Professional Documents
Culture Documents
Abstract. Based on the strong correlation between nondimensional wave height and wave age,
we propose a quartic polynomial model by piecewise method according to significant wave
height. In order to validate our new model, a validation dataset was created by interpolating
the mean wave period data of ERA-40 to 23 cycles counterpart measuring points of Jason-1
altimeter with bilinear interpolation. We found that (1) the previous inversion models based
on altimeter measurements cannot provide enough accuracy with our global validation dataset;
(2) compared with H98 model, the RMS error and bias were reduced by 0.8949 s and 0.3759 s,
respectively, by piecewise quartic polynomial model. Furthermore, the bias of the new model
is 0.0197 s, which indicates that it offers enough accuracy in the global range. 2012 Society of
Photo-Optical Instrumentation Engineers (SPIE). [DOI: 10.1117/1.JRS.6.063591]
Keywords: mean wave period; bilinear interpolation; nondimensional wave height; wave age;
piecewise quartic polynomial model.
Paper 12080 received Mar. 28, 2012; revised manuscript received Sep. 12, 2012; accepted for
publication Sep. 26, 2012; published online Nov. 7, 2012.
1 Introduction
Satellite altimeters provide accurate sea state data with high spatial resolution globally. Global
descriptions of wave conditions can be obtained from satellite observations that can also be used
in the inversions of wave parameters. At present, the inversion of significant wave height (H S ) by
satellite altimeters is widely used due to the high precision. However, H S provides only limited
information about the wave spectra, namely about its height and energy. Mean wave period (T),
the mean time that the full wave takes to pass a given point, is another important wave parameter.
However, the retrieval of T from altimeters has not been investigated adequately, and its data are
not contained in the altimeters geophysical data records (GDR) currently.1
Several model-related parameters contained in altimeter GDR data are used to represent the
mean wave period. The first attempt of an altimeter based on a T inversion model is proposed by
Challenor and Srokosz.2 Their model establishes a theoretical relationship between HS and backscatter coefficient ( 0 ) with T. An empirical model based on wave dynamics is advanced by
Hwang et al.3 and buoy observations are used to derive their specific model. New empirical
models based on heuristic arguments are proposed by Gommenginger et al.4 and by Mackay
et al.5 to obtain T measurements from collocated TOPEX altimeter and buoy dataset. A neural
network is used by Quilfen et al.6 to establish a relationship between altimeter H S and 0 with
buoy T based on a collocated data set of TOPEX/Poseidon and National Data Buoy Center
(NDBC) buoy measurements. Considering that the underlying relationship between the variables
is not so clear in the Ref. 6 algorithm and the specific inversion models for different altimeters are
quite different, the Ref. 6 model will not be mentioned here.
Although both the models from Refs. 3 and 4 are based on the NDBC buoy dataset, the
NDBC buoy dataset is restricted to only coastal regions, offering rather limited sea state
063591-1
Vol. 6, 2012
conditions. The European Centre for Medium-Range Weather Forecasts (ECMWF) has completed the ERA-40 reanalysis data based on numerical Wave Model (WAM) data. ERA-40
can provide T data with high precision and spatial resolution globally7. In this paper, a global
validation of the previous T models is performed based on a collocated ERA-40/Jason-1 validation dataset. A new empirical model based on strong correlation between nondimensional wave
height and wave age is given by using an independent development dataset.
Sf; fx dfd:
(1)
(2)
In this paper, we use the conversion relationship between T Z and T 0;1 , which was proposed
by Li9 [see Eq. (3)] to make the ERA-40s mean wave period data compatible with those of the
altimeter data:
T Z 0.8017T 0;1 :
(3)
In this paper, using the collocated ERA-40 and altimeter dataset, a global validation of the
previous wave period models based on altimeter measurements is performed, which is known for
the first time. The ERA-40 and altimeter data have to be processed to make their time and space
scales compatible. A total of 23 cycles of data of the Jason-1 GDR database, from January 2002
to August 2002, are chosen. In order to match the time of altimeter with ERA-40, the selected
times are before and after 3 min of 00:00, 06:00, 12:00, and 18:00, respectively. The ERA-40 T
data are then bilinearly interpolated to the selected altimeter data locations, and after that the time
and space scales of two databases are compatible. Because the T data at ice and land are constant
as 37267 s, it is easy for us to delete these exceptional data from the above interpolated data.
With this criterion and the applications of standard ice and rain flags from the AVISO Jason-1
GDR dataset, we have selected 73,640 data points acquired from the months of 01, 03, 05, and
07 as validation dataset, and 70,604 data points acquired from the months of 02, 04, 06, and 08 as
development dataset.
