Chinese Journal of Oceanology and Limnology
Vol. 27 No. 1, P. 112-116, 2009
DOI: 10.1007/s00343-009-0112-1
A new model to estimate significant wave heights with
ERS-1/2 scatterometer data*
GUO Jie, HE Yijun, William Perrie, SHEN Hui,
CHU Xiaoqing,
Institute of Oceanology, Chinese Academy of Sciences, Key Laboratory of Ocean Circulation and Wave, CAS,
Qingdao 266071, China
Graduate School of the Chinese Academy of Sciences, Beijing 100039, China
Bedford Institute of Oceanography, Dartmouth, NS, Canada
Received Oct. 10, 2008; revision accepted Nov. 17, 2008
Abstract A new model is proposed to estimate the significant wave heights with ERS-1/2
scatterometer data. The results show that the relationship between wave parameters and radar
backscattering cross section is similar to that between wind and the radar backscattering cross section.
Therefore, the relationship between significant wave height and the radar backscattering cross section is
established with a neural network algorithm, which is, if the average wave period is 7s, the root mean
square of significant wave height retrieved from ERS-1/2 data is 0.51 m, or 0.72 m if it is >7s otherwise.
Keyword: scatterometer; significant wave height; neural networks; wind waves; swell
1 INTRODUCTION and H1/3 with the buoys in the Gulf of Mexico,
estimating the wave periods from wind speeds and
A good description of ocean wave conditions is altimeter-measured H1/3. These wave parameters can
crucial to the safety of voyage as well as of marine be measured at 7 km resolution along the track.
constructions. Ocean waves can be described with However, the two-dimensional spatial resolution
the parameters such as significant wave height, wave quite low; therefore, its applications have been
period, and wave direction etc. In the past, ocean limited.
wave parameters are measured with wave gauges and A scatterometer is a specialized sensor for
buoys; however, it is almost not possible for the measuring sea surface wind vector in spatial
detail and constant measurement operations at sea in resolution of 50 km or less, and the precision of wind
a large marine area or global oceans. This situation speed is about 2 m/s. As QuikSCAT can cover 90%
remains unchanged until oceanic satellites were of the global oceans every day, the data have been
launched. applied widely. The wind vector can be measured by
At present, synthetic aperture radar (SAR) a scatterometer because the NRCS is wind-vector
mounted in a satellite is the only sensor used to dependent. In fact, the NRCS represents the radar
remotely measure ocean wave spectra. However, return intensity at sea surface forced by wind. Ocean
because of the azimuthal wave cutoff and the lack of waves almost always include wind waves and swell.
power, it is unable to measure the high-frequency Therefore, the NRCS depends not only on wind
part of a wave spectrum. The application is therefore vectors, but also on ocean waves. For TRMM
limited. On the other hand, significant wave height (tropical rainfall mapping mission), the NRCS
(H1/3) can be measured globally with such a satellite depends on the waves and wind speed, having one
radar altimeter by inferring directly from the shape of precipitation radar (PR) and two satellite altimeters
radar pulse returning to the nadir-looking altimeter (Jason-1 and ENVISAT) installed (Tran et al., 2007).
with the assumption of the Gaussian surface The spectrum model by Stephen et al. (1985) is used
elevations. Moreover, wind speed can also be
retrieved from the altimeter-measured normalized * Supported by the National High Technology Research and Development
radar cross section (NRCS). Hwang et al. (1998) Program of China (863 Program) (No.2008AA09Z102), and by the
Canadian Space Agency (CSA) GRIP Program.
established the relation among wind speed, period
** Corresponding author: abpnxs@r.postjobfree.com
No.1 GUO et al.: A new model to estimate significant wave heights with ERS-1/2 scatterometer data 113
data of only 12 NDBC buoys can be used (Fig.1).
