 |
| Titel |
Ensemble generation in the MPI-ESM seasonal prediction system |
| VerfasserIn |
Johanna Baehr, Kristina Fröhlich, Michael Botzet, Wolfgang Müller |
| Konferenz |
EGU General Assembly 2014
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 16 (2014) |
| Datensatznummer |
250093885
|
| Publikation (Nr.) |
 EGU/EGU2014-9048.pdf |
|
|
|
|
|
| Zusammenfassung |
| In the seasonal forecast system based on the global coupled climate model MPI-ESM as used
for CMIP5 simulations, we describe the implementation of two ensemble generation
techniques. The system is initialized in the atmospheric, oceanic, and sea ice component of
the model from reanalysis/ observations with full field nudging in all three components. For
the perturbation of the ensemble, we compare ensembles generated from lagged initialization
and bred vectors, and demonstrate that the bred vector initialized ensemble can also yield
reliable predictions.
Here, bred vectors are implemented with a vertically varying norm in the ocean
component to generate initial perturbations of the full water column. Hindcast ensembles
initialized with either ensemble generation method reproduce the observed surface
temperature anomalies at 2-4 months lead time particularly in the tropics. Compared against
the spread of surface temperature in the ERAinterim reanalysis, the resulting ensemble
hindcasts are more reliable in the tropics than at higher latitudes, and more reliable over the
ocean than over land.
However, Talagrand diagrams indicate that both the lagged initialized ensemble and the
bred initialized ensemble are underdispersive at 2 to 4 months lead time, especially in the
tropics. The direct comparison of the lagged and bred initialized ensembles shows
distributions in the Talagrand diagram closer to ‘flat’ (quantified by β-scores) for more
regions for the bred initialized ensemble than for the lagged ensemble. We also assess the
reliability of the ensembles through the spread-error ratio. The bred initialized ensemble
shows a small improvement over the lagged initialized ensemble, in particular at high
latitudes. At mid-latitudes, the bred initialized ensemble shows smaller spread-error ratios
than the lagged initialized ensemble in several coastal regions, particularly in the respective
winter hemisphere. |
|
|
|
|
|
|