 |
| Titel |
Statistical emulation of the rapid response of a climate model to astronomical forcing variations |
| VerfasserIn |
Nabila Bounceur, Michel Crucifix |
| Konferenz |
EGU General Assembly 2011
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 13 (2011) |
| Datensatznummer |
250050475
|
|
|
|
|
|
| Zusammenfassung |
| We are interested in the response of a general circulation model of the atmosphere and the
ocean to variations of the astronomical forcing during the Pleistocene. However, the demand
on computing resources would be far too excessive for long time simulations. So, our aim is
to formulate a reduced order model for this response, by the construction of a statistical
model called an emulator, trained on the available runs.
Insolation during long time variations is influenced by the eccentricity e, the longitude
of the perihelion Ï and the obliquity É. To deal with this astronomical theory of
paleoclimate, input data are expressed in an adequate form and then consider the base
(É,e-
sin(Ï),e -
cos(Ï )).
The choice of a small number of experiments to develop the emulator is crucial. The
distribution of the experimental points was made following two plans in order to compare
them. A full factorial design and a Latin hypercube one. The latest maximize the
minimum distance between the design points. 27 experiments by plan were then
made.
Here, we have developed and designed an emulator of a three-dimensional Earth system
model of intermediate complexity (LOVECLIM, Goosse et al., 2010), considering the
principal components of its response (surface temperature) using a weighted principal
component analysis. The first three principal component account for 99% of the response
variance.
The emulator proposed here is based on a Gaussian process model. This is a stochastic
process for which any finite set of the simulated data has a joint multivariate Gaussian
distribution (Rasmussen, 1996). We further consider the combination of the Gaussian process
with a linear regressor on the one hand, and with a quadratic one on the other hand. Statistical
methods of model selection are used to choose the appropriate emulator, such as the
leave-one-out cross-validation, that is, using a single data point in the training set for
prediction.
The Gaussian process emulator provides the quantification of uncertainty about
evaluating the emulator at a limited number of input data. The surface responses obtained for
each emulator parameter allow us, considering the uncertainties, to study the influence of the
astronomical parameters variations on the output, and to detect of nonlinearities. |
|
|
|
|
|
|