Typical methods used to acquire wind profiles from Doppler radar measurements rely on plan position indicator (PPI) scans being performed at multiple elevation angles to utilize the velocity–azimuth display technique or to construct dual-Doppler synthesis. These techniques, as well as those employed by wind profilers, often produce wind profiles that lack the spatial or temporal resolution to resolve finescale features. If two radars perform range–height indicator (RHI) scans (constant azimuth, multiple elevations) along azimuths separated by approximately 90°, then the intersection of the coordinated RHI planes represents a vertical set of points where dual-Doppler wind syntheses are possible and wind speed and direction profiles can be retrieved. This method also allows for the generation of high-resolution wind time histories that can be compared to anemometer time histories. This study focuses on the use of the coordinated RHI scanning strategy by two high-resolution mobile Doppler radars in close proximity to a 200-m instrumented tower. In one of the first high-resolution, long-duration comparisons of dual-Doppler wind synthesis with in situ anemometry, the mean and turbulence states of the wind measured by each platform were compared in varying atmospheric conditions. Examination of mean wind speed and direction profiles in both clear-air (nonprecipitating) and precipitating environments revealed excellent agreement above approximately 50 m. Below this level, dual-Doppler wind speeds were still good but slightly overestimated as compared to the anemometer-measured wind speeds in heavy precipitation. Bulk turbulence parameters were also slightly underestimated by the dual-Doppler syntheses.
Instrumented towers are typically used to collect wind profile data, but their use is restricted by height and location. These limitations have spurred the development and use of remote sensing instruments in evaluating the vertical structure of wind. Radar wind profilers can typically acquire measurements through a greater depth than an instrumented tower and provide wind profiles at adequate revisit times to observe mean boundary layer characteristics (Martner et al. 1993). While the effectiveness and accuracy of Doppler wind profilers has frequently been demonstrated, the spatial resolutions are often coarse and the wind profile below 50 m is rarely sampled (Martner et al. 1993; Angevine et al. 1998). Both high temporal resolution and near-surface measurements are important for atmospheric phenomena, where knowledge of the wind profile is critical to understanding the driving meteorology and wind-structure interaction. For example, wind profiles and turbulence characteristics of thunderstorms outflow winds have not been incorporated into design codes partly due to a lack of high-resolution observations even though these phenomena are responsible for the majority of wind-induced damage outside of regions affected by tropical cyclones (Letchford et al. 2002; Holmes et al. 2008).
Mobile Doppler radars possess a unique advantage in collecting high-resolution data within rare, engineering-design-relevant events. Methods have been developed to obtain a wind profile from both single-Doppler (radial velocities only; Browning and Wexler 1968) and dual-Doppler (full horizontal wind vector) synthesized grids (e.g., Hirth et al. 2015). These methods, which require assumptions of homogeneity in the wind field (in the case of single-Doppler wind retrievals) and are all based on collecting horizontal plan position indicator (PPI; constant elevation, multiple azimuths) scans at multiple elevation angles, also lack the temporal and spatial resolution necessary to resolve finescale features in the rapidly evolving wind profile. In addition to PPIs, most radars are able to scan multiple elevation angles along a single azimuth, referred to as a range–height indicator (RHI) scan. If two radars perform RHIs along azimuths separated by approximately 90°, then the intersection of the coordinated RHI planes represents a vertical set of points where dual-Doppler wind syntheses are possible and horizontal wind speed and wind direction profiles can be retrieved. A similar method was previously used with two Doppler lidars (Calhoun et al. 2006). A limited, yet successful, comparison of the dual-Doppler lidar profiles with other remotely sensed data was presented, but near-surface data were not available. Using a similar technical approach, the objective of this study is to validate profiles of dual-Doppler horizontal wind speed, direction, and turbulence parameters estimated with coordinated RHIs through comparisons with wind profile data from anemometers on a 200-m instrumented tower.
While dual-Doppler techniques have been employed for years and the errors associated with dual-Doppler measurements are well understood to be a function of the angle between the beams of each radar (Miller and Strauch 1974; Doviak et al. 1976; Davies-Jones 1979), few comparisons of radar-derived wind fields to ground-based in situ point measurements have been performed historically. Dual-Doppler wind speeds collected in clear air (or a nonprecipitating environment) were reported to be within 0.5 m s−1 of in situ tower measurements despite a coarse 150–200-m grid resolution and suspected ground clutter influence (Kropfli and Hildebrand 1980; Schneider and Lilly 1999). The inherent volumetric averaging of the radar measurements was also suspected to contribute to reduced magnitudes of the dual-Doppler wind components as compared to the point measurements of the anemometer. Using data from two fixed radar platforms and a 444-m-tall tower, Dowell and Bluestein (1997) combined one volume of dual-Doppler wind fields and a time-to-space conversion of 10 min of tower data into a supercell thunderstorm. The horizontal wind fields compared well in terms of identifiable features, but the radar wind speeds at 500 m were 6 m s−1 lower than tower wind speeds at 444 m. Dual-Doppler data below 500 m were not available. While Doppler velocity errors and differences in measurement height potentially contributed to the observed differences, Dowell and Bluestein (1997) also noted that smoothing inherent in both the collection and analysis of the radar data likely impacted the scales resolved in the dual-Doppler analysis.
