1. Errors on SMOS retrieved SSS and their dependency to a priori wind speed X. Yin 1, J. Boutin 1, J. Vergely 2, P. Spurgeon 3, and F. Gaillard 4 1. LOCEAN 2. ACRI 3. ARGANS 4. IFREMER 1

2. SMOS SSS map using ISAS optimum interpolation 1. SSS from L2OS v5.5 reprocessing, August 2010 (“SSS OP ”) 2. In Situ Analysis System (ISAS) SSS 2

3. Why do we need a better estimation of wind speed? 1. ECMWF WS as prior and fixed σ WS = 0m/s 2. ECMWF WS as prior and tuned σ WS = 2m/s (L2OS OP) 3. SSMI WS as prior and tuned σ WS = 2m/s 3

4. Why do we need a better estimation of wind speed? ECMWF WS v38r2 – v37r2 1m/s -1m/s Retrieved WS1 v38r2 – v37r2 1m/s -1m/s The differences between two versions of ECMWF are partially but not totally reduced by the L2OS retrievals. The difference between two versions of retrieved WS and that of SSS are well collocated, and the correlation coefficient between the two is 0.859. Errors in ECMWF WS can lead to errors in SSS, since retrieved WS can not reduce the errors in WS a priori totally with the L2OS retrieval scheme. Retrieved SSS1 v38r2 – v37r2 0.2psu -0.2psu 4

5. The retrievals are based on the Levenberg and Marquardt iterative convergence method. The first guessed geophysical inputs (SSS, SST, WS and TEC) are adjusted so as to minimize a so called “cost function” χ 2 expressed by Introduction Can we reduce the biases in retrieved wind speed, if 1)Relax the error on a priori WS ( σ WS ) : 2m/s -> 5m/s (“OP, 5m/s”) 2)Two step scheme. 5

6. Introduction Two step scheme, objective: to reduce the biases in the retrieved SSS and WS without increasing noise in the SSS retrievals. 1)1 st step: a priori ECWMF WS with increasing error of a priori WS ( σ WS ) to 5m/s -> retrieved WS -> 2D spatial median filtering (50 km radius close to SMOS resolution) -> smoothed WS Objective: to relax the dependency of error/bias in retrieved WS to a priori estimate (ECMWF), and to reduce noise in retrieved WS used as a priori estimate for the next step. 2) 2 st step: smoothed WS from 1 st step (instead of ECMWF) used as a priori estimate with ( σ WS ) set back to be 2 m/s -> retrieved SSS and WS Note: for both steps, errors of a priori SSS, SST and TEC are the same as in the operational L2OS processor, i.e. 100psu, 1 °C and 10 tecu. 6

7. ECMWF WS SSMI WS SMOS operational retrieved WS Radiometer wind speeds lower than ECMWF WS in the eastern equatorial pacific ocean because of strong surface currents, but still higher than SSMI WS -> positive anomalies in retrieved SSS compared with ISAS SSS rSSS - ISAS 1. One ascending orbit in April 2013, with two versions of ECMWF WS(v38r2 and v37r2) uses as a priori estimate. Can the retrieved WS and SSS converge, with two different versions of a priori WS? 2. One ascending orbit in Aug, 2010 in the eastern equatorial pacific ocean, where we found large WS biases between ECWMF and SSMI. Comparisons among SMOS retrieved WS, ECWMF WS and SSMI WS, and comparisons of SSS. 3. All ascending orbits in Aug. 2010: performance over the global ocean Data 7

8. Results Can the retrieved WS converge to the same value with the two step scheme, using two different versions of a priori ECMWF WS? YES with some exceptions! Operational, rWS1 v38r2 – v37r2 1m/s -1m/s Two step, rWS1 v38r2 – v37r2 1m/s -1m/s Exceptions: 1) too large differences between two a priori WS; or 2) RFI 8

9. Results Exceptions: 1) too large differences between two a priori WS; 2) RFI 0.2psu -0.2psu 0.2psu -0.2psu Operational, rSSS1 v38r2 – v37r2 Two step, rSSS1 v38r2 – v37r2 Can the retrieved SSS converge to the same value with the two step scheme, using two different versions of a priori ECMWF WS? YES with some exceptions! 9

10. TEST: comparisons of different methods 3S-2N Wind speed Salinity 10

11. Monthly maps of std of SSS for each 0.5 * 0.5 grid Operational SSS Two-step SSS With mask: abs(diff) > 0.2 No mask Two-step scheme enhances problems near the coastal and RFI regions. diff 11

12. Monthly maps in August 2010 No mask With mask 12

13. Monthly maps in August 2010 Std of SSS is higher if we only increase error on a priori WS. Differences in SSS Differences in std of SSS 13

14. rWS OP - WS ECMWF rWS twostep - WS ECMWF retrieved WS Problems near the coastal and RFI regions 14

15. Conclusions 1.The two step scheme enhances the capability of retrieving WS using multi-angular MIRAS TB. 2.The retrieved SSS and WS converge to the same value with the two step scheme, regardless of different versions of ECMWF WS used as a priori estimates. 3.The retrieved WS with the two step scheme are closer to SSMI WS in Eastern Equatorial Pacific than the L2OS OP WS. 4.The retrieved SSS with the two step scheme are closer to in-situ SSS in Eastern Equatorial Pacific than the L2OS OP SSS. 5.Compared with L2OS OP retrievals, the two step scheme does not increase the noise in retrieved SSS in the open ocean at low and moderate latitude. 6.The two step scheme enhances problems near the coastal and RFI contaminated regions. 15

16. TEST: comparisons of different methods Latitude: 3S-2N rSSS – ISAS (psu)rWS – SSMI (m/s) Mean(median)stdMean(median)std L2OS OP (2m/s) 0.48(0.33)0.911.14(1.18)1.37 Two step 0.26(0.20)0.820.40(-0.03)1.90 No WS Smoothing 0.33(0.27)1.000.62(-0.09)2.57 L2OS OP (5m/s): 1 st step 0.38(0.31)0.950.71(0.24)2.24 Tests (one orbit in 2010/08/06,13h- 14h ): 1)L2OS, σ WS = 2 m/s 2)Two step: 1) σ WS = 5 m/s + WS smoothing; 2) σ WS = 2 m/s 3)Two step: 1) σ WS = 5m/s + no WS smoothing; 2) σ WS = 2 m/s 4)L2 OS, σ WS = 5 m/s 3S-2N 16

17. Monthly SSS3 maps in August 2010 No mask 17

18. Variances of retrieved parameters 18