534
IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 3, JULY 2005
Wind- and Wave-Field Measurements Using Marine
X-Band Radar-Image Sequences
Heiko Dankert, Jochen Horstmann, and Wolfgang Rosenthal
Abstract—This paper describes two algorithms for the retrieval
of high-resolution wind and wave fields from radar-image sequences acquired by a marine X-band radar. The wind-field
retrieval algorithm consists of two parts. In the first part, wind
directions are extracted from wind-induced streaks, which are
approximately in line with the mean surface wind direction. The
methodology is based on the retrieval of local gradients from the
mean radar backscatter image and assumes the surface wind
direction to be oriented normal to the local gradient. In the
second part, wind speeds are derived from the mean radar cross
section. Therefore, the dependence of the radar backscatter on
the wind vector and imaging geometry has to be determined.
Such a relationship is developed by using neural networks (NNs).
For the verification of the algorithm, wind directions and speeds
from nearly 3300 radar-image sequences are compared to in situ
data from a colocated wind sensor. The wave retrieval algorithm
is based on a methodology that, for the first time, enables the
inversion of marine radar-image sequences to an elevation-map
time series of the ocean surface without prior calibration of the
acquisition system, and therefore, independent of external sensors.
The retrieved ocean-surface elevation maps are validated by comparison of the resulting radar-derived significant wave heights,
with the significant wave heights acquired from three colocated in
situ sensors. It is shown that the accuracy of the radar-retrieved
significant wave height is consistent with the accuracy of the in
situ sensors.
Index Terms—Friction velocity, inversion, marine radar, RAR,
real aperture radar, surface, waves, wind, wind field.
I. INTRODUCTION
W
IND and waves are the most important environmental
phenomena that affect maritime structures and ships.
Their presence makes the design of those structures significantly different from structures on land. Since wind and waves
are very complex and strongly varying phenomena, it is not
easy to achieve a full understanding of their fundamental
character and behavior.
Wind and waves are typically measured at single points with
one-dimensional (1-D) or two-dimensional (2-D) in situ sensors, e.g., anemometers and wind vanes for wind measurements,
and wave buoys or laser sensors for wave measurements. These
sensors typically provide a time series with a high temporal resolution of wind vectors and ocean-surface elevation at a certain location. In situ wind measurements from towers, ships, and
buoys are often effected by blockage effects and turbulence, as
well as by the errors that arise due to measurements at different
Manuscript received December 15, 2004; revised February 24, 2005;
accepted April 1, 2005. Associate Editor: R. Garello.
The authors are with the GKSS Research Center, 21502 Geesthacht, Germany
(e-mail: [email protected]).
Digital Object Identifier 10.1109/JOE.2005.857524
Fig. 1. WaMoS system installed on Platform “2/4k” of the Ekofisk oil field in
the central North Sea. A wave-rider buoy is placed near the oil field, and the two
laser sensors are mounted on the main complex.
mast heights. In the case of buoy measurements, the tilt and displacement of the sensor, especially at high sea states, leads to
additional wind-measurement errors. In situ sensors for wave
measurements are adequate to measure areas with spatial homogeneous conditions, e.g., off shore. Close to the coast, inside harbors, or behind off- and near-shore buildings and structures, the
sea state can be strongly inhomogeneous. The sea-state parameters, which were measured by the 1-D and 2-D in situ sensors
in such areas, are often not entirely representative of the wave
situations in the neighborhood. In the case of 1-D in situ measurements, there is also a lack of directional information on the
sea state. Last but not least, the positioning of the wind and wave
in situ sensors is limited, e.g., a minimum water depth and maximum current speed constrain wave-rider-buoy deployment.
The shortcomings of the point sensors mentioned above can
be overcome by utilizing marine radars, which map the ocean
surface in both the temporal and spatial domains. The marine
radars used for wind and wave retrieval operate at the X-band
(9.5 GHz) with horizontal (HH) polarization. They have the capability of measuring the backscatter from the ocean surface in
space and time under most weather conditions, independent of
lightning conditions. The marine radar scans the ocean surface
at grazing incidence by rotating its antenna. During rotation, the
radar emits many very short electromagnetic pulses and receives
the backscattered electromagnetic energy from the ocean surface. With each antenna revolution, the radar collects an intensity image of the backscatter of the ocean surface, producing a
temporal radar-image sequence (Fig. 1).
