Accuracy evaluation of the crop-weather yield predictive models of Italian ryegrass and forage rye using cross-validation

Abstract

The objective of this study was to evaluate the accuracy of the yield predictive models of Italian ryegrass (IRG, Lolium multiflorum Lam.) and forage rye (FR, Secale cereale L.) reported in previous studies through K-fold cross-validation method. In previous studies, statistical models were constructed for dry matter yield prediction of IRG and FR using general linear model based on climatic data by locations in the Republic of Korea. The yield predictive model for IRG cultivated in the southern region of the Korean Peninsula and Jeju Island were DMY = 78.178AGD–254.622MTJ + 64.156SGD–76.954PAT150 + 4.711SAP + 1028.295 + Location and DMY =–8.044AAT + 18.640SDS–7.542SAT + 9.610SAP + 17282.191, respectively. The yield predictive model for FR was as follows: DMY = 20.999AGD + 163.705LTJ + 113.716SGD + 64.379PAT100–4964.728 + Location. However, accuracy evaluation was not performed in the previous research. In this study, the reported models and the data set used for model construction were investigated. Subsequently, K-fold cross-validation was performed to assess the accuracy of the models. The results showed that the yield predictive models fit to the data sets well, while the accuracy of these models was in the common level since the data sources might keep major variances in cultivars, climatic conditions, and cultivated locations. Therefore, models with better fitness and accuracy might be constructed based on a data set with smaller variance. Hence, the standardization of the crop cultivation experiments is very necessary to decrease the variance in the historical data used for future crop yield modeling.