Abstract

VHF wind profiler measurements of zonal and meridional winds are compared with the NCEP–NCAR reanalysis at sites in the tropical Pacific. By December 1999 the profilers at Darwin, Australia, and Biak, Indonesia, in the western Pacific; Christmas Island, Kiribati, in the central Pacific; and Piura Peru, in the eastern Pacific had collected between 8 and 13 yr of nearly continuous data. While these profilers routinely observe winds up to about 20 km, only winds at Christmas Island are assimilated into the reanalysis. The long period of profiler operation provides an opportunity to study differences between the profiler and reanalysis winds in the equatorial Pacific, a region with geographically sparse observations. Mean and seasonal mean zonal and meridional winds are used to identify differences in the profiler and reanalysis winds. Two potential causes for the discrepancy between profiler and reanalysis winds are identified. The first of these is related to different spatial and temporal characteristics of the reanalysis and profiler data. The second cause is the geographical sparseness of rawinsonde data, and not assimilating wind profiler observations. The closest agreement between the mean and seasonal mean zonal winds was found at Christmas Island, a site at which profiler winds are assimilated. A good agreement between reanalysis and profiler meridional and zonal winds is also shown at Darwin, where nearby rawinsonde observations are available. The poorest agreement was found at Piura (where profiler winds are not assimilated), the closest rawinsonde is almost 2000 km from the profiler site, and topography is not adequately resolved in the reanalysis.

Introduction

In this paper we present a comparison of horizontal winds from the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis and those measured using VHF (50 MHz) wind profilers over the tropical Pacific. Reanalysis winds and observations were compared at four sites: Darwin, Australia, Biak, Indonesia, Christmas Island (republic of Kiribati), and Piura, Peru. Profilers at these sites have monitored winds since the early 1990s, and the long study period combined with a good time–height resolution of the data provides an opportunity to study differences between the two datasets.

A number of authors have studied the accuracy of wind profiler–measured horizontal winds. Strauch et al. (1987) compared independent profiler measurements of horizontal winds over 1 month and found a difference standard deviation of about 1.3 m s−1 over all heights. This error, which they attributed to errors in the radial velocity measurements, was reduced to 0.9 m s−1 when only measurements with the highest signal-to-noise ratios were used. Weber and Wuertz (1990) compared about 2 yr of wind profiler and rawinsonde horizontal winds near the Rocky Mountains in Colorado. They showed a standard deviation of about 2.5 m s−1 in the differences between profiler and rawinsonde wind measurements, which they attributed to meteorological variability. May (1993) compared horizontal winds measured by a VHF profiler with 120 rawinsonde ascents over the central-western Pacific. He found root-mean-square differences of 2.3 m s−1 for measurements at heights below 10 km. Jasperson (1982) studied wind variability by comparing rawindsondes launched simultaneously, and also separated in time and space. For radiosondes launched within 20 m of each other, but separated in time by 30 min, he found mean wind differences of about 1.6 m s−1. The difference values were larger when the separation between the radiosonde launch sites increased and when the difference in launch time increased. Considering the spatial and temporal differences between profiler and rawinsonde measurements (e.g., the rawindsonde takes about 20 min to reach 5 km, while profiler observations are almost instantaneous), these studies suggest that wind profilers provide good estimates of winds that are comparable with those measured by rawinsonde.

Wind profiler observations have been used to assess the quality of the reanalysis and operational wind analyses. Pauley et al. (1994) compared VHF profiler observations with operational regional analyses at a site free of significant orographic features (Illinois). They found correlations between the profiler winds and operational regional analyses of about 0.95 and difference standard deviations of about 2.3 m s−1, which are similar to the standard deviations found in other wind profiler quality studies. Their study suggested that where there is a good agreement between profiler and model winds, the relatively high temporal resolution of profiler winds compared to other measurement sources might offer an improvement to model analyses. Gage et al. 1988 examined the impact of incorporating Christmas Island wind profiler data in the reanalysis, and the European Centre for Medium-Range Weather Forecasts (ECMWF) analysis. Using 10 months of data (November 1986–August 1987), they found that the assimilation of profiler data reduced the bias between observations and model analyses from about 1–3 to 0.5 m s−1. This directly showed the improvement in model analyses that can result by assimilating profiler observations. Our study builds on the Gage et al. (1988) analysis by comparing the NCEP–NCAR reanalysis winds with observations at each of the wind profiler sites across the tropical Pacific, and over a longer study period.

We expect a greater difference between the reanalysis and profiler horizontal winds in observationally sparse areas and where profiler winds are not assimilated. However, even if profiler winds are assimilated, significant differences may exist if the profilers observe local atmospheric processes that are not resolved by the spatial or temporal resolution of the reanalysis. In addition, observations are rejected (not assimilated into the reanalysis) if they differ too much from the model (Kalnay et al. 1996; Boutier 2001). The assimilation system attempts to form an analysis that best fits the observations and, therefore, will not draw exactly to a particular observation, even if that observation is unaffected by local processes.

The datasets and the processing methods used in this study are described in the next section. In section 3 we estimate the accuracy of the wind profiler measurement using independent observations of horizontal winds from opposing wind profiler beams. This is followed by a comparison of the reanalysis and profiler meridional and zonal winds at each site. The paper concludes with a discussion of the potential causes for the differences in reanalysis and profiler winds.

Data and methods

Wind profiler observations

Horizontal wind profiles are typically measured using wind-profiling radar or rawinsondes. However, in the Tropics there are few observing sites, which are widely spaced. Figure 1 shows the locations of VHF (50 MHz) wind profilers operating in the tropical Pacific. Also shown are the locations of the nearest rawinsonde launch sites. These rawinsonde observations are typically available through the Global Telecommunications System (GTS) for assimilation into atmospheric models. Table 1 lists the name, latitude, longitude, and elevation of closest rawinsonde sites, as well as their distance from the wind profiler sites. With the exception of Darwin, the rawinsonde sites are generally over 1000 km (several degrees in latitude and longitude) from the profiler sites.

Fig. 1.

The Trans-Pacific Profiler Network. Antenna symbols mark the locations of wind profiling radars at Darwin, Biak, Christmas Island, and Piura. Balloon symbols mark the nearest rawinsonde launch sites. The crosses mark the closest four reanalysis grid points around the profiler site. A solid line marks the equator, and dashed lines mark the tropic of Capricorn and the tropic of Cancer. The profiler, rawinsonde, and reanalysis gridpoint names and detailed locations are given in Table 1 

Fig. 1.

The Trans-Pacific Profiler Network. Antenna symbols mark the locations of wind profiling radars at Darwin, Biak, Christmas Island, and Piura. Balloon symbols mark the nearest rawinsonde launch sites. The crosses mark the closest four reanalysis grid points around the profiler site. A solid line marks the equator, and dashed lines mark the tropic of Capricorn and the tropic of Cancer. The profiler, rawinsonde, and reanalysis gridpoint names and detailed locations are given in Table 1 

Table 1.

