 |
| Titel |
Bayesian design of control space in inverse modelling: Application to mesoscale carbon dioxide inversion |
| VerfasserIn |
Lin Wu, Marc Bocquet, Frédéric Chevallier |
| Konferenz |
EGU General Assembly 2011
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 13 (2011) |
| Datensatznummer |
250048493
|
|
|
|
|
|
| Zusammenfassung |
| In geophysical inverse modelling, the control space is the set of parameters resolved through
the assimilation of observations. We propose a consistent Bayesian formalism to design the
discretization of control space over a large dictionary of adaptive multiscale grid
representations described by several types of trees. Scale-dependent errors, such as
aggregation errors (that lead to representativity errors) are introduced and formulated
explicitly. The optimal representation of control space is constructed by optimizing a criterion
that accounts for the inversion performance, e.g. the reduction of uncertainty, or the number
of degrees of freedom for the signal (DFS) that measures the information gain from
observations to resolve the unknown parameters.
The spatiotemporal resolution of carbon source-sink fluxes is a crucial issue in carbon
dioxide inversions because of the ill-posedness of the carbon inverse problem. The CO2
concentration observations are sparse as compared to the high dimensional carbon fluxes.
However, this resolution issue is seldom investigated due to the lack of a multiscale
framework for analysis. Based on our proposed multiscale formalism, we construct optimal
multiscale representations of carbon fluxes for mesoscale inversion and perform
inversions using synthetic CO2 concentration data. Compared with the regular grid
at finest scale, optimal representations can have similar inversion performances
with much less grid-cells (e.g. 6% of the total number of grid-cells at finest scale).
These optimal representations are obtained by maximizing the DFS criterion. DFS is
found to be consistent with the root mean square error (RMSE) of carbon fluxes,
provided the correlations of the errors of a priori fluxes are physically realistic.
Scale-dependent representativity errors are considered for more reliable carbon
inversions. |
|
|
|
|
|
|