Aquarius / SMAP Ocean Roughness and SSS
Alex Fore, Simon Yueh, Wenqing Tang, Akiko Hayashi
L2B and L3 data are available at: h2p://ourocean.jpl.nasa.gov
© 2016 California InsItute of Technology, Government Sponsorship acknowledged
SMAP AND OCEAN ROUGHNESS
0 5 10 15 200
0.005
0.01
0.015
0.02
0.025
0.03
0.035
WindSAT Wind Speed [m/s]
Exce
ss S
urfa
ce E
mis
sivi
tySMAP GMF vs Aquarius GMF: A0; T12323
SMAP A0 eHAQ A0 eHSMAP A0 eVAQ A0 eV
SMAP and Aquarius roughness model agree well
0 5 10 15 20−0.5
0
0.5
1
1.5
2
2.5x 10−3
WindSAT Wind Speed [m/s]
Exce
ss S
urfa
ce E
mis
sivi
ty C
osin
e Am
plitu
deSMAP GMF vs Aquarius GMF: A1; T12323
SMAP A1 eHAQ A1 eHSMAP A1 eVAQ A1 eV
0 5 10 15 20−3
−2.5
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2x 10−3
WindSAT Wind Speed [m/s]
Exce
ss S
urfa
ce E
mis
sivi
ty C
osin
e Am
plitu
deSMAP GMF vs Aquarius GMF: A2; T12323
SMAP A2 eHAQ A2 eHSMAP A2 eVAQ A2 eV
0 5 10 15 20−30
−28
−26
−24
−22
−20
−18
−16
−14
−12
−10
WindSAT Wind Speed [m/s]
SMAP GMF vs Aquarius GMF: HH bias: −0.62; VV bias: −0.32 SMAP HH vs PALSAR: −0.30; R12170
SMAP A0 HHAQ A0 HHPALSAR A0 HHSMAP A0 VVAQ A0 VV
0 5 10 15 200
0.05
0.1
0.15
0.2
0.25
0.3
0.35
WindSAT Wind Speed [m/s]
SMAP GMF vs Aquarius GMF: A1/A0; R12170
SMAP A1/A0 HHAQ A1/A0 HHSMAP A1/A0 VVAQ A1/A0 VV
0 5 10 15 20−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
WindSAT Wind Speed [m/s]
SMAP GMF vs Aquarius GMF: A2/A0; R12170
SMAP A2/A0 HHAQ A2/A0 HHSMAP A2/A0 VVAQ A2/A0 VV
0 5 10 15 20−38
−36
−34
−32
−30
−28
−26
WindSAT Wind Speed [m/s]
SMAP GMF vs Aquarius GMF: HV bias: 0.37VH bias: 0.14; R12170
SMAP A0 HVSMAP A0 VHAQ A0 HV
SMAP SAR data extends L-‐band roughness model to extreme winds
Blanca 6/4/2015 1300Z; NHC advisory indicates 50 m/s sustained 60 m/s gusts Image is VIIRS Suomi 6/4/2015 0830 UTC
−150 −100 −50 0 50 100 150−20
−19.5−19
−18.5−18
−17.5−17
−16.5−16
−15.5−15
−14.5−14
−13.5−13
−12.5−12
−11.5−11
−10.5−10−9.5−9
−8.5−8
−7.5−7
−6.5−6
Tangential Wind Direction
SMAP L1C Blanca at > 25 km & < 45 km from eye location
HHVVHH GMFVV GMF
Upwind (wdir opposite look dir)
−150 −100 −50 0 50 100 150−30
−29
−28
−27
−26
−25
−24
Tangential Wind Direction
SMAP L1C Blanca at > 25 km & < 45 km from eye location
XPOLXPOL GMF
Upwind (wdir opposite look dir)
AQUARIUS AND SMAP SSS
10
(a) ARGO May 2015 (b) HYCOM May 2015
(c) Aquarius May 2015 (d) SMAP TB-only May 2015
Fig. 7. (a) ARGO SSS map for May 2015, (b) same for HYCOM, (c) same for Aquarius, and (d) same for SMAP. The increase in detail from
the ARGO map to the SMAP SSS map is striking, we see far more detail and small-scale SSS events than in the monthly buoy maps and than
in the Aquarius map. Even the HYCOM salinity map vastly underestimates the freshening events that occur in the Amazon freshwater plume.
