Please note that this is an author-produced PDF of an article accepted for publication following peer review. The definitive publisher-authenticated version is available on the publisher Web site
1
Geophysical Research Letters
November 2
008;Volume 35 (22) : Pages 1-6
http://dx.doi.org/10.1029/2008GL035709
© 2008 American Geophysical Union
An edited version of this paper was published by AGU.
Archimer http://www.ifremer.fr/docelec/
Archive Institutionnelle de l’Ifremer
Direct ocean surface velocity measurements from space: Improved
quantitative interpretation of Envisat ASAR observations
J.A.Johannessen
1,2, *
, B. Chapron
3
, F. Collard
4
, V. Kudryavtsev
5,6,1
, A. Mouche
4
, D. Akimov
5
, and
K.-F. Dagestad
1
1
Nansen Environmental and Remote Sensing Center, Bergen, Norway Thormoehlensgate 47, N-5006, Bergen
Norway
2
Geophysical Institute, University of Bergen, Norway
3
Institute Francais de Recherche pour l´Exploitation de la Mer, Plouzané, France
4
CLS – Direction of Radar Applications, Plouzané, France
5
Nansen International Environmental and Remote Sensing Center, St. Petersburg, Russia
6
Marine Hydrophysical Institute, Sebastopol, Ukraine
*: Corresponding author : J.A.Johannessen, email address : johnny.johannessen@nersc.no
Abstract:
Previous analysis of Advanced Synthetic Aperture Radar (ASAR) signals collected by ESA's Envisat
has demonstrated a very valuable source of high-resolution information, namely, the line-of-sight
velocity of the moving ocean surface. This velocity is estimated from a Doppler frequency shift,
consistently extracted within the ASAR scenes. The Doppler shift results from the combined action of
near surface wind on shorter waves, longer wave motion, wave breaking and surface current. Both
kinematic and dynamic properties of the moving ocean surface roughness can therefore be derived
from the ASAR observations. The observations are compared to simulations using a radar imaging
model extended to include a Doppler shift module. The results are promising. Comparisons to
coincident altimetry data suggest that regular account of this combined information would advance the
use of SAR in quantitative studies of ocean currents.
1. Introduction
SAR measurements offer a potential to map current divergence and convergence zones,
where distinct upper layer dynamics, changes in wave properties and coupling to
biogeochemical processes occur. Kudryavtsev et al. (2005) and Johannessen et al. (2005)
proposed a practical radar imaging model (RIM) to advance the quantitative interpretation of
high resolution radar measurements of surface current features. This model explicitly builds
on a particular decomposition of the sea surface into a background of regular small wave
slopes and heights covering most of the surface, and fewer isolated very rough patches of
intermittent steep waves with large curvature and breaking waves.
Using the SAR high resolution processing principle, Chapron et al. (2005) pioneered the
method to retrieve the line-of-sight radar-detected ocean surface roughness velocity from
single antenna satellite SAR measurements. Regular access to Doppler shift measurements
from ASAR Wave Mode (WM) and Wide Swath Mode (WSM) images has been possible only
since mid 2007, providing an increasing data set of both kinematic and dynamic properties of
the radar-detected moving ocean surface roughness. This is demonstrated in Figure 1 where
the influence of the greater Agulhas Current is visible in the line-of-sight (ground range)
Doppler velocity captured by the ASAR sensor.
The single-antenna Doppler shift anomalies are obtained by subtracting the predicted from
the measured Doppler centroids. The method works best for images with quasi-uniform radar
cross-section at moderate to higher winds, predominantly used in this study, and yield
estimates with a resolution (azimuth, range) of about 10km by 6km for WM imagettes and
about 8km by 4km for WSM images with 30% overlap in azimuth. For WSM products, prior to
geophysical interpretations, corrections are applied to compensate along-track large cross
section variations and biases are further removed using land surface references. For WM
products, biases are removed for each orbit. The resulting Doppler anomalies are then
obtained with an RMS error up to 5 Hz, equivalent to respectively 0.35 m/s and 0.21 m/s in
range directed surface Doppler velocity at 23° and 33° incidence angles.
The Agulhas Current regime has been described as one of the strongest western boundary
currents (up to 2 m/s) in the world's oceans. The estimated radial Doppler velocity reaching
up towards 2 m/s (Figure 1) appears to map the expression of this current. Passing the
retroflection region centered at 16° E, the Agulhas return current meanders eastward back
into the South Indian Ocean between 38° - 40° S. This reversal of the mean flow translates
into opposite sign radial surface Doppler velocities reaching up to 1.5 m/s (Figure 1). The
persistent manifestation of these Doppler velocity signatures and the apparent agreement to
the location of the core geostrophic current derived from weakly map of altimetry are certainly
striking.
Although the Doppler velocity is not a direct surface current measurement, it inevitably
suggests that the use of Doppler observations can help to derive new and innovative
estimates of the mesoscale dynamics. To reach consistent quantitative results, a semi-
empirical model is highly preferable to guide quantitative interpretation based on both surface
roughness variation and Doppler anomaly analyses. In this paper, the RIM model extended
with the Doppler module is used to predict the expected Doppler shift. The approach is
described in section 2. Model results are compared to Envisat C-band ASAR WM and WSM
Doppler frequency shift measurements in section 3, followed by a summary in section 4.
