A comprehensive quality control of the SMOS Level 2 data is essential for a successful retrieval of sea surface salinity (SSS) maps. In particular, a poor filtering at Level 2 (L2) will in turn negatively impact the quality of the spatio-temporally averaged SSS maps (i.e., Level 3 or L3 maps). On the other hand, overfiltering (i.e., having a high false alarm rate) will substantially reduce the number of SSS data available for averaging at L3, thus increasing the noise level of L3 SSS maps. In this study, the impact of the different quality flags provided in the operational SMOS L2 User Data Product (UDP) in the quality of the SSS retrievals is assessed. SMOS data has been filtered following several combinations of flags. The flags can be grouped in geophysical filters (filters related to geophysical conditions prevailing in a given area), retrieval filters (filters related to the reliability of the SSS retrieval method) and geometrical filters (filters related to the geometry of the snapshots taken by SMOS). Filtered data in such a way, have been compared with ARGO buoys data, both at L2 (i.e., at swath grid level) and L3. On the other hand, the distribution of the SSS Bayesian-based inversion residuals (which indicate the consistency of the measured brightness temperatures (TBs), the forward model, and some a priori or background information) have also been analysed. These residuals, from a theoretical point of view, should follow a chi square distribution. The L2 products provide two quality flags that are based on the theoretical distribution of the residual. However, since measurement errors are not Gaussian nor uncorrelated, the real residual distributions are not always close to the theoretical one and, as such, the residual theoretically-based flags/thresholds become less effective. Moreover, the theoretical chi square distribution depends on the degrees of freedom of the problem. In the case of a SMOS grid point, the degrees of freedom correspond to the number of valid measurements used for retrieving SSS plus the (relatively small) number of geophysical parameters present in the inversion a priori or background term (typically SSS, sea surface wind speed, sea surface temperature, and total electronic content) in every point. The number of measurements used in the computation of salinity changes for each grid point (depends on the position on the Field Of View). However, one of the quality flags provided by L2 UDP and related with the chi square distribution of the residual, assumes that the number of measurements (i.e. the degrees of freedom of the theoretical distribution) is constant. This produces an inconsistency in the quality control of the SMOS data. A new filtering approach consists of adjusting the real distribution of residuals to the desired chi square distribution. This filter is based on geometric assumptions about the number of retrieved measures across swath. Alternatively, one can estimate the shape of the distribution of residuals as a function of the degrees of freedom and empirically set a threshold value for each distribution or for each interval of degrees of freedom (when the distribution of residuals does not vary significantly over a specific interval). Preliminary results of these two new approaches will be presented at the conference

SMOS & Aquarius Science Workshop, 15-17 April 2013, Brest, France

Peer Reviewed