Optimal adjustment of the atmospheric forcing …...M. Meinvielle et al., 2011 : Optimally improving...
1
Atmospheric variables (Ta, qa, Ua, radsw, radlw, precip) Model output CONCEPT Data assimilation We use a sequential method based on the SEEK filter, with an ensemble experiment of 200 members to evaluate parameter uncertainties. To better isolate forcing errors, we have to minimize the other sources of error such as initial condition, and model errors. Atmospheric parameters perturbations are calculated from multivariate EOF of monthly means parameters over the whole ERAinterim period. Sea surface temperature (SST) is more precisely observed from space than near-surface atmospheric variables and air-sea fluxes. But ocean general circulation models that carry out simulations of the recent ocean variability use, as surface boundary conditions, bulk formulae which do not involve the observed SST. In brief, models do not take advantage directly in their forcing of one of the best observed ocean surface variable, except when specifically assimilated. The objective of this research is to develop new approaches based on ensemble data assimilation methods that use SST satellite observations (and when available SMOS or AQUARIUS satellite sea surface salinity data) to constrain (within observation-based air-sea flux uncertainties) the surface forcing function (surface atmospheric input variables) of long-term ocean circulation simulations. The problem of the correction of atmospheric fluxes by data assimilation has already been approached in other studies and projects (Skachko et Al., 2009, Skandrani et Al., 2009). The main goal of this work is to adapt the methodology to a different experimental context. The idea is to evaluate a set of corrections for the atmospheric data of the ERAinterim reanalysis, that cover the period from 1989 to 2007, assimilating SST (Hurrel, 2008) and SSS (Levitus climatology) data. Model runs with these new atmospheric parameters are used for assesment. CONTEXT METHOD FOR ATMOSPHERIC PARAMETER ESTIMATION METHOD STEPS : 1. Ensemble forecast : model response to parameters uncertainties • Using reduced initial condition error • Using reduced model error Forecast error covariance in augmented space 2. Parameter estimation : Kalman Filter for an augmented control vector • Small observation error • Truncation of the prior gaussian distribution Correction of atmospheric parameters 3. Model run with new parameters : model response vs analysis efficiency The method permits to isolate the forcing errors minimizing other error sources. Analysis step and direct application of corrected parameters in the model Comparable effects Optimal adjustment of the atmospheric forcing parameters of ocean models derived from ERAinterim Reanalysis using sea surface temperature data assimilation in long term (1989-2007) simulations of the global ocean circulation. M. Meinvielle ([email protected]), J-M. Brankart, P. Brasseur, B.Barnier, J.Verron LEGI/MEOM – Grenoble, France RESULTS AND CONCLUSIONS References : C. Skandrani et al., 2009 : Controlling atmospheric forcing parameters of global ocean models : sequential assimilation of sea surface Mercator-Ocean reanalysis data, Ocean Sci., 5,403-419. S. Skachko et al., 2009 : Improved turbulent air-sea flux bulk parameters for the control of the ocean mixed layer : a sequential data assimilation approach, J.Atmos. Ocean. Tech.,26,538-555.. M. Meinvielle et al., 2011 : Optimally improving the atmospheric forcing of long term global ocean simulations with sea surface temperature observations, Mercator Quaterly Newsletter, 42, 24-32 M. Meinvielle et al., Computing long term atmospheric forcing corrections using sequential SST data assimilation in a realistic case, Ocean Sci. (in prep.) EXPERIMENTAL CONTEXT: - Model: NEMO, 2° global simulation ORCA2 - First guess forcing: ERAinterim reanalysis atmospheric parameters (1989-2007) - Objective: monthly forcing corrections Correct the forcing function by SST data assimilation ERAinterim forcing Corrected forcing : ERAcor 1 month C.I. Corrected SST Reference Observed SST 1 month C.I. ERROR FORCING, MODEL ERRORS … P f = MP a M T + Q C.I. PERTURBATED FORCING P f P a X a = X b + K ( Y – HX b ) SST and forcing parameters Our method reduces significantly the intertropical band warm bias classically observable in forced simulations like the one forced by ERAinterim data. Consistence with ocean dynamics Valuable information to correct atmospheric variables The heat excess observed in ERAinterim is reduced by increasing