Impact of parameter uncertainty and water stress parameterization on wheat growth simulations using CERES-Wheat with GLUE

This study used the Crop Environment Resource Synthesis (CERES)-Wheat model to explore the impact of parameter uncertainty and model structure on model output. We obtained observational and management data for winter wheat growth from an experiment conducted in Yangling, China, over 2012/13 and 2013/14, which we used for model input and evaluation. This experiment was conducted with full irrigation to avoid the effect of water stress on winter wheat growth. The Generalized Likelihood Uncertainty Estimation (GLUE) was used to generate 10,000 random parameter sets, including one calibrated parameter set, to explore the effects of parameter uncertainty on model output. Our results showed that GLUE-calibrated parameters were significantly different from observations. Further analysis indicated that frequent water stress occurred in the modeling results, even though no water stress actually occurred with full irrigation. This disagreement resulted mainly from the unrealistic water stress parameterization. GLUE-calibrated parameters matched very well with observations when this parameterization was closed in the CERES-Wheat model. Thus, the unrealistic water stress parameterization strongly affected the GLUE algorithm in locating calibrated parameters. The parameter sensitivity analysis demonstrated that the model errors produced by the water stress parameterization were compensated mainly by the parameters most sensitive to the winter wheat growth and yield simulations, such as the standard kernel weight and phyllochron. Hence, special attention should be paid to these parameters to identify possible structural defects in the model. In addition, results of the parameter uncertainty effect on model output showed that phenology-related simulations could better capture observations when multiple sets of parameters were used with and without water stress conditions. For the yield, maximum leaf area index, and final aboveground biomass, the model generally produced smaller biases without water stress than with water stress due to the unrealistic water stress parameterization. This study provides a better way for improving crop simulations and predictions.