Expansion of Cultivars Applicable as a Soybean Crop Model and Evaluation of Model Errors in Yield Simulation

Abstract

To expand the cultivars applicable as a soybean crop model, we evaluated the accuracy of the model predicted yield for three cultivars: ‘Yukihomare’, ‘Satonohohoemi’ and ‘Sachiyutaka’. Varietal differences were incorporated into the genetic parameters of the phenological model as differences in earliness. The five major domestic cultivars, including the three newly applied cultivars, were grown by sowing in 2017 and 2018 in Tsukubamirai, Ibaraki, Japan. We measured the above-ground biomass at the beginning of seed filling (R5), as well as the seed yield and stem and pod biomass at maturity (R8) to evaluate the model outputs. The average of canopy cover ratio and radiation use efficiency from germination (VE) to R5 and the harvest index at R8 were also calculated. The relative root mean square error (rRMSE) for the seed yield of the three newly applied cultivars in 2017 was 15.8%, which is more accurate than previous calculations. However, the model overestimated the above-ground biomass at R5 and underestimated the harvest index at R8 for ‘Yukihomare’. In 2018, when there was little rainfall during the soybean growing period, the model severely overestimated the seed yield, with an rRMSE of 60.1%. This overestimation is attributed to the overestimations of the canopy cover ratio and the harvest index. For these traits, we need to account for the impacts of drought to improve the model performance.