at ocean surface, in which two-scale scattering theory
They covered the North Pacific Ocean and the North
is applied, and can well predict the observed
Atlantic Ocean. The distances of buoys to coastal
dependence of NRCS on radar frequency,
regions are over 50 km. Table 1 lists the property of
polarization, incidence angle, and wind velocity in an
the data set, including the buoy station number,
wide incidence angle range of 0 70 . The spectrum
coordinates, water depth, and the number of data
model is integrated with swell effect examination on
point in each data set. In total, 10 485 data from
a radar cross section. The effect is significant in low
radar frequency (L band) at normal incidence, and ERS-1/2 scatterometer are collocated with the
can be nearly eliminated in high frequency (Ku band) NDBC buoy data in this study, of which 5 300 are
at a large angle of incidence (about 50 ). With neural used to built the model and 5 185 data are used for
networks and large high-quality collocated datasets, validation.
Quilfen and Chapron (2003) studied the relation The NRCS, azimuth angles, and incidence
between the C-band scatterometer NRCS angles are measured by the ERS-1/2
measurement by ERS (European Remote Sensing scatterometer. The wind directions, wind speed
Satellite) and integrated sea state parameters (i.e., the (V), average wave periods (T), and significant wave
mean wave period and significant wave height) heights (H1/3) are measured in NDBC buoy stations.
measured by buoys, and found that the NRCS is The comparison period spans from January 1991 to
affected by wave parameters. In the following December 2000. For comparisons between
sections, our methodology, results, and conclusions scatterometer and buoy data, the maximum
to retrieve wave parameters from ERS-1/2 differences in longitude is 0.15, 0.15 in latitude,
scatterometer data are presented. and 0.5 h in time.
2 METHODS AND RESULTS
2.1 Data
The ERS missions consist of two remote sensing
satellites launched in the 1990s by the European
Space Agency. The first series, ERS-1, was launched
in July 1991, and ERS-2 in April 1995 in order to
ensure long term continuity of data, which is
essential for researches and applications. ERS-1/2
(ERS-1 and ERS-2 in short) scatterometer data are Fig.1 Locations (triangles) of 12 buoys in the North Pacific
used in this paper Ocean and the North Atlantic Ocean
Because National Data Buoy Center (NDBC)
2.2 The effect of ocean wave on the radar
buoy collects wave data hourly, for each ERS-1/2
scattering cross section
scatterometer data point, any two buoy data points
For the TRMM PR (active), Tran et al. (2007)
before and after the ERS-1/2 scatterometer time are
discussed the effect of H1/3 on NRCS. Their results
selected for comparison. The wave data from NDBC
are consistent with a previous analysis at higher
buoys are acquired and reported hourly. H1/3, the
incidence angles (20, 30, 40 and 60 ) (Nghiem et
significant wave height (in meter) is the 1/3 of the
al., 1995). Here, ERS-1/2 scatterometer data are used
highest waves during the 20 min sampling period,
with NDBC buoy data to determine the relation
and the average wave period (T, in second) is the H1/3
between wave parameters and the NRCS. Fig.2
that observed during the same period. The wind
shows the relationship between the NRCS and
direction is the degrees clockwise off the true
relative azimuth angle at an incidence angle of 30
geographical North. The significant wave height and
when H1/3/(gT2) is 0.005. Fig.3 shows the NRCS
average wave period are derived from the buoy heave
versus relative azimuth angle curve with an incidence
motion spectrum measured over a 20-min acquisition
angle of 45 when H1/3 is 2 m. The relationship is
period starting at 30 min after the hour. The wind and
close to that between NRCS and wind. The blue
wind direction data are collected with a wind sensor
asterisks are the results from our model and the red
located on the buoy s mast. The elevation of the wind
dots represent the buoy and scatterometer
sensor is 5 m or 10 m above sea level. In this research,
observations (Figs.2 and 3).
