Premium
Comparison of the Generalized Likelihood Uncertainty Estimation and Markov Chain Monte Carlo Methods for Uncertainty Analysis of the ORYZA_V3 Model
Author(s) -
Tan Junwei,
Cao Jingjing,
Cui Yuanlai,
Duan Qingyun,
Gong Wei
Publication year - 2019
Publication title -
agronomy journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.752
H-Index - 131
eISSN - 1435-0645
pISSN - 0002-1962
DOI - 10.2134/agronj2018.05.0336
Subject(s) - glue , markov chain monte carlo , statistics , monte carlo method , mathematics , bayesian probability , posterior probability , econometrics , sampling (signal processing) , computer science , engineering , mechanical engineering , filter (signal processing) , computer vision
Uncertainties in crop model have attracted many attentions in recent years. The Generalized Likelihood Uncertainty Estimation (GLUE) and the Markov Chain Monte Carlo (MCMC) methods have been widely used to quantify model uncertainties for hydrological models. While few papers have focused on the comparison of these two methods for a crop model, in this study the GLUE and MCMC were applied for parameter uncertainty analysis of the rice growth model ORYZA_V3. We examined the influence of subjective factors for the GLUE method, and made a comparison of results for the two methods. In the GLUE method, sample size of parameter sets exceeding 30,000 had negligible effects on the results, whereas the accepted sampling rates (ASR) had a pronounced influence on the posterior parameter distributions and the derived 95% confidence interval (95CI) of biomass simulations. Furthermore, the GLUE method failed to construct the posterior distributions for some less sensitive parameters. Due to the large dependence on ASR, the GLUE method might easily lead to deceptive results, and should be used with caution. Because the MCMC has a well‐documented statistical background and it can obtain clear and stable posterior distributions of parameters, this method is strongly recommended for crop model users in parameter uncertainty analysis.