Latitude, longitude, and elevation of VHF wind profilers in the Trans-Pacific Profiler Network, and the closest regular rawinsonde launch sites and reanalysis grid points. Also listed are the latitudinal (Y ), longitudinal (X ), and straight line (R) distance in kilometers of the rawinsonde and reanalysis grid points from the profiler sites. The reanalysis grid points are the closest grid points to the northwest (0, 0), northeast (0, 1), southwest (1, 0) and southeast (1, 1) of the profiler sites. For the rawinsonde sites, the station identifier and station number is included. For the profiler sites, the beams, vertical (V), east (E), west (W), north (N), south (S), and the years they became operational are listed. A boldface “S” and “R” indicate the closest rawinsonde and reanalysis grid point to the profiler site

Latitude, longitude, and elevation of VHF wind profilers in the Trans-Pacific Profiler Network, and the closest regular rawinsonde launch sites and reanalysis grid points. Also listed are the latitudinal (Y ), longitudinal (X ), and straight line (R) distance in kilometers of the rawinsonde and reanalysis grid points from the profiler sites. The reanalysis grid points are the closest grid points to the northwest (0, 0), northeast (0, 1), southwest (1, 0) and southeast (1, 1) of the profiler sites. For the rawinsonde sites, the station identifier and station number is included. For the profiler sites, the beams, vertical (V), east (E), west (W), north (N), south (S), and the years they became operational are listed. A boldface “S” and “R” indicate the closest rawinsonde and reanalysis grid point to the profiler site
Latitude, longitude, and elevation of VHF wind profilers in the Trans-Pacific Profiler Network, and the closest regular rawinsonde launch sites and reanalysis grid points. Also listed are the latitudinal (Y ), longitudinal (X ), and straight line (R) distance in kilometers of the rawinsonde and reanalysis grid points from the profiler sites. The reanalysis grid points are the closest grid points to the northwest (0, 0), northeast (0, 1), southwest (1, 0) and southeast (1, 1) of the profiler sites. For the rawinsonde sites, the station identifier and station number is included. For the profiler sites, the beams, vertical (V), east (E), west (W), north (N), south (S), and the years they became operational are listed. A boldface “S” and “R” indicate the closest rawinsonde and reanalysis grid point to the profiler site

Profiler measurements have been made at Biak since 1992, at Darwin since 1990, at Christmas Island since 1986, and at Piura since 1991. These profilers were initially built with the capability of measuring radial velocity in three beam directions: a vertical beam and two oblique beams, separated by an azimuth of 90° and inclined 14.3° from the vertical. Using the geometry of these three beams and a suitable averaging period, horizontal and vertical winds can be derived. Later, some of these profilers were upgraded to produce radial velocity measurements along five beam directions (a vertical beam, and beams inclined toward the north, south, east, and west at 14.3° from vertical). Table 1 indicates the year that profiler operations started, and which beams were operational. Although five beam directions are required for momentum flux measurements, only three beam directions are required for a horizontal velocity estimate (e.g., Vincent and Reid 1983). Wind profiler–derived horizontal winds from these systems are routinely observed from a minimum height of about 2 km to a maximum height of about 20 km at a 300–500-m vertical resolution.

For this comparison, wind profiler radial velocities were averaged using multiple averaging periods (1, 3, 6, 12, 24, and 48 h) and a 3-h sampling interval, using a consensus-averaging technique (Strauch et al. 1984). This method calculates an average value by examining all of the measurements within a particular averaging period. The average is then calculated using the largest number of measurements that are within a specified velocity deviation window. Because radial velocities were expected to vary significantly over the averaging periods, we set the velocity deviation window to ±4 m s−1 (representing a window for horizontal winds of about ±16 m s−1), and only calculated an average if over 60% of the measurements were within this deviation window. If an average could not be calculated, a missing value was noted at that time and height.

In this study, we have assumed a zero mean vertical velocity. Vertical velocities were typically small in comparison with the mean radial velocities measure by the off-vertical beams. Gage et al. (1991) showed average vertical winds at Christmas Island with magnitudes less than several centimeters per second. The assumption of zero vertical velocity is also consistent with other VHF profiler studies (e.g., Pauley et al. 1994). The average radial velocities were combined to form zonal and meridional winds. We limited the maximum height to 12 km, because low signal-to-noise ratios above this level often limit the quality of the measurements.

NCEP–NCAR reanalysis model

Wind observations over the tropical Pacific are geographically sparse and are rarely made at multiple heights. These observations often have a low time resolution and may not be made on a routine basis. The NCEP–NCAR reanalysis takes an analysis/forecast model to perform data assimilation using data from 1957 through to the present (Kalnay et al. 1996). The assimilation system uses a three-dimensional variational scheme to optimally combine observations into a forecast model (Parrish and Derber 1992). The assimilation output includes zonal and meridional winds, temperature, geopotential height, as well as other parameters, on a global 2.5° latitude × 2.5° longitude grid (90°–90°N, and 0°–357.5°E). The data cover 17 pressure levels from 1000 to 10 hPa, corresponding to heights of about 0–25 km above sea level. Only six of these pressure levels are in the 2–12-km range of the profiler wind observations. A 6-h time resolution is used in the reanalysis with data at 0000, 0600, 1200, and 1800 UTC. Because of the geographical sparseness of observations, in particular over oceans and in the Tropics, the global coverage and long record of the reanalysis make it a crucial dataset for atmospheric research. This study uses the original NCEP–NCAR reanalysis (R1). Kanamitsu et al. (2002) describe an updated reanalysis (R2), which includes, among other changes, corrections for known observation errors and a smoothed surface topology. Wind corrections were applied to some extratropical observations, and the smoothed topology removed oscillations near steep topography. We believe these changes should not significantly affect the tropical wind data.

Observations from a variety of sources are assimilated into the reanalysis. Global rawinsonde observations are a primary source of data. Other observation sources include surface marine data (ships, buoys, etc.), surface land data (e.g., automatic weather stations), aircraft data, Special Sensing Microwave Imager (SSM/I) surface winds, and satellite temperature, radiances, and cloud drift winds. Currently the only VHF profiler observations in the tropical Pacific available on the GTS, and subsequent assimilation into the reanalysis, are those from Christmas Island. An optimal interpolation quality control screens observations that contain gross errors, or that may be accurate, but represent spatial and temporal scales that cannot be resolved in the analysis/forecast system (Kalnay et al. 1996).

Using geopotential height at the 17 levels of the reanalysis, the reanalysis zonal and meridional winds were interpolated to the wind profiler observation heights. The reanalysis data were also interpolated in latitude and longitude to match the locations of the wind profilers. Four reanalysis grid points forming the smallest box around the wind profiler location were used in the interpolation (Fig. 1). This method produced a reanalysis dataset of meridional and zonal winds with times, heights, and locations that correspond as closely as possible to the wind profiler datasets.

