Prepared by

For

May 17, 1999

As described by McPhaden et al [1998], the TAO array consists of approximately 70 buoys spaced fairly regularly across the tropical Pacific Ocean within 9 degrees north and south of the equator. This array measures upper-ocean properties and surface weather conditions, including wind speed and direction, thus providing a large-scale picture of the wind field over the Pacific. The winds are measured 4 meters above the ocean at a sample rate of 2 Hz, over period of 6 minutes centered at the top of each hour. Nominal wind speed and direction sample resolutions are 0.2 ms^-1 and 1.4° , respectively. The TAO data used were obtained at two different sample rates. Hourly data from 1990 through 1996 were used for the TAO/pseudo-NSCAT bias evaluation and 10-day averaged values derived from daily winds between September 1996 to June 1997 were used for the speed/direction bias estimates as well as the EOF/CCA analysis.

Ten-day averaged NSCAT winds, computed with the NSCAT-1 model function [NASA/JPL, 1997] and interpolated to a 1-degree grid, were provided by Wenqing Tang (NASA/JPL). Data from the tropical Pacific region within 10 degrees of the equator were extracted for this analysis.

Because of the twice daily sampling of NSCAT and the evidence for semi-diurnal and higher frequency wind variability in the tropical Pacific [Deser, 1994], we tested the effect of aliasing these signals into the computation of the daily and longer-term means. The goal of this analysis was to quantify the potential for aliasing given the sampling regime, not necessarily to quantify aliasing during the limited period of NSCAT. Thus, it is not necessary that TAO and NSCAT data overlap in time for this analysis, so for more robust statistics, TAO records over several years (1990-96) were used. The hourly winds from the TAO buoy array (69 buoys) were subsampled at the NSCAT crossing times. The daily averages of these pseudo-NSCAT data were compared with daily averages computed using all the hourly wind data. For each TAO buoy location, the hourly wind components (u and v) before and after the local NSCAT crossing times (1030 and 2230) were averaged to produce a semi-diurnal time series that would simulate the NSCAT sampling protocol. Daily and 10-day means were calculated with both the TAO and pseudo-NSCAT data sets and the differences were analyzed. Ten-day averages were chosen for this analysis to be consistent with the 10-day averaged NSCAT data utilized in the EOF and CCA studies discussed below. A 10-day average was necessary to create the global coverage in the NSCAT data set and was not considered limiting since the emphasis in the study is on climatological response time scales.

A representative time series of the TAO daily mean and the pseudo-NSCAT daily mean for both zonal (Figure 1) and meridional (Figure 2) wind components, along with histograms of the daily differences and the 10-day mean daily differences, over the entire data set, are presented. The time series track each other very well, with the general shape and character consistently represented by the pseudo-NSCAT daily means. The range of the distribution of the 10-day mean differences for both components is less than ± 1.0 ms^-1. The range of the mean daily differences for both computational methods was generally less than the NSCAT measurement threshold of 2 ms^-1. The means of these distributions showed very small biases (mean and SD were -0.12 ± 0.23 ms^-1 for zonal and -0.03 ± 0.24 ms^-1 for meridional components for the 10-day data), negative as expected. Our conclusion is that sampling interval aliasing of semi-diurnal signals in the NSCAT data is not a significant error source for daily and longer time scales.

We compared large scale and low frequency modes in TAO buoy data with those from NSCAT data using Empirical Orthogonal Functions (EOF's) in order to evaluate the NSCAT measurement of the wind field. The NSCAT wind vectors were averaged to a 10-day, 1x1 degree global data set (courtesy of Wendy Tang, NASA/JPL), from which the tropical Pacific region of interest (10S to 10N and 160E to 100W) was extracted. A subset of this data, produced by linearly interpolating in 2-D to the locations of the TAO buoys, allowed as direct a comparison as possible with the more sparsely sampled TAO data set, thus adjusting for any spatial sampling bias. The TAO array data had gaps of various lengths in most of the buoy records during the operational period of NSCAT (September 1996 through June 1997). Daily wind data from the TAO array were averaged into 10-day intervals with no average calculated unless at least 5 of the 10 days had data. Continuous 10-day time series were recovered for 34 of the 69 TAO buoys. Their locations were clustered in the central basin between 140E and 110W (Figure 3). Both data sets were interpolated to a common 10-day time base and divided into three distinct sets for comparative analysis: 1) TAO winds, 2) NSCAT winds at TAO locations (NSCAT@TAO) and 3) 1° x 1° resolution NSCAT winds. The EOFs of each were resolved for the first five modes (Table 1).