data from 34 NDBC buoys are collected; however,
CHIN. J. OCEANOL. LIMNOL., 27(1), 2009 Vol.27
114
Table 1 Buoys data and the data collocated with ERS-1/2 scatterometer data in
the Pacific and Atlantic Oceans (1991 2000)
Station Location Water depth (m) Number of data points
The Pacific
460**-**-**-** N, 130 16 19 W 3 374 67
46005 46 03 N,131 01 12 W 2 779.8 642
460**-**-**-** N, 177 34 35 W 3 662.3 140
510**-**-**-** N, 162 12 28 W 3 252 866
510**-**-**-** N, 152 28 51 W 5 303.5 340
460**-**-**-** N, 120 26 54 W 447.1 12
460**-**-**-** N,123 19 00 W 126.5 6
The Atlantic
410**-**-**-** N, 75 21 36 W 3 316.2 843
440**-**-**-** N, 66 34 47 W 88.4 143
440**-**-**-** N, 69 09 48 W 195.7 642
42002 25 10 N, 94 25 00 W 3566.16 46
420**-**-**-** N, 88 12 48 W 443.6 123
Fig.2 Relationship between relative azimuth angle and NRCS for incidence angle 30 when H1/3/(gT2) is 0.005
Fig.3 Relationship between relative azimuth angle and NRCS for incidence angle 45 when the H1/3 is 2 m
of modeling a large variety of physical phenomena.
2.3 The algorithm of deriving ocean wave
Here, adaptive means the method is able to process a
parameters
large number of data or deal with new relevant
The significant wave height (H1/3) and H1/3/(gT2) variables. Second, even if the learning phase of the
are retrieved by ERS-1/2 scatterometer data using a network takes a long times, the operational phase is
neural network (NN) algorithm. very efficient. This phase requires few calculations
NN offer interesting possibilities for solving and can be performed with personal computers.
problems involved in transfer functions. First, the Moreover, NN architecture can be easily
NN are adaptive, providing a flexible and easy way implemented on dedicated hardware using parallel
No.1 GUO et al.: A new model to estimate significant wave heights with ERS-1/2 scatterometer data 115
algorithms, and further saving the processing time. 2.4 Comparison between derived and buoy wave
In this paper, H1/3 is retrieved from scatterometer data
data using neural networks technology.
If a wave period of buoy is less than 7 s (T 7s),
The learning data include incidence angles,
the case is defined as wind-wave domination. The
cos, NRCS, wind speed (V) and H1/3 from
learning data include 4 100 collocated pairs by
buoys. The module structure consists of a multi-layer
random choice. An additional 4 048 ones are
perception (MLP) that includes one hidden layers.
randomly taken as test data and not used in the
The transfer function of the input hidden layer is a
learning phase. Fig.4 shows H1/3 for buoy data in
sigmoid function f(x) =2/[1+exp(-2x)], and that of the
comparison with the retrieved H1/3 from ERS-1/2
output layer assumes the linear function f(x)=x. The
scatterometer. Fig.5 compares the H1/3/(gT2) from
input data are the incidence angles and cos,
buoy data with the retrieved ones from ERS-1/2
and NRCS, and wind speed (V) while the output
scatterometer. Table 1 displays the detail.
data are H1/3 or H1/3/(gT2). In the following equation,
d
Aj == f ( ij xi ) j=1,, n.
i =1
Aj is the output of the jth neuron, d is the amount
of input, and n is the neuron amount of the hidden
layer. The connection weights are determined during
the learning phase using the back-propagation
network (Lin et al., 2006).
The retrieved H1/3 and H1/3/(gT2) from
scatterometer data are compared with the H1/3 and
H1/3/(gT2) values from buoy data in Figs.4 7 and
Table 1.
