.../...

4.2 Comparison with buoy data

The aim of this section is to estimate the accuracy of the weekly and monthly wind speed and direction in comparison with buoy wind data. This is achieved by using : the National Data Buoy Center (NDBC), the Tropical Atmosphere Ocean (TAO), and the European Buoys (ODAS) buoy networks (Figure 9). More than 90 buoys covering Atlantic and Pacific ocean areas betwenn 10°S and 57°N.

Figure 9 : Buoy network location

For the validation of the scatterometer average wind field, the buoy wind data are referenced to 10m height, assuming a logarithm wind profile, Von Karman's constant of 0.4, neutral stratification and, a wind speed dependent drag coefficient (Ezraty 1987).

For each week and each month, mean values of buoy wind speed, zonal and meridional components are computed arithmetically. Weekly and monthly means are computed for all ERS-1, ERS-2 and NSCAT periods for which at least 3.5 days and 15 days buoy measurements are collected, respectively. For each averaging period, the closest scatterometer grid point (1° ´ 1° ) to each buoy location is selected. Therefore, a collocated data sets between scatterometer gridded wind fields (averaging objective method) and buoy averaged winds are performed for NDBC, TAO and ODAS buoy networks. Results are then compared using the following standard statistic data analysis :

The wind speed, zonal component and merdional component are assumed as a random variables wich could be characterized by their moments. For this purpose, the four conventional (C moments) and linear moments (L moments) of each variable are estimated.

Let is W a wind variable (wind speed, zonal component, merdional component or wind difference). The corresponding four C moments are determined as :

                               (4)

, _W, S_W and K_W are the W mean (bias), standart deviation, skweness and kurtosis, respectively. Variance and rms values are dereived from and _W estimates.

The L moments (Hosking, 1990} are defined by :

               (5)

n is the nth linear moment of W

is the shifted nth Legendre polynomial. It is related to Legendre polynomial by :

                      (6)

F is the probability function of wind variable W

Q(F), called quantile function, is provided by the following equation :

      (7)

The meaning of C moments and L moment are similar as can be shown through the equations. The main advantage of L moments is their relative small sensitivity to data errors generaly producing outliers in data series.

The statistical significance of the fisrt and second moment is evaluated by Student test (T-test) and Fisher test (F-Test), respectivelly. Troughout this paper, the significance is estimated for 95% confidence.

Moreever, the linear regression parameters are estimated to assess the comparisons between satellite gridded wind fields and buoy averaged winds. In this paper we provide the following parameters :

                 (8)

Where 

x and y denote the buoy and scatterometer wind estimates, respectively. b is the slope and a is the intercept on the y axis : y = bx + a. bs is the slope of symmetric regression line. is the correlation coefficient. Its calculation involves the residual, , between y and linear regression model. p₁, and p₂ are the rms deviations of the first and second principal component of x and y distribution. They provide a measurement of the major and minor axis of the elliptical x and y distribution.

4.3 Global comparisons

Table 2, 3, and 4 provide the main statistical parameters characterizing wind speed comparisons. The wind speed correlation coefficients ranging from 0.85 to 0.89 indicate a good consistency between satellite and buoy averaged winds. The rms values of the differences buoy-satellite wind speeds do not exceed 1.16m/s over NDBC and TAO networks. Results derived from ODAS/satellite comparisons show higher rms values : 1.48m/s for NSCAT, and 1.66m/s for ERS-2. The latter are mainly due to a poor number of comparison data points, and to the high wind variability in ODAS area (Figure 9). Furthermore, the statistics calculated by several meteorological centers (ECMWF, CMM, UKMet) indicate that ODAS buoy wind speed tend to be underestimated according to meteorological wind analysis (see ftp://ftp.shom.fr/meteo/qc-stats, site maintained by P. Blouch).

The results of the regression analyses carried out on collocated data, show that the slopes calculated over each buoy network and against buoy wind estimates, are quite similar for the three averaged scatterometer wind speeds. In NDBC area (Table 2), buoy and scatterometer wind speeds agree quite closely, which is expressed by slopes of about 1 and intercepts of about zero. Comparisons between buoy and scatterometer winds in Pacific tropical ocean give regression line slopes of about 0.80, suggesting an overestimation of low wind speed and underestimation of high wind speed by scatterometer wind fields compared to TAO winds. In north Atlantic area, the slopes are very close to 1, whereas the intercepts are of about 0.50, indicating that the scatterometer wind fields are consistently high compared to ODAS week-averaged wind speeds. The calculation of the statistical parameters according to the buoy wind speed ranges, show that their values are made variable by the outlying points at low and high wind speeds.

For the wind direction, no systematic bias is found, and the overall bias and standard deviation about the mean angular difference are less 8° and 38° , respectively. These results are consistent with the calibration/validation of the scatterometers against buoy (Graber et al, 1996 and 1997; Caruso et al, 1999). For instance, in Pacific tropical area, where the wind direction is quite steady, the standard deviation calculated for buoy wind speed higher than 5m/s, does not exceed 17° .

A

B

C

Figure 10 :	a/ Scatter plot and frequency of wind speeds, x-axis shows averaged TAO wind speed,
	y-axis shows the gridded scatterometer wind speed
	b/ as Figure 12a but for zonal component
	c/ as Figure 12a but for meridional component

The geographical features of the difference between scatterometer and buoy weekly wind estimates have been investigated. At each TAO buoy location, statistical parameters of the difference series are computed. For instance the rms values of the wind speed and direction differences are shown in Figure 11. The main result is that the rms values are generally higher within the 150E and 170E Pacific band. A study of the correlation between in-situ and sensor matchups indicates that there is a significant difference between the correlation coefficients computed in the eastern Pacific and those computed in the western Pacific with 95 % confidence. Using TAO buoy measurements, it was determined that in the western Pacific there is 6 times more energy compared to the eastern Pacific (Mangum, 1992) . This high variability of the wind is mainly explained by the high surface temperature and the convective activity of this zone. Averaging procedures give less representative results in these wind conditions.

a)

b)

[R]

.../...