Daily crop water uptake was simulated using algorithms from three crop simulation models, CropSyst,CERES, and EPIC (listed in order of decreasing process detail). Simulated results were compared,vith measurements of sap flow and soil water content far maize (Zea mays L.) growing at Prosser, WA, under a wet and a dry irrigation treatment, and with sail water content measurements for nonirrigated maize at Davis, CA. At Presser, the dry treatment imposed only a mild stress; at Davis, the stress was severe. Simulation variables such its maximum crop evapotranspiration, root density by soil layer, and green leaf area index were provided as daily input. At Presser, all algorithms performed similarly when simulating crop water uptake. For the met treatment, the root mean square error (RMSE) was 0.27 to 0.28 mm d(-1), and the relative error [RE = 100 (RMSE/Measured average)] was 7.0 to 7.2%. For the dry treatment, simulation accuracy decreased (RMSE = 0.33-0.38 mm d(-1); RE = 9.0-10.5%), The time evolution of mater uptake simulated by CropSyst better depicted the measured sap flow (water uptake) difference between wet and dry treatments. Simulations of soil water content by layer for the wet treatment, compared with measurements available for 17 d, yielded RMSEs from 0.022 to 0.024 m(3) m(-3) and REs from 8.5 to 9.2%, For the dry treatment (12 d of measurements), the best simulations were obtained with the water uptake algorithms from CropSyst and CERES, with RMSE = 0.015 m(3) m(-3) (both models) and RE = 6.4% (CropSyst) and 6.6% (CERES), compared with RMSE = 0.019 m(3) m(-3) and RE = 8.1% for EPIC, Under the severe water stress at Davis, CropSyst had the best performance. This algorithm simulated changes in soil wafer content by layer (8 d of measurements available) with RMSE of 0.011 m(3) m(-3) and RE of 5.0%, while the RMSE and RE values For CERES and EPIC were 0.016 and 0.019 m(3) m(-3) and 7.6 and 9.0%, respectively. The more process-oriented algorithm (CropSyst) showed an increasing advantage as water stress severity increased. The EPIC algorithm had the poorest performance under water stress. This could be improved by modifying the value of the water extraction distribution parameter in EPIC, but with this change the wet treatment simulations at Prosser deteriorated substantially, indicative of limitations in EPIC's simple approach.
