Use of eigendecomposition in a parameter sensitivity analysis of the Community Land Model

摘要

This study explores the use of eigendecomposition in a sensitivity analysis of the Community Land Model CLM, revision 3.5, with respect to its parametrization. Latent heat, sensible heat, and photosynthesis are used as target variables. The eigendecomposition of a sensitivity matrix, containing numerically derived sensitivity measures, can be used to study parameter significance. Existing parameter ranking and selection methods are examined. Furthermore, a new parameter significance ranking index is proposed which is working in concert with a new proposed selection criterion. This methodology explicitly takes parameter covariations into account. The results are consistent and similar to the most elaborate method tested in this study, but the new method has fewer assumptions. The number of significant parameters depends on the degree of variation that a single parameter is allowed to generate in the cost function. The method declares two thirds out of 66 parameters to be significant model parameters for an allowed change of 1% and only 10 parameters for an allowed change of 10% of the cost function. The sensible heat flux is shown to be the least sensitive model output in comparison with latent heat or photosynthesis. Parameters that determine maximum carboxylation and the slope of stomatal conductance are very sensitive for photosynthesis, whereas soil water parameters are significant for latent heat and C_4photosynthesis. It is concluded that the proposed procedure is parsimonious, can analyze sensitivities of more than one model output simultaneously, and helps to identify significant parameters while taking parameter interactions into account. Key PointsEigendecomposition is a valuable method for analyzing complex model sensitivityProvide new sensitivity index together with selection criteriaPhotosynthesis is the model output integrating most information to heat fluxes