320 likes | 533 Views
AN UPDATE ON LAND CONTAMINATION, AF-FOV BIAS DRIFT WITH LATITUDE, AND COHERENCE BETWEEN L1PP/L1OP AND MTS. Joe Tenerelli , CLS, France February 2, 2011 Contact for Cal Meeting Feb 3-4 2010: Email: jtenerelli@cls.fr Office: +33 02 98 22 44 97 Cell : +33 06 83 42 39 72. SUMMARY.
E N D
AN UPDATE ON LAND CONTAMINATION, AF-FOV BIAS DRIFT WITH LATITUDE, AND COHERENCE BETWEEN L1PP/L1OP AND MTS Joe Tenerelli, CLS, France February 2, 2011 Contact for Cal Meeting Feb 3-4 2010: Email: jtenerelli@cls.fr Office: +33 02 98 22 44 97 Cell: +33 06 83 42 39 72
SUMMARY • Since commissioning phase we have been dealing with several difficult problems for ocean surface salinity retrieval. Other than RFI and bias differences between ascending and descending passes, we have noticed the following in DPGS Level 1B brightness temperature solutions: • SSS bias drift with latitude in descending passes after October 2010 • land contamination very far from coastlines • Moreover, Ignasi Corbella has provided solutions, based on his MTS software, that seem significantly better than those of DPGS in terms of both reduced bias variation with latitude and reduced land contamination. • Here we show that it is possible to obtain solutions very close to those of MTS, both in terms of SSS and in terms of bias variation with latitude, using exactly the same L1A, J+, baseline weights, and image reconstruction method used by DPGS. Based on this result it is suggested that there is a problem in the DPGS processing of L1A to L1B unrelated to the overall reconstruction methodology, the choice of baseline weights or the formulation of G and J+. This problem is possibly linked to the foreign sources removal (such as direct sun correction) or the implementation of the FTT.
LATITUDINAL DRIFT AND LAND CONTAMINATION Below we show the mean bias between SMOS retrieved SSS and that of WOA Climatology for Nov 1-15, 2010 for descending (upper left) and ascending (upper right) passes separately and for their difference (lower right panel). A noticeable gradient with latitude in the bias is apparent in descending passes (upper left). Also evident is the large halos of land contamination around large land masses. The halo positions change with pass direction as has been seen for some time. DPGS descending DPGS ascending SSS bias decreases with increasing latitude (overall ~2 psu change). DPGS desc - asc
LATITUDINAL BIAS DRIFT SSS bias decrease with latitude for descending passes is associated with (Tx+Ty)/2 increase by about 1 K (red curve).
APPROACH In order to examine these problems in the simplest possible context we developed a simple off-line image reconstruction breadboard (a combination of C++ MEX code and a MATLAB driver script) as an extension to the code we have been using for processing Level 1B products. The breadboard is able to model ocean scenes over the entire front half-space, compute corresponding calibrated visibilities (including the three sets of NIR antenna temperature Stokes vectors) using an L1-generated G matrix, and to use an L1-generated J+ matrix and baseline weights file to reconstruct images corresponding to both modeled and measured (MIRAS) calibrated visibilities. This breadboard will allow us to investigate the leading problems (drift, land contamination, foreign source removal including sun glint) in the simplest context that includes effects of image reconstruction. The breadboard image reconstruction (here termed JRECON in results) is essentially the simplest possible and incorporates no FTT; no foreign source removal (and no sky removal). In using images derived using this breadboard we only use the alias-free field of view. Also, any bias associated with the Corbella term is effectively incorporated (with all other sources of bias) directly into a fixed OTT for each polarization. Thus, it is implicitly assumed that the set of all receivers’ physical temperatures remain fixed in time (in this case about 295 K), though no explicit assumption is made regarding their equality.
