|
Titel |
Sensitivity analysis using the variational data assimilation software DassFlow-Ice |
VerfasserIn |
Nathan Martin, Jérôme Monnier, Jitendra Singh, Ronan Madec, Olivier Gagliardini |
Konferenz |
EGU General Assembly 2011
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 13 (2011) |
Datensatznummer |
250052848
|
|
|
|
Zusammenfassung |
To be confident in the accuracy of the modelling of free surface ice-flows requires to confront
numerical experiments to actual observations. This type of geophysical flows is strongly
sensitive to their input parameters and boundary conditions and shows multiscale behaviours.
Thus, sensitivity analysis is a useful tool to improve our understanding of ice-flows. To
achieve this goal, we are developing at the Mathematical Institute of Toulouse a
computational software named DassFlow (cf. N. Martin et al., Internal Report IMT, 2011 and
M. Honnorat et al., RR INRIA, 2007) following several rules of implementation and
algorithmic structure. Then, the direct computational code has been algorithmically
differentiated in order to obtain its adjoint part.
The direct component of the code is an ice-flow modelling tool implementing the
full-Stokes equations with ALE (Arbitrary Lagrangian Eulerian) representation of
the free-surface. It is based on an order 2 high-precision finite-element kernel. It
allows computations of 2D-flowline domains and a 3D version is currently under
development. Some of the features are friction boundary condition on the bedrock,
imposed lateral flux, non-linear rheology solver and general steady or quasi-static
free-surface simulations can be performed. The software has been validated on
analytical test-cases and successfully compared with the results published in the
Ice-Sheet Model Intercomparison Project (cf. F. Pattyn et al., The Cryosphere Discuss,
2008).
Using the Automatic Differentiation tool Tapenade, we obtained the adjoint
code of DassFlow-Ice (including non-newtonian rheology solver). The code has
been derivated according to several control variables such as imposed lateral flux,
exponent and rheological parameter of Glen’s law, accumulation at the free surface,
friction coefficient, shape factor for 3D friction effects. The performance has been
optimized to reach a time of computation around 4 times a direct run for an adjoint
run.
We present an application of our tool on the Variegated Glacier in Alaska. This glacier
has been monitored for many years because of its surge behaviour which occurs more or less
periodically. The computation is made on a 2D-domain where topography of the
bedrock and of the surface has been measured along the central flowline. The specific
behaviour of this glacier is mainly driven by its basal conditions and quantifying
the sensitivity of the friction coefficient along the bedrock could be of primary
interest.
First runs confirmed that the main driving parameters of the flow are the basal
conditions and specific area of the glacier has been highlighted. These first sensitivity
tests are a first step toward a complete data assimilation procedure which would
allow to infer the basal drag coefficient from the knowledge of surface velocities. |
|
|
|
|
|