Wind speed from TOPEX B side Ku Altimeter

D J T Carter

Satellite Observing Systems Ltd.

6 January 2002

Download the Word version here

Introduction

The TOPEX CD of Geophysical Data Records includes estimates of 1 Hz s0 values and of wind speeds derived from s0.  Wind speed is calculated from an algorithm obtained by AVISO (1996) by fitting a table, given by Witter & Chelton (1991), of wind speed:s0 values from Geosat data.   To adjust for differences in observed values of s0 from Geosat and from TOPEX Ku-band, 0.63 dB is subtracted from the TOPEX value before entering the algorithm.  The wind speed written to the CD is rounded to the nearest 0.1 m/s.

Recently, improvements to the algorithm have been sort by including the altimeter estimates of significant wave height (Hs) as well as s0; for example by Gourrion et al. (2000) who derive algorithms for wind speed from an analysis of altimeter and scatterometer cross-overs.   Gommenginger et al. (in press) compare values from a range of algorithms against buoy wind speeds and conclude that Gourrion et al's algorithm derived using a multilayer percepton (MLP) network design (see Annex A below) gives the 'best overall results for a two-parameter algorithm'.

This note compares co-located wind speeds from US NDBC buoys and from the TOPEX B transmitter (operational since February 1999) from the AVISO (1996) algorithm with s0 - 0.63 (i.e. the wind speed on the CD but not rounded to 0.1 m/s) and from the Gourrion et a/. (2000) MLP algorithm with the Ku-band estimates of s0 and Hs (neither adjusted).

 

Data Set

The derivation of the data set of 992 pairs of US NDBC buoy and TOPEX data are described in Cotton (2001).   The criteria for "coincident" altimeter and buoy data were 50 km and 30 minutes; for each event, the single altimeter 1 Hz data record closest to the buoy location was used.   Buoy speeds were adjusted, when necessary, to that at 10 m above the sea surface.

 

Results

AVISO winds

Figure 1 shows the buoy wind speeds plotted against those from TOPEX using the one-parameter algorithm with s0 reduced by 0.63 dB, together with the line obtained by orthogonal distance regression (assuming both buoy and TOPEX winds have errors of equal magnitude).

Figure 1

 

The regression equation is:

   U(TOPEX) = 1.135 U(buoy) - 0.721          (1)

The residual standard deviation is 0.89 m/s, and slope and intercept are significantly different from 1 and 0 respectively. 95% confidence limits are included in Table 1 (below).   (Using the wind speeds on the TOPEX CDROM, specified to 0.1 m/s, gives slope and intercepts of 1.136 and -0.725 respectively.)

MLP winds

Figure 2 shows the buoy wind speeds plotted against those from TOPEX using the two-parameter MLP algorithm with TOPEX s0 and Hs.  The regression equation is:

   U(MLP) = 1.073 U(buoy) - 0.539              (2)

with residual standard deviation of 0.83 m/s.   Again, both slope and intercept are significant, but closer to 1 and 0 than in Equation 1.

Figure 2

 

However, the plotted data do not appear to be distributed quite linearly; the low wind speeds do not seem to fit.   So regressions were carried out fitting only data pairs with both buoy and MLP wind speeds greater or equal to a cut-off ranging from 2.5 m/s to 10 m/s. Results are included in Table 1, and show that with increasing cut-off the slope decreases towards 1 and the intercept approaches 0.   For cut-off at 5 m/s the 95% confidence limits of slope and intercept cover 1 and 0 respectively, and the residual standard deviation is reduced to 0.74 m/s.

The regression of the 197 pairs with both buoy and TOPEX MLP values less than 5 m/s is shown in Figure 3. The regression equation is:

   U(MLP) = 1.203 U(buoy) - 1.090                   (3)

But note the scatter in the figure and the rather wide confidence limits given in Table 1.  (Some of this scatter may be because of relatively large variations over 50 km and 0.5 hr when winds are light, or because of the high sensitivity of U to s0 at low U.)

