Sample size, range of parameters and time-dependent effects on global sensitivity analysis in sugarcane modelling

Abstract

AbstractProcess-based crop models (PBCM) for sugarcane present many genotype parameters compared to other crops, making it harder to calibrate. Global sensitivity analysis (GSA) has thus become an important tool for understanding, calibrating and further developing PBCMs. This work used a recently updated Agronomic Modular Simulator for Sugarcane (SAMUCA) to simulate crop growth and development, conducted with two treatments: with green cane trash blanket (GCTB) and under bare soil (BARE). Using the extended Fourier Amplitude Sensitivity (eFAST) algorithm, GSA was performed on the 24 genotype parameters of SAMUCA. The objective was to determine the sample size (SZ): how many samplings are necessary to quantify the sensitivity indices. Additionally, we aimed to assess the influence of parameter range variation and identify which genotype parameters explain the highest variance in simulations of the SAMUCA model under BARE and GCTB conditions. The results showed that SZ greatly affected the convergence and sensitivity indices, and the SZ required here needed to be >2049 for the analysis to cover all variables. Two sets of parameter ranges were used for analysis (the first used maximum and integer values of each parameter reported in the literature; the second applied 25% perturbation to the previously calibrated values). The results indicated that the parameter range affected the parameters' order of importance. Furthermore, we identified that at different phenological stages of sugarcane development, distinct parameters were responsible for explaining the most variance of the output. However, there was no difference among ratoons or interference in the results of BARE or GCTB.