The advent of high-resolution mobile Doppler radars has facilitated the collection of dual-Doppler data over in situ anemometry (as well as very near the ground), yet few comparisons between the two platforms exist. Single-Doppler radar data have been collected near the ground in tornadoes to measure the near-surface wind field and relate radar-derived wind speeds to observed wind damage (Wurman et al. 2007; Kosiba and Wurman 2013; Wurman et al. 2013). Given the complexities of such deployments, coordinated comparisons between single-Doppler wind speeds and near-surface anemometers were limited and often complicated by terrain differences. Hurricane environments have also provided an opportunity to compare radar wind measurements to in situ data (Lorsolo et al. 2008; Kosiba et al. 2013). These environments can support longer durations of data collection and better placement of assets to facilitate more thorough comparisons of data from different types of instrumentation. In a one-to-one comparison of data from Hurricane Isabel (2003), single-Doppler radial velocities collected at 30-m wind speeds compared very well with 10-m wind speeds. The former were slightly overestimated but that may have been due to differences in measurement height (Lorsolo et al. 2008). Similarly, Kosiba et al. (2013) compared 10-m wind speeds with spatially averaged single-Doppler radar data collected ~150 m AGL over two different roughness regimes during the landfall of Hurricane Rita (2005). Greater differences between the 1-min averaged single-Doppler velocities and the 1-min averaged 10-m wind speeds were evident over the rougher terrain. A correlation coefficient of approximately 0.5 was computed between the single-Doppler and anemometer fluctuations of the 1-min mean wind speed about the 5-min mean. After correcting for differences in measurement height and roughness, the 10-min averaged wind speeds and turbulence intensities (computed from 10-min independent segments of the single-Doppler and anemometer time histories) showed good agreement with the single-Doppler data.
2. Experimental methodology
Radar data were collected using the Texas Tech University Ka-band (TTUKa) mobile Doppler radars (Weiss et al. 2009; Hirth et al. 2015). The TTUKa radars employ pulse compression to combine the sensitivity achieved using a relatively long pulse without sacrificing range resolution, which was 15 m for this study. The increased sensitivity afforded by pulse compression also allows for data collection in many clear-air environments. For the data presented herein, a 1.22-m antenna results in a half-power beamwidth of 0.49°. RHI and PPI measurements are collected with scanning rates of 6° s−1 and 30° s−1 and are oversampled with 0.1° and 0.352° angular resolution, respectively. The accuracy of the velocity measurement is 0.03 m s−1 (Hirth et al. 2015). One consequence of transmitting within the Ka band is the tendency of the transmitted signal to attenuate in regions of significant precipitation. This effect was seen during data collection for the present study and occasionally resulted in brief data voids.
Data collection occurred at Texas Tech University’s National Wind Institute (NWI) field site located at Reese Technology Center (RTC) in Lubbock, Texas. This study will focus on data collected by anemometers mounted on a triangular, open-lattice tower 200 m in height. The tower is surrounded by flat terrain characterized by roughness lengths between 0.009 and 0.01 m and populated by few obstructions. The tower is instrumented at 10 different heights (0.9, 2.4, 4, 10, 17, 47, 75, 116, 158, and 200 m) using boom arms that extend from the center of the tower at an orientation of 300° north relative. All 10 levels originally employed R. M. Young 81000 ultrasonic anemometers to measure the u, υ, and w wind components with an accuracy of ±1% RMS, but they were upgraded with Gill R3-50 ultrasonic anemometers in January 2012. The upgraded ultrasonic anemometers also measure all three wind components but with an accuracy of <1% RMS. The data acquisition sampling rate was also increased from 20 to 50 Hz during this instrument upgrade. Gill 27005 UVW anemometers with carbon fiber thermoplastic (CFT) propellers are also mounted at 8 of the 10 levels (4, 10, 17, 47, 75, 116, 158, and 200 m). The UVW anemometers are characterized by a distance constant of 2.1 m, and the data are cosine corrected according to the recommendations of the manufacturer. The sonic and UVW anemometers are mounted on each boom arm at 4.26 and 3.65 m from the tower, which is equivalent to 3.5 and 3 tower diameters, respectively. The TTU 200-m tower undergoes routine maintenance and provides robust data for comparison to other platforms for wind directions where tower shading is not evident.
b. Radar data collection
Four validation datasets collected on 5 Dec 2011, 4 June 2012, 15 June 2012, and 28 June 2013 are discussed. For each dataset, the TTUKa radars were deployed to the NWI field site and positioned such that RHI planes intersected near the 200-m tower. Despite attempts to maintain consistency in the radar deployment locations, the distance between the 200-m tower and the RHI intersection location varied between datasets. However, the separation was never greater than 400 m (Fig. 1). The data collection strategies employed by the TTUKa radars also varied between the events, but they fall into one of three types:
Shallow RHIs—This scanning strategy consisted of coordinated RHIs being performed from 0° to 6° in elevation. This elevation sequence was chosen to maximize the temporal resolution of the radar data while ensuring dual-Doppler observations through the depth of the 200-m tower. PPIs were not included in this scanning strategy. Upon quality control and temporal alignment of the data from each radar, this strategy resulted in dual-Doppler profile intervals of approximately 2.5 s.