Figure 3

 

The above analyses use uncalibrated TOPEX Hs values.   Regressing TOPEX and buoy Hs (Figure 4) gives:

   Hs(TOPEX) = 0.9638 Hs(buoy) + 0.0650         (4)

i.e

   Hs(buoy) = 1.038 Hs(TOPEX) - 0.0674           (5)

Figure 4

 

Using TOPEX Hs corrected by Equation 5 in the MLP algorithm, instead of the unadjusted values, makes very little difference to the results.  For example, the slope and intercept of U(MLP) on U(buoy) of all the 992 data pairs become 1.070 and -0.521, compared to 1.073 and -0.539 in Equation 2.

 

Conclusion

The good agreement between wind speeds from US NDBC buoys and those from TOPEX obtained from the MLP algorithm of Gourrion et al (2000) using Transmitter B s0 and Hs values indicates that for wind speeds above 5 m/s no calibration is required.

At wind speeds less than 5 m/s the situation is not so clear - see Figure 3.   Equation 3 suggests that the algorithm underestimates the true (buoy) wind speed, increasingly so with decreasing wind speed, but the use of equation 3 would result in no estimated speed less than 0.9 m/s.   Further research is clearly needed if altimeter measurements of low wind speeds are to be better resolved; meanwhile no large errors will result if, as for higher wind speeds, no calibrations are applied to the values from the MLP algorithm.

Note that if the MLP algorithm is used with data from Geosat or ERS altimeters then the s0 value entered into it should be the measured value plus 0.63 dB.

 

References

AVISO 1996
  AVISO User Handbook: Merged TOPEX/POSEIDON Products
  AVI-NT-02-101-CN, Edition 3.0, July 1996. CNES, Toulouse.

Cotton P. D. 2001
  TOPEX B Side Altimeter Wind Speed and Significant Wave Height Calibration (1999-2001).
  Satellite Observing Systems Unpublished note.

Gommenginger C. P., Srokosz M. A., Challenor P. G. & Cotton P. D. [In press]
  Development and validation of altimeter wind speed algorithms using an extended collocated buoy/Topex dataset.
  IEEE TGRS.

Gourrion, J., Vandemark D., Bailey S. & Chapron B. 2000
  Satellite altimeter models for surface wind speed developed using ocean satellite crossovers.
  Ifremer Tech. Report No. IFREMER-DROOS-2000-02.

 

Eq.

Number

Intercept/
slope

95%ile

range

Res. s.d.

U(s0):buoy

1

992

-0.7206

-0.9461

-0.4950

0.894

1.1355

1,0461

1.09951

U(s0,Hs):buoy

2

992

-0.5386

-0.7415

-0.3357

0.833

1.0728

1.0461

1.0995

U(s0,Hs):buoy >2.5

900

-0.3709

-0.6053

-0.1366

0.795

1.0536

1.0241

1.0832

U(s0,Hs):buoy >5.0

684

-0.1728

-0.4960

0.1505

0.741

1.0320

0.9950

1.0690

U(s0,Hs):buoy >7.5

375

-0.1592

-0.7460

0.4276

0.734

1.0270

0.9689

1.0851

U(s0,Hs):buoy >10.0

113

0.1047

-1.2364

1.4459

0.769

1.0079

0.8977

1.1182

U(s0,Hs):buoy <5.0

3

197

-1.0900

-1.6609

-0.5191

0.729

1.2034

1.0346

1.3722

Hs:buoy

4

992

0.0650

0.0348

0.0952

0.172

0.9638

0.9499

0.9777

Table 1: Results of orthogonal distance regressions.

(U:wind speeds in m/s. s0:backscatter (dB), Hs:significant wave ht in m.)

 

 

Annex A: Ku-band MLP solution algorithm from Gourrion et al (2000)

   

where

         

          and

where

  and 

where

Then wind speed (in metre/sec) is given by:

   

where

                       and