News-Blog‎ > ‎

Tropical Atlantic SSS- 2010 seasonal cycle as seen from SMOS using Level 1 data from the first ESA reprocessing

posted Jun 9, 2011, 7:29 AM by Salinity CERSAT   [ updated Jun 20, 2011, 3:18 AM ]
To assess the quality of the first version of the ESA operational Level 2 SSS swath products, satellite/in situ SSS data match-ups were collected over the second half of 2010. This database reveals that 95% of the SMOS Level 2 products show a global error standard deviation on the order of ~1.3 practical salinity scale (pss). Simple spatio-temporal aggregation of the Level 2 products to generate monthly SSS maps at 1°x1° spatial resolution reduces the error down to about 0.6 pss globally and 0.4 pss in the tropics for 90% of the data. Several major local problems were however detected in the products. Systematically, SMOS operational Level 2 SSS data are too salty within a ~1500 km wide belt along the world coasts and sea ice edges, with a contamination intensity and spread varying from ascending to descending passes. This spurious signal in the OP products was found to be  present when brighter land or ice masses are located in the extended field of view domain of the antenna. Numerous world ocean area are moreover permanently or intermittently contaminated by Radio-Frequency Interferences (rfi), particularly in the northern high latitudes and following Asia coastlines. Moreover, critical spatio-temporal drifts in the retrieved SSS fields were found with varying signatures in ascending and descending passes. The detailed quality assessment of these SMOS first operational SSS products for the second half of 2010 can be found in Reul et al, 2011.


Recently, the complete year 2010 of SMOS L1A (calibrated visibilities) and L1B (brightness temperatures) data have been reprocessed by ESA with fixed calibration parameters to tentatively better understand the sources for the several problems encountered in SMOS Level 2 products, in particular over the oceans.  During the analysis of this dataset, it was found that the major sources of errors in SMOS Level 2 products :i.e.: spatio-temporal drifts, biases between ascending and descending passes and land-contamination were mostly induced by several bugs in the L1A and L1B processors. While some residual differences between ascending and descending passes may be attributed to badly modeled geophysical contributions at level 2 (such as galactic signal reflections), most of the spatio-temporal drifts observed in the SSS at Level 2 can be attributed to an erroneous correction of the direct sun impact at L1 and an unaccurate thermal model for the instrument used for this version of the processing. In addition, a bug in the weight of the visibility at zero baseline (V(0,0)) was also found to be responsible for most of the very significant land-contamination observed in the operationnal L2 products. New SSS products were therefore derived at CATDS/C-EC based on the reprocessed L1A data but using a simplified L1A to L1B processor in which the direct sun correction was omitted and the V(0,0) bug corrected.
Because the direct sun aliases contamines the Extended Field of View of the antenna, in our re-processing, the SSS retrievals were limited to the Alias-free Field of view. 
Level 3 daily and monthly SSS fields were produced from this new L1B data set at global scale with a spatial resolution of 25 km. A 10-days and 50 km moving-window average was further applied to the daily data. In the animation below, we show the new SSS products evolution over the Tropical Atlantic region combining both ascending and descending passes from January to December 2010. Note that only L1 data acquired in Full-polarisation were used to generate these new Level 3 SSS which  explains the numerous gaps in the SSS data during the commissionning phase which last from January to end of May. During that perido, the instrument was indeed operated sequentially in dual and full polarisation modes.


Legend: 10-days ruuning mean averaged SSS from SMOS full-polarisation mode data for year 2010.
 
A simple and preliminary Radio Frequency Interference filter was as well applied to the data, so that RFI contaminated data may still be present, particularly along the west coasts of Africa. In addition, the direct sun aliases are sometimes located at the border of the Alias Free domain, particulary at the end of the year (November to December) in descending passes. Note that these solar contributions that were not corrected for in our processing  produce some spurious fresh stripes in the SSS data at the end of the year.

Comparison SMOS with Thermosalinograph data

Preliminary comparison of the new monthly-averaged SMOS SSS with Thermosalinograph data acquired from Research Vessels and ships of opportunity that crused in the Tropical Atlantic in 2010 are provided for each month in the figures below:

