Date post: | 31-Dec-2015 |
Category: |
Documents |
Upload: | george-erickson |
View: | 33 times |
Download: | 3 times |
QWG-10 ESRIN 4-6 February 2013
Quality control study for SMOS data / Flags analysis
C. Gabarró, J. Martínez, E. OlmedoM. Portabella, J. Font and BEC team
J. Boutin & N. Martin, LOCEANJ.L. Vergely, ACRI-st
SMOS-BEC Activities Overview, 12-June-2008, Barcelona SPAIN
Analysis of the best combination of L2 flags on L2 and L3 data quality: focus on the usefulness of Chi2 probability flag and galactic noise flag.
L2 comparison with ARGO floats data by selecting data with: +/- 50 Km between SMOS & ARGO +/- 5 days between SMOS & ARGO
L3 SMOS SSS with/without flag sorting compared with in situ optimal interpolation maps (ISAS/IFREMER)
Four different flags combinations have been tested at L2 and L3. The RMS (mean/std) and also number of points are important.
Three periods of time: February -> low galaxy effect March 2011 ascending -> large effect of galaxy August -> large galaxy effect
2 / 10
Flagging combination Cases
SMOS-BEC Activities Overview, 12-June-2008, Barcelona SPAIN
Fg_ctrl_chi2_P: main goodness of fit indicator; flag on SSS is raised if the probablility that an anomaly occurs about the fit is >95% or less than 5% (too good fit adjustment)
Fg_ctrl_gal_noise: galactic noise flag; flag on SSS is raised if 10% of Tb along a dwell are discarded from the SSS retrieval (e.g. because they are affected by a scattered galactic noise >4K) = since these Tbs are removed from SSS retrieval, retrieved SSS should remain usable (although noisier)
In L2OS v5 Fg_gal_noise included in the Fg_ctrl_poor_geophysical and Fg_ctrl_chi2_P in Fg_ctrl_poor_retrieval
3 / 10
Tested flags
QWG-10 ESRIN 4-6 February 2013 4 / 10
Fg_ctrl_chi2 and Fg_ctrl_chi2_P filters description
L2P handles two different flags related with distribution:
())+
(
Where:
are the Nm observations performed at different angles
T represents the transposition operation
is the variance/covariance matrix for Tb
are different parameters to be retrieved
are the a priori knowledge of parameters (ontained from models or satellites, auxiliary information
is the variance/covariance matrix for these parameters
QWG-10 ESRIN 4-6 February 2013 5 / 10
Fg_ctrl_chi2 and Fg_ctrl_chi2_P filters description
Two different flags are associated to the the residual of the cost function:
Fg_ctrl_chi2Fg_ctrl_chi2_P
When then Fg_ctrl_chi2 is set to 1=>BAD QUALITY; (Tg_Chi2 = 1.35 (corresponding to Fg_ctrl_chi2_P for N=50) => redundant with Fg_Chi2_P for N>50 and not statistically relevant because it does not take into account theoretical distribution of Chi2 as a function of Nm => Better to use Chi2_P
Fg_ctrl_chi2_P based on the theoretical expected distribution of Chi2 (gamma function). By default 5%<Tg_chi2_P<95% (i.e. with level 2 processor scaling 50 <Dg_Chi2_P<950)
Tested thresholds: Low thresholds: 0%, 5% (default), 25%, 50%, 75% High thresholds: 25%,50%,75%,95% (default), 100%
QWG-10 ESRIN 4-6 February 2013 6 / 10
Dg_chi2_P_X=GAMMq(Nm/2, Chi2*Nm /2)
If Tg_chi2_P_min < Dg_chi2_P_X < Tg_chi2_P_max =>Fg_ctrl_chi2_P=0 => GOOD QUALITY
Otherwise Fg_ctrl_chi2_P=1 => BAD QUALITY
Sometimes the Chi2 distribution does not match well with the theoretical one => useful indicator of problems for future studies.....