Comparison of independent profiler beam observations

To provide meaningful estimates of horizontal velocity, we must assume that the wind field across the wind profiler beams is homogeneous over some averaging period (e.g., Strauch et al. 1987). If we make a further assumption that the vertical velocity is zero over this averaging period, independent estimates of the zonal wind can be made using the east and west beams, while independent estimates of the meridional wind can be made from the north and south beams (e.g., Pauley et al. 1994). Differences in these estimates can be used to estimate the error of the wind profiler measurements, and are attributed to 1) random errors in the radial velocity measurements, 2) a nonhomogeneous wind field, or 3) significant vertical velocity.

For this error analysis, we used 3-h horizontal (zonal and meridional) profiler winds at the six averaging periods outlined in section 2a. Horizontal winds were used rather than radial velocities so that differences could be related to the reanalysis comparison in section 4, and to other studies. The error analysis was applied to the period of five-beam operation at Biak (1995–99), Christmas Island (1993–99), and Piura (1994–99). We could not assess the quality of the Darwin profiler winds because it has operated using only three beams over the study period.

Biak

Examination of radial velocities at Biak revealed that on around 3 December 1998, the relay switching between the north and south beams failed, leaving the state in a south-beam mode. North-beam observations after this time were in reality south-beam observations. The horizontal velocities calculated from the north beam over the period 1992–99 were corrected by reversing the sign of the radial velocities after 3 December 1998. However, as a result of the north and south beams being the same, these two beams are only compared for the period 1995 to 3 December 1998.

The mean difference between the profiler zonal horizontal winds calculated from the east (UE) and west (UW) beams are shown in Fig. 2a, while the mean difference for the meridional winds calculated from the north (VN) and south (VS) beams are shown in Fig. 2d. There are large differences in the zonal and meridional estimates below 4 km and for the zonal estimates above 10 km. Between 4 and 10 km, the zonal wind mean difference is positive and has a magnitude of less than 0.5 m s−1, while the meridional wind difference is negative with a magnitude less than 0.2 m s−1. The averaging period has little impact on the mean difference. The averaging period does, however, have a significant influence on the difference standard deviation, in particular, near the surface (Figs. 2b and 2e). As the averaging period is increased, the variability of the wind field across the profiler beams is reduced, and random radial velocity measurement errors contribute less to the mean. The mean difference then represents a systematic bias between the beams. The difference standard deviation from 4 to 10 km of the 3-h averages is less than about 2 and 1 m s−1 for the zonal and meridional winds, respectively. The leftward shift of the difference standard deviation profile as a result of increased averaging period is largest between the 1- and 3-h profiles. This indicates that there should be an improvement in the horizontal wind estimates when using 3- rather than 1-h averages. Figures 2c and 2f show the ratio of variances for the independent zonal and meridional wind estimates. The meridional wind variance ratio shows a large difference below 4 km with the north-beam variance larger than that of the south beam. However, above 4 km the 99% confidence intervals for the variance ratios include 1 (i.e., the variances are equal). The zonal winds show east-beam variances that are smaller than those of the west beam. Averaging appears to exacerbate the problem with the difference in variances increasing with increasing averaging period. The cause of these differences in particular in the zonal estimates is not understood, but may be caused by ground clutter and limited dynamic range (P. E. Johnston 2003, personal communication).

Fig. 2.

Statistics describing the difference between the east beam (UE) and west beam (UW), and between the north beam (VN) and south beam (VS) at Biak for 1- (black), 3- (purple), 6- (blue), 12- (green), 24- (yellow), and 48-h (red) averaging periods. (a) Mean difference (UEUW) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (b) Standard deviation of the differences S(UEUW) with 99% confidence intervals determined from a chi-square distribution. (c) Ratio of variances of UE and UW with 99% confidence intervals determined from the f distribution. (d) Mean difference (VNVS) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (e) Standard deviation of the differences S(VNVS) with 99% confidence intervals determined from a chi-square distribution. (f) Ratio of variances of VN and VS with 99% confidence intervals determined from the f distribution. Statistics are calculated for the period from 1995 to 1998

Fig. 2.

Statistics describing the difference between the east beam (UE) and west beam (UW), and between the north beam (VN) and south beam (VS) at Biak for 1- (black), 3- (purple), 6- (blue), 12- (green), 24- (yellow), and 48-h (red) averaging periods. (a) Mean difference (UEUW) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (b) Standard deviation of the differences S(UEUW) with 99% confidence intervals determined from a chi-square distribution. (c) Ratio of variances of UE and UW with 99% confidence intervals determined from the f distribution. (d) Mean difference (VNVS) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (e) Standard deviation of the differences S(VNVS) with 99% confidence intervals determined from a chi-square distribution. (f) Ratio of variances of VN and VS with 99% confidence intervals determined from the f distribution. Statistics are calculated for the period from 1995 to 1998

Christmas Island

The mean differences of the zonal (Fig. 3a) and meridional (Fig. 3d) wind estimates at Christmas Island are generally negative, with magnitudes of less than 0.3 m s−1. As in the case of Biak, the averaging period has little influence on the mean difference. The difference standard deviation below 4 km is significantly smaller than that for Biak (Figs. 3b and 3e). However, above 4 km, the difference standard deviation rapidly increases. Again, the largest leftward shift of the difference standard deviation profile occurs when using 3- rather than 1-h averages. The ratios of variances generally suggest equal variances for the zonal estimates up to 11 km (Fig. 3c) and the meridional estimates up to 8.5 km (Fig. 3f).

Fig. 3.

As in Fig. 2, but for Christmas Island and the period from 1993 to 1999

Fig. 3.

As in Fig. 2, but for Christmas Island and the period from 1993 to 1999

Piura

At Piura, the magnitudes of the mean difference in the zonal (Fig. 4a) and meridional (Fig. 4d) wind estimates are less than about 0.2 m s−1 up to a height of 10 km. While the zonal differences are consistently negative, the meridional bias changes sign from negative to positive above 6 km. The mean meridional winds at Piura (described in section 4) are northerlies (have negative velocities). This suggests that above 6 km, the mean northerlies that are derived from the south beam are stronger than those for the north beam. Similar to Biak and Christmas Island, the leftward shift of the difference standard deviation profiles is greatest between the 1- and 3-h averages. The 3-h difference standard deviation profiles show magnitudes of less than 1 m s−1 up to 7.5 km for the zonal winds (Fig. 4b) and 7 km for the meridional winds (Fig. 4e). Ratios of the variances for the zonal wind estimates suggest equal variances up to a height of 8 km, above which east-beam variances are larger than those of west beam variances (Fig. 4c). The meridional wind, however, shows a large difference in variance above 4 km that increases with height (Fig. 4f). Figure 5 shows a scatterplot of south- and north-beam horizontal velocities. This diagram shows that points that are not on the diagonal (i.e., north-beam estimates that are not equal to the south-beam values) tend to lie between the diagonal and the line of zero north-beam velocity and along the zero line. This distribution of differences suggests that the north-beam velocities are biased toward smaller velocities or zero and, therefore, indicate a problem with the north-beam velocity estimates. The bias is consistent with the mean difference values above 6 km. This bias in the north beam will not affect the comparison with the reanalysis winds because the south beam has operated for the longest period and is used for creating meridional wind estimates.