Repetitive RHIs—Coordinated RHIs were performed from 0° to 45° in elevation to provide dual-Doppler profiles well above the top of the tower. With this scanning strategy, dual-Doppler profiles were separated by approximately 9 s.
RHI–PPI combo—RHI elevation series were performed from 0° to 30° in elevation and PPIs were interwoven at regular intervals, resulting in decreased temporal resolution of the dual-Doppler RHI wind profiles. In addition to providing wind profiles, this scanning sequence was used to assess the horizontal extent and structure of thunderstorm outflow gust fronts for other objectives. The RHIs resulted in dual-Doppler profiles that were separated by 9 s, but intervals of 20 RHIs were separated by 56 s, during which time two PPIs were performed.
c. Data processing and quality control
TTUKa radial velocity fields were initially inspected for range and velocity folding, noise, and attenuation. Further quality control procedures included heading verification via alignment of clutter with known ground targets and correcting signal processing “jitter.” The jitter is a phenomenon caused by hysteresis in the vertical offset of successive RHIs by approximately 0.5°. The correction of the radar jitter resulted in some data loss below the 0.5° elevation for alternating RHI scans. The latitude, longitude, and north-relative RHI angles of both radars were used to geometrically determine the RHI intersection point and the distance between the RHI intersection point and each radar. Only data corresponding to the range bin closest to the intersection point (defined at 0° elevation) were used. To account for sight differences in the RHI elevation angle sequence between the radars, a grid with a vertical resolution of 20 m was constructed, and the mean of the radial velocity and elevation angle within each 20-m increment was assigned to the grid point. Approximately four radial velocity samples contributed to the mean value at each grid point. The height of each grid point was assigned as the mean height of the contributing observations (Table 1). Though necessary, this gridding scheme resulted in a slight height difference between the assigned gridpoint heights and the 200-m tower measurement heights. To compute the horizontal dual-Doppler wind speed at each grid point, the radial velocities of each radar were related to the Cartesian wind components through
where Vr is the mean radial velocity assigned to the grid point, is the mean elevation angle of the radial velocities contributing to the grid point, is the north-relative azimuth angle of the RHI. and Vt is the hydrometer fall speed (Ray et al. 1978; Markowski and Richardson 2010). In precipitating environments, the fall speed is typically estimated through a power-law relationship with reflectivity (Miller and Strauch 1974; Doviak and Zrnić 1993), but it was assumed to have minimal influence given the low (<6°) elevation angles used in the analysis and therefore neglected. The impact of this assumption will be investigated in a later section. The vertical velocity (w) contribution was also ignored. Not satisfying mass continuity will introduce error in the dual-Doppler retrieved winds, but these errors are expected to be small given the shallow domain. With these assumptions, Eq. (1) and an identical equation incorporating the information from the second radar were used to solve a system of equations for the u and υ wind components at each grid point for each pair of intersecting RHIs. Though variable, the crossing angles of the RHI azimuths for each deployment (Fig. 1) were within the bounds of a typical dual-Doppler lobe (Davies-Jones 1979).
For comparisons with TTUKa radar data, 8 of the 10 tower levels were used when data were available. The lowest two levels—0.9 and 2.4 m—were omitted from the analysis, as they are not instrumented with UVW anemometers (Table 1). Wind speed and direction data from the UVW and sonic anemometers of the eight identified tower levels were examined for erroneous values (spikes, dropouts, etc.), which were subsequently removed. The effects of direct tower shading were also considered, such that data between 115° and 160° were removed. This quality control restraint resulted in a loss of only 6% of the raw data. Also considered was the possibility of flow stagnation for wind directions aligning with the boom arm orientation (300°). Wind tunnel studies with a similar tower configuration suggest less than ~5% reduction in wind speed (for wind directions within 15° of the boom arm orientation) due to stagnation for anemometers mounted within 1 tower diameter of the tower (Gill et al. 1967). While both sonic and UVW anemometers are mounted well outside this range (3.5 and 3 tower diameters, respectively), numerical simulations indicate the potential for slight (less than 1.5%) wind speed reduction within 4 tower diameters of the upwind side of the tower (IEC 2005). However, this reduction was based on minimum tower porosity and may not be representative of the NWI 200-m tower and was assumed negligible. For one-to-one comparisons of the 0.4-Hz dual-Doppler wind speed and direction time histories of the shallow RHI datasets with the 50-Hz anemometer time histories, the anemometer time histories were reduced by identifying and using the instantaneous tower measurement closest to the time of the dual-Doppler measurement. For turbulence comparisons, the raw 50-Hz tower data were employed.