The radar backscatter from the ocean surface is mainly caused
by the small-scale roughness of the sea surface (in the order of
the electromagnetic wavelength of the emitting radar 3 cm),
0364-9059/$20.00 © 2005 IEEE
DANKERT et al.: WIND- AND WAVE-FIELD MEASUREMENTS USING MARINE X-BAND RADAR-IMAGE SEQUENCES
which is mostly generated by the local surface wind [1]. It has
been shown that the radar cross section (RCS) is strongly dependent on wind speed ([1] and [2]) and wind direction ([2] and
[3]), which enables the retrieval of the wind vector from radar
70
images of the ocean surface [4]. At grazing incidence
and vertical (VV) polarization, the main backscatter mechanism
at the ocean surface is Bragg scattering [5]. At grazing incidence and HH polarization, as considered in our study, the RCS
predicted from Bragg theory is too low [6]. Lyzenga et al. [7]
add the effect of wedge scattering as an important additional
backscatter mechanism at grazing incidence and HH polarization. Trizna and Carlson [3] noted differences between HH and
VV polarized radar returns. The value of the RCS for VV polarization can be explained with Bragg scattering in the composite
surface model. In contrast to spiky echoes due to breaking waves
and small-scale bores induced by wave breaking, which are very
important for imaging at HH polarization and low grazing angles ([8] and [9]). For a detailed description of radar scattering
at grazing incidence, refer to Wetzel [10] and Brown [11]. In
the presence of long ocean-surface waves, the small-scale surface roughness, and therefore, the RCS, is modulated. At mod70 , the modulation is
erate incidence angles 20
mainly due to the tilt and hydrodynamic modulation [12], while
at grazing incidence, the modulation stems in addition from
shadowing off the radar beam due to the ocean-surface waves
[10]. These modulation mechanisms of the small-scale surface
roughness lead to the imaging of ocean-surface waves that are
greater than twice the radar resolution ( 10 m).
On an operational basis, marine radar-image sequences are
used to determine spectral and integral ocean-wave parameters
[13], e.g., peak period, wave direction, and significant wave
height, as well as mean near-surface currents [14]. The significant wave height is statistically determined from the radarimage spectrum using an empirical function
(1)
which relates the signal-to-noise ratio (SNR) to the significant
wave height ([15]). The function has to be calibrated for each
radar system, by tuning the constants and using external
wave measurements, which are typically acquired by buoys.
Recently, two methods have been developed to retrieve individual ocean-surface waves from radar-image sequences. The
first method introduced by Borge et al. [16] is based on the statistics of ocean-surface waves, and requires that the significant
wave height be retrieved from the empirical function (1). The
second methodology was introduced by Dankert and Rosenthal
[17] and is independent of in situ measurements as well as the
empirical function (1). This method is very robust and assumes
that the main modulation of the RCS is due to the local surface slope in the antenna look direction (tilt modulation). Their
approach neglects hydrodynamic modulation as well as the geometrical effect of shadowing.
In this paper, two marine radar-based remote-sensing techniques are described and verified, which enable the measurement of ocean-surface winds and waves in the spatial and temporal domain. In Section II, the radar and in situ data used for
this study are described. Section III describes the method for retrieving the wind direction as well as the wind speed from ma-
535
Fig. 2. Radar-image sequence of 32 images of the ocean surface taken from
the Central North Sea from Ekofisk platform “2/4k” in February 2001. The wind
speed during measurement was 15 m 1 s , the wind direction 185 , and
the mean wave-propagation direction 180 . The directional spread of the
30 .
wave-propagation direction s
rine radar-image sequences. The methodology to retrieve the individual wave height from radar-image sequences in space and
time is introduced in Section IV. Finally, in Section V, the conclusion and outlook are given.
II. INVESTIGATED DATA
The radar-image sequences investigated in this study were
all acquired with the Wave Monitoring System (WaMoS II),
which was developed at the GKSS Research Center. WaMoS
II consists of a standard marine radar and a personal computer
equipped with an analog-to-digital converter. This system can
store and process the acquired radar-image sequences. The marine radar system operates in the X-Band (9.5 GHz) with HH
polarization in transmission and reception. The radar antenna
covers an area within a radius of 2000 m at a resolution of
12 m in range (antenna look direction). The antenna rotation
period is 2.6 s. The radar system is mounted aboard the platform 2/4k, located at 56.5 N and 3.2 E in the central North Sea
within the Norwegian oil field Ekofisk (Fig. 1). The radar was
mounted at a height of 74 m facing north west, which enables
the radar to image the ocean surface between 155 (SSE) and
25 (NNE). The water depth in the imaged area is fairly homogeneous with a depth of about 70 m. A standard radar-image
sequence consists of 32 images, representing a time span of
82 s. Fig. 2 shows a typical radar-data set with a wave field propagating in the northerly direction. The shadows originate from
the equipment of the measurement platform and the big scatter
in the South from the main field. The investigated radar-image
sequences cover a time period of 8 months between February
and September 2001. They represent more than 3200 acquisition times with wind speeds of up to 17 m s and significant
wave heights of up to 6 m.