Fig. 4.

As in Fig. 2, but for Piura and the period from 1994 to 1999

Fig. 4.

As in Fig. 2, but for Piura and the period from 1994 to 1999

Fig. 5.

Scatterplot of the horizontal meridional wind estimates from the north beam VN and the south beam VS for a single altitude. Solid lines mark the diagonal where VN = VS, and the lines VN = 0 and VS = 0

Fig. 5.

Scatterplot of the horizontal meridional wind estimates from the north beam VN and the south beam VS for a single altitude. Solid lines mark the diagonal where VN = VS, and the lines VN = 0 and VS = 0

Comparison of wind profiler and reanalysis winds

The previous section focused on the accuracy of profiler measurements and highlighted some of the errors associated with assumptions on the uniformity of the wind field and hardware problems. In this section we compare the profiler wind estimates to independent estimates from the NCEP–NCAR reanalysis. We reduce the 3-h sampling in the profiler data (described in section 2a) by taking observations every 6 h to correspond with the reanalysis times. The reanalysis dataset has been interpolated over height, latitude and longitude, as described in section 2b. While the profiler winds are averaged using multiple averaging periods, the reanalysis winds are not. The multiple profiler averaging periods are compared with the original 6-hourly reanalysis. Comparison statistics, however, are created from both datasets using only times at which data points exist in both datasets. Missing values in the profiler datasets are also treated as missing values in the reanalysis dataset. We begin by describing mean difference statistics, followed by a discussion on seasonal differences. Each site is discussed, beginning on the western side of the Pacific at Darwin and continuing to the eastern side of the Pacific at Piura.

Darwin

Mean zonal and meridional winds for the period 1990–99 at Darwin are shown in Fig. 6a and 6f, respectively. The profiles derived from the profiler and reanalysis appear to correspond well. The zonal winds in both datasets are easterlies below and westerlies above 6 km. The meridional winds transition from southerlies below to northerlies above 8 km. The maximum differences in the zonal wind occur at 4 and above 11 km (Fig. 6b). The meridional wind differences are also larger at 4 and above 11 km, but also near 2 km (the lowest observation height). The standard deviation of the differences SUPUR is influenced by the averaging period, but by a smaller amount than was shown for Biak, Christmas Island, and Piura in the beam-to-beam comparison (section 3). Zonal SUPUR is about 2.5 m s−1 at 2 km, increasing to a local maximum at 6 km of 3.2 m s−1 (Fig. 6c). For the meridional winds, SVPVR is about 2.3 m s−1 near 2 km and increases with height (Fig. 6h). Standard deviations of the zonal and meridional winds are shown in Figs. 6d and 6i, respectively. Both zonal and meridional profiler winds show similar profiles, with local minimums near 2 and 6 km, and a local maximum at 4 km. There are, however, differences between the reanalysis and profiler standard deviations. These differences are more clearly shown in the ratio of variances for the zonal (Fig. 6e) and meridional (Fig. 6j) winds. Below 5 km, the profiler observations with longer averaging periods (reduced variance) have variances that are closer to the reanalysis. Above 5 km, shorter profiler-averaging periods give a variance that corresponds more closely to the reanalysis. The mean statistics suggest that the profiler and reanalysis winds correspond reasonably well. The difference in variances, however, suggests that some atmospheric processes may not be fully resolved in both datasets.

Fig. 6.

Statistics describing differences between the profiler and reanalysis zonal winds (UP, UR), and between the profiler and reanalysis meridional winds (VP, VR) at Darwin for 1- (black), 3- (purple), 6- (blue), 12- (green), 24- (yellow), and 48-h (red) averaging periods. (a) Mean zonal wind for the profiler (solid) and reanalysis (dashed) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (b) Mean difference (UPUR) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (c) Standard deviation of the zonal differences with 99% confidence intervals determined from a chi-squared distribution. (d) Standard deviation of the zonal profiler (solid) and reanalysis winds (dashed). (e) Ratio of variances of UP and UR with 99% confidence intervals determined from the f distribution. (f) Mean meridional wind for the profiler (solid) and reanalysis (dashed) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (g) Mean difference (VPVR) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (h) Standard deviation of the meridional differences with 99% confidence intervals determined from a chi-squared distribution. (i) Standard deviation of the meridional profiler (solid) and reanalysis winds (dashed). (j) Ratio of variances of VP and VR with 99% confidence intervals determined from the f distribution. Statistics are calculated for the period from 1990 to 1999

Fig. 6.

Statistics describing differences between the profiler and reanalysis zonal winds (UP, UR), and between the profiler and reanalysis meridional winds (VP, VR) at Darwin for 1- (black), 3- (purple), 6- (blue), 12- (green), 24- (yellow), and 48-h (red) averaging periods. (a) Mean zonal wind for the profiler (solid) and reanalysis (dashed) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (b) Mean difference (UPUR) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (c) Standard deviation of the zonal differences with 99% confidence intervals determined from a chi-squared distribution. (d) Standard deviation of the zonal profiler (solid) and reanalysis winds (dashed). (e) Ratio of variances of UP and UR with 99% confidence intervals determined from the f distribution. (f) Mean meridional wind for the profiler (solid) and reanalysis (dashed) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (g) Mean difference (VPVR) with 99% confidence intervals determined from the t distribution and the standard error of the mean of the differences. (h) Standard deviation of the meridional differences with 99% confidence intervals determined from a chi-squared distribution. (i) Standard deviation of the meridional profiler (solid) and reanalysis winds (dashed). (j) Ratio of variances of VP and VR with 99% confidence intervals determined from the f distribution. Statistics are calculated for the period from 1990 to 1999

Seasonal contour plots for the zonal and meridional winds at Darwin are shown in Fig. 7. These are time–height cross sections, where all of the 6-hourly data are averaged to form monthly means. The seasons at Darwin can be described in terms of monsoon break, transition, and monsoon periods. The zonal flow is strongly influenced by the movement of the monsoon trough and development of the northwest Australian heat low. During the Southern Hemisphere summer (December–February), when the monsoon trough lies over northern Australia, monsoon westerlies bring rain to northern Australia (e.g., McBride 1987). During the Northern Hemisphere summer, when the monsoon trough lies north of the equator, easterly trade winds dominate the region near Darwin. This cycle is clearly seen in the profiler winds with westerly flow beginning at low levels in December, and reaching a maximum depth and strength in late January (Fig. 7a). During the latter part of February, a transition occurs back to a monsoon break period and easterly trade wind flow. Maximums in the low-level easterly flow occur between March and May, and also in a deeper layer from September to November. These maximums in the easterlies correspond with maximums in southerlies for the meridional wind component (Fig. 7d). Maximums in the upper-level westerlies occur from April to July, and from September to November, with a period of weaker easterlies from July to September. These mean profiler winds agree well with a 36-yr study of rawinsonde winds at Darwin by Drosdowsky (1996). In his study, the mean westerlies during Late January are shown extending to about 350 hPa, a similar depth as that of the profiler observed westerlies. This is deeper than other climatological descriptions of the Australian summer monsoon, which only show westerlies extending to 500 hPa (e.g., McBride 1987).