3. Wind profile comparisons
a. 5 December 2011
The TTUKa radars were deployed to the NWI field site as moderate snow fell across the region on 5 December 2011. Approximately 30 min of coordinated RHIs were acquired using the repetitive RHI scanning sequence. Radial velocity data from TTUKa1 (Fig. 2a) and TTUKa2 (Fig. 2b) suggest substantial directional shear in the lower atmosphere with the 0 m s−1 isodop located at approximately 1.5 km in height from the perspective of TTUKa1, while the dual-Doppler wind profiles computed at the RHI intersection point indicate veering wind direction with height, as well as wind speeds between 8 and 12 m s−1 through the depth of the profile. Dual-Doppler wind profiles collected over the entire 30-min period were averaged and compared to available sonic and UVW wind profiles from the 200-m tower averaged over the same time period (the 74-m tower level was not operating for this event). The tower was located 264 m north-northwest of the RHI intersection location. Despite the separation, the dual-Doppler profiles showed good qualitative agreement with both the sonic and UVW profiles (Fig. 3). Mean dual-Doppler wind speeds were slightly greater than the mean sonic wind speeds through the depth of the profile with the greatest difference (7.3%) observed at the 10-m level (Fig. 3a). Above this level, mean wind speed differences were less than 4.5% of the dual-Doppler wind speeds. Dual-Doppler wind speeds at 10 m were also 3.5% greater than the UVW mean wind speed. The difference changed with height, such that the 200-m UVW mean wind speeds exceed the 190-m dual-Doppler mean wind speeds by 3.5%. The mean dual-Doppler wind direction also demonstrated slight deviations from that of the tower with the 190-m mean dual-Doppler wind direction backed 7° as compared to the mean 200-m sonic wind direction. However, the general profile shape was well replicated by the dual-Doppler wind direction profile (Fig. 3b). Profiles of root-mean-square error (RMSE) between the dual-Doppler and sonic wind speeds were also computed and included in Fig. 3. For wind speed, RMSE values were greatest at the 10-m level with a value of 1.68 m s−1 and decreased to 1.09 m s−1 at the 200-m comparison level. The wind direction comparison between the sonic and dual-Doppler wind speeds demonstrated RMSE values between 6.8 and 8.5°.
b. 4 June 2012
A thunderstorm outflow dataset was obtained on 4 June 2012 as slow-moving, multicellular thunderstorms moved across the NWI field site and produced a large swath of outflow winds and stratiform precipitation, yielding over 3 h of coordinated radar data collection. For this deployment, the dual-Doppler intersection point was located 391 m southwest of the tower. The shallow RHI scanning sequence was employed, which resulted in a dual-Doppler profile every 2.5 s (after quality control and temporal alignment). These data were then compared with the available 200-m tower data during this time (the 17- and 200-m levels were not operating during this event). Wind speeds were generally between 10 and 25 m s−1 through the depth of the tower, as measured by the UVW wind sensors. The wind direction initially veered from west to northwest, and then backed from northwest to southwest near the end of data collection. Time histories of dual-Doppler and 200-m tower UVW wind speed (Figs. 4a and 4b) and direction (Figs. 4c and 4d) profiles from the first hour of data collection illustrate the evolution of the wind profile during this time period. Because of limited data availability, dual-Doppler data below 30 m were not included in the time histories. From these time–height sequences, it is evident that many of the transient features recorded in the tower data were also resolved by the dual-Doppler synthesis. For instance, a wind speed lull and sharp veering of the wind direction to approximately 350° just after 2358 UTC were similarly recorded by both platforms (Fig. 4). There was, however, a slight offset in the measurement of this feature, such that the lull was experienced at the tower between 40 and 50 s before it was measured at the RHI intersection location. This difference was attributed to the larger separation distance between the tower and the RHI intersection location. The dual-Doppler wind speeds were slightly overestimated compared to the sonic and UVW wind speeds. This bias was evident in the mean wind speed profiles for the entire event, in which the mean dual-Doppler wind speeds were consistently greater than both the UVW and sonic mean wind speeds through the depth of the profile (Fig. 5a). The magnitude of the difference between the two platforms was larger below 50 m, exceeding 16% of the mean dual-Doppler wind speed at 10 m. Above 50 m, wind speed differences between the two platforms were less than 5%. The RMSE between the sonic and dual-Doppler wind speeds was greatest at the 10-m level with a value of 2.8 m s−1. Above 50 m, RMSE values were generally less than 1.3 m s−1. While the mean dual-Doppler wind direction profiles demonstrated small deviations from the tower profiles, RMSE values exceeded 7° at the 10-m level and were generally above 3° through the 150-m level (Fig. 5b).
To investigate the impact of assuming negligible fall speed, calibrated reflectivity (Z) from TTUKa1 was used in a power-law relationship with constants recommended by Miller and Strauch (1974) and Doviak and Zrnić (1993):
where A = 2.6, B = 0.1, and C = 0.4. To correct for air density, γ0 was taken as the density at sea level and γ as the density based on the 100-m tower level (1121 m MSL). The results of this analysis are presented in Table 2. The computed fall speeds ranged from −4.7 to −1.9 m s−1. As expected, the bias profiles computed with the dual-Doppler wind data corrected for fall speed demonstrated a slight improvement over those in Fig. 5a. The greatest improvement was above the 47-m comparison level, where the bias was reduced by 11% (Table 2). At and below the 47-m level, the correction had little impact with only a 0.3% improvement in the 10-m bias. This was also expected given the dependence of the fall speed correction on the sine of the elevation angle in Eq. (1). There was also a slight improvement in the RMSE for the dual-Doppler data with the fall speed correction (Table 2).
