Debiased SMOS SSS L3 v2 maps generated by LOCEAN/IPSL/ACRI-ST
J. Boutin (firstname.lastname@example.org)– JL. Vergely (Jean-Luc.Vergely@acri-st.fr) – S. Marchand (email@example.com)
19 May 2017
In order to propose improved methodologies to be implemented in future CATDS-CPDC versions, LOCEAN/IPSL (UMR CNRS/UPMC/IRD/MNHN) and ACRI-st have derived a methodology for correcting systematic SSS biases. This second version uses an improved debiasing technique that is also implemented in the CATDS CPDC processing; only maps generation (filtering and smoothing) and geographical coverage differ from CATDS CPDC version. We welcome feedbacks from users about the quality of these new products, as they are experimental.
We have shown that, when considering monthly SSS anomalies, with respect to a SMOS monthly climatology (built from the SMOS_LOCEAN_v2013 products available at CATDS CEC LOCEAN), the precision of SMOS SSS monthly anomalies is on the order of 0.2 (see Boutin et al. 2016); in particular, working in terms of anomalies, removes most of the biases occurring around continents. In view of these good results, we have developed a first attempt to correct SMOS SSS systematic biases by preserving the temporal SMOS SSS dynamic (Kolodziejczyk et al., 2016). This method was the basis for the debiased products v0 delivered at CATDS CEC LOCEAN in March 2015. It has been updated, in CATDS CEC LOCEAN debiased products v1 to be applied on CATDS RE04 reprocessed products (similar to ESA L2OS version 622 products) and to include a latitudinal bias removal correction. In this new update, the outlier filtering before the computation of biases has been refined, in particular taking into account SSS natural variability as inferred from SMOS data themselves, the bias estimate is now done over 7 years (2010-2017), and 9day and 18day products are delivered. We recall at the end of this note the principle of the method. It has been validated using ship measurements from French Sea Surface Salinity Observation Service (http://www.legos.obs-mip.fr/observations/sss) and using SMAP data. Examples of the comparisons are shown below:
Figure 1 : top) mean difference of SSSsmos minus SSSship as a function to the distance to coast for various SSS products; bottom) standard deviation of the difference. Statistics of SMOS v2 debiased products are reported in blue. With respect to the non debiased SMOS products (magenta), mean difference as well as standard deviation of the difference are reduced whatever the distance from the coast and they become, between 100 and 800km from the coast, less than the one of the Argo interpolated product (ISAS, green).
Figure 2 : Example of comparison of SMOS (9 days product) and SMAP SSS (8 day product – JPL CAP) in the Bay of Bengale.
These maps are provided every 4 days from 01/2010 to 12/2016 and are derived from a combination of ascending and descending orbits. Debiased SSS are temporally averaged using a slipping Gaussian kernel with a full width at half maximum of 9 days (9 day product) and of 18 days (18 days product). Maps are at a spatial resolution 25x25km2; a median filtering over nearest neighbors is applied. They also contain an estimation of the mean error of the salinities (field eSSS) obtained from the spatial standard deviation of the SSS in the 50km radius around each grid node. This error estimate also contains spatial natural variability and should only be considered as a qualitative indicator (e.g. larger error expected in areas contaminated by RFI). With respect to debias_v1 version, at high latitudes, an ice mask derived from SMOS Acard retrieved parameter is applied. Nevertheless, because the ice edge varies from one month to the other, SSS biases are not stable at high latitude: hence no bias correction is applied north of 45N and south of 47S. Users must be very cautious poleward of these regions: they have been removed from the CATDS CPDC debiased (L3Q) products. We chose to keep them in this experimental product because very large variations can still be detectable.
Summary of the methodology:
The SMOS sea surface salinities (SSS) are affected by biases coming from various unphysical contaminations such as the so-called land contamination and latitudinal biases likely due to the thermal drift of the instrument. These biases are relatively weak and have almost no impact on soil moisture retrieval. On the contrary, for salinity estimation, the impact is non negligible and can reach more than 1 salinity unit in some regions close to the coasts.
These biases are not easy to characterize because they exhibit very strong spatial gradients and they depend on the coast orientation in the Field Of View (FOV). Moreover, these biases are dependent on the position on the swath.