All five modes showed similar patterns among the respective data sets. The first two modes account for 58 to 68% of the total variance, while each of the subsequent modes are 10% or less of the total variance. Accordingly, our comparisons are focused on the first two modes. The EOF amplitude and eigenvectors for modes 1and 2 are shown in Figures 3 to 5 for each of the three data sets.

The EOFs of the TAO array winds (Figure 3) illustrate the two distinct wind patterns. Mode 1 consists of an equatorial jet-like feature. Since the data have had the mean removed for this analysis, this represents a mode with enhanced eastward equatorial winds, or weakened westward winds, compared to the mean field. The amplitude function for this mode increases in March of 1997 and represents the equatorial, eastward wind anomaly pattern associated with the onset of the 1997-98 El Niño [Liu et al, 1998b]. The second mode indicates a strengthening of southwestward wind vectors, or equivalently, a strengthening of the northeast Trade Winds in the east-central basin. The seasonal cycle and the Monsoon patterns in the western Pacific are also evident in this mode, but the 9-month record length limits the resolution of these phenomena. The associated amplitude function shows the mode strengthening gradually from October 1996, reaching a maximum in March 1997 and subsequently decreasing in magnitude through early June 1997. The remaining EOF modes have much weaker amplitudes but do represent organized large scale geophysical signals. These patterns are represented in the CCA described in Section 4. The EOF analysis of the TAO array winds thus represents two dominant orthogonal modes of the equatorial wind field during this time; the jet-like equatorial flow and the Trade Winds-Monsoon pattern.

The results of the EOF analysis using the subset of NSCAT wind vectors at the TAO buoy locations (NSCAT@TAO wind) are very similar (Figure 4). The first mode again represents the jet-like equatorial flow and the amplitude function is almost identical to the corresponding TAO data. The percentage of variance accounted for by the mode is slightly less than the TAO winds EOF mode 1 (35% as opposed to 41%), whereas, the respective mode 2 variances are nearly identical.

EOF analysis of the full resolution tropical Pacific NSCAT field (Figure 5) produces wind field modes similar to those from the less densely sampled data sets, except that the order of the two modes is reversed. The Trade Winds (mode 1) have a greater proportion of variance (33% vs. 25%) than the equatorial jet (mode 2). The reversal of the order of these modes is a result of the selection criteria for including TAO data in the analysis. NSCAT coverage extends well east of 125W, allowing the Trade Winds mode to represent a larger percentage of the variance since this variability is more prevalent in this region. Although the TAO array also extends into this region, gaps in the data at these locations precluded their inclusion. The resulting reversal of the order of the modes was tested by calculating the EOF using only the NSCAT data west of 95W (not shown), which resulted in the same order of modes seen in the TAO data. The full NSCAT wind fields show more detail in the structure due to the denser nature of the data set, which is a benefit of the NSCAT instrument. The amplitude functions for the high resolution modes are quite similar to the corresponding TAO located EOFs, demonstrating that the wind fields from these two disparate measurement systems represent the same large scale variability.

The relationship between the two data sets can be further quantified using Canonical Correlation Analysis (CCA) [Barnett and Preisendorfer, 1987]. CCA modes represent ordered patterns of correlation between the two arrays, which have the same time axis, but may have different spatial coordinates or may be different variables entirely. The first five EOF modes of the respective data sets were used as input to the CCA in the following discussions.

The CCA between the TAO wind field and the NSCAT@TAO winds had the highest correlation in a mode resembling the Trade Wind pattern (Figure 6a). This accounts for approximately the same amount of variance in both data sets (29.1% and 27.4%, respectively, Table 2). The CCA amplitude functions are nearly identical and are similar to those of the Trade Wind mode identified in the EOF analysis, increasing to a maximum near March 1997 and then decreasing with the onset of El Niño conditions. This mode reflects a pattern of increasing Trade Winds in the eastern central Pacific during late Northern Hemisphere winter and early spring, 1997, followed by a weakening trend in late spring. The second mode of the NSCAT@TAO CCA (Figure 6b) reflects the eastward equatorial jet with the TAO data accounting for slightly more variance than the NSCAT@TAO winds (35.3% verses 29.4%). Again, the amplitude functions track each other very well; showing a consistent increase after March 1997, reflecting strengthening, eastward equatorial winds during that period of time.

The remaining modes of the CCA (Figures 6c,6d, and 6e) show that the correlations remain high, (r>0.91) although much smaller percentages of the variance are explained (Table 2). These modes, which are less dominant for the large scales and probably reflect less climatically important variability, are nevertheless coherently represented in both the TAO and NSCAT observations.