c. 15 June 2012
An additional thunderstorm outflow event was sampled at the NWI field site on 15 June 2012 as a mesoscale convective system formed along the higher terrain of New Mexico (west of Lubbock) and propagated east. Within the outflow, 10-m wind speeds exceed 20 m s−1, while a low-level jet centered at approximately 330 m exhibited wind speeds over 30 m s−1 as sampled by TTUKa1 (Fig. 6a) and TTUKa2 (Fig. 6b). The dual-Doppler profile location was 114 m south of the tower. The 158-m level of the tower was not operational for this event, and data from the 200-m level were rendered unusable due to noise. Over 1.5 h of coordinated radar data were collected, but the use of the RHI/PPI scanning sequence coupled with the significant quality control necessary for the tower data reduced the usable dataset to approximately 30 min. Dual-Doppler and 200-m tower data within that time period were averaged to produce the mean profiles in Fig. 7. As with the previous thunderstorm outflow dataset, the mean dual-Doppler wind speeds exceed both the UVW and sonic mean wind speeds through the depth of the tower measurements (Fig. 7a). The largest differences were once again noted near the surface, where the mean dual-Doppler wind speed at 10 m exceeded the mean sonic wind speed by 15.2%. RMSE values were higher for this dataset than the 4 June 2012, such that the 10-m level demonstrated an RMSE of 3.78 m s−1 but decreased 47% at the 116-m level. The mean dual-Doppler wind direction profile showed slight backing when compared to the sonic and UVW mean wind direction profiles. However, differences between the dual-Doppler and sonic wind direction profiles were less than 5° on average with an RMSE greater at the 10-m level (7.59°) and slowly decreasing with height (Fig. 7b).
Fall speeds were also computed for the 15 June 2012 dataset using Eq. (2). Similar to the 4 June 2012 dataset, fall speeds ranged from −4.3 to −1.7 m s−1. Incorporating the fall speed improved the bias between the dual-Doppler and sonic wind speeds by 13% at the 116-m measurement level, but only by 0.2% at the 10-m level. As with the 4 June 2012 dataset, the RMSE slightly improved as well, with the greatest impact being at the 116-m level (Table 2).
d. 28 June 2013
A long-duration (over 8.5 h) clear-air dataset was collected on 28 June 2013 using the shallow RHI scanning sequence. The dual-Doppler profile location was 193 m south of the tower. All eight fully instrumented levels of the tower were operational. Unlike the outflow events, wind speeds through the depth of the tower were generally light (between 3 and 10 m s−1). As boundary layer convective mixing increased throughout the afternoon, wind speeds generally decreased and gust–lull features became more prominent (Figs. 8a and 8b). The wind direction generally veered from north to southeast through the duration of data collection (Figs. 8c and 8d). The mean wind profiles computed over the entire period demonstrate that the magnitude of the difference between the dual-Doppler and UVW wind speed profiles at each comparable level was less than 0.5 m s−1 (Fig. 9a). Whereas the mean wind speed profiles of the thunderstorm datasets exhibited the greatest differences below 47 m, the differences between the clear-air profiles in the same layer were relatively small (less than 6%). Above 47 m, the differences between the clear-air wind speed profiles were similar to differences noted between the thunderstorm profiles, such that the mean dual-Doppler wind speeds exceed the mean sonic wind speeds by 6% on average. Unlike the thunderstorm outflow, the RMSE wind speed profiles of the clear-air dataset were fairly consistent with height. Values ranged from 1.53 m s−1 at the 10-m level to between 1.35 and 1.45 m s−1 though the rest of the profile. With the exception of the lowest profile level (10 m), the dual-Doppler wind directions were slightly backed relative to the tower measurements (Fig. 9b). It should be noted that a backed bias appeared to be present in the 10-m wind direction of the 200-m tower data. This bias was also evident in the wind direction profiles of the events previously discussed and is of unknown origin. The magnitude of the difference between the dual-Doppler and sonic wind directions at other profile levels was between 6° and 9°. The wind direction comparison for the clear-air case demonstrated substantially greater RMSE values than in other cases with values generally greater than 15°.
4. Turbulence profile comparisons
The higher temporal resolution of the coordinated RHIs collected using the shallow RHI scanning sequence (used in the 4 June 2012 thunderstorm outflow validation experiment and the 28 June 2013 clear-air validation experiment) can be leveraged to generate dual-Doppler wind time histories at constant heights. These time histories can be compared to the UVW and sonic wind time histories at corresponding levels of the tower. Raw (50 Hz) tower and corresponding dual-Doppler wind speed time histories for each level of the tower were divided into independent segments 10 min in length. Data within these segments were averaged and used to investigate the turbulence properties of shallow RHI datasets.
a. Turbulence intensity
Turbulence intensity (TI) is defined as the ratio between the standard deviation of a given 10-min segment and the mean wind speed of the same segment (Simiu and Scanlan 1986):
For this dataset, only the along-wind (longitudinal) turbulence was considered. Longitudinal TI values for each segment were averaged together to produce a mean TI for each platform level. The TI profiles generated from the 200-m tower data reflected expectations of greater TI values near the surface due to the effects of surface roughness. The dual-Doppler-derived TI profile from the outflow dataset (Fig. 10a) displayed a similar trend but with reduced values of TI and a much less pronounced decrease in TI with height compared to the anemometer-derived TI values. The mean 10-m dual-Doppler TI value underestimated the 10-m sonic (UVW) TI value by 44% (39%). Above the 74-m tower level, both the dual-Doppler- and anemometer-derived TI displayed similar values between 0.045 and 0.05. Mean TI profiles collected by both platforms in clear air demonstrated greater values of turbulence than what were seen in the thunderstorm dataset (Fig. 10b). At the 10-m level, the dual-Doppler TI value showed little difference from the sonic and UVW TI values, with the platforms measuring 0.28, 0.29, and 0.29, respectively. Differences between the platforms increased above this level, where the dual-Doppler TI values underestimated those of the tower by 9% on average.
