.../...

4.3 Comparison with model

The global accuracy of the weekly and monthly wind fields derived from ERS-1 scatterometer measurements is evaluated in comparison with the European Center for Medium Range Weather Forecasts (ECMWF) wind estimates. The latter are provided at synoptic time (00h, 06h, 12h, 18h) with a spatial resolution of 1.125° in longitude and latitude. The weekly and monthly averaged wind fields are computed from ECMWF analysis. Figure 14 shows the annual mean difference of winds between scatterometer and ECMWF calculated on a 1° grid. The agreement between the two wind fields is quite good. The mean and standard deviation values of the difference are 0.53 m/s and 1.15 m/s, for wind speed, 0.22 m/s and 1.34 m/s, for the zonal component, 0.05 m/s and 1.26 m/s, for the meridional component. In the subtropical regions the difference values are mainly negative, but they do not exceed 0.5 m/s. In the rest of the world, the difference values are mainly positive, indicating that the wind speeds calculated from the scatterometer are larger than those estimated from ECMWF analysis. Large-scale differences are found in the Southern Hemisphere (SH). For instance the difference in wind speeds reaches 2 m/s north-east of Australia. Zonal and meridional annual mean differences are typically less than 0.5 m/s. Such substantial errors are only found near continental margins and in the Tropical Pacific area located between 130E and 180E. In order to investigate further, the wind speed, zonal component and meridional component derived from scatterometer and ECMWF are averaged over the longitudinal range of three ocean basins and during the ERS-1 period. The results are shown in Figure 15. In comparison with previous climatological studies, we note that the zonal winds over each ocean basin are well represented by the global zonal means. The correlation between the two averaged winds is high and significant with 95% confidence. However, the zonal wind speeds calculated from the scatterometer are slightly weaker in the North Atlantic and North Pacific, and higher in the Southern hemisphere high latitude compared to ECMWF zonal wind speed values. The discrepancy is larger in the Southe Pacific than in the South Atlantic. For the two basins, the most substantial differences between the two data sets are located south of 60 S, exceeding 1.50 m/s. In the Indian ocean, the scatterometer provides higher zonal winds.

Furthermore, the difference between the scatterometer and ECMWF wind fields is not consistent from year to year. For instance, the zonal mean of wind speeds calculated for the years, 1992, 1993, 1994 and 1995 over the Atlantic basin is represented in Figure 16. It is obvious that after 1993, the wind speed derived from ECMWF analysis has increased, especially in the Northern hemisphere. This could be due to the change in the ECMWF numerical model used to estimate surface wind (Ritchie et Al., 1995).

a

b

c

Atlantic

Pacific

Indian

Figure 16 :	Four years of annual mean zonal of wind speed (m/s) from gridded
	scatterometer (solid line) and ECMWF (dashed line) wind fields.

Another comparisons performed between scatterometer and ECMWF wind fields concerns the spatial scales. To illustrate the result, the zonal correlation function of the zonal and meridional components are calculated according to distance using the following formula :

Where C is the autocorrelation function, f and x represent wind variables and distance, respectively.

This calculation is possible, since the dates of the analyzed fields from the scatterometer and ECMWF are available between 1992 and 1995. The homogeneity of wind fields could be assumed (Wickert et al, 1971).

The zonal correlation function and confidence intervals for the zonal and meridional components are estimated in various regions. Figure 17 shows the results of these calculations in three areas of the Atlantic basin. The agreement between each pair of zonal correlation functions is good. However, ECMWF wind components exhibit higher correlation coefficients at small distances, indicating the smooth nature of small scale variability when using the ECMWF numerical model.

[R]