 |
| Titel |
Use of eigendecomposition in a parameter sensitivity analysis of the Community Land Model |
| VerfasserIn |
Maren Göhler, Matthias Cuntz, Juliane Mai |
| Konferenz |
EGU General Assembly 2013
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 15 (2013) |
| Datensatznummer |
250083456
|
|
|
|
|
|
| Zusammenfassung |
| Sensitivity Analysis is a widely used tool for model identification and calibration, and to
quantify model output uncertainty. This paper explores the use of eigendecomposition in a
global sensitivity analysis of a complex land surface model, the Community Land Model
CLM, revision 3.5, with respect to its parametrization. We use different fluxes from and to the
atmosphere as target variables, namely the fluxes of latent and sensible heat, and
photosynthesis. We identified 66 parameters in the stand-alone version of CLM3.5. The
parameter sensitivity measures from global parameters are arranged in a sensitivity matrix S.
The eigendecomposition of ST S tells parameter importance while taking into account
interactions between the parameters.
A main focus of sensitivity analysis is the selection of important parameters, e.g. for
parameter estimations. We therefore examine existing ranking and selection methods to
determine parameter importance from the sensitivity matrix S. We propose a new parameter
importance ranking index which takes parameter interactions into account and determine its
uncertainty with bootstrapping. Furthermore, we propose a new selection method
working in concert with this index for detecting important parameters. The most
elaborate selection method which we tested marks comparatively more parameters as
relevant compared to the other methods. We show that our selection method performs
very similar to the elaborate method. The number of important parameters detected
with the new selection procedure depends on the amount of variability in the cost
function one wants to conserve. It retains two thirds of the 66 parameters when
conserving 99% and only 10 parameters when conserving 90% variability in the
cost function when analyzing three model outputs simultaneously. But it can be
shown that the sensible heat flux is the least sensitive model output compared to
latent heat and photosynthesis when disentangling the three model outputs. The C3
evergreen vegetation type has less sensitive parameters compared to the both deciduous
types with C3and C4 photosynthetic pathways since soil parameters play less a role
during the year. The soil evaporation resistance description is over-sensitive to all
analyzed fluxes and vegetation classes and is therefore excluded from the analysis. The
parameters which determine Vcmax and the slope of the stomatal resistance model
become very important if sensitivity is determined with respect to photosynthesis.
The soil water parameters are important for the latent heat and C4 photosynthesis.
We conclude that our new proposed parameter selection procedure can analyze
sensitivities of more than one model output simultaneously, helps to identify important
parameters while taking their interactions into account and is inexpensive compared to
other. |
|
|
|
|
|
|