b. Gust factor
The gust factor (GF) estimates the gustiness of the wind by comparing the magnitude of a maximum gust wind speed to a mean wind speed. For this research, the GF at each level was defined as the ratio of the maximum 2-s wind speed (determined from a moving average through the 10-min segment) and the mean wind speed of the 10-min segment:
where the circumflex indicates a peak value and the overbar indicates the mean. Given the lower sampling rate (approximately 2.5 s), the dual-Doppler-derived gust factor at a constant height was defined as the maximum instantaneous wind speed of the 10-min segment divided by the mean segment dual-Doppler wind speed:
As with the TI profiles, the GFs were also higher near the bottom of the profile, where surface roughness effects dominate. However, the 10-m mean outflow GF computed with the dual-Doppler data was underestimated when compared to the 10-m sonic and UVW mean outflow mean GFs with values of 1.16, 1.30, and 1.31, respectively (Fig. 10c). The clear-air GFs at 10 m also showed slight differences between the platforms with the sonic GF exceeding the dual-Doppler GF by 8.3% (Fig. 10d). At and above the 75-m level of the tower, differences between the dual-Doppler and anemometer GFs were minimal for both datasets. The trends in the GF profiles echo those seen in the TI profiles, in that greater near-surface deviations were noted in the data from the thunderstorm outflow event.
c. Longitudinal integral scales
Longitudinal integral scales () estimate the average gust size in the along-wind dimension and were computed for each 10-min segment using the method and equations employed in Schroeder and Smith (2003), such that a time integral scale is obtained by integrating an exponential curve fit to autocorrelation function. The time integral scale is then multiplied by the mean wind speed of the 10-min segment to obtain the integral length scale. Mean dual-Doppler values were greater than both the mean sonic and UVW values through the depth of the outflow profile and at most levels of the clear-air profile (Figs. 10e and 10f). This difference indicates that the dual-Doppler synthesis overestimated the scales of turbulence associated with each environment but was still able to differentiate between the larger scales associated with the stable outflow environment and the smaller scales associated with the convectively mixed clear-air environment. As with the TI and GF profiles, the greatest differences between the mean dual-Doppler and anemometer profiles were found at the 10-m level of the outflow profiles, where dual-Doppler integral scales were estimated to be 83.8% larger than the sonic anemometer integral scales (Fig. 10e). While the magnitude of the integral scales generally increased with height, the differences between the dual-Doppler and tower values decreased with the 150-m dual-Doppler level, measuring a mean value of the 1102.4- and the 158-m sonic and UVW measuring mean values of 842 and 874 m, respectively. The mean clear-air profiles were consistent with the implications of the mean clear-air TI and GF profiles—namely, smaller scales of turbulence were present and produced a gustier, more turbulent wind field than the thunderstorm outflow (Fig. 10f). The smaller scales of turbulence indicated by the tower were also reflected in the dual-Doppler-derived profiles, where the mean 10-m values were 165, 112, and 119 m for the dual-Doppler, sonic, and UVW measurements, respectively. The magnitude of the differences varied with height, but it was generally smaller than the differences noted in the outflow profiles, most likely owing to the smaller scales of turbulence of the convectively mixed boundary layer.
d. Variation of averaging time
The length and quality of the clear-air dataset also permitted an examination of the effect of different averaging times on the dual-Doppler time histories. In addition to the 10-min averaging time, 2-, 5-, 15-, and 30-min averaging times were also used to generate mean wind speed, wind direction, and turbulence statistics for the dual-Doppler and 200-m tower sonic data at each height. The same methods as previously described were used for the computation of each parameter, only the lengths of the independent segments were altered to reflect the different averaging intervals. A linear regression analysis comparing the dual-Doppler and tower data was performed for each of the averaged datasets at the 10-, 47-, 75-, 116-, 158-, and 200-m levels (Fig. 11). The results of the linear regression analysis demonstrate an improvement in the correlation coefficient (R) with longer averaging times and higher altitudes for most parameters. Wind speed values exhibited relatively high correlations with the exception of the 10-m level, which required a 30-min averaging time to reach a 0.9 correlation (Fig. 11a). Wind direction exhibited R values above 0.9 for all levels and averaging times except for the 2-min averaged data at 10 and 47 m. Of the turbulence parameters, the TI correlation coefficients exhibited the greatest increase with averaging time. The R value between the 116-m dual-Doppler and sonic TI increased from 0.56 using a 2-min average to 0.95 using a 30-min average (Fig. 11d). Regardless of averaging time or height, all correlations between the sonic and dual-Doppler wind speed, wind direction, and TI were significant at the 0.05 level. Statistically significant correlations for GF and were also noted above the 47-m level. The correlation coefficients for GF and were generally lower as compared to wind speed, wind direction, and TI. The R values for GF did not exceed 0.8 at any level or averaging time. The lower GF values may be due to the differing definition of the gust wind speed between the tower and dual-Doppler data. The correlations were equally low with the exception of the 30-min averaging time, where four of the six levels exhibited statistically significant correlations above 0.9. Integral scales have also been shown to be highly dependent upon averaging time and vary widely across datasets and segments within the same dataset (Schroeder and Smith 2003; Simiu and Scanlan 1986).
