Assessment of uncertainty and sensitivity analyses for ORYZA model under different ranges of parameter variation

• Published in: European journal of agronomy Vol. 91; pp. 54 - 62
• Main Authors: Tan, Junwei, Cui, Yuanlai, Luo, Yufeng
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2017
• Subjects: Crop model; LHS; Regression-based method; TDCC; Crop model; Regression-based method; LHS; TDCC
• DOI: 10.1016/j.eja.2017.09.001
• ISSN: 1161-0301

•Effect of ranges of parameter variation (RPV) on SA was conducted.•LHS technique was employed to generate parameter sets.•RPV has no effect on CV’s change rule of model outputs over time.•Too small or too large RPV makes some parameters lose sensitivity.•RPV was suggested as ±30% perturbation of default values. We explore the effects of different ranges of parameter variation (RPV) on sensitivity and uncertainty analyses for ORYZA_V3 model. In this study, a latin hypercube sampling (LHS) technique is used to generate parameter sample sets, and a regression-based method is employed for the sensitivity analysis on 16 crop parameters. Then, a top-down concordance coefficient (TDCC) is calculated to assess the stability of parameter sensitivity rankings across diverse RPV. Furthermore, coefficients of variation (CV) and 90% confidence intervals (90CI) of daily model outputs are analyzed by considering uncertainty in observations. We find that the increasing RPV multiplies the CV of daily model outputs, whereas the RPV has no effect on the CV’s change rule over time. The 90CI of model outputs include most of the observations when the RPV is more than ±30% perturbation. The standardized regression coefficient (SRC) of some parameters are obviously minified when the RPV is ±5% or ±50% perturbation. The results highlights the importance of RPV selection in the sensitivity and uncertainty analysis of crop model, and ±30% perturbation was suggested when the RPV cannot be specifically obtained.