063591-2
(4)
Vol. 6, 2012
where cP is the phase speed of the dominant wave, U is the wind speed, and is a measurement
of the degree of wave development. According to the dispersion relationship for deep water
gravity waves, follows the expression
cP
g
;
U wP U
(5)
where g is the acceleration of gravity, and wP is the peak frequency of spectrum. Based on
wP 2T P , where T p is the peak wave period (T p ), follows the expression
gT P
:
2U
(6)
However, T p is an unstable parameter, especially for large periods, so it is not a good sea state
descriptor.4 Alternatively, the zero-crossing period T Z is an integrated property. It is more robust
than T p and can be used in place of T p . Thus, we use the zero-crossing period T Z as our
mean wave period in this paper. The empirical conversion relationship between T Z and T p
is T p 1.44 T Z .10 Thus the is
1.44
gT Z
:
2U
(7)
Nondimensional wave height (Hn ) is another important parameter of the degree of wave
development:
Hn
gH S
;
U2
(8)
a
:
(9)
a
1.44 Z a0 a1
2
3
4
2U
U2
U2
U2
U2
Using our development dataset, we can determine the model described by Eq. (9) and obtain
the coefficients of the quartic polynomial. Results show that all the coefficients are in line with
the 95% confidence intervals:
Fig. 1 Nondimensional wave height versus wave age for collocated development dataset.
Red line shows quartic polynomial fitting.
Journal of Applied Remote Sensing
063591-3
Vol. 6, 2012
a0 0.617 0.00463;
a3 0.3108 0.099;
a1 4.257 0.0395;
a2 1.416 0.106;
a4 0.02718 0.0253:
In order to improve the quartic polynomial model, the scatter diagrams of the bias of
T Z against significant wave height (H S ), nondimensional wave height (H n ), wave age (),
and wind speed (U) are performed. The bias of T Z is calculated as the mean of
TERA-40 Talt. The correlation between the bias and H S is the highest in the above relationships, and Fig. 2 shows the scatter diagram of the bias against H S . As is shown, the correlation is 0.6581. Thus, an obvious bias trend against H S is noticed after applying the quartic
polynomial model to determine mean wave period. The piecewise fitting method is used to
improve our quartic polynomial model [see Eq. (9)]. We divide the development dataset into
eight segments with 1 m intervals of H S , and calculate the coefficients of the quartic polynomial,
respectively. Results are given in Table 1.
R-square and RMSE of eight segments in the piecewise fitting are given in Table 2.
It is found that the fitting is less effective in the region of 0 < H s 1 and H S > 6 (see
Table 2). Although RMSE of H S > 6 section is much smaller than other intervals, R-square
of H S > 6 section is less effective than that of 1 < H s < 6. Figure 3 shows the occurrences
of the mean wave period by applying the piecewise quartic polynomial model to the development dataset, which indicates that the vast majority of sea state data focuses on 1 < H s < 6. Thus,
the fitting effectiveness in 0 < H s 1 and H s > 6 sections is relatively poor. However, the data
Table 1 Coefficients for piecewise quartic polynomial model, with 95% confidence intervals
given below the coefficients.
0 < Hs 1
1 < Hs 2
2 < Hs 3
3 < Hs 4
4 < Hs 5
5 < Hs 6
6 < Hs 7
7 < Hs 8
a0
0.9672
0.1718
0.6957
0.0318
0.5747
0.0201
0.6256
0.0229
0.5804
0.0286
0.5321
0.03
0.401
0.1027
0.03445
0.26435
a1
6.486
0.664
5.086
0.114
4.563
0.078
3.544
0.094
3.186
0.129
3.035
0.149
3.458
0.696
5.566
2.176
a2
2.858
0.697
1.777
0.116
1.782
0.084
1.22
0.107
1.319
0.157
1.563
0.198
2.674
1.493
7.135
6.045
a3
0.6494
0.2539
0.3556
0.0415
0.4082
0.0316
0.3027
0.0414
0.4533
0.0636
0.5078
0.0887
1.421
1.2208
5.365
6.757
a4
0.05695
0.02904
0.02691
0.00473
0.03479
0.00372
0.02917
0.00498
0.05438
0.00796
0.05328
0.01198
0.3219
0.3205
1.799
2.598
063591-4
Vol. 6, 2012
0.4988
0.8116
0.8602
0.8405
0.8351
0.8121
0.7446
0.7405
RMSE
1.537
0.6932
0.4860
0.4225
0.3543
0.2470
0.2376
0.2138
Table 3 RMS error and the bias for Refs. 2, 3, and 4 and PQP models.
Ref. 2
Ref. 3
Ref. 4
PQP
RMS (s)
2.4833
1.8922
1.8565
0.9973
Bias (s)
1.9015
0.3956
0.7344
0.0197
quantity in these sections is rather small (Fig. 3), which indicates that the piecewise fitting
performs well as a whole.