Colocated with the radar acquisitions, in situ wind data were
collected by a wind anemometer and a wind vane mounted
aboard the mast at a height of 80 m above mean sea level aboard
the oil platform. Wind speeds represent 10-min means and were
converted to a measurement height of 10 m, considering the
measured air–sea temperature differences. The in situ wave
536
IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 3, JULY 2005
data were measured by a wave-rider buoy, which was located
1.5-km southwest of the radar system. The buoy measurements
represent 20-min mean wave parameters, e.g., peak period and
significant wave height. In addition to the wave rider, two laser
sensors were available. The sensors were mounted aboard the
main platform located 1.9-km south of 2/4k (see Fig. 1). In
contrast to the buoy data, the laser data represent continuous
measurements of the sea-surface elevation with a sampling
frequency of 2 Hz. The air and sea-surface temperatures were
collected by the wave-rider buoy.
III. WIND-VECTOR RETRIEVAL
The wind-vector retrieval algorithm consists of two steps. In
the first step, wind directions are retrieved from wind-induced
streaks, which are visible at scales of typically 200 m. These
wind directions are used as input to the second step, where the
wind speed is derived from the dependence of the RCS on wind
speed, wind direction, and imaging geometry.
Fig. 3. Scatterplot of in situ wind directions versus radar-retrieved wind
directions.
A. Wind Direction From Wind-Induced Streaks
The method for retrieving wind directions is based on the
imaging of wind-induced streaks in radar images. These streaks
were first observed in synthetic-aperture-radar (SAR) images
and are very well aligned with the mean surface wind direction [18]. In marine-radar imagery, the streaks have a typical
spacing of 200 m and are most likely caused by the local wind
field in the lower boundary layer [4]. However, in marine-radar
images, these wind streaks are superimposed by other ocean features and are barely visible. By integrating a radar-image sequence over time (typically 32 images representing 1 min of
data), signatures with higher variability in time, e.g., surface
waves, are averaged out. Only static and quasi-static signatures
with frequencies below the integration time, like wind streaks,
remain visible. To automatically measure the orientation of the
wind streaks, the local gradients from radar images are retrieved,
which were previously smoothed and reduced to 100-m resolution. The orientation of a wind streak, and therefore, wind orientation, is defined to be oriented normal to the local gradient.
This resolves the wind direction with a 180 ambiguity, and follows the methodology for retrieving wind directions developed
for SAR imagery [19] and [20].
Simply using the unambiguous wind-direction dependence of
the RCS at HH polarization ([3]) is problematic, because the
radar is often affected by surrounding equipment that lead to
disturbed areas in the radar images (Fig. 1), and therefore, in
difficulties finding the peak of the RCS, which is located upwind
(wind is blowing towards the antenna). This is especially the
case for radar systems based at the coast as well as for systems
aboard oil rigs, such as the Ekofisk setup.
In this case, the 180 directional ambiguity is removed by automatically extracting the movement of wind gusts visible in the
radar-image sequence. The radar-image sequence is subdivided
into two or more subsequences (typically 24 images), which
may overlap each other in time. Each subsequence is integrated
over time to remove signatures with higher temporal variability
such as ocean-surface waves. The movement of wind gusts is retrieved from these mean RCS images. For details, refer to [21].
Fig. 4. Local wind directions at the ocean surface (solid arrows) retrieved from
the mean RCS of a radar-image sequence of 32 images taken at the Ekofisk 2/4k
platform on February 10, 2001. The in situ wind direction was 335 (dashed
arrows) and the wind speed 8 m 1 s . The shaded area is not considered. The
polar image is divided into subareas for NN training.
Fig. 3 gives the scatter plot of in situ measurements of wind
directions against the marine-radar-retrieved directions for each
of the 3271 data sets. The standard statistical parameters result
in a correlation of 0.99, a bias of 0.6 , and a standard deviation
of 14 .
In Fig. 4, the resulting local mean directions are plotted for
one sample scale. They agree well with the wind direction measured at the radar platform at 80-m height.
B. Wind Speed Using Neural Networks (NNs)
It is well known that the RCS is strongly dependent on local
wind conditions ([2] and [22]). To find a transfer function that
describes the dependence of RCS on the ocean-surface wind
speed, wind direction, and radar-imaging geometry, a feed-forward back-propagation NN was used as a multiple nonlinear regression technique. For the training of the NN, the data set that
consisted of 3271 radar-image sequences was subdivided into
a training and a test data set with a ratio of 2:1. To include the
dependencies of RCS on wind direction and range distance, the
DANKERT et al.: WIND- AND WAVE-FIELD MEASUREMENTS USING MARINE X-BAND RADAR-IMAGE SEQUENCES
mean RCS image (integrated over time) were subdivided in several range and azimuth bins, as depicted in Fig. 4. Each radar
image is divided into subareas of four 300-m range intervals
starting at a distance of 900 m and in azimuth sectors of 5 . In
the first step, NNs were trained using the mean radar intensity
of the range–azimuth cells and the mean radar-retrieved wind
direction with respect to the antenna look direction as input and
the in situ wind speed converted to 10-m height as output. In
the training of the NNs, all areas with shadows due to the platform construction or areas with static patterns (blue area) were
masked and excluded from the training and later measurements.