When the TAO winds are compared to the full resolution NSCAT wind field via CCA, the highest correlated mode remained the Trade Winds mode (Figure 7a). The data were highly correlated (r=1.00 compared to 0.99), with almost an equal partition of variance between them (28.1% and 29.2%, Table 2). The spatial pattern also indicates a slightly greater influence of zonal winds along the equator than in the corresponding EOFs from the individual data sets. The second mode of the TAO and full resolution NSCAT (Figure 7b) represents a zonal equatorial flow to the east, or weakening of flow to the west. The TAO winds have significantly more variance attributed this mode than the NSCAT winds (34.8% vs. 21.6%). We attribute this difference to the sampling distribution, since the TAO buoys are relatively more concentrated near the equator than the evenly distributed NSCAT data. The NSCAT winds also show a deflection of the axis of this flow to the south in the central portion of the field and a reduction of the core flow towards the east. The amplitudes of this mode in the two data sets are well correlated and indicate a stronger presence of this mode after March 1997.

In order to quantify the effect of spatial sampling bias due to the equatorial concentration of TAO data we examined the results of CCA between the NSCAT data at TAO locations and the full NSCAT data. Since these sets originate from the same data, differences would be due to biases from the spatial subsampling of the NSCAT@ TAO locations. The CCA between NSCAT@TAO winds and the full NSCAT field (Table 2) finds a difference in the percent variance accounted for in mode 2 of 34.1% and 25.4% respectively. This is similar to the TAO/full NSCAT CCA, and is consistent with the conclusion that the buoy sampling locations enhance the percent variance accounted for by the TAO winds in this mode.

The CCA spatial pattern, which represents the optimal vector relationship between the data sets, can also be used to test for directional bias. Examination of the NSCAT vectors in these patterns shows them to be directed slightly to the right of the TAO vectors (Figures 6a-e). The differences in direction of the CCA eigenvectors (in degrees) for all the modes are shown as a function of latitude in Figure 9. The data are clustered at specific lines of latitude and show a good deal of scatter. The lower window in Figure 9 compares the means and 90% confidence intervals for each latitude cluster. The NSCAT vectors had a mean offset to the right of 14° (Std Dev=28° with a 90% CI of ± 4° over all data). This value is consistent with that found by other investigators [14° , Bourassa, et al, 1997, 9.8° , Caruso and Kelly, tech report, in preparation] for NSCAT wind vector comparison with direct surface measurements from ships and buoys. Only one cluster mean is different from 14° at the 90% confidence level. The lack of any significant trend or sign reversal in this offset across the equator eliminates the Ekman planetary boundary layer as a possible source. The magnitude of this directional error is smaller than the NSCAT mission requirement of ± 20° rms error for individual measurements. Nevertheless, the bias is likely to influence climatological averages and users should consider a rotational correction, particularly if the data are to be applied to computing divergence and curl.

The NSCAT wind field is highly representative of the winds over the Tropical Pacific when compared to the wind fields measured by the TAO buoy array. Although the NSCAT sampling is twice daily, there is no significant aliasing effect resulting from the diurnal and semi-diurnal wind variability seen in the Tropical Pacific. An EOF analysis of the different data sets resulted in excellent agreement of the principal modes of variability. The first two modes reflect equatorial jet and Trade Wind variability, respectively, together attributing for 60-70% of the total variance in the wind field. The reversal of the order of the two modes for the full resolution NSCAT wind field is a result of the coverage of the NSCAT data compared to the subset of TAO locations used. CCA analysis quantifies the high correlation between the data sets and the reasonableness of the TAO buoy sampling to defining the first two modes of the wind field. A 14° clockwise rotation of the NSCAT winds relative to the TAO winds was found irrespective of hemisphere.

Acknowledgments. NASA/JPL Contract 960594 to Earth and Space Research funded this research. Wenqing Tang, NASA/JPL, provided the 10-day averaged NSCAT data. Michael McPhaden, NOAA/PMEL, made the TAO buoy data available and provided helpful editing and suggestions

Barrnett, T. P., and R. Preisendorfer, Origins and levels of monthly and seasonal forecast skill for United States surface air temperatures determined by canonical correlation analysis. Mon. Wea Rev., 115, 1825-1850. 1987.

Bourassa, M.A., M.H. Freilich, D.M. Legler, W.T. Liu and J.J. O'Brien, Wind observations from new satellite and research vessels agree. EOS Trans. AGU, 78(51), p597 & 602, 1997.

Caruso, M. and K. Kelly, technical report, in preparation.

Liu, W.T., W. Tang, and P.S. Polito, NASA scatterometer provides global ocean-surface wind fields with more structures than numerical weather prediction. Geophys. Res. Letters, 25(6), 761-764, 1998a.