Sensitivity analysis for a forest growth model: A statistical and time-dependent point of view

Abstract

Increasing number of computer models are being used to simulate and predict the state of certain systems, in which parameter calibration and model output uncertainties co-exist due to the incomplete understanding of the system under simulation and biased model structure. In this paper, we demonstrated the ability of using sensitivity analysis, which include both screening and variance-based methods, to explore model structure and behavior. A forest growth model 3-PG2 and 141 plots of Corymbia maculata and Eucalyptus cladocalyx are used. Two model outputs, leaf area index and root biomass, were evaluated. Comparability between screening and variance-based methods and the change in sensitivities over time were assessed. High consistency was found and the variance-based method exhibited excellent convergence and stable sensitivity rankings. The results show that for each model output, the methods presented here can effectively identify the relative sensitivities of each input parameter. The results present some instructive hints about the model structure and underlying model behavior evolution features as simulation period becoming longer. This study shows that the sensitivity analysis methods are effective tools in model calibration and identification.