In order to validate the piecewise fitting effectiveness, the bias trend is performed. In Fig. 4,
we compare the bias before and after the piecewise fitting is applied in our models. The nonpiecewise quartic polynomial [QP model, see Eq. (9)] and piecewise quartic polynomial model
Fig. 4 The bias trend before and after piecewise fitting applied.
Journal of Applied Remote Sensing
063591-5
Vol. 6, 2012
(PQP model, see Table 1) are used to calculate T(alt). It is evident that the bias of PQP model
decreases to 0 and its bias trend disappears compared with the QP model.
The bias trend clearly disappears in the piecewise fitting method, which indicates that the
piecewise fitting model performs better than the nonpiecewise fitting model.
Fig. 5 Histogram of retrieved mean wave period from PQP and Ref. 3 models, against ERA-40
wave period histogram.
Journal of Applied Remote Sensing
063591-6
Vol. 6, 2012
Fig. 6 The occurrence of differences between retrieved mean wave period from PQP and
Ref. 3 models.
Finally, Fig. 6 shows the occurrence of differences between T(ERA-40) and T(alt) derived
from the Ref. 3 and PQP models with the validation dataset. The result in Fig. 6 reveals that the
differences based on the PQP model mainly focus on the interval of (1 s, 1 s), and the absolute
value of difference that is greater than 1 s is in the minority. But the differences derived from the
Ref. 3 model behave worse because the sample quantity in the (1 s, 1 s) interval is obviously
lesser than that of the PQP model. We can conclude that the PQP model can reach better
agreement with ERA-40 data than that of the Ref. 3 model in the global range.
5 Conclusion
The previous mean wave period inversion models are validated by the global ERA-40/Jason-1
collocated validation dataset. Because the previous empirical inversion models are all deduced
from the sea state conditions limited buoy dataset, the previous models do not perform well in
terms of the RMS error and the bias of TERA-40 Talt. Thus, the results of the mean
wave period calculated with the previous models cannot have a good agreement with the
ERA-40 one.
A piecewise quartic polynomial model of mean wave period based on the strong correlation
between nondimensional wave height and wave age is proposed. The global ERA-40/Jason-1
collocated development dataset is used to establish the model. It is demonstrated that the new
model developed performs better than the previous models in all metrics that we have considered. Because our new mean wave period model can adapt to global sea state conditions, it is
more effective than the previous models in the global range.
Acknowledgments
This work was supported by the Natural Science Foundation of China (NSFC, 41176157).
References
1. AVISO and PODAAC User Handbook for IGDR and GDR Jason Products (Ed. 4.2), NASA
(2012).
2. P. G. Challenor and M. A. Srokosz, Wave studies with the radar altimeter, Int. J. Remote
Sens. 12(8), 16711686 (1991), http://dx.doi.org/10.1080/01431169108955200.
3. P. A. Hwang et al., A statistical comparison of wind speed, wave height, and wave period
derived from satellite altimeter and ocean buoys in Gulf Mexico region, J. Geophys. Res.
103(C5), 1045110468 (1998), http://dx.doi.org/10.1029/98JC00197.
Journal of Applied Remote Sensing
063591-7
Vol. 6, 2012
4. C. P. Gommenginger et al., Measuring ocean wave period with satellite altimeters: a simple
empirical model, Geophys. Res. Lett. 30(22), 15 (2003), http://dx.doi.org/10.1029/
2003GL017743.
5. E. B. L. Mackay et al., A parametric model for ocean wave period from Ku band altimeter data, J. Geophys. Res. 113(03029), 116 (2008), http://dx.doi.org/10.1029/
2007JC004438.
6. Y. Quilfen et al., Calibration/validation of an altimeter wave period model and application
to TOPEX/Poseidon and Jason-1 altimeters, Marine Geodesy 27(34), 535549 (2004),
http://dx.doi.org/10.1080/01490410490902025.
7. A. J. Simmons, Development of the ERA-40 data assimilation system, in Proceedings of
the ECMWF Workshop on Re-analysis, ERA-40 Proj. Rep. Ser., Vol. 3, pp. 1130,
Eur. Cent. for Medium-Range Weather Forecasts, Reading, United Kingdom (2001).
8. S. Caires et al., Global ocean mean wave period data: validation and description, J. Geophys. Res. 110(C02003), 112 (2005), http://dx.doi.org/10.1029/2004JC002631.
9. R. Li, Study on the relationships of various wind wave periods, Trans. Oceanol. Limnol.
1003(2), 1318 (2007).
10. S. C. Wen et al., Form of deep-water wind-wave frequency spectrum, I. Derivation of
spectrum, Prog. Nat. Sci. 4(4), 407427 (1994).
Hongli Miao is a professor in College of Information Science & Engineering, Ocean University of China. He obtained his BSc and ME in physics.
His current research interests mainly include data processing in altimetry
and LED illumination.
063591-8
Vol. 6, 2012