To take into account the dependence of the RCS on the stratification conditions in the lower marine atmospheric boundary
layer (MABL) [23], the air–sea temperature difference as measured by the wave-rider buoy was considered as additional input
to the NN. This inclusion resulted in a significant improvement
of the wind-speed retrieval.
For radar setups aboard ships or platforms that are standing
alone and well off the coast, the antenna look direction can be
given with respect to the wind direction. If the radar platform
is situated at the coast or in the neighborhood of a larger object, e.g., another platform, as is the case for the investigated
example, the input of wind direction to the NN has to be differentiated. In these cases, the radar look direction as well as the
radar-retrieved wind direction are used as input to the NN. This
allows the inclusion of the influence of the platforms’ neighborhood on the wind-speed estimate, e.g., wind shadowing due to
a neighboring platform.
For the given data set, the best results were obtained with the
input of the mean RCS in the cross-wind direction at four different ranges, air–sea temperature difference, and the radar-retrieved differentiated wind direction. In the cross-wind case, the
wind shadowing or blockage effect of the platform itself does
not influence the wind field. The differentiated wind direction
is needed to correct for the shadowing effect of the neighboring
platform, which are significant with respect to the southerly
wind and would lead to a significant underestimation of the
mean wind speed. Fig. 5 gives the resulting scatterplot of the
in situ wind speeds versus radar-retrieved wind speeds. In comparison to in situ wind speeds measured at the platform and converted to 10-m height, the correlation is 0.97 with a bias of 0.03
m s and the standard deviation is 0.85 m s . The resulting
parameterization enables the retrieval of wind speeds as low as
0.75 m s .
For the retrieval of high-resolution wind fields from radarimage sequences, an NN was trained considering the mean RCS,
distance to antenna as well as local wind direction, and antenna
look direction versus North. Fig. 6 shows the resulting wind
field with a resolution of 120 m. However, the validation of such
highly resolved wind fields is a difficult and extremely expensive task that has to be tackled in the future.
Aside from this method, another new technique for wind-field
retrieval with spatially and temporally high resolution using marine radar-image sequences has recently been introduced. The
method is based on analyzing the movement of wind gusts,
which become visible in radar-image sequences after filtering.
In contrast to other methods, this new technique requires no calibration phase for the radar system. For details, refer to [21].
537
Fig. 5. Scatterplot of wind-anemometer wind speeds versus wind speeds
retrieved from colocated marine-radar image sequences.
Fig. 6. High-resolution ocean wind field retrieved at Ekofisk 2/4k on March
23, 2001 using the determined local wind directions, together with an NN that
parameterizes the wind speed spatially.
IV. OCEAN-SURFACE RETRIEVAL
A. Imaging Mechanisms
Ocean waves are imaged by a radar, because the long oceansurface waves modulate the RCS. The modulation process is a
sum of four contributing processes: the geometrical effects of
shadowing and tilt, hydrodynamic modulation, and wind modulation. The empirical method described here assumes that the
main modulation mechanisms are wind and tilt modulation:
(2)
Hydrodynamic modulation is neglected. For the given data sets
from Ekofisk, with a radar-antenna installation height of 74 m,
shadowing appears only in the far range. It is therefore assumed
that shadowing only has a minor contribution to the RCS.
The modulation process is mathematically described by a
modulation transfer function (MTF), which is commonly defined as the expansion of the RCS for the spectral amplitudes of
538
IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 3, JULY 2005
the wave field
[12]. For a local facet at location and time
, the local RCS is given by [17]
1
e
(3)
at location and time
describing the deviation of the RCS
as a product of the local modulation function
and the
. The local wave field is a product of a
local wave field
carrier wave with wavenumber and a slowly varying amplitude
. This translation of the RCS to the local surface
function
tilt is a local, spatial, and temporal description of the modulation
process. The MTF is therefore a spatial and temporal function
that has to be determined for each location with respect to the
radar. This is different from the description of the MTF in the
spectral domain.
The inversion method is based on this local, spatial, and temporal description of the modulation process by determination of
the surface tilt angle in the antenna look direction at each pixel
of the radar-image sequences. The mean RCS is dependent on
the local depression angle at the X-band with HH polarization [3]. This (linear or nonlinear) dependence is parameterized
from the mean
as a look-up table. The deviation of the RCS
represents a local temporal change of the depression
RCS
angle, which is assumed to be equal to the local ocean-surface
tilt.
