Global sensitivity analysis by means of EFAST and Sobol’ methods and calibration of reduced state-variable TOMGRO model using genetic algorithms

Abstract

One common constraint for using crop models for decision making in precise greenhouse crop management is the need for accurate values of model parameters depending on climate conditions, crop varieties, and management. Estimating these parameters from observed data on the crop, using a crop model, is an interesting possibility. Nevertheless, the accuracy of estimations depends on the sensitivity of the model output variables to the parameters. Therefore, this paper proposes the use of the reduced state variable TOMGRO model which describes nodes, leaf area index, total plant weight, total fruit weight, and mature fruit weight as states variables. The objective of this work was to compare EFAST and Sobol’ sensitivity analysis methods to determine the most sensitive parameters for TOMGRO model outputs. A former sensitivity analysis showed that 8 parameters were the most sensitive and they were calibrated using genetic algorithms (GAs) to adapt the model to semi-arid weather conditions of Central Mexico. Genetic algorithms are important heuristic search algorithms for optimization problems and have been used to calibrate non-linear models related to control of greenhouse climate conditions. Simulation and analysis of the TOMGRO model showed that the estimations for the state variables are close to the measured data. The model could be adapted for simulating other greenhouse crops by means of sensitivity analysis and calibration.