Other noticeable differences between the maps include the Mississippi river fresh water plume and the east Pacific.
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
WindSat / SSMI/S Wind Speed [m/s]
Salin
ity D
iffer
ence
[psu
]
SMAP Salinity Difference to HYCOM
CAP Bias: 0.05TB−Only Bias: 0.05CAP RMS: 0.74TB−Only RMS: 0.80
Fig. 8. SSS bias (diamonds) and STD (squares) with respect to HYCOM as a function of WindSat / SSMI/S wind speed. The CAP SSS product
out-performs the TB-only product for high wind speeds in particular, where the ancillary wind speed product used is likely to be biased low.
The inclusion of �0 allows for an independent estimate of the wind speed, removing the ambiguity inherent in TB-only processing between
excess TB due to wind roughness of the surface and excess TB due to low SSS.
March 15, 2016 DRAFT
TB SSS Processing • Compute delta TB using ancillary data and model – Average over each day; use 8 day median filtered value
– Decimated by fore/a` x asc/dec • Grid into a 25 km L2A swath grid a la RapidSCAT – Gridding method oversamples observaIons onto the grid.
– EffecIve resoluIon is somewhat larger than 40 km • EsImate wind speed and salinity using constrained objecIve funcIon minimizaIon
2.4 Level 2B Algorithms
The inputs to the L2B algorithms are the averaged “four-flavor” (H-fore, H-aft, V-fore, V-aft) TB observa-tions computed in the L2A algorithm with the �TB corrections computed in Section 2.2 applied for eachflavor and ascending / descending portion.
2.4.1 Combined SSS/WSPD Retrieval
Due to the way in which salinity and wind speed a↵ect the sea surface emissivity, we are not able to fullyseparate the e↵ects of surface roughness and salinity. In the combined SSS/WSPD processing we allow thewind speed to vary within a region about the ancillary wind speed via the objective function while leavingthe salinity unconstrained. We use a maximum likelihood method with the following objective function
F (spd, sss) =X
i
TB,i � Tm
B,i (spd, sss, anc dir, anc swh, anc sst)
NEDTi
�2+
✓spd� spd anc
1.5m/s
◆2
, (2.1)
where TB,i is one of the four flavors of TB, TmB,i is the model value of TB, and we use the GMFs developed
in [6, 7]. Additionally we constrain the wind speed to be greater than zero and less than 50 m/s andthe salinity to be greater than zero and less than 40 psu. We use NLopt and constrained optimizationby linear approximations method [3, 4] to minimize this objective function. WSPD and SSS minimumobjective function solutions to this problem are the final L2B retrievals. The combined WSPD and SSSprocessing generates the L2B datasets “smap sss” and “smap spd”.
2.4.2 High Wind Speed Retrieval
If we fix the salinity at the ancillary HYCOM value, we are able to solve for the wind speed without anyconstraints using the following objective function:
F (spd) =X
i
TB,i � Tm
B,i (spd, anc sss, anc dir, anc swh, anc sst)
NEDTi
�2. (2.2)
The main di↵erences between the high wind speed processing and the combined processing are the fixingof salinity at the ancillary value and the removal of the wind speed term in the objective function. Thehigh wind speed processing generates the L2B datasets “smap high spd”.
Users should be aware that errors in the ancillary salinity will map to errors in the wind speed retrievedusing this method. Typically one will see erroneously high wind speeds in regions such as the Amazonriver outflow and other major rivers. This wind speed product is intended for use only in high wind speedconditions such as tropical storms.
2.5 Level 3 Algorithms
A Level 3 (L3) product is also produced at JPL, which contains the map-gridded SSS, WSPD, and a high-wind version of WSPD from L2B products. The map grid resolution is 0.25� in latitude and longitude.We use Gaussian weighting to interpolate the L2B estimates onto the map grid with a search radius ofapprox. 45 km and a half-power radius of 30 km. Bit 0 of the L2B “quality flag” dataset is used to filterthe data before aggregation into the L3 map product.
4
j-1
j
j+1
i-1 i i+1
Fig. 3. An example of the L2A gridding algorithm: the solid black grid lines represent the boundaries between the SWCs while the two ellipses
represent two sequential L1B footprint observations. i represents the cross-track coordinate while j represents the along-track coordinate. The
dashed boxes within each SWC indicate the size of the “overlap” region. Any L1B observation whose footprint falls within the dashed “overlap”
region for each SWC will be included in that SWC for salinity processing. For example, the dark gray footprint will be assigned to SWCs
{(i, j � 1) , (i, j) , (i+ 1, j � 1) , (i+ 1, j)}.