B. Method
An overview of the method for the determination of the time
series of ocean-surface elevation maps is given by the inversion
scheme in Fig. 7. The method requires raw polar radar-image
, as shown in Fig. 1, as input.
sequences
Marine radar antennas are directional antennas that radiate
radio-frequency energy in patterns of lobes that extend outward
from the radar antenna in the antenna look direction. The radiation pattern also contains weak minor lobes. Because of the radiation pattern, each radar antenna has a typical receiving pattern.
Before analyzing the radar images, this receiving pattern has to
be determined for correction of the data. Therefore, in the first
step, each radar image is corrected with the characteristic antenna receiving pattern. For details regarding the measurement
of the antenna receiving pattern. refer to [17].
In the next step, the dependence between the local variation of
and the local tilt has to be
the RCS from the mean RCS
determined. The mean RCS is parameterized for each antenna
by
look direction
(4)
gives the parameterization function and , the
where
resulting 2-D parameterization. The range-dependent depression angle is given by
(5)
with giving the distance from the antenna and the given installation height of the radar antenna
74 m.
Fig. 7. Inversion scheme for the determination of ocean-surface elevation.
The ocean-surface waves cause a local change of the depres, the tilt modulation of the RCS, and thereby, a
sion angle
local deviation of the RCS
from the mean value . With
the given parameterizations, (4) and (5), and the known deviafrom its mean value, the local change of
tion of the RCS
is determined, which is assumed to be
the depression angle
equal to the local ocean-surface tilt (see Fig. 7). The local
tilt angles are determined for each location in space and time.
. For details, refer
The result is a sequence of tilt images
to [17].
is determined from the
The ocean-surface elevation
by direct integration. Thereby, tilt-image setilt angles
is transformed into the wavenumber frequency
quence
domain by performing a 3-D fast Fourier transform (FFT). The
is integrated by
resulting complex 3-D tilt spectrum
multiplying with an integration transfer function , which is
complex and shifts the phase of all wavenumber components in
.
the Fourier space by
(6)
The result is a complex 3-D wave spectrum of the oceansurface-elevation field.
The integration process causes an amplification of small
wavenumbers. Therefore, a posterior filtering process is necessary to retrieve only the signal of the ocean-surface wave
field. The dispersion relation of linear surface-gravity waves is
used, which connects wavenumbers with their corresponding
frequency coordinates [13]:
(7)
where indicates the absolute frequency, the gravitational acceleration, the water depth, and , the velocity of encounter.
The filtering is done by fitting the theoretical dispersion relation
to the signal coordinates in the complex wavenumber-frequency
spectrum [14]. Thereby, the dispersion shell is strongly broadened in and to also get components of the wave spectrum
DANKERT et al.: WIND- AND WAVE-FIELD MEASUREMENTS USING MARINE X-BAND RADAR-IMAGE SEQUENCES
Fig. 8. Ocean-surface-elevation sequence of a radar-image sequence recorded on March 28, 2001 at Ekofisk 2/4k (see Fig. 1). The determined
compared to 4.47 m retrieved from a colocated time series of the wave-rider buoy.
that do not lie exactly on this function. To suppress those Fourier
coefficients with small wavenumbers and those with noise from
nonrelevant spectral components, a bandpass filter is also apis determined by transplied (see [17]). The wave field
forming the retrieved complex 3-D wave spectrum
into the spatio–temporal domain by an inverse 3-D FFT.
Fig. 8 shows a resulting ocean-surface-elevation image sequence, recorded on March 28, 2001 at Ekofisk 2/4k. An azimuthal dependence of the ocean-surface elevation is clearly
visible. This dependence is explained as a geometrical projection factor as follows: only the tilt component of the water surface in the antenna viewing direction affects the modulation of
the RCS. Therefore, the RCS is not modulated if the radar is
looking parallel to the wave crests. The significant wave height
in the area around the given wave-travel direction for the whole
4.47 m, which is in excellent agreeimage sequences is
ment with
4.47 m, retrieved from a colocated buoy time
series.
C. Validation
This section is focused on the statistical comparison of the
ocean-surface-elevation image sequences with the colocated
2-Hz elevation time series (20 min each) of three in situ sensors,
one wave-rider buoy, and two laser sensors [“Flare North” (FN)
and “Flare South” (FS)]. All in situ sensors are situated within
the radar measuring range.
The comparison is focused on the significant wave height
as integral statistical parameter, the most important quan-
539
H
is 4.47 m
is directly determined from
tity used to describe a sea state.
the standard deviation of the spatio–temporal wave elevation:
4
(8)
denotes the expecwhere gives the population mean, and
is determined only for the area within
tation value of .
22.5 of the wave-travel direction
180 , due to the fact
that the radar is mainly imaging waves that travel towards and
is also diaway from the radar. For the in situ time series,
rectly determined from the standard deviation of the elevation
time series.
up to 6 m are proA total of 1535 radar data sets with
cessed, together with their colocated time series, from the in situ
sensors. Only data sets, recorded under wind-speed conditions
above 4 m s are considered. This wind speed is necessary
for a measurable modulation of the RCS.
beFig. 9(a) gives an example comparing the value of
tween the wave-rider buoy and the marine radar, and Fig. 9(b)
between the buoy and the laser FN. In both cases, a correlation
of 0.93 and a standard deviation of 0.35 m could be achieved.