turbulent heat losses. Mathematically optimal method To guide classical empirical corrections - Consistence of fluxes correction with uncertainty characterized by heat fluxes datasets discrepancies. - Forcing the model with corrected parameters (estimated for each month of 1989-2007) : reduced warm bias in the intertropical band with respect to observations. - Diagnostic of the net heat flux computed with observed SST : sensible reduction as expected to correct ERAinterim forcing set, correction of the negative trend observed in ERAinterim dataset, better heat balance over the 1989-2007 period. - An objective method to correct atmospheric reanalysis variables by taking advantage of their consistency with ocean dynamics: an alternative to « ad hoc » forcing corrections. An negative trend (over -1W/m²) is observed in ERAinterim net heat flux timeseries. It is unconsistent with the global warming observed in the 90s. Our method equilibrates the heat budget and detrends the net heat flux. Objective not explicitely prescribed in the method itself. More realism in the forcing data Model SST Model SST Observed SST Model Bulk formulae Model dependant fluxes (evaporation, wind stress, turbulent heat fluxes, radiative heat flux) radiations The control vector is extended to correct forcing parameters (air temperature, air humidity, longwave and shortwave downward radiations, precipitation, wind velocity). The assimilation step is realized « off-line », that is to say that we don't correct the model state. We obtain atmospheric parameters corrections that we can apply to the model in free runs. Impact in the model : Better agreement with SST observations. Partitioning of the correction over the different fluxes Net heat flux balance: Equilibrated budget Turbulent fluxes modification mostly due to wind speed correction. (Monthly mean in red, 1989-2007 mean in black) Large discrepancies between the different forcing/flux dataset : e.g. in the 20N-20S latitudes band: 20 to 40 W/m²
Transcript of Optimal adjustment of the atmospheric forcing …...M. Meinvielle et al., 2011 : Optimally improving...
Atmospheric variables (Ta, qa, Ua, radsw, radlw, precip)
Model output
CONCEPT
Data assimilation
We use a sequential method based on the SEEK filter, with an ensemble experiment of 200 members to evaluate parameter uncertainties. To better isolate forcing errors, we have tominimize the other sources of error such as initial condition, and model errors. Atmospheric parameters perturbations are calculated from multivariate EOF of monthly meansparameters over the whole ERAinterim period.
Sea surface temperature (SST) is more precisely observed from space than near-surface atmosphericvariables and air-sea fluxes. But ocean general circulation models that carry out simulations of therecent ocean variability use, as surface boundary conditions, bulk formulae which do not involve theobserved SST. In brief, models do not take advantage directly in their forcing of one of the bestobserved ocean surface variable, except when specifically assimilated.The objective of this research is to develop new approaches based on ensemble data assimilationmethods that use SST satellite observations (and when available SMOS or AQUARIUS satellite seasurface salinity data) to constrain (within observation-based air-sea flux uncertainties) the surfaceforcing function (surface atmospheric input variables) of long-term ocean circulation simulations.The problem of the correction of atmospheric fluxes by data assimilation has already beenapproached in other studies and projects (Skachko et Al., 2009, Skandrani et Al., 2009). The main goalof this work is to adapt the methodology to a different experimental context.
The idea is to evaluate a set of corrections for the atmosphericdata of the ERAinterim reanalysis, that cover the period from1989 to 2007, assimilating SST (Hurrel, 2008) and SSS (Levitusclimatology) data. Model runs with these new atmosphericparameters are used for assesment.
CONTEXT
METHOD FOR ATMOSPHERIC PARAMETER ESTIMATION
METHOD STEPS :1. Ensemble forecast : model response to parameters uncertainties
• Using reduced initial condition error• Using reduced model error
Forecast error covariance in augmented space2. Parameter estimation : Kalman Filter for an augmented control vector
• Small observation error• Truncation of the prior gaussian distribution
Correction of atmospheric parameters3. Model run with new parameters : model response vs analysis efficiency
The method permits to isolatethe forcing errors minimizingother error sources.
