Evaluation of EPIC for assessing crop yield, runoff, sediment and nutrient losses from watersheds with poultry litter fertilization

Faced with limited comprehensive data on the economic, agronomic, and environment effects of land-applying animal wastes, water quality models are increasingly used to explore management and policy alternatives. However, thorough evaluation of these models is needed to assess their predictive ability for this resource issue. The EPIC (Environmental Policy Integrated Climate) model version 3060 was evaluated using data collected from six cultivated small watersheds (4.0 to 8.4 ha) near Riesel, Texas. The study watersheds were fallow in 2001, cropped with corn (Zea mays L.) in 2002 and 2003, and planted to winter wheat (Triticum aestivum L.) in 2004. A target poultry litter application rate from 0 to 13.4 Mg ha(-1) was randomly assigned to each of the watersheds. Monthly data of runoff, sediment, and soluble P for 2001-2002 from one watershed (Y13) were used to calibrate the initial CN2, erosion control practice factor, RUSLE C factor coefficient, and phosphorus sorption ratio. The modeling efficiency (EF) for the calibrated period was 0.90 for runoff, 0.65 for sediment, and 0.94 for soluble P. EPIC was validated using the 2001-2004 measured data for the other five watersheds and the remaining data for Y13. It successfully predicted surface runoff on an annual, monthly, and daily basis for all watersheds, with EF values larger than 0.5 and R2 larger than 0.7. The sediment, organic N and P, soluble P, and NO3-N losses simulated by EPIC were satisfactory, with EF values ranging from 0.59 to 0.87 based on annual comparisons and larger than 0.4 (in 25 out of 30 tests) based on monthly comparisons. EF was 0.96 for crop yields. Paired t-tests based on monthly comparisons of runoff, sediment and nutrient losses, and annual crop yields indicated that the differences between predicted and observed values were not significantly different from zero at the significance level of alpha = 0.05, except for soluble P losses for the control watershed. Both parametric and nonparametric statistical tests for EF values of monthly comparisons of runoff, sediment and nutrient losses, and percent errors of crop yields indicated that the reliability of the model was not significantly different among the poultry litter application watersheds and the control watershed, with the exception of soluble P losses for the control watershed. These statistical tests indicate that EPIC was able to replicate the runoff, water quality, and crop yield impacts of poultry litter application.