“latitudes” are linearly scaled to generate the Salinity Wind Cell (SWC) grid indices which are approximately 25
km in spacing [6].
After computing the SOM coordinates for all TB footprints, we assign each TB footprint to every SWC that
the footprint 3 dB contour overlaps a configurable portion of. This gridding algorithm was developed for Version
3 of the QuikSCAT data products and is currently used for processing RapidScat data [7], and is known as the
overlap method. This gridding algorithm over-samples the TB observations onto the SWC swath in a way that is
consistent with the measurement geometry. In Figure 3 we have an example of the L2A gridding algorithm. In this
figure the solid black lines represent the boundaries of the SWCs while the dashed lines indicate the size of the
“overlap” region, which is set to 0.75 the size of the SWC. Any L1B TB observation whose footprint falls within
the dashed “overlap” region for each SWC will be included in that SWC for salinity processing. For example, the
dark gray footprint will be assigned to SWCs {(i, j � 1) , (i, j) , (i+ 1, j � 1) , (i+ 1, j)}. The data are posted at
approximately 25 km, however, the intrinsic resolution of the L2A data is somewhat larger than the resolution of
the L1B footprints which is 40 km.
After assigning every L1B TB observation to SWCs we apply land and ice flagging to the individual TB
measurements and remove all observations that are flagged as land/ice from each SWC. Any SWC containing an
observation that is flagged as land/ice and was removed is then flagged as having possible land/ice contamination
in the quality flag. We then average the H-pol and V-pol TB for fore and aft looks separately to obtain up to four
looks for each SWC. We refer to these four looks as “flavors” of TB (fore H-pol, aft H-pol, fore V-pol, aft V-pol).
The reason we must aggregate the fore and aft looks separately is that the wind directional response is a function
March 15, 2016 DRAFT
15 20 25 30 35 40 45 50 55 600
2
4
6
8
10
12
14
16
18
20
Cross Track Index
Mea
n N
umbe
r of L
1B T
Bs p
er S
WC
Mean Number of L1B TBs per look in a SWC
H/ForeH/AftV/ForeV/Aft
15 20 25 30 35 40 45 50 55 600.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Cross Track Index
Aggr
egat
e N
EDT
By F
lavo
r
Aggregate NEDT By Flavor
H/ForeH/AftV/ForeV/Aft
L2A Gridding
25 km
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
WindSat / SSMI/S Wind Speed [m/s]
Salin
ity D
iffer
ence
[psu
]SMAP Salinity Difference to HYCOM
CAP Bias: 0.05TB−Only Bias: 0.05CAP RMS: 0.74TB−Only RMS: 0.80
L2B SWC
−60 −45 −30 −15 0 15 30 45 60−0.7−0.6−0.5−0.4−0.3−0.2−0.1
00.10.20.30.40.50.60.7
Latitude [deg]
SSS
Bias
[psu
]SMAP/AQ − APDRC Stats for May 2015
AQ: 0.06AQ/CAP: 0.09SMAP/TB: −0.04SMAP/TBADJ: −0.01SMAP/CAP: 0.02SMAP/RSS: 0.01
−60 −45 −30 −15 0 15 30 45 600
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
Latitude [deg]
SSS
STD
[psu
]SMAP/AQ − APDRC Stats for May 2015
AQ: 0.20AQ/CAP: 0.19SMAP/TB: 0.28SMAP/TBADJ: 0.24SMAP/CAP: 0.24SMAP/RSS: 0.27
100 101
0.1
0.12
0.14
0.16
0.18
0.2
0.22
0.24
0.26
Window Size Squared
SSS
STD
[psu
]SMAP/AQ − APDRC Stats for May 2015 vs Averaging Window
AQAQ/CAPSMAP/TBSMAP/TBADJSMAP/CAPSMAP/RSS
Summary • SMAP provides addiIonal source of L-‐band ocean roughness models
– SMAP-‐based models agree very well with Aquarius for normal winds (say up to 20 m/s).
– SMAP sampling enables pushing the models to hurricane/extreme winds, where some evidence that Aquarius roughness model isn’t quite right.