Fig. 7.

Monthly time–height composites for Darwin describing (a) profiler mean zonal winds, (b) reanalysis mean zonal winds, (c) mean difference between the profiler and reanalysis zonal wind (UPUN), (d) profiler mean meridional winds, (e) reanalysis mean meridional winds, and (f) mean difference between the profiler and reanalysis meridional wind (VPVN). The zero contours in each composite are shown as thick black lines. The region in each composite for which the 99% confidence interval includes zero is enclosed by line filled contours. The contour interval is 2 m s−1 for the mean zonal winds and 0.5 m s−1 for the meridional winds and the wind differences

Fig. 7.

Monthly time–height composites for Darwin describing (a) profiler mean zonal winds, (b) reanalysis mean zonal winds, (c) mean difference between the profiler and reanalysis zonal wind (UPUN), (d) profiler mean meridional winds, (e) reanalysis mean meridional winds, and (f) mean difference between the profiler and reanalysis meridional wind (VPVN). The zero contours in each composite are shown as thick black lines. The region in each composite for which the 99% confidence interval includes zero is enclosed by line filled contours. The contour interval is 2 m s−1 for the mean zonal winds and 0.5 m s−1 for the meridional winds and the wind differences

The reanalysis seasonal zonal winds only show a shallow layer of monsoon westerlies, although the time of maximum strength does correspond to the maximum in the profiler westerlies (Fig. 7b). The trade wind flow appears to be well represented in the reanalysis, although the low-level easterlies appear weaker in the reanalysis. In contrast, maximums in the upper-level westerlies are larger in the reanalysis. The stronger low-level southerlies, which occur in tandem with the stronger low-level easterlies, are significantly weaker than the profiler southerlies. The magnitude of the differences between the profiler and reanalysis zonal winds, and meridional winds are shown in Figs. 7c and 7f, respectively. Maximum differences occur during the period of low-level southeasterlies from August to October near 4 km and also in the period of upper-level westerlies from April to June near 11 km. These large seasonal differences appear to be the cause of the relatively larger differences in the mean zonal and meridional winds at 4 and 11 km in Fig. 6. The reanalysis appears to capture the transition from break period to monsoon well, although the depth of the monsoon circulation is not well represented.

Biak

Mean zonal and meridional winds for the period 1992–99 at Biak are shown in Figs. 8a and 8f, respectively. While the profiler and reanalysis zonal wind profiles appear to show similarity, the meridional wind profiles do not. Both the reanalysis and profiler zonal winds transition from westerlies below to easterlies above 4 km. The largest differences (of about 1.2 m s−1) are above 5 km (Fig. 8b). The meridional profiler winds are southerlies below and weak northerlies above 4 km. The reanalysis winds, however, transition from southerlies below 5 km, to northerlies between 5 and 7.5 km, and back to southerlies above 7.5 km. The maximum differences are near 2 and 12 km, where the profiler and reanalysis winds have their largest magnitudes, respectively. The maximum meridional differences are similar in magnitude to the zonal differences, however, the meridional wind speed magnitudes are significantly smaller. The magnitudes of the zonal and meridional differences below 3 km, are smaller than those found in the independent beam comparison (section 3a). This suggests that errors in the horizontal wind speeds determined from the individual beams may be correlated, while the errors in the independent profiler and reanalysis datasets are not. Zonal SUPUR and meridional SVPVR are shown in Figs. 8c and 8h, respectively. The smallest values occur near 4 km, with SUPUR about 2.7 m s−1, and SVPVR about 2.3 m s−1. With the exception of observation heights below 3 km, profiles of SUPUR and SVPVR for Biak and Darwin are similar. The profiler zonal wind standard deviation (Fig. 8d) is similar to that at Darwin, with a local maximum near 4 km. The meridional standard deviation profile is more uniform over height between 3 and 10 km (Fig. 8i). Ratios of the zonal variances show the largest differences related to the averaging period below 3 km. The profiler 1-h averages have a larger variance than the reanalysis, while the longer averaging periods give variance ratios closer to or less than 1. Between 3 and 7 km, the profiler zonal variances are larger than the reanalysis variances, while above 9 km they are typically smaller. It is interesting to note that between 4 and 6 km, even 2-day averages of the profiler winds produce variances greater than those in the reanalysis. The meridional variance ratio profiles are more uniform over height (Fig. 8j). The largest difference in the variances are found below 3 km. Above 9 km, 6- and 12-h-averaged profiler winds generally produce variances closest to that of the reanalysis.

Fig. 8.

As in Fig. 6, but for Biak

Fig. 8.

As in Fig. 6, but for Biak

Seasonal contour plots for the zonal and meridional winds at Biak are shown in Fig. 9. East–west overturning circulations over the Indian and Pacific Oceans influence the zonal winds at Biak. The northward and southward movement of the monsoon trough and associated convection influences both the zonal and meridional winds at Biak. When the monsoon trough is north of the equator during the Northern Hemisphere summer, easterlies are observed at Biak, while during the Southern Hemisphere summer when the monsoon trough lies over northern Australia, westerlies are observed. Convection in the monsoon trough also drives the Hadley circulation. The Madden–Julian oscillation and Southern Oscillation have a significant influence on the winds at Biak through modulation of the Walker circulation, however, their influence is on timescales not addressed in the climatology, which are beyond the scope of this current study.

Fig. 9.

As in Fig. 7, but for Biak

Fig. 9.

