ARCNESS automatic differentiation project
Overview
This page describes the project algorithmic differentiation for data assimilation
in terrestrial carbon models funded by the
Terrestrial node of ARCNESS, the
Australian Research Council Network for Earth
System Science.
The project is about using automatic differentiation, implemented by
operator overloading. This is being applied to problems of
initialisation, calibration and data assimilation in the
land-surface component of ACCESS (Australian Community
Climate and Earth System Simulator).
The initial MASCOS work on automatic differentation was applied to
analysing the Brazilian Proposal. Work on the Brazilian Proposal is
not part of the ARCNESS project. However the analysis of the Brazilian Proposal
provides an interesting example of automatic differentiation
and so presentations and publications are noted here.
Activities
-
12/2/07. Mark Fielding commenced at MASCOS are programmer supported
by ARCNESS funding
Immediate work plan
-
Development of test suite for checking validity of derivative
operations.
-
Development for Fortran-90 code for second derivatives.
-
Development of Fortran-90 code for first derivative calculations
in vector models.
Wider goals
-
Deployment of automatic differentiation in initialisation of carbon pool model(s).
-
Deployment of automatic differentiation in calibration of carbon pool model(s).
Development status
-
Fortran-90 code for overloading to carry first derivatives, (most operations and common functions)
developed mid 2006.
-
C++ code for overloading to carry first and second derivatives, (most operations and common functions)
developed 2005.
Presentations
-
Carbon data assimilation: AMOS 2007, Adelaide, February 2007.
-
Adjoint sensitivity analysis for attribution of responsibility
for climate change. ANZIAM 2007, Fremantle, January 2007
-
Presentation at MSRI-NCAR workshop on carbon data assimilation. July 2006
-
Automatic differentation in the analysis of strategies for mitigation of global change.
at MODSIM 2005, Melbourne, December 2005.
Disclaimer
This page, its contents and style, are the responsibility of the author and do not represent the
views, policies or opinions of The University of Melbourne.
Ian Enting: last change 15/2/07.