Inhalt des Dokuments
Preprint 29-2006
Algorithmic Sensitivity Analysis in the Climate Model Climber 2
Author(s) :
Thomas Slawig
Preprint series of the Institute of Mathematics, Technische Universität Berlin
Preprint 29-2006
MSC 2000
- 65K10 Optimization and variational techniques
-
86A10 Meteorology and atmospheric physics
Abstract :
This report summarizes the results of the project
Algosense performed at the Institut für Mathematik, Technische Universität Berlin,
from July 2001 to June 2002.
Aim of the project was to analyze the applicability of tools for Algorithmic (or Automatic) Differentiation (AD)
to two climate models developed at the Potsdam Institute for Climate Impact Research (PIK).
These were the so-called Box Model, a small model of the
North Atlantic stream, and the more complex model Climber 2
which is a so-called model of intermediate complexity
consisting of atmosphere, ocean, ice, and vegetation components.
Applications that are considered start from pure sensitivity
calculations over uncertainty estimations to optimization runs.
First and higher order derivatives are of interest.
The outline of this report is the following:
In the next section we describe the basic tools and techniques of
Algorithmic
Differentiation. The following two sections deal with the two models studied in this project.
In each of them the corresponding model and its special features
important for Algorithmic Differentiation
are briefly introduced. Then the used Algorithmic Differentiation tools and
technical details of the AD process are presented. At last numerical results are given.
Further emphasis is put on the necessary code preparations to apply
the AD tools.
The last section of the report gives a summary and deals with the perspectives and opportunities of the application of Algorithmic Differentiation to these and maybe other climate models.
Keywords :
algorithmic differentiation, parameter studies, sensitivity analysis, climate modeling