Analysis step and direct application of correctedparameters in the model
Comparable effects
Optimal adjustment of the atmospheric forcing parameters of ocean models derived from ERAinterim Reanalysisusing sea surface temperature data assimilation in long term (1989-2007) simulations of the global ocean circulation.
M. Meinvielle ([email protected]), J-M. Brankart, P. Brasseur, B.Barnier, J.VerronLEGI/MEOM – Grenoble, France
RESULTS AND CONCLUSIONS
References : C. Skandrani et al., 2009 : Controlling atmospheric forcing parameters of global ocean models : sequential assimilation of sea surface Mercator-Ocean reanalysis data, Ocean Sci., 5,403-419.S. Skachko et al., 2009 : Improved turbulent air-sea flux bulk parameters for the control of the ocean mixed layer : a sequential data assimilation approach, J.Atmos. Ocean. Tech.,26,538-555..M. Meinvielle et al., 2011 : Optimally improving the atmospheric forcing of long term global ocean simulations with sea surface temperature observations, Mercator Quaterly Newsletter, 42, 24-32M. Meinvielle et al., Computing long term atmospheric forcing corrections using sequential SST data assimilation in a realistic case, Ocean Sci. (in prep.)
EXPERIMENTAL CONTEXT:- Model: NEMO, 2° global simulation ORCA2- First guess forcing: ERAinterim reanalysisatmospheric parameters (1989-2007)- Objective: monthly forcing corrections
Correct the forcing function by SST data assimilation
ERAinterim forcing Corrected forcing : ERAcor
1 monthC.I.
Corrected SST
Reference
Observed SST1 month
C.I. ERROR FORCING, MODEL ERRORS …
Pf = MPaMT + Q
C.I.
PERTURBATED FORCING
Pf
Pa
Xa = Xb + K ( Y – HXb )
SST and forcing parameters
Our method reduces significantly theintertropical band warm bias classicallyobservable in forced simulations likethe one forced by ERAinterim data.
Consistence with ocean dynamics
Valuable information to correct atmospheric variables
The heat excess observed in ERAinterim isreduced by increasing turbulent heat losses.
Mathematically optimal method
To guide classical empirical corrections
- Consistence of fluxes correction with uncertainty characterized by heat fluxes datasets discrepancies.- Forcing the model with corrected parameters (estimated for each month of 1989-2007) : reduced warm bias in the intertropicalband with respect to observations.- Diagnostic of the net heat flux computed with observed SST : sensible reduction as expected to correct ERAinterim forcing set,correction of the negative trend observed in ERAinterim dataset, better heat balance over the 1989-2007 period.- An objective method to correct atmospheric reanalysis variables by taking advantage of their consistency with ocean dynamics:an alternative to « ad hoc » forcing corrections.
An negative trend (over -1W/m²) isobserved in ERAinterim net heat fluxtimeseries. It is unconsistent with theglobal warming observed in the 90s.Our method equilibrates the heatbudget and detrends the net heat flux.
Objective not explicitely prescribed in the method itself.
More realism in the forcing data
Model SST Model SST
ObservedSST
Mo
del
Bulkformulae
Model dependant fluxes (evaporation, wind stress, turbulent heat fluxes, radiative heat flux)
radiatio
ns
The control vector is extended to correctforcing parameters (air temperature, airhumidity, longwave and shortwavedownward radiations, precipitation, windvelocity). The assimilation step is realized« off-line », that is to say that we don'tcorrect the model state. We obtainatmospheric parameters corrections that wecan apply to the model in free runs.
Impact in the model :Better agreement with SST
observations.
Partitioning of the correction over
the different fluxes
Net heat flux balance: Equilibrated budget
Turbulent fluxes modification mostly due towind speed correction. (Monthly mean in red, 1989-2007 mean in black)
Large discrepanciesbetween the different forcing/flux dataset :
e.g. in the 20N-20S latitudes band: 20 to 40 W/m²