With a bias of 15 cm, the radar is slightly underestimating the
buoy measurements.
Table I shows a full intercomparison between all four sensors. Thereby, the upper three rows give the comparison between
the radar measurements and those from the in situ sensors. The
lower three rows of Table I give an intercomparison between all
three in situ sensors. Correlation, bias, and standard deviation
between the in situ sensors are within the same range as between
540
IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 3, JULY 2005
Fig. 9. Comparison of
H
derived from 1535 wave-rider records with (a) radar
TABLE I
MAIN STATISTICAL PARAMETERS RESULTING FROM THE COMPARISON OF 1535
VALUES BETWEEN TWO DIFFERENT SENSORS
H
the in situ sensors and the radar. Different from that of the buoy,
the bias between the radar and the laser FS is negligible. It is obvious that the accuracy of the in situ sensors is decreasing with
increasing wave height. For the radar measurements, it is nearly
constant (see Fig. 9).
V. CONCLUSION AND OUTLOOK
Two methodologies for the retrieval of high-resolution wind
and wave fields from marine radar-image sequences acquired at
the X-band with horizontal (HH) polarization in transmission
and reception are introduced and validated. It is demonstrated
that radar-image sequences of the ocean surface provide reliable
information on ocean winds and waves, as well as new opportunities to investigate the spatial and temporal behavior of ocean
wind and wave fields.
A common marine X-band radar is utilized for providing a
time series of radar backscatter images from the ocean surface.
The radar technique thereby allows measurement under most
weather conditions. With the preexisting installations of radar
systems on marine structures, harbors, platforms, and ships, the
measurements can be done in a very cost-efficient way.
The radar cross section (RCS) of the ocean surface at the
X-band with HH polarization at grazing incidence is strongly
dependent on the surface wind speed, wind direction, range distance, and stratification conditions. Furthermore, the RCS is
modulated by long ocean-surface waves. This provides opportunities for the development of algorithms for remote measurements of surface wind vectors and surface waves from radar
images.
The first method for wind-field retrieval consists of two steps:
one for wind direction and another for wind-speed retrieval.
H
values and (b)
H
derived from wave records of laser FN.
Wind directions are derived from wind-induced streaks, which
are oriented in direction of the wind. This is done by determining
the local gradients in the mean RCS image, which give the orientation of the wind streaks with an ambiguity of 180 . The
ambiguity is removed by analyzing the movement of wind-gust
patterns in the radar-image sequences [21]. Comparison to in
situ measurements resulted in a correlation of 0.99 with a bias
of 0.6 and a standard deviation of 14.2 .
Wind speeds are retrieved, in the second step, from the dependence of the RCS on wind speed and wind direction. This dependence is parameterized by the training of a neural network (NN)
considering different input parameters: the mean RCSs in the
cross-wind direction at four different ranges (at cross wind, the
wind field is not disturbed by the platform itself), the radar-retrieved wind direction, and the antenna look direction, to take
into account the influence of the platforms’ neighborhood on
the wind field. This results in very good and practicable results.
The radar system can therefore be installed without any other
additional sensors for wind measurements. By taking the air–sea
temperature difference as additional input parameter to the NN,
the wind-speed retrieval is significantly improved. Thereby, the
dependence of the RCS on the stability in the lower marine atmospheric boundary layer (MABL) is considered. Comparison
of radar-derived wind speeds (considering all these parameters)
to in situ wind speeds measured at the platform and converted to
10-m height resulted in a correlation of 0.97 with a bias of 0.03
m s and a standard deviation of 0.85 m s . In contrast to
typical in situ sensors like anemometers, the measurements of
the radar system are not influenced by movements of ships or
platforms, and local turbulence due to installations.
The second part of this paper describes an empirical inversion scheme that allows the derivation of ocean-surface-elevation map sequences from radar-image sequences. A calibration
procedure is not necessary. The radar system can act and make
measurements as a stand-alone device. The ocean-surface elevation is derived from the tilt angle of the ocean surface, which is
determined in each pixel of all the radar images. Thereby, the dependence of the mean RCS on the mean local depression angle
is parameterized as a look-up table. The spatio–temporal change
of the RCS and the depression angle, respectively, is assumed
to be related to the local ocean-surface tilt. The modulation of
DANKERT et al.: WIND- AND WAVE-FIELD MEASUREMENTS USING MARINE X-BAND RADAR-IMAGE SEQUENCES
the RCS by the ocean-surface tilt is described locally, spatially,
and temporally. Hydrodynamic modulation and the geometrical
effect of shadowing are neglected.
