Скачать презентацию The Impact of the Assimilation of Aquarius SSS Скачать презентацию The Impact of the Assimilation of Aquarius SSS

00b86f181ecbd8bbd170d6f76b957fce.ppt

  • Количество слайдов: 29

The Impact of the Assimilation of Aquarius SSS Data (level 2, v 2. 0) The Impact of the Assimilation of Aquarius SSS Data (level 2, v 2. 0) in the GEOS Ocean Data Assimilation System Guillaume Vernieres, Robin Kovach, Christian Keppenne, Santha Akella, Ludovic Brucker, Emmanuel Dinnat, Yury Vikhliaev, Bin Zhao, Anna Borovikov, … Aquarius SSS Retrieval Correction & Preprocessing OSE Summary/Future work Vernieres, G. , R. Kovach, C. Keppenne, S. Akella, L. Brucker, and E. Dinnat (2014), The impact of the assimilation of Aquarius sea surface salinity data in the GEOS ocean data assimilation system, J. Geophys. Res. Oceans, 119, doi: 10. 1002/2014 JC 010006.

Basics of the Retrieval Algorithm Overview of SAC-D/Aquarius observation scheme (image credit: NASA, CONAE) Basics of the Retrieval Algorithm Overview of SAC-D/Aquarius observation scheme (image credit: NASA, CONAE) SSS Tb Estimated from the 3 Aquarius radiometer operating at 1. 4 Ghz NOAA OISST (Reynolds)

Aq vs in-situ bulk salinity (Avg S between 0 and 10 m) Bias: <Aq Aq vs in-situ bulk salinity (Avg S between 0 and 10 m) Bias:

Aq vs in-situ bulk salinity (Avg S between 0 and 10 m) RMS (Aq Aq vs in-situ bulk salinity (Avg S between 0 and 10 m) RMS (Aq SSS – Argo bulk S)

Preprocessing of the Retrieval Reasons for the differences: Biases/Errors in retrieval Bulk vs Skin Preprocessing of the Retrieval Reasons for the differences: Biases/Errors in retrieval Bulk vs Skin Collocation/Binning …

Preprocessing of the Retrieval In house Feed Forward Artificial Neural Net Targets consist of Preprocessing of the Retrieval In house Feed Forward Artificial Neural Net Targets consist of in-situ bulk observations (2011 -2012) Minimization distributed across as many cpu's as necessary IN OUT

Preprocessing of the Retrieval Training set using 2011 -2012 data OUT Aquarius SSS [psu] Preprocessing of the Retrieval Training set using 2011 -2012 data OUT Aquarius SSS [psu] FFANN SSS [psu] In-situ bulk salinity [psu] IN

Preprocessing of the Retrieval Validation set using 2013 (JAN-AUG) data OUT Aquarius SSS [psu] Preprocessing of the Retrieval Validation set using 2013 (JAN-AUG) data OUT Aquarius SSS [psu] FFANN SSS [psu] In-situ bulk salinity [psu] IN

Preprocessing of the Retrieval (BIAS) BEFORE Preprocessing of the Retrieval (BIAS) BEFORE

Preprocessing of the Retrieval (BIAS) AFTER Preprocessing of the Retrieval (BIAS) AFTER

Preprocessing of the Retrieval (RMS) BEFORE Preprocessing of the Retrieval (RMS) BEFORE

Preprocessing of the Retrieval (RMS) AFTER Preprocessing of the Retrieval (RMS) AFTER

Assimilation and experimental design Model • MOM 4 p 1, CICE, MERRA Forcing Ocean Assimilation and experimental design Model • MOM 4 p 1, CICE, MERRA Forcing Ocean Assimilation System: IODAS using En. OI configuration (Christian Keppenne) 24 hour assimilation window Experiments starts from MERRA Ocean re-analysis Observations: In-situ salinity Jan 1 2012 +/-12 hrs Aquarius salinity Jan 1 2012 +/-12 hrs

Assimilation and experimental design Data withholding Experiments (2011 -2012): OMF=In-situ bulk S – S Assimilation and experimental design Data withholding Experiments (2011 -2012): OMF=In-situ bulk S – S from 24 hr lead Forecast Experiment name Observation assimilated Rms OMF’s [psu] (70 S-70 N, 0 -10 m) BASENODA None 0. 268 BASEDA In-situ (0 -100 m) 0. 186 RAW Aquarius SSS (L 2 v 2. 0) 0. 239 ANN Reprocessed Aquarius SSS 0. 195 ALL In-situ (0 -100 m) and reprocessed Aquarius SSS 0. 166

Results Results

Results RMS OMF Results RMS OMF

Results (Upper 100 m salinity content difference) BASENODA - BASEDA ANN - BASEDA Results (Upper 100 m salinity content difference) BASENODA - BASEDA ANN - BASEDA

Independent verification V 3. 0 with sst correction) Independent verification V 3. 0 with sst correction)

Summary Aq SSS + In-situ Salinity Aq SSS + In-situ bulk S improves SSS Summary Aq SSS + In-situ Salinity Aq SSS + In-situ bulk S improves SSS forecasts at in-situ locations improves S forecasts at in-situ locations Aquarius provides added info to the ocean observing system

Future work Quantify the importance of SSS in our seasonal forecast Refine DA system Future work Quantify the importance of SSS in our seasonal forecast Refine DA system for SSS assimilation (observation errors, background error covariances, …) Some regional problems need to be addressed: (Sea-Ice & land contamination) Fresh water fluxes? SMOS? Aquarius Version 3. 0

Aquarius v 3. 0, BIAS V 3. 0 Aquarius v 3. 0, BIAS V 3. 0

Aquarius v 2. 0, BIAS V 2. 0 Aquarius v 2. 0, BIAS V 2. 0

Aquarius v 3. 0, RMSE V 3. 0 Aquarius v 3. 0, RMSE V 3. 0

Aquarius v 2. 0, RMSE V 2. 0 Aquarius v 2. 0, RMSE V 2. 0

EXTRAS … EXTRAS …

V 3. 0 proposed correction V 3. 0 proposed correction

Feed Forward Artificial Neural Network: Sigmoid Cost function, Jacobian and minimization methodology: Preconditioned Newton Feed Forward Artificial Neural Network: Sigmoid Cost function, Jacobian and minimization methodology: Preconditioned Newton Conjugate Gradient for the minimization of F VERY SLOW Convergence Summations in F and J distributed over many cpu's

Preprocessing of the Retrieval 2011 -2012 <Aq SSS – Argo bulk S> std(Aq SSS Preprocessing of the Retrieval 2011 -2012 std(Aq SSS – Argo bulk S) std(NN SSS – Argo bulk S)

Preprocessing of the Retrieval 2013 <Aq SSS – Argo bulk S> std(Aq SSS – Preprocessing of the Retrieval 2013 std(Aq SSS – Argo bulk S) std(NN SSS – Argo bulk S)