A multiple and ensembling approach for calibration and evaluation of genetic coefficients of CERES-Maize to simulate maize phenology and yield in Michigan

Abstract

The phenological and growth parameters of CERES-Maize were estimated using data from field variety trials in 2017 and 2018 to simulate hybrid maize (Zea mays L.) grown in Michigan. Multiple calibration methods used include GENCALC (Genotype Coefficient Calculator), GLUE (Generalized Likelihood Uncertainty Estimate), NMCGA (Noisy Monte Carlo Genetic Algorithm) and ensembling approach. Three irrigated sites were used for calibration while six rainfed sites for evaluation. Better results were obtained when using multiple years of data in calibration than using only a single year. Model evaluation also suggests that fixed soil root growth factor (SRGF) used in calibration (irrigated condition) tended to restrict root dynamics under rainfed condition. This resulted in substantial yield mismatch due to poorly modeled yields, although phenology was better predicted. Adjusting SRGF under rainfed condition resulted in better model evaluation for both years. Moreover, weighted averaging of genetic coefficients resulted in better predictions of phenology and yields.