Bayesian Calibration of AquaCrop Model

Abstract

The AquaCrop simulation model, modelling the dynamic change of crop growth status, is an important crop management tool for quantifying crop yield response to water. To effectively simulate the soil water balance and the crop growth process, a number of system parameters and canopy state variables are inevitably adopted. As a result, certain key parameters need to be calibrated so that the AquaCrop model can achieve a better performance of prediction for various scales of regions. This paper aims to apply Bayesian technique to calibrate the AquaCrop model. In this approach, the prior information regarding the system parameters is expressed in the form of a uniform probability distribution. Then with the advent of output variable measurement (e.g. biomass) by remote sensing techniques, the parameter distributions are iteratively updated by using Bayesian Markov Chain Monte Carlo (MCMC) method. The calibrated system parameters are expressed by the posterior distributions and gained by distribution mean value. Finally, the Bayesian calibration is compared with the conventional optimisation based calibration in terms of biomass and canopy cover, where simulated annealing is chosen as the optimisation approach, indicating a better calibration performance can be achieved by using Bayesian methods. Consequently, it is recommended that Bayesian calibration is one promising approach to the problem of crop model calibration.