Fg_ctrl_chi2_P filter description
Nm=33 Nm=97 Nm=137
Comparison of density functions (real->red theoretical->blue):
>95%
<5%
BAD QUALITY
BAD QUALITY
GOOD QUALITY
GOOD QUALITY
GOOD QUALITY
>95% >95%
<5% <5%
BAD QUALITY
BAD QUALITY
BAD QUALITY
QWG-10 ESRIN 4-6 February 2013
Fg_ctrl_chi2 & Fg_ctrl_chi2_P
7 / 10
Most of the points filtered out by Fg_ctrl_chi2 are at the edges of the swath (for N<50)
% of filtered points
march june desember 2011
Fg_ctrl_chi2 17.58 16.24 27.47
Fg_ctrl_chi2_P 29.55 27.99 39.13
<SSSsmos-SSSargo> becomes very negative for Chi2P>95%
Chi2P>95%QWG10, 4-6 Feb. 2013
Std(SSSsmos-SSSargo)
Chi2P>95%Chi2P<95%Chi2P<75%
Std increases when Chi2P>95% => flag is useful for removing bad retrieval
Std larger for 75%<Chi2P<95%=>Chi2_P good indicator of SSS quality
Std remains at low values for Chi2P<75%
QWG10, 4-6 Feb. 2013
Number of SMOS measurements: about 20% with Chi2P>95%
Chi2P>95%
All Chi2P
75%<Chi2_P<95%
QWG10, 4-6 Feb. 2013
Galactic noise flag: Qualitative effect on SMOS
SSS maps:
Application of the flag suppress numerous SSS in latitudinal bands => large
noise and missing pixels in these areas (e.g. In
Southern hemisphere March 2011, ascending
orbits)
No obvious SSS bias when removing the flag
QWG10, 4-6 Feb. 2013
SMOS SSSMarch 2011 asc orbits
Galactic noise flag: Qualitative effect on SMOS
SSS maps:
Application of the flag suppress numerous SSS in latitudinal bands => large
noise and missing pixels in these areas (e.g. In
Southern hemisphere March 2011, ascending
orbits)
No obvious SSS bias when removing the flag
QWG10, 4-6 Feb. 2013
SMOS-ISAS SSSMarch 2011 asc orbits
Quantitative effect of Galactic noise flag on SMOS SSS: comparison with ARGO SSS over global Ocean
SSS averaged over +/-5days, +/-50km around ARGO)
Flag testedFlag not tested
Number of colocations
QWG10, 4-6 Feb. 2013
Quantitative effect of Galactic noise flag on SMOS SSS: comparison with ARGO SSS (Global Ocean)
(SMOS SSS averaged over +/-5days, +/-50km around ARGO)
Flag testedFlag not tested
Mean difference SSSsmos-SSSargo
QWG10, 4-6 Feb. 2013
Quantitative effect of Galactic noise flag on SMOS SSS: comparison with ARGO SSS (Global Ocean)
(SMOS SSS averaged over +/-5days, +/-50km around ARGO)
Flag testedFlag not tested
Std difference SSSsmos-SSSargo
QWG10, 4-6 Feb. 2013
Quantitative effect of Galactic noise flag on SMOS SSS: comparison with ARGO SSS in S. subtropical Pacific (10°S-30°S)
(SMOS SSS averaged over +/-5days, +/-50km around ARGO)
Flag testedFlag not tested
Number of colocations
QWG10, 4-6 Feb. 2013
Quantitative effect of Galactic noise flag on SMOS SSS: comparison with ARGO SSS in S. subtropical Pacific (10°S-30°S)
(SMOS SSS averaged over +/-5days, +/-50km around ARGO)
Flag testedFlag not tested
Mean difference SSSsmos-SSSargo
QWG10, 4-6 Feb. 2013
Quantitative effect of Galactic noise flag on SMOS SSS: comparison with ARGO SSS in S. subtropical Pacific (10°S-30°S)