APPROACH For the forward scene modeling of visibilities, we go no further than computing the set of 15996 calibrated visibilities and so we do not simulate the MIRAS mode switching. This should not pose a problem for ocean scenes (at least for the basic problems we are dealing with now). The breadboard image reconstruction from the MIRAS measurements involves combining calibrated visibilities from eight arm polarization modes (XXX, YYY, three XYY and three YXX modes), mapping them from the ASIC Correlators Board matrix domain into the vector of 15,996 calibrated visibilities (simply averaging the few correlations that map onto the same 15,996-vector element). After creating the visibility vector, we apply the baseline weights and then J+ to obtain the brightness temperature Fourier components. We then apply the standard Blackman filter (as we have always done for processing L1B data) to obtain physical space brightness temperature maps over the fundamental hexagon. In doing this we do not explicitly account for the Corbella term using some representative receiver physical temperature and we do not apply an FTT; instead, we implicitly incorporate these and other bias corrections into the OTTs, which is therefore different (and generally larger) than DPGS OTTs. The fixed OTTs are then applied to all reconstructed Stokes vectors and the SSS in computed using the same linearization I have used in the past.
APPROACH • Importantly, in the results that follow I have used exactly the same L1A calibrated visibilities, baseline weights and J+ matrix to compute images as used by DPGS for both the November orbits and for the reprocessing campaign. Therefore, any differences between my solutions and those of DPGS should not be related to problems in the calibrated visibilities, J+ or baseline weights themselves. • The most striking results I have found are • In terms of AF-FOV mean bias relative to the model, the JRECON breadboard solutions behave more consistently like those of MTS than those of DPGS, especially for descending passes in November. In particular, the large linear drift with latitude in the DPGS November descending pass solutions does not exist in JRECON and MTS solutions, even though JRECON uses the same L1A, J+ matrix and baseline weights as DPGS. • The large positive SSS bias halos around land areas in the DPGS solutions are absent in the JRECON and MTS salinity maps, even though JRECON uses the same L1A, J+ matrix and baseline weights as DPGS. • These two results point to a potential problem in the DPGS processing of L1A to L1B that may originate with either the foreign sources removal algorithms or perhaps in the implementation FTT (FTR rescaling?).
Two Sets of Ocean Target Transformations (OTTs): DPGS and JRECON for 2010-11-10 Pacific Ascending Pass DPGS OTTs Txx Tyy DPGS OTT maps for the four Stokes vector components (Txx,Tyy,Uxy,Vxy) based upon the half-orbit shown above. Uxy Vxy
Two Sets of Ocean Target Transformations (OTTs): DPGS and JRECON for 2010-11-10 Pacific Ascending Pass JRECON OTTs Txx Tyy JRECON OTT maps for the four Stokes vector components (Txx,Tyy,Uxy,Vxy) based upon the same half-orbit shown above. OTTs for Uxy and Vxy are similar to those for DPGS, but those for Txx and Tyy are significantly different, owing to the fact that for JRECON we do not apply the FTT, any foreign sources corrections, or any correction for the nonzero receivers’ physical temperatures. Therefore, these OTTs contain implicitly all sources of bias, including that associated with the Corbella term in the fundamental equation relating visibilities to scene brightness temperature. With this approach we cannot account for possible changes in the overall physical receiver temperatures but this does not pose a problem for the period considered. Uxy Vxy
Two Sets of Ocean Target Transformations (OTTs): DPGS and JRECON for 2010-11-10 Pacific Ascending Pass JRECON OTTs Txx Tyy It is important to note that these OTTs should be considered valid only so long as the receivers’ physical temperatures (which are not generally identical) do not change ‘too much’ and the scene brightness distribution does not change ‘too much’, where ‘too much’ has to be determined by extensive testing. Uxy Vxy
BIAS VARIATION WITH LATITUDE In the following slides we shall examine the AF-FOV mean bias between reconstructed SMOS brightness temperatures (as obtained using DPGS L1OP, JRECON, and MTS) and those predicted by our best ocean scene model. To focus attention on differences in bias variation with latitude among the solutions I have adjusted all curves to the same mean level. Overall level differences may be possibly attributed to slight differences in FTT implementation (including choice of mean scene Tb) or, in the case of JRECON, no use of the FTT. The main concern here is only differences in bias trends with latitude as we are especially concerned with trying to find the origin of the bias variation with latitude in the DPGS solutions.