The inversion scheme has been applied to more than 1500
radar-image sequences. Additionally, colocated measurements
from three in-situ sensors, one wave rider buoy, and two laser
sensors, were taken. The validation has been performed through
, derived from
the comparison of significant wave heights
values
the time series of ocean-surface elevation maps, to
from the colocated in situ wave records. Comparison resulted
in a correlation of 0.93 and a standard deviation of 0.35 m. An
intercomparison between all four sensors was performed. Correlation, bias, and standard deviation are within the same range
for all sensors.
Recently, a new research platform called “FINO” has been
built in the German bight at 54 N and 6.6 E, at a water depth
of 30 m. The platform is located within a potential offshore
windmill park. It provides a marine-radar system and various
meteorological and hydrological in situ sensors, e.g., wind and
temperature sensors, a water gauge, and a wave-rider buoy. Data
from these instruments are being received at the moment and are
used for further validation of the developed methods.
Because the radar system measures the wind-induced
roughness at the ocean-surface boundary layer, it in fact gives
a measure of the wind-induced surface stress, and therefore,
. Based on this, the wind-retrieval
the friction velocity
method will be adjusted for measuring . The main advantage is that no additional air–water temperature measurements
are required.
ACKNOWLEDGMENT
The authors were supported by the European Commission, in
the framework of the European project MaxWave, and by the
Bundesministerium für Bildung und Forschung (BMBF), in the
framework of the Envisat Oceanography (ENVOC) project. All
radar-image sequences used were made available through the
kindness of the company OceanWaves.
541
[8] L. Wetzel, Wave Dynamics and Radio Probing of the Ocean Surface.
New York: Plenum, 1986, ch. On microwave scattering by breaking
waves, pp. 273–284.
[9] D. Trizna, “A model for Brewster angle effects on sea surface illumination for sea scatter studies,” IEEE Trans. Geosci. Remote Sens., vol. 35,
no. 5, pp. 1232–1244, 1997.
[10] L. Wetzel, “Electromagnetic scattering from the sea at low grazing
angles,” in Surface Waves and Fluxes, L. Geernaert and W. J. Plant,
Eds. The Netherlands: Kluwer, 1990, vol. II, pp. 109–171.
[11] G. Brown, “Special issue on low-grazing angle backscatter from rough
surfaces,” IEEE Trans. Antennas Propag., vol. 46, no. 1, pp. 1–2, Jan.
1998.
[12] W. R. Alpers, D. B. Ross, and C. L. Rufenach, “On the detectability of
ocean surface waves by real and synthetic aperture radar,” J. Geophys.
Res., vol. 86, no. C7, pp. 6481–6498, Jul. 1981.
[13] I. Young, W. Rosenthal, and F. Ziemer, “A three–dimensional analysis of
marine radar images for the determination of ocean wave directionality
and surface currents,” J. Geophys. Res., vol. 90, no. C1, pp. 1049–1059,
1985.
[14] C. Senet, J. Seemann, and F. Ziemer, “The near-surface current velocity
determined from image sequences of the sea surface,” IEEE Trans.
Geosci. Remote Sens., vol. 39, no. 3, pp. 492–505, Mar. 2001.
[15] J. N. Borge, K. Hessner, and K. Reichert, “Estimation of the significant wave height with X-band nautical radars,” in Proc. 18th Int. Conf.
Offshore Mechanics and Arctic Engineering (OMAE), Newfoundland,
Canada, 1999.
[16] J. N. Borge, G. Rodríguez, K. Hessner, and P. González, “Inversion
of marine radar images for surface wave analysis,” J. Atmos. Ocean
Technol., vol. 21, no. 8, pp. 1291–1300, Aug. 2004.
[17] H. Dankert and W. Rosenthal, “Ocean surface determination from
X-band radar-image sequences,” J. Geophys. Res., vol. 109, no.
C04016, 2004.
[18] J. Horstmann, W. Koch, and S. Lehner, “Ocean wind fields retrieved
from the advanced synthetic aperture radar aboard ENVISAT,” Ocean
Dyn., vol. 54, no. 6, pp. 570–576, Dec. 2004.
[19] J. Horstmann, W. Koch, S. Lehner, and R. Tonboe, “Ocean winds from
RADARSAT-1 ScanSAR,” Can. J. Remote Sens., vol. 28, no. 3, pp.
524–533, 2002.
[20] W. Koch, “Directional analysis of SAR images aiming at wind direction,” IEEE Trans. Geosci. Remote Sens., vol. 42, no. 4, pp. 702–710,
Apr. 2004.
[21] H. Dankert, J. Horstmann, and W. Rosenthal, “Ocean surface winds retrieved from marine radar-image sequences,” in Proc. Int. Geoscience
Remote Sensing Symp., Anchorage, Alaska, 2004, pp. 1903–1906.