The zero order bias is the so-called Ocean Target Transformation (OTT) which is a correction applied at brightness temperature level. Here, we consider remaining biases on the SSS retrieved from brightness temperatures corrected with an OTT. SSS maps are obtained from a correction applied at salinity level. This correction is determined using simultaneously the January 2010-December 2016 period of SMOS observations. Indeed, it is possible to build salinity time series for each grid point depending on the observation conditions (for instance depending on the orbit direction) and check, from a statistical point of view, the consistency of the salinities.
The first step of this empirical approach is to characterize as accurately as possible these biases as a function of the dwell line position. We first characterize the seasonal variation of the latitudinal biases using SSS in the Pacific Ocean further than 800km from the coast. We look for the dwell line (i.e. across track position) the least affected by latitudinal biases (at XXkm for the center of the swath) and we adjust all the SSS for a latitude and time varying bias estimated from biases averages with respect to the reference dwell line in the Pacific Ocean. The second step is to correct for biases in the vicinity of land. We have found that these biases vary little in time, and can be characterized according to the grid point geographical location (latitude, longitude) and to its location across track. If we assume that the salinity at a given grid point varies very slowly during a given period, then, the different satellite passes crossing the same pixel during the given period should give consistent salinities. Additionally, assuming that the bias does not vary temporally for a given grid point implies that the relative salinity variation over the whole period should be the same whatever the distance to the center of the track. It is then possible to estimate the relative biases between the various distances across track and to obtain, with a least squares approach, a time series of relative salinity variations obtained from all the passes. The main difference between the debias_v1 version and the debias_v2 version, is the variability accepted between the SSS measured within 18 days at the same latitude, longitude: in debias_v2 version, we take into account an estimate of the natural variability expected from SMOS observed SSS while in debias_v1 version a geographical constant noise on SMOS SSS was considered. Hence the debias_v2 version better preserves SSS natural variability especially close to river plumes. Note that the across track relative bias estimate does not use any external climatology. It allows checking that all the dwell-lines and orbit types (ascending or descending) give consistent results.
These relative salinity variations are then converted, in a last step, to salinities by adding a single constant determined, in each pixel, using an average SSS climatology over the whole period (ISAS data). This last step, because it uses only one SSS climatology per grid point as reference totally preserves the SMOS temporal dynamic.
Alory G., T. Delcroix, P. Téchiné, D. Diverrès, D. Varillon, S. Cravatte, Y. Gouriou, J. Grelet, S. Jacquin, E. Kestenare, C. Maes, R. Morrow, J. Perrier, G. Reverdin and F. Roubaud, 2015. The French contribution to the Voluntary Observing Ships network of Sea Surface Salinity. Deep Sea Res., 105, 1-18, doi:10.1016/j.DSR.2015.08.005.
Boutin, J., N. Martin, N. Kolodziejczyk, and G. Reverdin, 2016, Interannual anomalies of SMOS sea surface salinity, Remote Sensing of Environment, doi:http://dx.doi.org/10.1016/j.rse.2016.02.053.
Kolodziejczyk, N., J. Boutin, J.-L. Vergely, S. Marchand, N. Martin, and G. Reverdin Mitigation of systematic errors in SMOS sea surface salinity, 2016, Remote Sensing of Environment, doi:http://dx.doi.org/10.1016/j.rse.2016.02.061.
The CATDS data are freely distributed. However, when using these data in a publication, please use the following reference and acknowledgement :
Boutin Jacqueline, Vergely Jean-Luc, Marchand Stéphane (2017). SMOS SSS L3 debias v2 maps generated by CATDS CEC LOCEAN. V2.1. SEANOE. http://doi.org/10.17882/52804#54823
"The L3_DEBIAS_LOCEAN_v2 Sea Surface Salinity maps have been produced by LOCEAN/IPSL (UMR CNRS/UPMC/IRD/MNHN) laboratory and ACRI-st company that participate to the Ocean Salinity Expertise Center (CECOS) of Centre Aval de Traitement des Donnees SMOS (CATDS). This product is distributed by the Ocean Salinity Expertise Center (CECOS) of the CNES-IFREMER Centre Aval de Traitement des Donnees SMOS (CATDS), at IFREMER, Plouzane (France)."
The CATDS-CEC Locean research products are freely available on FTP :
user : ext-catds-cecos-locean
password : catds2010
If you want to register to our mailing list to be informed about major events, please send us a message at firstname.lastname@example.org (less than 10 messages / year).