Table 1 H1/3 (m) and H1/3/(gT2) RMS for NN inversion
Fig.4 Comparison in H1/3 (m) between buoy data and those
retrieved ones from ERS-1/2 scatterometer data
2
Item H1/3 H1/3/(gT )
Wind-wave Domination
Corr1 0.75 0.83
RMS1 0.51 m 0.000 97
Error1 0.41 m 0.000 74
Bias1 -0.002 8 m 1.29e-006
Swell domination
Corr2 0.84 0.92
RMS2 0.72 m 0.000 77
Error2 0.55 m 0.000 6
Bias2 0.002 m 6.06e-006
In Table 1, Corr is correlation coefficient, RMS
for root mean square, and
1 N
( k2 )2
1
RMS= (1) Fig.5 Comparison in H1/3/(gT2) between buoy data and
k
N k =1
retrieved ones from ERS-1/2 scatterometer data
1 N
k2
1
Error = (2) In total, 10 485 the ERS-1/2 scatterometer-yielded
k
N data that collocated with NDBC buoy readings are
k =1
used in this study, of which 8 148 are 7 s and 2 337
1 N
( k2 )
1
are >7 s in wave period. If a wave period of buoy is
Bias= (3)
k
N >7 s, swell domination is assumed; otherwise,
k =1
wind-wave domination. The learning data include
where N is the number of test data. 1 200 collocated pairs by random choice. Additional
CHIN. J. OCEANOL. LIMNOL., 27(1), 2009 Vol.27
116
1 137 collocated pairs are randomly taken as test data In wind-wave domination, the RMS of H1/3 is 0.51 m,
and not used in the learning phase. Fig.6 shows the while that in swell domination case, 0.72 m. The H1/3
comparison in H1/3 between buoy data and RMS values of the wind-wave domination in this
scatterometer data, and Fig.7 is for H1/3/(gT2) ones in study are consistent with those of Ebuchi and
the same manner. Kawamura s paper (1994), while in the swell
The bias, average absolute error and root mean domination, the RMS results are on the high end of
square (RMS) of the H1/3 and H1/3/(gT2) retrievals the Ebuchi and Kawamura (1994) results. It shows
with the buoy-measured values are given in Table 1. that the effect of swell on the radar cross section is
significant. It is practical to retrieve H1/3 and
H1/3/(gT2) values from the ERS-1/2 scatterometer
with neural networks methods. In the future, the
results from H1/3/(gT2) retrievals shall be used to
calculate T and wave lengths.
References
Ebuchi, N. and H. Kawamura, 1994. Validation of wing
speeds and significant wave heights observed by the
TOPEX Altimeter around Japan. Journal of
Oceanography 50: 479-487.
Hwang, P. A., W. J. Teague, G. A. Jacobs and D. W. Wang,
1998. A statistical comparison of wind speed, wave
height, and wave period derived from satellite altimeters
and ocean buoys in the Gulf of Mexico region. Journal
of Geophys. Res. 103(C5): 10 451-10 468.
Fig. 6 Comparison in H1/3 between buoy data and retrieved
Lin, M. S., X. G. Song and X. W. Jiang, 2006. Neural
ones that from ERS-1/2 scatterometer data
network wind retrieval from ERS-1/2 scatterometer data.
Acta Oceanologic Sinica 25(3): 35-39.
Nghiem, S. V., K. Li, S. H. Lou, G. Neumann, R. E.
McIntosh, S. C. Carson, J. R. Carswell, E. J. Walsh, M.
A. Donelan, and W. M. Drennan, 1995. Observations of
ocean radar backscatter at Ku and C-bands in the
presence of large waves during the surface wave
dynamics experiment. IEEE Trans. Geosci. Remote Sens.
33(3): 708-721.
Quilfen, Y. and B. Chapron, 2003. Relationship between ERS
Scatterometer Measurement and Integrated Wind and
Wave Parameters. Journal of Atmospheric and Oceanic
Technology 21: 368-373.
Stephen L. Durden and John F. Vesecky, 1985. A Physical
Radar Cross-Section Model for a Wind-Driven Sea with
Fig.7 The comparison in H1/3/(gT2) between the buoys Swell. IEEE Journal of Oceanic Engineering. 10(4):
records and those retrieved from ERS-1/2
445-451.
scatterometer data
Tran, N., B. Chapron and D. Vandemark, 2007. Effect of
Long Waves on Ku-Band Ocean Radar Backscatter at
3 CONCLUSIONS
Low Incidence Angles Using TRMM and Altimeter
A neural networks algorithm is developed by the Data. IEEE Geosci. & Remote Sensing Lett. 4(4):
authors for retrieving H1/3 and H1/3/(gT2) (as 542-546.
described in 2.3) from ERS-1/2 scatterometer data.