150 likes | 299 Views
Observation impact studies with ocean reanalysis. Elisabeth REMY, Nicolas FERRY, Laurent PARENT, Marie DREVILLON, Eric GREINER and the Mercator-Ocean team. Use of reanalysis in observation impact/sensitivity studies.
E N D
Observation impact studies with ocean reanalysis Elisabeth REMY, Nicolas FERRY, Laurent PARENT, Marie DREVILLON, Eric GREINER and the Mercator-Ocean team
Use of reanalysis in observation impact/sensitivity studies • Reanalysis provide ocean state estimates over longer time period than the operational systems that are regularly evolving. Their analysis can reveal impact of the changes in the observation system. • Different diagnostics can be usefull to identify observations impact: • - physical diagnostics • - assimilation diagnostics: innovation, increment, residual. • Recent reanalysis at Mercator-Océan: • Glorys1v1 (ORCA025/SAM2): 2002-2008, • ORCA2/SAM2 : 1980-2008, ORCA2/SAM3 : 1960-2005 (Ctrl, in situ only, different MDTs for 1993-2005)
Sensitivity of the global analysis error to the number of assimilated in-situ observations ORCA2/SAM3 1960-2005 Number of assimilated observations Global mean misfit to T observations in°C Global rms misfit to T observations in °C 1960-1992 1993-2005
Sensitivity of the analysis to the number of assimilated in-situ observations Glorys (ORCA025/SAM2)2002 - 2008
OSE experiments for in situ observations • Planned experiments: • One simulation 2002-2009 of Glorys without the ARGO data, • Tests of dispersion of virtual floats using the code « ARIANE » (lagrangian diagnostics tool) with the reanalysis outputs.
DT = MDT + SLA SSH = MSSH + SLA geoid Ellipsoid of reference MDT sensitivity • The assimilation of Sea Level Anomalies requires the use of a Mean Dynamical Topography. • Inconsistencies can exist between the dynamical height deduced from the in-situ observations and from the « SLA + MDT » infomation, if the prescribed MDT differs from the « real one » or the model is not able to represent it. • - OSE with ORCA2 : synthetic « Rio » MDT and model MDT • Currently looking at MDT errors impact with Glorys • No bias correction
Cumulative trend of the mean temperature and salinity 0-300 m ORCA2/SAM3 reanalysis 1993-2001 Assim. with the model MDT Assim. with the Rio MDT trend T 0-300 m (°C) trend S 0-300m (PSU)
Trend of the mean temperature 0-300 m and deep salinity Glorys1v1 reanalysis 2002-2008, Rio MDT Trend T 0-300 m (°C/year) Trend S 700-1500m (PSU/year)
Mean Atlantic Meridional Overturning Streamfunction Mdt rio ORCA2/SAM3 with Rio MDT ORCA2 without assimilation ORCA2/SAM3 with model MDT ORCA025 without assimilation Glorys1v1 with Rio MDT
MDT sensitivity : comments • Need of realistic mean dynamical topography to assimilate SLA + in situ observations : better position of the gyres. • the analysis with ORCA2 or ORCA025 follows the prescribed MDT (Rio or model), • still problems when using the Rio MDT: • effect on unconstrained regions/variables of the system (deep T-S fields), • local problems can have large impact trhrough the ocean dynamic. (The MOC anomaly reveals a regional unrealistic meridional cirulation cell located around 35°O. This response is probably due to a pressure gradient, linked to the MDT constrain (work in progress)). • Planned experiments : • test of the new Rio MDT based on GRACE geoid.
Real Time and re-processed observations • SLA observations Delayed Time /Near Real Time • To look at climatological signal, re-processed data are necessary. Mean sea level evolution from NRT observations (blue), DT observations (red), ORCA2°/SAM3 reanalysis (green) and PSY3v2 (black) • In situ observations data sets • Use of CORA in GLORYS and CORIOLIS in PSY3v2 : differences are seen in the analysis (M. Drevillon presentation) • → Comparison of the reanalysis and real time outputs.
Conclusion • Long term Impact of changes in the observation systems can be difficult to follow in an operational system which is rapidly evolving. • We can see impact of observation system changes/incoherencies between data sets in our reanalysis. • To evaluate precisely their impact, it requires the setup of dedicated diagnostics/experiments. • The conclusions of those experiments will partly depend on the system (model configuration, assimilation scheme, error specification…), light configurations can be usefull. • We are still trying to make use of the assimilated observations in an optimal way (observation operator, observation error covariance matrix estimation…) • The use of the nemo adjoint is planned to identify model bias sources. Can also be used for observation array design?
Trend T 0-300 m In situ, Rio MDT + SLA In situ, MDT model + SLA In situ only