• SMAP can conInue the record of global SSS maps from space. – Accuracy is not as good as Aquarius but is close. – The STD with respect to APDRC is mostly between 0.1 to 0.2 psu
• Loss of Radar affects the salinity retrievals significantly. • Zonal biases sIll an issue for SMAP, possible due to TB calibraIon • Data available at: ourocean.jpl.nasa.gov
– V1.0 based on validated data release (ongoing) – V1.1-‐beta based on most recent radiometer data (3/15-‐2/16) – V2 expected to be release near end of April, based on v1.1-‐beta.
Flow Chart of TB-‐only SSS Processing
Chapter 2
Science Algorithm Overview
The SMAP TB-only salinity processing inherits experience from the NSCAT, QuikSCAT, RapidScat, andAquarius projects as well as L-band Geophysical Model Functions (GMFs) developed for the CombinedActive / Passive Aquarius products produced at the Jet Propulsion Laboratory (JPL). Su�cient informa-tion has been included in the L2B SSS product for users to perform their own geophysical retrievals overocean. In Figure 2.1 we give a simple overview of the data flow for the SMAP SSS/WSPD processing flow.
L1B TBAncillary data
match-upprocessing
L1B datamatch-ups
L2B datamatch-ups
l1b to dtb
8 daymedian filter
�TBby revl1b to l2a
L2A
l2a to l2b�TB
vs time
L2B SSS/WSPD
Figure 2.1: Flow chart of the SMAP SSS/WSPD processing. The two nodes in the top row are the inputsto the entire algorithm, while the bottom-most node is the output.
2.1 Pre-Processing
First we generate collocations of HYCOM SSS, NCEP wind speed and direction, NOAA optimum interpo-lation Sea Surface Temperature (SST), and NOAA WaveWatch III Significant Wave Height (SWH) withboth the L1B TB radiometer data product as well as the L2B swath grid discussed in Section 2.3.1.
6
L3 Map Data Mask • Within +/-‐ 50 degrees laItude. • Not in a few regions: – Amazon / Congo / Ganges river ouilows. – China RFI region. – East Pacific region.
• More than 300 km from coast for zonal stats. • All of various window averages completely full for plots versus window size (i.e. same map pixels for all). – EffecIvely dilates excluded regions out to max of +3 pixels away from land / or other regions
• Use exactly the same map pixels for all data products!
SMAP TB CorrecIon
• EsImated delta TB correcIon as a funcIon of laItude and Ime.
• Smoothed over 8 day repeat period.
Ascending / Left / DTBH
Q2−15 Q3−15 Q4−15 Q1−16 Q2−16
−50
0
50
−1
−0.5
0
0.5
1
Decending / Left / DTBH
Q2−15 Q3−15 Q4−15 Q1−16 Q2−16
−50
0
50
−1
−0.5
0
0.5
1
Ascending / Right / DTBH
Q2−15 Q3−15 Q4−15 Q1−16 Q2−16
−50
0
50
−1
−0.5
0
0.5
1
Decending / Right / DTBH
Q2−15 Q3−15 Q4−15 Q1−16 Q2−16
−50
0
50
−1
−0.5
0
0.5
1
Ascending / Left / DTBV
Q2−15 Q3−15 Q4−15 Q1−16 Q2−16
−50
0
50
−1
−0.5
0
0.5
1
Decending / Left / DTBV
Q2−15 Q3−15 Q4−15 Q1−16 Q2−16
−50
0
50
−1
−0.5
0
0.5
1
Ascending / Right / DTBV
Q2−15 Q3−15 Q4−15 Q1−16 Q2−16
−50
0
50
−1
−0.5
0
0.5
1
Decending / Right / DTBV
Q2−15 Q3−15 Q4−15 Q1−16 Q2−16
−50
0
50
−1
−0.5
0
0.5
1
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5−1
−0.5
0
0.5
1
1.5
2
2.5
3
3.5
4
WindSat / SSMI/S Wind Speed [m/s]
Spee
d D
iffer
ence
[m/s
]SMAP Speed Difference to WindSat / SSMI/S
TB Bias: 0.11TB RMS: 1.09
L2B SWC
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
SSMI/S Wind Speed [m/s]
Salin
ity D
iffer
ence
[psu
]AQ Salinity Difference to HYCOM
ADPS Bias: 0.09ADPS RMS: 0.40
L2 Footprints