2. Approach
The RIM builds on a two-scale asymptotic decomposition and derivation of the Doppler
velocity is straightforward (Appendix B in Chapron et al., 2005). Accordingly, a sea surface
normalized radar cross-section, (NRCS,
0
), is defined locally. It is then modulated and
experiences local vertical and horizontal movements due to longer surface waves. Over an
ocean imaged scene, the Doppler frequency f
D
becomes a mean quantity,
2
f
D
k
R
(usin
wcos
)
0
(
)
0
(
)
(1)
Here k
R
is the radar wavenumber,
u
and are the horizontal and vertical velocities of the
scattering facets, and
w
is the modification of the incidence angle
due to the local tilt
induced by the longer waves. This two-scale assumption helps to consider the NRCS
variations caused by both the change of the local surface tilt (
) and the hydrodynamic
modulation (
˜
0
h
) of the scattering facets, as
h
0
0
0
~ ~
, where
)(cos
y
sin
RxR
,
R
is the radar look direction, and
yx
, are the local
components of the sea surface slope. We ignore effects of surface tilt out of the incidence
plane. To the second order in steepness, the radial velocity
D
V
of the target (assumed
positive if directed away from the radar) writes
/sin
TH
D
DR f S f
Vfk cu
c
(2)
where
c
f
is the mean velocity of the scattering facets, and is the radial surface current
velocity. As hypothesized, facets travel along large-scale surface waves composed from a
wide spectrum of waves with k < k
L
(where k
L
is a spectral cutoff linked to the scale of the
facets), and , is the contribution due to tilting and hydrodynamic modulation of the facets.
can be expressed as:
S
u
c
f
TH
c
f
TH
c
f
TH
cot
M
f
t
M
1f
h
cos(
R
)cot
M
2f
h
ck
2
B(k)
kk
L
dk
(3)
where
M
f
t
ln(
0
)/
is the tilt modulation transfer function (MTF),
is the hydrodynamic MTF (real and imaginary part describes correlation of a
scattering facets modu
lations with elevations and slopes of the modulating waves), B(k) is the
2D saturation spectrum, and
M
f
h
M
1f
h
i M
2 f
h
M
1f
h
h
f
M
2
is the direction of k. The two first terms in (3) provide
changes of n i
TH
f
c
if
R
sig n
turns from down- to up-wind, while effect of facets-slopes
correlation (third term in (3) is not pendent on
R
de
and provides up- and down-wind
asym
RIM es
m of two
bution . The partial contri
met
ry in
D
V
.
assum the NRCS to be represented by the sum:
qq
b
p
R
p
000
)1(
where
p
R0
and
b0
are the NRCS of the regular surface (at p = vv- or hh-p rization) and the non
regular surface of breakers covering a fraction q of the sea surface.
p
R0
follows a composite
model leading to the su terms, i.e. the so-called two-scale Bragg and si-specular
s to
P
0
become
pp
br
p
br
qP
0
/)1(
,
p
sp
p
sp
qP
0
/)1(
, and
p
b
p
wb
qP
00
/
. As a key aspect, the RIM
polarization ratio becomes controlled by the non-Bragg scalar scattering contribution. The
RIM predictions are in good agreement with experimental data (e.g. by Mouche et al., 2006).
ola
q
ua
butioncontri s
,
p
brsp
p
R
0
3
The systemat
ic and significant deviation between a standard composite-Bragg scattering
model prediction and observations proved that the scalar term plays a crucial role,
comparable to the sea surface curvature effect in advanced scattering model (e.g., Mouche et
l., 2007a, 2007b). The radial Doppler velocity thus becomes:
a
V
D
u
s
P
j
P
(c
j
c
j
TH
)
(4)
r m
ith
of each o
f the types of scattering facet can be found in
udryavtsev et al. (2003b).
with the subscript j representing Bra
gg waves (br), specula irror points (sp) and breakers
(wb). For the Bragg-facets the spectral cutoff wavenumber
Lbr
k is defined as k
Lbr
= d k
R
(w
d = 1/4), while the range of longer waves modulating the breaker-facets is limited to k < k
Lwb
= d k
wb
= d k
R
/10. For specular mirror points the dominant modulating waves k
Lsp
are assumed
to be equal to the peak wavenumber in the wind wave spectrum. Explicit expressions for the
hydrodynamic modulations
K
The mean line-of-sight velocity of the scattering facets
c
j
in eq. (4) is represented as sum
of the phase speed of the Bragg waves (
a
c
b
r
), advection speed of “mirror points” (
c
sp
) and
speed of breakers (
c
wb
). The advection speed of the “mirror points” is expressed following
onguet-Higgins (1957)
L
c
sp
cos
R
cos
ck
2
B(k)dk
kdk
R
/s
up
2
sin
R
sin
ck
2
B(k)dk
kdk
R
/s
cr
2
(5)
where up- and cross-wind mean squared slopes of the large-scale surface ( and ) are
defined as
2
up
s
2
cr
s
22 2 2 2
[,] [cos,sin] ()
Lbr
up cr w
kk
s
sk
kk
.
