Sensitivity and Uncertainty Analyses of Crop Yields and Soil Organic Carbon Simulated With Epic

Modeling biophysical processes is a complex endeavor because of large data requirements and uncertainty inmodel parameters. Model predictions should incorporate, when possible, analyses of their uncertainty and sensitivity. Thestudy incorporated uncertainty analysis on EPIC (Environmental Policy Impact Calculator) predictions of corn (Zea maysL.) yield and soil organic carbon (SOC) using generalized likelihood uncertainty estimation (GLUE). An automatic parameteroptimization procedure was developed at the conclusion of sensitivity analysis, which was conducted using the extendedFourier amplitude sensitivity test (FAST). The analyses were based on an experimental field under 34-year continuous cornwith five N treatments at the Arlington Agricultural Research Station in Wisconsin. The observed average annual yields pertreatment during 1958 to 1991 fell well within the 90% confidence interval (CI) of the annually averaged predictions. Thewidth of the 90% CI bands of predicted average yields ranged from 0.31 to 1.6 Mg ha-1. The predicted means per treatmentover simulations were 3.26 to 6.37 Mg ha-1, with observations from 3.28 to 6.4 Mg ha-1. The predicted means of yearly yieldover simulations were 1.77 to 9.22 Mg ha-1, with observations from 1.35 to 10.22 Mg ha-1. The 90% confidence width forpredicted yearly SOC in the top 0.2 m soil was 285 to 625 g C m-2, while predicted means were 5122 to 6564 g C m-2 andobservations were 5645 to 6733 g C m-2. The optimal parameter set identified through the automatic parameter optimizationprocedure gave an R2 of 0.96 for average corn yield predictions and 0.89 for yearly SOC. EPIC was dependable, from astatistical point of view, in predicting average yield and SOC dynamics.