As in Fig. 7, but for Biak

Westerlies are shown in the profiler zonal winds in the lower troposphere from November to May (Fig. 9a). The westerlies reach a maximum depth of 6 km in April and December. Above the westerlies, the easterly flow strengthens during December, shifting downward the transition zone of easterly and westerly flow. The upper easterlies weaken from March to May, but strengthen again during the Northern Hemisphere summer monsoon. The pattern of profiler zonal winds is consistent with that of the reanalysis, although, the lower-level westerlies in the reanalysis are less deep and have a slightly later onset and earlier transition to easterly flow (Fig. 9b). The largest difference in the zonal winds is during the period of upper-level easterlies from August to September (Fig. 9c). Above 4 km, the profiler meridional winds present evidence for a Hadley circulation (Fig. 9d). During the period when the monsoon trough is in the Southern Hemisphere (south of Biak), the profiler meridional winds show a southward flow from about 4 to 9 km, and northward flow above that. The circulation is reversed during the Northern Hemisphere summer monsoon. Below 4 km, the meridional winds remain southerly, with little variation. This may be a result of the problem identified with lower-level observations in section 3a. The reanalysis winds also suggest a Hadley circulation, with transitions occurring between the Southern and Northern Hemisphere summer monsoons. However, the circulation pattern appears more complex. The circulation shows the same meridional direction of flow in the upper (above 10 km) and the lower (below 5 km) levels, with meridional flow in the opposite direction between. The largest differences in the meridional winds occur below 4 km and may be associated with errors in the profiler observations. The largest difference below 4 km occurs from June to July, a period when the profiler winds are southerlies and reanalysis winds show northerlies. Larger differences also occur during the Southern Hemisphere summer, when profiler upper-level southerlies are deeper and stronger than those in the reanalysis. While the overall circulation patterns appear consistent in the profiler and reanalysis data, there are differences in the depth and onset time of transitions.

Christmas Island

Mean zonal and meridional winds for the period 1986–99 at Christmas Island are shown in Figs. 10a and 10f, respectively. The mean zonal profiles show a good agreement between the profiler and reanalysis, with easterlies below and westerlies above 9 km. The largest difference between the reanalysis and profiler zonal winds occurs above 10 km (Fig. 10b). The mean meridional winds show less agreement. The mean meridional reanalysis winds are smaller in magnitude than the profiler winds. Also, the mean meridional reanalysis winds are southerlies below 3 km, while the profiler winds are northerlies. The magnitude of the difference is about 0.5 m s−1 from 3 to 8 km, which is large compared to the magnitude of the mean meridional winds (Fig. 10g). The zonal and meridional difference standard deviations SUPUR (Fig. 10c) and SVPVR (Fig. 10h) are smallest near 2 km with values of about 2.5 m s−1 and 1.9 m s−1, respectively. Profiles of SUPUR below 7 km are similar to those at Darwin, while above 7 km, values at Christmas Island are typically larger. Profiles of SVPVR are similar to those at Darwin over most levels except below 3 km where the Christmas Island values are smaller. The zonal wind standard deviation, in contrast to Darwin and Biak, shows a minimum at 4 km (Fig. 10d). Above 4 km, the zonal standard deviation increases with height up to about 11 km. The meridional standard deviation is relatively constant from 2 to 3 km and then increases with height (Fig. 10i). The maximum difference in the zonal variances is below 3 and above 9 km (Fig. 10d). From 3 to about 8.5 km, 12- and 24-h-averaged zonal profiler winds give variances closest to that of the reanalysis, while above 9 km, 1-h-averaged profiler winds give the closest agreement. At most observation levels below 10 km, 24- and 48-h averages of meridional profiler winds give variances closest to those in the reanalysis (Fig. 10j). The closer agreement of variances from profiler winds with 12-h or longer averaging periods suggests that diurnal variations may be a significant cause of the difference. The reanalysis grid points near Christmas Island represent winds over the ocean, while the profiler winds are affected by friction and diurnal heating over an island.

Fig. 10.

As in Fig. 6, but for Christmas Island

Fig. 10.

As in Fig. 6, but for Christmas Island

Seasonal contour plots for the zonal and meridional winds at Christmas Island are shown in Fig. 11. One of the dominant processes believed to influence the seasonal zonal winds in this region is a contraction and expansion of the Walker circulation as a result of the movements of the heating zone, associated with the Madden–Julian oscillation and El Niño–La Niña (e.g., Gage et al. 1996). Both the profiler (Fig. 11a) and reanalysis (Fig. 11b) datasets show a transition from deep easterlies in June–October to a circulation pattern with upper-level westerlies and lower-level easterlies. During the Southern Hemisphere summer, convection over the Maritime Continent and Pacific warm pool drives the Walker circulation. This circulation is shown in the profiler and reanalysis winds as the easterly flow, with westerlies above from October to June. During the Northern Hemisphere summer, the intertropical convergence zone (ITCZ) in the western Pacific is displaced northward, and convection over the Maritime Continent is suppressed. This weakens the Walker circulation over the central-equatorial Pacific, with the upper-level westerlies giving way to easterlies. The onset time and depth of the upper-level westerlies corresponds well in both datasets. The westerlies begin in early October, reaching a maximum depth in late October–early December. However, above about 9 km and after November, the reanalysis westerlies are stronger than the profiler westerlies. From about 3 to 8 km the transition over time from westerlies to strong easterlies appears more rapid in the reanalysis. This is shown in the difference composite, with relatively larger positive values at 6 km in April, indicating stronger reanalysis easterlies.

Fig. 11.

As in Fig. 7, but for Christmas Island

Fig. 11.

As in Fig. 7, but for Christmas Island

The meridional wind composites for the profiler and zonal winds show some similarities in the circulation patterns, but the profiler circulation patterns are typically stronger. In both the reanalysis and profiler winds, there is a deep layer of southerly flow from about March to May. Both also show southerly flow near the lowest levels from about May to October, with two regions of stronger northerlies above. The two northerly cores, however, are much more distinct in the profiler data. Around November–February there are cores of southerly winds in the upper levels, although, they are higher and more intense in the profiler data. The magnitudes of these differences are shown in Fig. 11f. The greatest differences are associated with the two cores of northerly flow in the profiler data and the upper-level southerlies from October to March. Clearly the profiler observes a significant meridional circulation that is not well resolved in the reanalysis. It appears that the circulation may be related to the position of the ITCZ in the central-equatorial Pacific. During the Northern Hemisphere summer, when the ITCZ is north of Christmas Island, low-level (below 3 km) southerlies are a result of inflow to the ITCZ, while the two stronger cores of northerly flow are outflows from the region of enhanced convection in the ITCZ. Trenberth et al. (2000) investigated the global monsoon by applying complex empirical orthogonal function analysis to reanalysis, and to the ECMWF reanalysis. The principal EOF represented the familiar overturning associated with the Hadley and Walker circulations, with a flow that peaked near 850 hPa (about 1.5 km) in one direction, and near 150 hPa (about 13 km) in the opposite direction. The second EOF, however, represented a shallower overturning that peaked in one direction near 925 (about 1 km) and at 700 hPa (about 3 km) in the opposite direction. Both EOFs showed a minimum near 350 hPa (about 8 km). These two modes of overturning may explain the observed circulation in the profiler meridional winds. The two cores of stronger northerlies (from June to November) may represent the deeper and shallower meridional overturning modes. The minimum between the cores at 9 km may correspond to the minimums in the two modes.