Despite the correlation improvements, the use of longer averaging times had little effect on the wind and turbulence biases noted with the 10-min averaging time (not shown). For most levels, increasing the averaging window only slightly reduced the wind speed and direction differences. Most of the turbulence statistics also benefitted little from longer averaging windows, such that TI and GF were still slightly underestimated at most heights. In fact, the difference between the dual-Doppler and sonic generally increased up to a 10-min averaging time before displaying substantial variability thereafter. However, these trends in may be related to the greater influence of nonstationarities at longer averaging times (Schroeder and Smith 2003).
The differences noted in the presented wind speed comparisons between the clear-air, snow, and the two outflow cases suggest that scatterer type may impact the accuracy of the low-level dual-Doppler measurements. At and below the 47-m level, the differences between the dual-Doppler wind speeds and the 200-m tower wind speed were dependent on the dataset. For instance, the clear-air dual-Doppler wind speeds demonstrated very little deviation from the 200-m tower wind speeds at 10 m, while outflow dual-Doppler wind speeds were, on average, 15.6% greater than the tower wind speeds at the same level. Considering that a radial velocity measurement reflects the weighted average of the radial velocity of scatterers within a resolution element, scatterers with a greater terminal fall velocity may not experience the full deceleration of the low-level wind (due to surface roughness effects), and thus the radial velocity measurement may demonstrate greater deviations from air velocity. A similar effect was documented in Dowell et al. (2005), in which particle tangential velocities were compared to air tangential velocities in a simulated tornado vortex. It was noted that in regions of strong velocity gradients, particularly in a shallow layer near the surface, particle velocities exceeded air velocities, as particles did not adjust instantaneously to the increased surface drag. Dowell et al. (2005) also observed that the difference between particle and air tangential velocity was greater for larger objects with a faster fall speed. With respect to the full horizontal wind vector in nonrotating flow, the observations in Fig. 12 suggest a similar concept. Dual-Doppler velocities based on scatterers with larger fall speeds (liquid hydrometers) show a greater deviation from sonic (Fig. 12a) and UVW (Fig. 12b) wind speeds than dual-Doppler velocities based on scatters with slower fall speeds (snow and clear-air scatters).
The properties of dual-Doppler data based on different scatterers were further explored in the time histories generated from the 28 June clear-air and 4 June thunderstorm outflow shallow RHI datasets. The time histories from the 4 June outflow event indeed reveal a consistent overestimation of the 10-min mean dual-Doppler wind speed at 10 m throughout the record (Fig. 13a). While the mean profile comparisons for this event suggest a similar, yet reduced, wind speed overestimation at higher levels in the profile, the 150-m time histories of the 10-min mean wind speed demonstrate that these differences were not consistent through the record. After approximately 0100 UTC, the 10-min mean wind speed nearly matched the 10-min mean UVW and sonic wind speeds (Fig. 13b). A similar convergence of the time histories was also noted at the 75- and 116-m comparison levels for this event. These changes could be related to a decrease in hydrometer size and fall speed accordingly. Both reflectivity from the 0.5° tilt (centered on the 200-m tower) of the KLBB WSR-88D and calibrated reflectivity from TTUKa1 and TTUKa2 generally decreased through 0130 UTC (not shown). This trend in reflectivity suggests a general decrease in average hydrometer size over the field site through the course of data collection. Similar fluctuations in the differences between the dual-Doppler, UVW, and sonic mean wind speeds were also noted in the clear-air dataset. However, unlike the 4 June outflow dataset, there were periods where the 10-min mean dual-Doppler wind speeds at 10 m were aligned with the UVW and sonic wind speeds (Fig. 13c). The lack of a consistent low bias in the dual-Doppler wind speeds at this height also provides a measure of confidence that ground clutter (high return power; zero radial velocity) had minimal influence on the data. Above the 10-m level, the dual-Doppler mean wind speeds showed little deviation from the UVW and sonic mean winds speeds during the first hour of data collection as exemplified by the 150-m time histories (Fig. 13d). Large differences occurred between the dual-Doppler and anemometer mean wind speeds after this time period as convective mixing increased. The effects driving the differences between the dual-Doppler and anemometer clear-air time histories would likely be better understood with the aid of a particulate monitor in future research.