(SMOS SSS averaged over +/-5days, +/-50km around ARGO)
Flag testedFlag not tested
Std difference SSSsmos-SSSargo
QWG10, 4-6 Feb. 2013=> We propose to remove Fg_Ctrl_gal_noise from Fg_poor_geophysical
QWG-10 ESRIN 4-6 February 2013
The following cases have been analyzed at L2 & L3:
Case 1: NO filtering by anything & Xswath=300km
Case 2: RFI=1, High_wind=1 (>12m/s), Poor_ret=1, Poor_geo=1, Xswath=300km: v5 Ctrl_poor_.. flags filtering
Case 3: RFI=1, High_wind=1, Poor_ret=1 , Poor_geo=1 (- gal_noise) , Xswath=300km
Case 4: RFI=1, High_wind=1, Poor_ret=1 (-chi2), Poor_geo=1 (- gal_noise ) , Xswath=300km
19 / 10
Flagging combination Cases
QWG-10 ESRIN 4-6 February 2013 20 / 10
ASC/DES orbits 21-30 August [-60,60]
DES
ASC
number of points decrease for fg_ctrl_gal_noise ¬ better RMS
Same number of points only 300km of swath considered
1: No filter2: poor_geo +poor_ret3: poor_ret=1, poor_geo=1 (- gal_noise )4: poor_ret=1 (-chi2), poor_geo=1 (- gal_noise )
L2-ARGO
Since only 300 Km are consideredhere GP with CHI2 set are already filtered by CHI2_P.
QWG-10 ESRIN 4-6 February 2013 21 / 10
ASC/DES orbits 21-30 August [-60,60]
DES
ASC
number of points decrease for fg_ctrl_gal_noise ->not better values
Same number of points only 300km of swath considered
1: No filter2: poor_geo +poor_ret3: poor_ret=1, poor_geo=1 (- gal_noise )4: poor_ret=1 (-chi2), poor_geo=1 (- gal_noise )
L2-ARGO
mean & STD (L2-ARGO)mean & STD (mean(L2)-ARGO)
QWG-10 ESRIN 4-6 February 2013
ASC+DES 21-30 August [-60,60]
22 / 10
1: No filter2: poor_geo +poor_ret3: poor_ret=1, poor_geo=1 (- gal_noise )4: poor_ret=1 (-chi2), poor_geo=1 (- gal_noise )
ASC+DES
mean & STD (L2-ARGO)mean & STD (mean(L2)-ARGO)
L2-ARGO
QWG-10 ESRIN 4-6 February 2013 23 / 10
ASC/DES orbits 21 Feb-2 Mar [-60,60]
ASC
DES
1: No filter2: poor_geo +poor_ret3: poor_ret=1, poor_geo=1 (- gal_noise )4: poor_ret=1 (-chi2), poor_geo=1 (- gal_noise )
L2-ARGO
QWG-10 ESRIN 4-6 February 2013
ASC+DES orbits 21 Feb-2 Mar [-60,60]
24 / 10
1: No filter2: poor_geo +poor_ret3: poor_ret=1, poor_geo=1 (- gal_noise )4: poor_ret=1 (-chi2), poor_geo=1 (- gal_noise )
L2-ARGO ASC+DES
Similar results for region [-30,30]
QWG-10 ESRIN 4-6 February 2013
ASC/DES orbits August/Feb [-10,10]
25 / 10
DES
ASC
August February
NOW pseudo L3 similar BIAS & less STD
1: No filter2: poor_geo +poor_ret3: poor_ret=1, poor_geo=1 (- gal_noise )4: poor_ret=1 (-chi2), poor_geo=1 (- gal_noise )
QWG-10 ESRIN 4-6 February 2013
CONCLUSIONS
Better results are obtained when NOT using Fg_ctrl_gal_noise, both at Level 2 and Level 3 analysis -> this filters out many good points during some periods of time.
We consider flag Fg_ctrl_Chi2 should not be used because is theoretically incorrect (should depend on Nm) and GP with bad retrieved are already filtered by Fg_ctrl_Chi2_P.