Piura

Mean zonal and meridional winds for the period 1991–99 at Piura are shown in Figs. 12a and 12f, respectively. For both the profiler winds and reanalysis winds, the mean zonal flow is easterly, however, there are differences in magnitude. The reanalysis easterlies are stronger than the profiler easterlies below 6 km, but weaker above. The magnitude of the zonal differences is larger than that at the other comparison sites (Fig. 12b). The largest zonal wind differences occur below 4 km. Meridional winds appear to have a good agreement above 6 km, while the difference is significantly larger below 5 km (Fig. 12g). For both the meridional and zonal winds, the largest difference is at about 3 km. Profiles of SUPUR and SVPVR typically show larger values than those at the other comparison sites (Figs. 12c and 12h). The minimum value for SUPUR is about 3.3 m s−1 near 2 km, while the minimum value for SVPVR is about 2.6 m s−1 at 5 km. The standard deviation of the profiler zonal winds increases with height up to 4 km, then remains relatively constant from 4 to 6.5 km (Fig. 12d). The reanalysis zonal wind standard deviation is larger than the profiler standard deviation below 4 km, and is similar from about 4 to 6.5 km. Above 6.5 km, both the reanalysis zonal variance and the profiler zonal variance increase with height, but the reanalysis variance increases at a faster rate. The profiler meridional wind standard deviation decreases with height to about 3.5 km and increases with height above that. Below 3 km the reanalysis meridional wind variance is smaller than the profiler variance, but it increases more rapidly with height until it follows the variance profile of the 1-h-averaged profiler winds (Fig. 12i). Ratios of profiler to reanalysis wind variance are shown for the zonal variances and meridional variances in Figs. 12e and 12j, respectively. The profiles of zonal variance ratio are different to those found at the other comparison sites. This is the only site where the zonal reanalysis variances are closer to the zonal profiler variances with shorter observation-averaging periods. The profiles of meridional variance ratio, however, are more typical of the other sites with lower levels favoring longer profiler observation averaging periods and upper levels favoring shorter averaging periods.

Fig. 12.

As in Fig. 6, but for Piura

Fig. 12.

As in Fig. 6, but for Piura

Seasonal contour plots for the zonal and meridional winds at Piura are shown in Fig. 13. The profiler zonal winds are consistently easterly, although, the strength varies through the year (Fig. 13a). The easterlies are a result of the global momentum balance-favoring easterlies near the equator. The easterlies are weaker during the Southern Hemisphere summer, when heating over the tropical desert between the Andes and the Peruvian coast leads to convergence of the zonal flow. The easterlies are also consistently weaker below about 3.5 km. The Andes rise to about 3.9 km to the east of Piura, and likely impede the easterly flow. The meridional winds are influenced by the movement of an anticyclone over the southeastern Pacific, and the meridional movement of the intertropical convergence zone. Typically the ITCZ lies to the north of Piura, and the pressure gradient between the anticyclone in the southeastern Pacific and the ITCZ favor northward flow. The profiler meridional winds show predominantly northward flow below 4 km. However, during February–June, profiler low-level winds are northerlies. Lietzke et al. 2001, studied the structure and evolution of the double ITCZ in the eastern tropical Pacific. They found that the double ITCZ was a short-lived phenomena occurring during March and April, with convective branches symmetric about the equator at latitudes from about 5° to 7°. The low-level northerlies seen in the mean profiler winds from February to June may be evidence of inflow to the ITCZ south of Piura.

Fig. 13.

As in Fig. 7, but for Piura

Fig. 13.

As in Fig. 7, but for Piura

The reanalysis zonal winds show a considerably different pattern to the profiler zonal winds (Fig. 13b). The region of maximum easterlies is about 2 km lower and the period of strong easterlies is longer. Consistent with the profiler winds, however, there is a weakening in the zonal winds from October to January. The maximum easterlies in August are stronger than those in the profiler dataset. Figure 13c indicates that the differences in zonal winds have large magnitudes, in particular, below 4 km. The blocking of the easterly flow by the Andes, as is suggested by the weaker profiler easterlies in this layer, is not evident in the reanalysis winds. Figure 14 compares the surface topology in the reanalysis to the digital elevation model (DEM) of the U.S. Geological Survey (USGS) Earth Resources Observation Systems (EROS) data center. The reanalysis topology gives a maximum height of the Andes to the east of Piura of only 1.4 km, while the DEM indicates a maximum height of about 3.4 km. In addition, the reanalysis cannot resolve the 150-km stretch of low, relatively flat topography from the Peruvian coast to the foot of the Andes. Because the topography near Piura is poorly resolved, the reduction in the strength of the easterlies may be an atmospheric process that cannot be resolved in the current 2.5° resolution of the reanalysis. The meridional winds also show large differences between the profiler and reanalysis (Fig. 13f). In particular, the meridional winds below 4 km are in the opposite direction. The reanalysis meridional winds appear inconsistent with the forcing of the meridional flow by the locations of the southeastern Pacific anticyclone, and northeast equatorial ITCZ.

Fig. 14.

Comparison of topology in the NCEP–NCAR reanalysis and topographic data from the USGS EROS data center as a function of longitude near Piura. The solid line shows USGS EROS topographic data. The square symbols show the reanalysis topology on the model's approximately 1.9° longitudinally spaced Gaussian grid. The plus symbols show the reanalysis topology on the 2.5° uniform grid (output grid) interpolated to the exact latitude of the Piura profiler. Horizontal and vertical dotted lines mark the height and longitude of the profiler

Fig. 14.

Comparison of topology in the NCEP–NCAR reanalysis and topographic data from the USGS EROS data center as a function of longitude near Piura. The solid line shows USGS EROS topographic data. The square symbols show the reanalysis topology on the model's approximately 1.9° longitudinally spaced Gaussian grid. The plus symbols show the reanalysis topology on the 2.5° uniform grid (output grid) interpolated to the exact latitude of the Piura profiler. Horizontal and vertical dotted lines mark the height and longitude of the profiler

Conclusions

Measurements in the tropical Pacific are geographically sparse, and some wind profiler measurements are not currently assimilated into the NCEP–NCAR reanalysis. As shown in Fig. 1 and Table 1, this problem is clearest at Biak and Piura, where the only assimilated wind profile measurements are from rawinsondes several degrees in longitude and latitude (over 1000 km) from the profiler sites. In contrast, horizontal winds from the Christmas Island profiler are assimilated into the reanalysis and rawinsonde observations are available at Darwin. At Christmas Island, 1-hourly averages of profiler winds are available on the global telecommunications system.

The profiler and reanalysis zonal winds show the closest agreement at Christmas Island. The dominant zonal circulation in this region is the Walker circulation and variations in its strength over time are well represented in the reanalysis. In contrast, the meridional winds at Christmas Island show a poor agreement. The profiler meridional winds indicate a much stronger seasonal variation. While the profiler meridional winds are available for assimilation into the reanalysis, they may be rejected if there is a large error of representativeness. We believe that the meridional circulation might not be resolvable by the 2.5° resolution of the reanalysis. A better agreement is shown between the profiler and reanalysis meridional winds at Darwin where the reanalysis resolves the southeasterly trade wind regime well.