JANUARY 2010
  
FEBRUARY 2010

MARCH 2010
APRIL 2010

MAY 2010
 
 
JUNE 2010

JULY 2010
AUGUST 2010
SEPTEMBER 2010

OCTOBER 2010


NOVEMBER 2010
 
DECEMBER 2010
 
Note that along the west-coast of Africa between 10°N and 25°N North (Senegal and Mauritania), SMOS data are most of the time too fresh with respect in situ data. Identically, strong drops in SMOS SSS are very often visible for data co-located with the transects of the Rio-Blanco across the Cape Verde and Canarian Islands  (e.g., see drops at ~18°N and ~28°N in Sep). Same comments hold for the R/V transects accross the North Caribbean Islands. These  too fresh signal  in SMOS could be caused by a remaining land contamination effect or by local RFI badly corrected for. Surprisingly, a consistent and coherent SMOS/TSG signal is found for the transect from Cuba to Panama, where a strong land-contamination could be expected due the SMOS geometry of observation of land masses in this zone. Moreover, the non-systematic effects and the significant variability in the amplitude of the Island effects (see for example the Rio-blanco transect in Apr showing a ~6 psu drop across Cape verde and the same transect showing a 2 psu drop in May, 3 psu drop in June, ~1 psu drop in August  and ~8 psu on September) suggest a potential effect of RFI. Further investigation will be required to elucidate the source for these spurious signals.
 
Some local differences between in situ TSG and monthly-averaged SMOS data, particularly around the highly variable fresh water plumes (e.g. Amazon) can  clearly be attributed  to temporal variability of the position of the plume. For example, in June, the colibri transect accross the plume (performed around the 19th of June) see the strong SSS gradient associated with the plume lying closer to the coast than the monthly-average SMOS product. Opposingly, the transect performed by the Toucan at the end of June (29th of June) exhibit  a latitudinal SSS gradient in good agreement with the one evaluated from SMOS data.
 
At the end of the year (months of Nov-Dec),  SMOS SSS retrievals appear to be systematically too fresh. This is particularly evident in december for the RV transect across the congo Plume. Both the TSG and SMOS exibit a very similar SSS gradient structure. However, SMOS SSS is fresher by about 1.5 psu. The exact source for this effect is currently unknown.
 
Considering all TSG and SMOS monthly-averaged SSS data at 25 km resolution aquired within the period from 1st of April to end of December (for which the number of full-pol SMOS data per month is considered sufficient to get reliable monthly-averages) , we found an overall root mean square error between SMOS and TSG data of about 0.73 pss, SMOS SSS being ~0.2 psu fresher than in situ TSG data in the mean.
 
 
 
The temporal eveolution of the mean and standard deviation of the SSS differences (TSG-SMOS) for each month are illustrated here below:
 
Clearly, a remaining seasonnal cycle in the mean differences between SMOS and TSG SSS is observed. This cycle is probably related to the non-corrected direct sun radiations in our algorithm for reconstructing the Tbs, which shall enhance the level of the Tbs from September to March, and therefore translate into a fresher SMOS SSS.  The error standard deviation is seen to drop from ~1.2 in january to about 0.6-0.7 pss in May, and then keep approximately stable until the end of the year. The drop in the error standard deviation from the beginning of the year to the end of May is very likely a result of the increasing number of available data acquired in full-polarisation mode, which stabilizes after may. Note however a local enhancement in the  error standard deviation  around september which is a month with strong galactic reflections in descending passes: the forward model used to correct for this effect in our algorithm is known to underestimate the galactic contributions (see previous post on these news). This may explain the local enhancement in the SSS error std in Sep. 
 

Comparison SMOS with Pirata Mooring Data

The  Pirata Mooring  array is constituted of 17 moorings distributed as shown in the Figure below and it provides times series of SSS data at 1 m depth. Note that the full year period is not covered by all moorings.  Detailed time series of SMOS and in situ data for each mooring can be seen in individual plots by clicking on the mooring number in the figure below:



 Legend: location of Pirata moorings in the Tropical Atlantic.click on the mooring symbol to see the SMOS and in situ SSS times series

We have as well considered the real-time data from the NTAS buoy-IX located at 15°N, 51°W (Northwest Tropical Atlantic Station denoted by a star in the plot shown above).  The overall comparison between SMOS SSS products at 25 km resolution and temporally averaged over 10-days or monthly periods are given in the figures below, considering data from all buoys over the year:
 
Legend: Regression plots between Pirata SSS and SMOS SSS. Left: 10-days averages. Right: monthly averages.
 
The overall root mean square errors are found to be 0.35 and 0.51 pss for monthly and 10-days averaged products, respectively. Note that in the low SSS regime (SSS<35.5), SMOS SSS is in general found slightly fresher by ~0.2 to 0.5 pss than the in situ measurements. It is not yet clear wether this bias in the satellite products is related to an algorithm issue, a calibration issue or to a measure of the vertical SSS gradients from the surface freshwater skin layer probed by the satellite to the 1 m depth in situ sensing.
 
Further discussions, results and additional comparison of these new SMOS SSS products with in situ data will be provided soon on this web site.
  



ą
Salinity CERSAT,
Jun 14, 2011, 5:36 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:38 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:38 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:38 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:39 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:39 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:39 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:39 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:39 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:39 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:36 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:36 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:37 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:38 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:38 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:38 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:38 AM
ą
Salinity CERSAT,
Jun 14, 2011, 5:38 AM
Comments