Simulation of water uptake in maize, using different levels of process detail

Daily crop water uptake was simulated using algorithms from three crop simu lation models, CropSyst,CERES, and EPIC (listed in order of decreasing proc ess detail). Simulated results were compared,vith measurements of sap flow and soil water content far maize (Zea mays L.) growing at Prosser, WA, unde r a wet and a dry irrigation treatment, and with sail water content measure ments for nonirrigated maize at Davis, CA. At Presser, the dry treatment im posed only a mild stress; at Davis, the stress was severe. Simulation varia bles such its maximum crop evapotranspiration, root density by soil layer, and green leaf area index were provided as daily input. At Presser, all alg orithms 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 CropSys t better depicted the measured sap flow (water uptake) difference between w et and dry treatments. Simulations of soil water content by layer for the w et 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 treatm ent (12 d of measurements), the best simulations were obtained with the wat er 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 a t Davis, CropSyst had the best performance. This algorithm simulated change s 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. Th e more process-oriented algorithm (CropSyst) showed an increasing advantage as water stress severity increased. The EPIC algorithm had the poorest per formance under water stress. This could be improved by modifying the value of the water extraction distribution parameter in EPIC, but with this chang e the wet treatment simulations at Prosser deteriorated substantially, indi cative of limitations in EPIC's simple approach.