO, N2O and N2 are challenges for all models.