Performance assessment of Hybrid-Maize model for rainfed, limited and full irrigation conditions

Abstract

Hybrid-Maize model was evaluated for its performance for simulating maize growth, grain yield, soil water content (SWC) and crop evapotranspiration (ETc) under full irrigation treatment (FIT), limited irrigation (50, 60 and 75 % of FIT) and rainfed using long-term (2005–2010) and field-measured data in Nebraska, USA. Current version of the Hybrid-Maize model was found to be unsatisfactory for estimating maize growth and water balance components or grain yield. In most cases, significant differences (p<0.05) were observed between model-estimated vs. field-measured variables. Leaf area index (LAI) was simulated with modest accuracy for most of the irrigated treatments [with a normalized root mean squared error (NRMSE) ≤ 16 % in 2009 and ≤ 30 % in 2010] in which a trend of underestimation was observed. The model simulated early season biomass well in 2010, but overpredicted in 2009 and underpredicted in 2010 during the late season. The total seasonal biomass for both years was simulated with very high overall NRMSE of 20.2 % and mean prediction error (MPE) of 0.28. The SWC estimates by the model were highly unsatisfactory in almost all treatments and years with negative EF values and high NRMSEs, ranging from 47 % to 62 %. This bias in SWC prediction resulted in significant (p<0.05) underestimation of seasonal ETc by the model in all years and treatments. The difference in simulated and field-measured ETc varied from 46 mm to as high as 167 mm, corresponding to the prediction error (Pe) range of 7.7–28 % with pooled data NRMSE of 16.2 %. These are very high ETc estimation errors that can result in significant challenges if the model is used for predicting water supply, demand and use analyses or in-season irrigation management. The final grain yield was simulated with moderate accuracy in some cases and poorly in other cases with up 1.2 MG/ha of underestimation for FIT, 2 Mg/ha of overestimation for limited irrigation treatment and up to 5.7 Mg/ha underestimation for rainfed conditions with a pooled data R2 of 0.69, NRMSE of 14 % and EF value of 0.57. Overall, model estimates were inconsistent for the same variables/treatments between the years. Additional improvements in the model to add additional soil layers input capability may aid in more representative/accurate estimates of SWC. The highest discrepancies between simulated and measured data were observed for rainfed treatments, indicating unsuitability of the model in rainfed/dry regions (and in irrigated conditions for most variables) and the need for model improvements in conditions of water stress. For most of the variables investigated, additional model improvements using carefully measured field data are suggested.