Profiler observations with multiple averaging periods were compared with the reanalysis. In general, at each of the sites, variances calculated from observations with longer averaging periods corresponded more closely to reanalysis wind variances at lower heights, while the reverse was true at upper levels. This could suggest both temporal, and spatial errors of representativeness, where local processes, such as sea breezes, and diurnal and semidiurnal tides are not represented in reanalysis. Diurnal variations in the zonal and meridional winds were not addressed in this current study.

The poorest agreement between reanalysis and profiler winds was shown to occur at Piura. This poor agreement is expected, because the only rawinsonde observations are over 1300 km from Piura. In addition, the closest rawinsonde site is on the eastern side of the Andes. The blocking of the easterly flow by the Andes evidenced in the profiler zonal winds is not shown in the reanalysis winds. In addition to the lack of assimilated observations, this may be a result of a poor representation of the topography near Piura. While the Andes to the east of Piura rise to about 3.4 km, the reanalysis topology has a maximum height of 1.4 km to the east of Piura. Also, the resolution of the model does not resolve the 150 km of relatively low and flat terrain between the Peruvian coast and the base of the Andes.

Differences outlined in this study suggest that some significant atmospheric processes across the equatorial Pacific may be better represented in the reanalysis, if a finer spatial resolution is used, and if profiler winds where available are assimilated.

Acknowledgments

Support for this work was partially provided by NSF Grant ATM 0116178 and the Cooperative Institute for Research in Environmental Sciences (CIRES). NCEP reanalysis data were provided by the NOAA–CIRES Climate Diagnostics Center in Boulder, Colorado. Unprocessed profiler data were provided by the NOAA Aeronomy Laboratory. Topography data near Piura were provided by the USGS EROS data center. The authors thank David Carter, Paul Johnston, and Tony Riddle for their contributions to the archival and quality control of the profiler data, helpful discussions on profiler data, and maintenance of the profiler systems. The Christmas Island profiler is supported by the NOAA Office of Global Programs. The Darwin profiler is operated by the Australian Bureau of Meteorology Research Centre. Profilers at Biak and Piura are partially supported by ATM0116178 and are operated in cooperation with the Indonesian National Institute of Aeronautics and Space (LAPAN) and the University of Piura, respectively.

REFERENCES

REFERENCES
Boutier
,
F.
2001
.
The use of profiler data at ECMWF.
Z. Meteor.
10
:
497
510
.
Drosdowsky
,
W.
1996
.
Variability of the Australian summer monsoon at Darwin: 1957–1992.
J. Climate
9
:
85
96
.
Gage
,
K. S.
,
J. R.
McAfee
,
W. G.
Collins
,
D.
Söderman
,
H.
Böttger
,
A.
Radford
, and
B.
Balsley
.
1988
.
A comparison of winds observed at Christmas Island using a wind-profiling Doppler radar with NMC and ECMWF analyses.
Bull. Amer. Meteor. Soc.
69
:
1041
1046
.
Gage
,
K. S.
,
J. R.
McAfee
,
D. A.
Carter
,
W. L.
Ecklund
,
A. C.
Riddle
,
G. C.
Reid
, and
B. B.
Balsley
.
1991
.
Long-term mean vertical motion over the tropical Pacific: Wind-profiling Doppler radar measurements.
Science
254
:
1771
1773
.
Gage
,
K. S.
,
J. R.
McAfee
, and
C. R.
Williams
.
1996
.
On the annual variation of tropospheric zonal winds observed above Christmas Island in the central equatorial Pacific.
J. Geophys. Res.
101
:
15061
15070
.
Jasperson
,
W. H.
1982
.
Mesoscale time and space wind variability.
J. Appl. Meteor.
21
:
831
839
.
Kalnay
,
E.
Coauthors,
.
1996
.
The NCEP/NCAR 40-Year Reanalysis Project.
Bull. Amer. Meteor. Soc.
77
:
437
471
.
Kanamitsu
,
M.
,
W.
Ebisuzaki
,
J.
Woollen
,
S.
Yang
,
J. J.
Hnilo
,
M.
Fiorino
, and
G. L.
Potter
.
2002
.
NCEP–DOE AMIP-II Reanalysis (R-2).
Bull. Amer. Meteor. Soc.
83
:
1631
1643
.
Lietzke
,
C. E.
,
C.
Deser
, and
T. H.
Vonder Haar
.
2001
.
Evolutionary structure of the eastern Pacific double ITCZ based on satellite moisture retrievals.
J. Climate
14
:
743
751
.
May
,
P.
1993
.
Comparison of wind-profiler and radiosonde measurements in the Tropics.
J. Atmos. Oceanic Technol.
10
:
122
127
.
McBride
,
J. L.
1987
.
The Australian summer monsoon.
Monsoon Meteorology, C. P. Chang and T. N. Krishnamurti, Eds., Clarendon Press, 60–92
.
Parrish
,
D. F.
and
J. C.
Derber
.
1992
.
The National Meteorological Center's spectral statistical–interpolation analysis system.
Mon. Wea. Rev.
120
:
1747
1763
.
Pauley
,
P. M.
,
R. L.
Creasey
,
W. L.
Clark
, and
G. D.
Nastrom
.
1994
.
Comparisons of horizontal winds measured by opposing beams with the Flatland ST radar and between Flatland measurements and NMC analyses.
J. Atmos. Oceanic Technol.
11
:
256
274
.
Strauch
,
R. G.
,
D. A.
Merritt
,
K. P.
Moran
,
K. B.
Earnshaw
, and
J. D.
van de Kamp
.
1984
.
The Colorado wind-profiling network.
J. Atmos. Oceanic Technol.
1
:
37
49
.
Strauch
,
R. G.
,
B. L.
Weber
,
A. S.
Frisch
,
C. G.
Little
,
D. A.
Merritt
,
K. P.
Moran
, and
D. C.
Welsh
.
1987
.
The precision and relative accuracy of profiler wind measurements.
J. Atmos. Oceanic Technol.
4
:
563
571
.
Trenberth
,
K. E.
,
D. P.
Stepaniak
, and
J. M.
Caron
.
2000
.
The global monsoon as seen through the divergent atmospheric circulation.
J. Climate
13
:
3969
3993
.
Vincent
,
R. A.
and
I. M.
Reid
.
1983
.
HF Doppler measurements of mesospheric gravity wave momentum fluxes.
J. Atmos. Sci.
40
:
1321
1333
.
Weber
,
B. L.
and
D. B.
Wuertz
.
1990
.
Comparison of rawinsonde and wind profiler radar measurements.
J. Atmos. Oceanic Technol.
7
:
157
174
.

Footnotes

Corresponding author address: Robert Schafer, University of Colorado, CIRES Building, Room 318, Boulder, CO 80309-0216. robert.schafer@colorado.edu