Large variability on the L2 SMOS data is observed at high latitudes (-40, -60).
Still some land sea contamination is observed.
26 / 10
SMOS-BEC Activities Overview, 12-June-2008, Barcelona SPAIN 27 / 10
mean(L2) – ARGO : Pseudo L3
L2 SMOS data with the sameARGO float measurementare averaged -> pseudo L3
5- 24 june 2011 (reprocessed data)
Large variability are high latitudes
Land Sea contaminationEffect or RFI?
STD (SMOS) per ARGO
(mean(L2 SSS) – ARGO)
Extra slides
QWG10, 4-6 Feb. 2013
SMOS-BEC Activities Overview, 12-June-2008, Barcelona SPAIN
Definition of Fg_ctrl_poor_retrieval & Fg_ctrl_poor_geophysical in v500
If
(Fg_ctrl_many_outliers(igp) = = true or Fg_ctrl_sunglint(igp) = = true or Fg_ctrl_moonglint(igp) = = true or Fg_ctrl_gal_noise(igp) = = true or Fg_ctrl_gal_noise_pol(igp) = = true or Fg_ctrl_num_meas_low(igp) = = true or Fg_sc_TEC_gradient(igp) = = true or Fg_sc_suspect_ice(igp) = = true or Fg_sc_rain(igp) = = true )
then Fg_ctrl_poor_geophysical (igp) = true
If (Fg_ctrl_retriev_fail(igp) = = true or Fg_ctrl_range(igp) = = true or Fg_ctrl_sigma(igp) = = true or Fg_ctrl_chi2(igp) = = true or Fg_ctrl_chi2_P(igp) = = true or Fg_ctrl_marq(igp) = = true or Fg_ctrl_reach_maxiter(igp) = = true)
Then Fg_ctrl_poor_retrieval(igp) = true
29 / 10
QWG-10 ESRIN 4-6 February 2013
ASC/DES orbits 21-30 August [-30,30]
30 / 10
1: No filter2: poor_geo +poor_ret3: poor_ret=1, poor_geo=1 (- gal_noise )4: poor_ret=1 (-chi2), poor_geo=1 (- gal_noise )
ASC
DES
QWG-10 ESRIN 4-6 February 2013
ASC/DES orbits 21 Feb-2 Mar [-30,30]
31 / 10
DES
ASC
1: No filter2: poor_geo +poor_ret3: poor_ret=1, poor_geo=1 (- gal_noise )4: poor_ret=1 (-chi2), poor_geo=1 (- gal_noise )
QWG-10 ESRIN 4-6 February 2013
Analysis on L3 MAPS ON DESCENDING
32 / 10
GALACTIC NOISE FLAG ON Descending & August
[-60,60] [-30,30]
-0. 50 0.9
0.4
1.0
0.55
L3 Maps 1ºx1ºSMOS-ARGO
Mean
(L3
-AR
GO
)S
TD
(L3
-AR
GO
)
-0.15
-0. 50
STD
(L3
-AR
GO
)M
ean
(L3
-AR
GO
)
-0.15
Mean
STD
Best results
SMOS-BEC Activities Overview, 12-June-2008, Barcelona SPAIN 33 / 10
Fg_ctrl_chi2 filter description
Xi2/Nm>1.35
ALL THESE POINTS ARE FILTERED
Nm=33 Nm=97 Nm=137
Xi2/Nm>1.35
ALL THESE POINTS ARE FILTERED
Xi2/Nm>1.35
ALL THESE POINTS ARE FILTERED
Comparison of distribution functions (real->grey theoretical->green):PORTION OF DISGARDED POINTS IS NOT ALLWAYS THE SAME
When then Fg_ctrl_chi2 is set to 0=>GOOD QUALITY
When then Fg_ctrl_chi2 is set to 1=>BAD QUALITY
PROBLEM: depends on Nm, but is constant for all Nm (Nm: number of measurements).