BIAS VARIATION WITH LATITUDE Consider the same Pacific orbits and bias curves we showed earlier, with the red (DPGS) and magenta (MTS) curves in the figure below showing the AF-FOV bias trend with latitude for a descending Pacific pass on November 10, 2010. The DPGS solution is generally aligned with the solution of MTS except for the addition of a distinct trend with latitude in DPGS. The ascending pass solutions agree well (adjust for a fixed offset).
BIAS VARIATION WITH LATITUDE By contrast, the JRECON solutions (red and blue) both agree very well with those of MTS except for differences in noise level, which stems from my use of baseline weights as well as more snapshots per reconstruction (combining 8 arm polarization modes to create one full-pol Tb scene). There is notable disagreement in the ascending pass near Hawaii but this is related to my removal of land from the bias calculation (presumably Ignasi included land in his bias average).
BIAS VARIATION WITH LATITUDE Examining the difference between the DPGS solution and that of JRECON for the descending pass, we see that the difference exhibits a distinct linear trend (overall rise of about 1 K in (Tx+Ty)/2 which is equivalent to a 2 psu drop) that is consistent with the trend seen in the SSS bias maps shown earlier.
SUMMARY Looking at the direct sun impact on the antenna temperatures Tx and Ty assuming Thsun=Tvsun=100,000 K, we see that for the ascending pass there is very little impact:
SUMMARY But for the descending pass the impact is much larger as the sun passes quite far into the front half-space of the array. In the lower right panel we show the modeled impact for all antennas as a function of boresight latitude. The trend is nearly linear and the variation among antennas is quite large. The variation decreases southward partially because I was obliged to use a mean antenna pattern within a few deg of the unit circle and in the back half space. The fact that this trend does not appear in the AF-FOV bias trend curves for the JRECON solutions suggests that the reconstruction without any direct sun removal effectively focuses the direct sun and keeps its mean impact on antenna temperatures outside of the AF-FOV. Why such a trend reappears in the DPGS solutions (corrected for direct sun) is an interesting question.
SUMMARY Maps of bias between SSS derived using DPGS L1B data and SSS derived using L1A and the JRECON breadboard are shown below. The strong land contamination halos in the DPGS solutions are not as evident in the JRECON solutions. For the descending passes, the linear bias gradient with latitude in the DPGS solutions (lower left) is not seen in the JRECON solutions (lower right). For all maps the sets of fixed OTTs shown earlier were used for both DPGS and JRECON solutions. DPGS ASC JRECON ASC DPGS DESC JRECON DESC
SUMMARY For the Nov descending passes, the DPGS SSS bias drift with latitude is consistent that the drift in AF-FOV bias shown earlier: DPGS DESC-ASC JRECON DESC-ASC DPGS (Tx+Ty)/2 increase With latitude
SUMMARY Subtracting the descending pass DPGS SSS from the JRECON SSS map, we see clearly both the dramatic difference between the JRECON solutions and the DPGS solutions that are affected by strong land contamination and the linear gradient in SSS bias with latitude. Again, exactly the same L1A files, J+ matrix and baseline weights were used by DPGS and JRECON so these should not be the source of these differences.
AUG 2010 SSS MAPS FROM DPGS AND MTS As Nicolas Reul has already shown, MTS SSS exhibits much less land contamination than DPGS SSS and the MTS SSS seems quite similar to what we obtain with JRECON solutions. DPGS MTS
CONCLUSIONS • By performing simple breadboard image reconstructions using the same methodology as in L1PP/L1OP and using the same L1A files, J+ matrix and baseline weights as in DPGS, we have been able to obtain solutions that are well-aligned with those of MTS in terms of AF-FOV mean bias. • Moreover, these breadboard solutions exhibit far less land contamination and much less if any unphysical bias variation with latitude than those of DPGS, at least for orbits tested in Nov 2010. • We are not performing any foreign sources corrections and we do not use an FTT. Instead, we simply apply fixed OTTs derived from a single ascending pass. • These results may help to narrow down the search from problems in DPGS processing from 1A to 1B.