Turbulence intensity also showed considerable variability between datasets. For both the outflow and clear-air shallow RHI datasets, the dual-Doppler TI computed over 10-min segments was subtracted from that of the sonic and UVW anemometers. At 10 m, the difference between the anemometer- and dual-Doppler-derived TI was always positive for the outflow dataset with magnitudes less than 0.1 (Fig. 14a). In clear air, there was substantial variability in the difference between the anemometer and dual-Doppler TI with periods of both under- and overestimation of TI by the dual-Doppler data and periods with almost no difference in TI between the two platforms (Fig. 14b). However, the range of the differences was much greater for the clear-air dataset with values falling ±0.2. While the larger differences in the clear-air TI are likely reflective of the smaller scales of turbulence associated with the atmospheric conditions during data collection, the consistent underestimation of the 10-m TI by the dual-Doppler data in the outflow event is likely due to enhanced differences between hydrometer scatterers and air motions near the surface. Clear-air scatterers—mostly aerosols, pollen, and small insects at Ka band—could more readily follow the surface-induced turbulent fluctuations than could liquid hydrometers.
While scatterer type make play a significant role in the difference between the dual-Doppler and anemometer TI values near the surface, time histories of the TI difference at 150 m suggest that the scales of turbulence present in the boundary layer may also contribute to this difference. At 150 m, differences between the dual-Doppler and anemometer TI time histories were minimal for the outflow dataset, less than 0.025 (Fig. 14c). While the dominant scatterer was assumed to still be liquid hydrometers, the intensity of turbulence was less and the scales of turbulence were larger at this height as compared to 10 m (see Figs. 12 and 14). Larger scales of turbulence may have been more easily resolved by the spatially averaged dual-Doppler data relative to smaller scales. Similarly, the first hour of the clear-air dataset was characterized by lower values of TI and larger integral scales prior to the mixing out of an inversion. During this period, the clear-air time histories at 150 m revealed little difference between the dual-Doppler and anemometer TI values (Fig. 14d). As convective mixing increased throughout the afternoon, TI magnitudes increased and turbulence scales became smaller. This mixing processes potentially contributed to the larger differences between dual-Doppler and anemometer TI values after approximately 1530 UTC as smaller scales of turbulence became more prevalent throughout the boundary layer. Considering the volumetric averaging inherent in assigning a velocity value to a radar bin, the scales of turbulence smaller than the radar bin (approximately 15 m in range and 25 m in azimuth for both shallow RHI datasets) were not fully resolved. Averaging of radial velocities to generate the gridded profiles most likely contributed to additional smoothing. Spectrum width could be used to investigate turbulence scales smaller than the dimensions of the radar bin, but the application of this radar moment to turbulence measurements is beyond the scope of this research.
A new method was introduced in which coordinated RHIs from two high-resolution mobile Doppler radars were used to acquire dual-Doppler horizontal wind speed and direction profiles. The high temporal resolution of the coordinated RHIs allows not only for the generation of wind profiles at intervals less than those of typical wind profilers but also the creation of wind time histories that can be used to compute turbulence statistics. Resulting coordinate RHI datasets from mobile radars can be useful for a variety of interests, such as assessing the kinematics of a prestorm environment or measuring the inflow of a single wind turbine. This method was successfully verified in comparisons with data from UVW and sonic anemometers at multiple levels of a 200-m meteorological tower in different atmospheric conditions. The coordinated RHI method employed by the high-resolution TTUKa mobile Doppler radars was able to produce dual-Doppler wind profiles and time histories that exhibited high correlations with and little bias from traditional anemometry. The differences between the dual-Doppler and 200-m tower wind speeds seemed to be driven by scatterer type, such that scatterers with larger fall speeds introduced a positive bias in dual-Doppler wind speeds. This effect is most pronounced near the surface, where the wind speed gradient is larger due to surface effects. Turbulence parameters generated using a 10-min averaging time were also well replicated by the dual-Doppler data. Turbulence intensity and gust factor were slightly underestimated near the surface, while longitudinal integral scales were overestimated, suggesting that the spectrum of turbulent scales resolved by the anemometers was not completely resolved by the dual-Doppler data. Increasing the averaging time improved the comparisons between some parameters, particularly turbulence intensity. The type of scatterer could also affect turbulence measurements as demonstrated in comparisons of dual-Doppler data from clear-air and precipitating environments.
The impact of scatterer type on near-surface dual-Doppler measurements, as suggested by this research, warrants further study, especially if radar measurements are to be used in design considerations or related to near-surface wind damage. The inclusion of a disdrometer and/or a particulate monitor near the RHI intersection point would aid in the diagnosis of scatterer size and fall speed. Additionally, data collection in a greater diversity of environments would also be beneficial in understanding the wind speed differences associated with different hydrometeors and/or clear-air scatterers. Methods to retrieve vertical velocities from the intersecting RHI planes and turbulence parameters from spectrum width fields could also be investigated. Regardless, this study illustrates both the utility and accuracy of high-resolution dual-Doppler wind profiles derived from coordinated RHIs while also providing valuable uncertainty estimates for near-surface dual-Doppler retrievals.
The authors thank Jerry Guynes for the design, construction, and maintenance of the TTUKa mobile Doppler radars and for help with data collection. The authors would also like to thank Dr. Stephen Morse for the construction and maintenance of the 200-m tower database and for the initial processing of the 200-m tower data. This research was supported by NSF Grant CMMI-1000160. The authors would also like to acknowledge the U.S. Department of Energy for providing funding for a portion of this research under the Congressionally Directed Project grant: Great Plains Wind Power Test Facility (Award DE-FG-06-GO86092). The comments from three anonymous reviewers greatly improved this manuscript.