Calibration of a crop growth model in APSIM for 15 publicly available corn hybrids in North America

Authors: WINN, CASSANDRA - Iowa State University; ARCHONTOULIS, SORTIRIOS - Iowa State University; Edwards, Jode

Submitted to: Crop Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/15/2022
Publication Date: 2/1/2023
Citation: Winn, C.A., Archontoulis, S.V., Edwards, J.W. 2023. Calibration of a crop growth model in APSIM for 15 publicly available corn hybrids in North America. Crop Science. 63(2): 511-534. https://doi.org/10.1002/csc2.20857.
DOI: https://doi.org/10.1002/csc2.20857

Interpretive Summary: Predicting which maize hybrids will be the most productive and stable performing in future production environments with varying weather and management regimes remains a very difficult problem in agriculture. Researchers in Ames, IA have tested a novel approach for evaluating differences among maize hybrids. Within a limited set of conditions, researchers were able to use crop growth models to predict up to 90% of the variation among maize hybrids. If the approach is validated across more variable environmental conditions, hybrids performance differences can be predicted in future years and with variable management regimes with much greater accuracy than currently possible.

Technical Abstract: Large-scale application of crop growth models (CGMs) in plant breeding is limited by the labor and cost required to measure cultivar-specific parameters. The objectives of this study were to 1) calibrate APSIM (Agricultural Production Systems sIMulator) for 15 publicly available maize hybrids and quantify prediction accuracy in modeling differences among genotypes; 2) better understand minimum data requirements for accurate model calibration; and 3) quantify simulated genotype by environment (GxE) interactions across many years for grain yield and five additional traits. We used two years of empirical field measurements to calibrate the 15 hybrids. Grain yield was simulated with an average R2 values of 0.89. Just over half of phenotypes measured were simulated with average R2 values over 0.8. Phenology parameters accounted for nearly half of the variability in grain yield and all of the variability in morphological traits such as leaf area index (LAI). Calibration with phenological data alone reduced normalized root mean square error (NRMSE) from 35 to 30% compared to an uncalibrated model on average across traits. Inclusion of additional physiological and N-related data further reduced average NRMSE to 20%. Reduction in NRMSE varied among hybrids and traits. Calibration reduced NRMSE for N-related traits from 42 to 17%, but only from 20 to 18% for LAI. Long-term simulations demonstrated distinct GxE interactions among the hybrids which accounted for 2-29% of the total genetic variation across traits. These model calibrations will be an important tool in bridging the gap between crop physiologists, plant breeders, and geneticists.