Global Sensitivity Analysis and Calibration by Differential Evolution Algorithm of HORTSYST Crop Model for Fertigation Management

Abstract

Sensitivity analysis is the first step in elucidating how the uncertainties in model parameters affect the uncertainty in model outputs. Calibration of dynamic models is another issue of considerable interest, which is usually carried out by optimizing an objective function. The first aim of this research was to perform a global sensitivity analysis (GSA) with Sobol’s method for the 16 parameters of the new HORTSYST nonlinear model that simulates photo–thermal time (PTI), daily dry matter production (DMP), nitrogen uptake (Nup), leaf area index (LAI), and crop transpiration (ETc). The second objective was to carry out the calibration of the HORTSYST model by applying a differential evolution (DE) algorithm as the global optimization method. Two tomato (Solanum lycopersicum L.) crops were established during the autumn–winter and spring–summer seasons under greenhouse and soilless culture conditions. Plants were distributed with a density of 3.5 plants m−2. Air temperature and relative humidity were measured with an S-THB-M008 model sensor. Global solar radiation was measured with an S-LIB-M003 sensor connected to a U-30-NRC datalogger. In the sensitivity analysis run in the two growth stages, it was observed that a greater number of parameters were more important at the beginning of fructification than at the end of crop growth for 10% and 20% of the variation of the parameters. The sensitivity analysis came up with nine parameters (RUE, a, b,\xa0c1\xa0,\xa0c2,\xa0A, Bd, Bn, and \xa0PTIini) as the most important of the HORTSYST model, which were included in the calibration process with the DE algorithm. The best fit, according to RMSE, was for LAI, followed by Nup, DMP, and ETc for both crop seasons; the RMSE was close to zero, indicating a good prediction of the model’s performance.