B
t
defines ts per unit
uantity is proportional to e
nhanced rou
ghness area, and
d
The breaker-fa
cet velocity
c is scale dependent and is described in terms of cc d)( hat
the length of wave breaking fron area with velocities ranging from
c to
cc d (Phillips, 1985). The q
the fraction of thcc dk )(
1
e
c
wb
cos(
R
)ck
1
(c)dc
kk
R
/10
/k
1
(c)dc
kk
R
/10
(6)
spectral distribution of the breaking f replaced by
where is the wind wave growth rate.
Assuming that the energy losses are proportional to the energy input from the wind, the
ronts in eq. (6) can be
kkcc dBkd )()(
1
2
*
)/( cu
3. Model results and comparisons
Results of the extended RIM - Doppler model (hereinafter DopRIM) are presented and
compared to Doppler anomalies obtained from the global Envisat ASAR WM data. Following
eq. (4), the partitioning of the scattering contributions plays an essential role to quantify the
individual contributions to the total Doppler velocity. Each weight is wind speed and direction
dependent, as well as incidence angle and polarization dependent. The specular point
4
velocity always dominates
V
D
at low incidence angle. With increasing incidence angles, this
part of the Doppler velocity becomes negligible. At moderate incidence angles, the simulation
of the total Doppler velocity predicts values that are about 35% of the wind speed. This is
significantly larger than expected from the phase speed of the Bragg waves and the wind
induced surface drift (about 3% of wind speed). The two-scale decomposition with tilting and
hydrodynamic effects explains this difference. More specifically, at moderate incidence
angles, the composite-Bragg facet velocity is larger for HH than for VV polarization. This is
anticipated from the larger tilting effects at HH than at VV. The composite non-Bragg facet
velocity has a relatively small weight for VV. On the other hand, for HH polarization, following
the RIM prescribed reduction of the polarization ratio, the composite-Bragg and singular
scattering contributions become equal. Consequently, at moderate to large incidence angles,
the breaking contribution cannot be neglected, and for HH, it eventually dominates V
D
at very
rge angles.
ber of ASAR Doppler frequency shift
bservations is growing, this will become feasible.
weekly mean surface
eostrophic map are assumed to explain some of this underestimation.
strophi
c current could consequently strengthen the ability to study surface current
ynamics.
la
Usi
ng WM da
ta the observed and simulated wind dependence of C-band Doppler shift for VV
and HH polarization are plotted for the 23° and 33° incidence angles in Figure 2. Overall the
simulated Doppler frequency shifts display a functional relationship versus wind speed in
good agreement with the observations, in particular up to a wind speed of +/- 15 m/s, with a
mean difference gradually increasing from about 2 Hz for VV at 23° to 5 Hz for HH at 33°. The
observed Doppler anomaly differences between HH and VV are generally small, and
assumingly related to the relatively weak NRCS polarization ratio measured at C-band. Under
the RIM decomposition, the scalar contributions must play a significant role. Further
investigations should therefore be directed to explain both the weak polarization ratio and the
small Doppler anomaly differences. As the num
o
In revisiting the expressions of the Agulhas Current captured in the WSM Doppler velocity
time series further quantitative analyses is now possible taking into account the relationship
presented above. The core position of the maximum surface geostrophic current derived from
the 7-day (15-22 September) composite altimeter map (Figure 3, left) is superimposed on the
full Doppler velocity map derived from ASAR (Figure 3, middle). The mean location and flow
direction of the southern part of the Agulhas Current and the evidence of the Agulhas return
current agrees very well. It is also worth noting that although the return current orientation is
rotated away from range direction, its radial component is clearly manifested. Comparison of
range directed velocities along the red-stippled line (Figure 3, middle) reveals, however,
distinct differences in magnitude (Figure 3, right). In particular at the core of the Agulhas
Current, where the maximum surface geostrophic current is only about 0.7 m/s compared to
the Doppler velocity that reaches nearly 2 m/s. This latter speed is also reported from surface
drifters trapped in the current (www.meds-sdmm.dfo-mpo.gc.ca). Effect of topographic
steering plus time-space averaging of the altimeter data superimposed on a smooth 200 km
resolution mean dynamic topography applied in the construction of the
g
By invoking the easterly, radial directed 4-10 m/s ECMWF wind speed into DopRIM the
simulated wind contribution to the Doppler velocity is found to be rather smooth with a speed
varying from 0.5 to 0.75 m/s in the ASAR look direction (Figure 3c). The 100 km wide and
opposite directed Doppler speed reaching nearly 2 m/s with an estimated accuracy of about
0.2 m/s and with a maximum shear of about 10
-4
s
-1
is therefore predominantly reflecting the
influence of the Agulhas Current on the Doppler velocity measurement. The same is also
valid for the 1.5 m/s Doppler speed of the Agulhas Return Current. This suggests that it is
possible to derive quantitative information of these intense surface currents from the radial
Doppler velocity. Using this method in combination with surface drifters and altimeter derived
surface geo
d
5
