The profiled parameter estimation method for dynamic systems applied to the calibration of dynamic crop growth models

Abstract

Parameter estimation, model calibration, or curve fitting is a crucial stage of the development process of dynamic models that account for crop growth and development. It consists of using measured data collected from a system or process to fit model predictions. Up to now, the frequentist and the Bayesian methods have been mostly applied to the calibration of crop growth dynamic models. A new approach named profiled estimation has been developed to try to overcome the limitations of the classical parameter estimation approaches. But crop growth modelers have not applied this method yet. Thus, the objectives of this work were: i) to show that the profiled estimation procedure (PEP) can be applied to the parameter estimation of crop growth models that use ordinary differential equations and ii) to compare the performance of PEP against a frequentist parameter estimation method that used a differential evolution (DE) algorithm. As both approaches were applied to a simple maize model with three state variables and seven parameters, the PEP method performed better than the frequentist method according to RMSE, MAE, and modeling efficiency statistics. Also, a dynamic model for the growth of a lettuce greenhouse crop with three state variables and twelve model parameters was calibrated by PEP and DE. In this case, the PEP performed poorly for the calibration of more influential model parameters but acceptably in the case of the estimation of all the parameters. Still, the results from this research show that crop growth modelers should consider the PEP in the calibration of their models.