Four years of soybean experimental data observed at Daxing, North China Plain, were used to assess the ability of the AquaCrop model to predict soybean final biomass and yield. The model was parameterized and calibrated using field data on leaf area index (LAI), available soil water, soil evaporation, biomass and final yield data. The model was assessed using calibrated and default parameters. Data on LAI were used to derive the fraction of ground cover and to calibrate the green canopy cover (CC) curve. An accurate calibration of the CC curve was performed, with low root mean square errors (RMSE<7.3%). Results relative to soil water balance simulations show a high variability of the predictions, thus a bias of the estimation, with R2 ranging 0.22–0.86 and low Nash-Sutcliffe efficiency EF, ranging between −0.47 and 0.82. The estimation errors were relatively high, with RMSE not exceeding 22.9mm. AquaCrop was compared with the soil water balance model SIMDualKc, that has shown better performance with R2≥0.83, EF generally greater than 0.75 and RMSE smaller than 12.5mm. The soil evaporation (Es) simulations were compared with the observations performed using microlysimeters; results for Aquacrop have shown a clear trend for under-estimation of Es, with “goodness-of-fit” results worse than for SIMDualKc (Wei et al., 2015). In general, AquaCrop has shown serious limitations to estimate crop transpiration or soil evaporation, which is likely due to abandoning the FAO dual Kc approach. However, the model performed well relative to biomass and yield predictions, with a yield RMSE of 302kgha−1. Overall, results show the adequacy of AquaCrop for estimating soybean biomass and yield when the model is appropriately parameterized. However, AquaCrop is not appropriate to support irrigation scheduling.
