Thumbnail Image

Modeling growth curves using nonlinear mixed effects models with applications in agriculture

Zhou, Biting
The dissertation research was initially motivated by a collaborative project aimed at analyzing longitudinal agricultural data to describe and predict crop growth curves. As the escalating global population and the ensuing demand for food, the significance of optimizing agricultural production to guarantee food security has been highlighted. In order to achieve an efficient crop management, resource allocation, and yield forecasting, the precise estimation and prediction of crop development is gradually becoming an indispensable step. However, the presence of inter- and intra- individual/group variability in longitudinal data introduces challenges in statistical modeling. To address this issue, we developed nonlinear mixed effects models to successfully estimate the crop growth trajectories over time for single trait as well as for multiple traits simultaneously. In addition, to improve the prediction accuracy, we proposed a novel calibration approach to adjust the predicted values for the current year by the growth curve estimated from the last year, assuming the strong correlation between growth rates across years. The predicted values after calibration was shown to have less mean squared error. The improved accuracy was demonstrated by different scenarios based on the availability of data.