[22] H. Dankert, J. Horstmann, S. Lehner, and W. Rosenthal, “Detection of
wave groups in SAR images and radar-image sequences,” IEEE Trans.
Geosci. Remote Sens., vol. 41, no. 6, pp. 1437–1446, Jun. 2003.
[23] M. R. Keller and W. C. Keller, “A wave tank study of the dependence
of X band cross sections on wind speed and water temperature,” J. Geophys. Res., vol. 97, no. C4, pp. 5771–5792, Apr. 1992.
REFERENCES
[1] P. Lee, J. Barter, K. Beach, C. Hindman, B. Lake, H. Rungaldier,
J. Shelton, A. Williams, R. Lee, and H. Yuen, “X-band microwave
backscattering from ocean waves,” J. Geophys. Res., vol. 100, no. C2,
pp. 2591–2611, Feb. 1995.
[2] H. Hatten, F. Ziemer, J. Seemann, and J. Nieto-Borge, “Correlation between the spectral background noise of a nautical radar and the wind
vector,” in Proc. 17th Int. Conf. Offshore Mechanics and Arctic Engineering (OMAE), Lisbon, Portugal, 1998, . .
[3] D. Trizna and D. Carlson, “Studies of dual polarized low grazing angle
radar sea scatter in nearshore regions,” IEEE Trans. Geosci. Remote
Sens., vol. 34, no. 3, pp. 747–757, May 1996.
[4] H. Dankert, J. Horstmann, and W. Rosenthal, “Ocean wind fields retrieved from radar-image sequences,” J. Geophys. Res., vol. 108, no.
C11, p. 3352, Nov. 2003.
[5] J. W. Wright, “Backscattering from capillary waves with applications
to sea clutter,” IEEE Trans. Antennas Propag., vol. AP-14, no. 6, pp.
749–754, Nov. 1966.
[6] F. Askari, T. Donato, and J. Morrison, “Detection of ocean wave fronts
at low grazing angles using an X-band real aperture radar,” J. Geophys.
Res., vol. 101, no. C9, pp. 20883–20898, Sep. 1996.
[7] D. Lyzenga, A. Maffet, and R. Shuchman, “The contribution of wedge
scattering to the radar cross section of the ocean surface,” IEEE Trans.
Geosci. Remote Sens., vol. GE-21, no. 4, pp. 502–505, 1983.
Heiko Dankert received the Diploma degree in
structural engineering from the University of Rostock, Rostock, Germany, in 2000, and the Ph.D.
degree in geophysics from the University of Hamburg, Hamburg, Germany, in 2003.
In 1999, he joined GKSS while writing his
Diploma thesis about the spatial analysis of diffraction of wave fields using optical-image sequences.
In 2000, he was a Visiting Scientist at the National
Cheng Kung University, Coastal Ocean Monitoring
Center, Tainan, Taiwan, R.O.C. Since 2001 he has
been a Research Scientist working in the German KFKI project MOSES in
the Model and Data Assimilation group, GKSS Research Center, Geesthacht,
Germany. From 2001 until 2003, he worked in the European project MaxWave
and since 2003 in the German KFKI project MOSES. His main research
interests are the development of algorithms to extract marine parameters,
wind fields and friction velocity from image sequences of optical and marine
radar systems, investigation of oceanographic and boundary layer processes,
numerical shallow-water wave- and hydrodynamic modelling.
542
Jochen Hortsmann received the Diploma degree in
physical oceanography in 1997 and the Ph.D. in earth
sciences in 2002, both from the University of Hamburg, Germany.
In 1995 he joined the Coupled Model System
Group, GKSS Research Center, Geesthacht, Germany. Since 2000, he has been a Research Scientist
at the GKSS Research Center at the Institute for
Coastal Research. In 2002, he was a Visiting Scientist at the Applied Physics Lab (APL), John Hopkins
University (JHU), MD, and the National Oceanic
and Atmospheric Administration (NOAA), National Environmental Satellite,
Data, and Information Service (NESDIS), Washington, DC. In 2004 and 2005,
he was a Visiting Scientist at the Rosenstiel School for Marine and Atmospheric
Sciences, University of Miami, FL. His main research interests are in extraction
of geophysical parameters from RAR, SAR as well as interferometric SAR.
IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 30, NO. 3, JULY 2005
Wolfgang Rosenthal received the Diploma degree in
physics and the Ph.D. degree in theoretical solid state
physics, both from Free University of Berlin, Germany, in 1966 and 1972, respectively.
He was a Research Scientist at the Institute for
Geophysics, Institute for Marine Research and at
Max-Planck Institute for Meteorology at the University of Hamburg, Germany. From 1985 to 2004, he
worked as a Research Scientist in the boundary layer
ocean-atmosphere branch at the Institute for Coastal
Research at GKSS Research Center.