Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
 Observations of convective precipitation using a 94-GHz cloud radar are presented. Due to Mie scattering, the Doppler power spectra collected at vertical incidence contains characteristics of the scatterers (hydrometeors). These characteristics are used for the retrieval of the vertical air motion and the associated raindrop size distribution in an attempt to accurately map the time-height structure of the vertical air motion and raindrop fields within intense convective precipitation. The data provide strong evidence of the interaction between draft intensity and raindrop size distribution and highlight the variability of convective precipitation at small scales. Horizontal sorting of the raindrops caused by the air motion is documented. Signal attenuation measured at 94 GHz is shown to be well correlated to rainfall rates. The observations demonstrate the capability of 94-GHz cloud radars for studies of precipitation processes at low altitudes even under intense convective conditions.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Accurate measurements of Drop-Size Distributions (DSD) are fundamental for understanding the processes governing cloud microphysics and improving precipitation representation in numerical models [e.g., Grabowski et al., 1999]. Especially at small scales, microphysical processes such as condensation of water vapor, collision and coalescence between the droplets, evaporation in unsaturated air, and droplet breakup contribute to the final product–the hydrometeor DSD [Feingold et al., 1988; Hu and Srivastava, 1995]. Furthermore, these processes are embedded in vertical air drafts that influence the final shapes of the DSD [Kollias et al., 2001]. As a result, and despite their importance, the small-scale variability of DSD and their interaction with air drafts remains unknown, especially at scales unresolved by most active and passive remote sensing instruments.
 Doppler radars are traditionally used to study precipitation. In a scanning mode they are excellent tools for monitoring the intensity and motion of precipitating systems, but the quantitative rainfall rate retrievals are based on numerous assumptions. Nevertheless, knowledge of the spatial distribution of rainfall rates can be useful for a wide range of applications. Unfortunately, scanning radars operating without polarization provide little information on the spatial variability of DSDs. Often, the radar reflectivity, which depends on the sixth power of the diameter (Rayleigh scattering), is used to characterize the spatial homogeneity of the precipitation field. Such interpretation can lead to erroneous conclusions about the nature and variability of the precipitation field. Thus, most observational efforts focus on the monitoring and classification of precipitating systems and surface rainfall measurements [e.g., Houze, 1993; Williams et al., 1995; Tokay et al., 1999; Cifelli et al., 2000; Kummerov et al., 2000; Williams et al., 2000]. Cloud modelers, however, require information on the evolution and modification of raindrop spectra under the action of physical processes operating on a very small scale in the presence of vertical drafts. Several researchers in the past [Srivastava, 1971; Carbone and Nelson, 1978; List et al., 1987; Hu and Srivastava, 1995] have investigated the form of the equilibrium DSD under the influence of various microphysical processes. Apart from the complexity of the microphysical processes, observations of these small-scale interactions in the presence of vertical motions are very difficult. As a result, these numerical studies cannot be compared with observations, since these scales are unresolved. Therefore, over the last 20 years a gap between observations and modeling of precipitation processes has developed.
 Vertically pointing radars can provide excellent vertical and temporal resolution. Furthermore, their mean Doppler velocity can be related to the sum of the air motion and raindrop's terminal velocity. Wind profilers, in particular, can detect both Bragg scattering (clear-air scattering) and Rayleigh scattering (hydrometeors) [e.g., Rogers et al., 1993] and therefore, under certain conditions, can successfully decompose the velocity measurements. When it is feasible, such decomposition [e.g., Wakasugi et al., 1986; Gossard, 1988], can provide more information and at higher resolution sampling compared with scanning radars. Under precipitating conditions, profilers can provide the wind profile at a single point and also map the vertical profile of the overhead cloud system, since the attenuation at these wavelengths is negligible. Despite the apparent potential for accurate measurements of vertical air motion and raindrop spectra, the retrievals from wind profilers are subject to many assumptions, therefore increasing the uncertainty of these measurements. Main sources of uncertainty are the assumption on the maximum raindrop size observed; wind shear induced Doppler spectra broadening and poor Doppler spectrum velocity resolution.
 Observations of both the vertical air velocities and raindrop size distributions are very rare. Aircraft observations [e.g., Beard et al., 1986; Szumowski et al., 1998; Atlas et al., 2000] are capable of this type of measurement, but they lack vertical coverage and are difficult to obtain near the surface. Further, the sampling volume of aircraft microphysical probes is very small, thus resulting in under sampling of large raindrops in the DSD due to their low concentration. In addition, the horizontal sampling path is relatively large (700–800 m) and the lack of another dimension makes the interpretation of the data difficult. Despite these sampling problems, aircraft penetrations are still the most reliable way to collect simultaneous measurements of the vertical air motion and raindrop size distributions.
 In this paper, a new technique using a 94-GHz Doppler radar and a 915-MHz wind profiler is applied in heavy convective rain [Lhermitte, 1987; Firda et al., 1999; Kollias et al., 1999, 2001]. Millimeter wave radars have been primarily used for cloud observations due to their high sensitivity to small droplets and their ability to make high-resolution observations of weak targets [Miller and Albrecht, 1995; Vali et al., 1998; Kollias and Albrecht, 2000]. Lhermitte  proposed an unexpected application of millimeter radar based on the presence of Mie scattering due to the very short wavelength (3-mm) of 94-GHz radars. This technique capitalizes on the modulation of the Doppler spectrum by the backscattering function that in the Mie regime oscillates between fixed maxima and minima. Under precipitating conditions at 94 GHz, these oscillations are apparent in the observed Doppler spectrum and can be used as reference points for the retrieval of the vertical air motion and subsequently the DSD. Using this 94-GHz technique, high spatial and temporal measurements of vertical air motion structures and DSDs in precipitating clouds were obtained. Unique observations of updrafts and downdrafts in convective systems are presented and their interaction with the raindrop size distributions is documented. Signal attenuation is a prohibiting factor and affects the penetration of the 94-GHz radiation under high rainfall rates. In this paper, measurements of attenuation of 94-GHz electromagnetic (EM) radiation in convective precipitation as a function of rainfall rates are presented. Thus, the performance of the technique under intense convective conditions is evaluated.
2.1. Millimeter-Wavelength Radars and Precipitation
 The use of millimeter wavelength radiation for precipitating system studies overcomes a significant obstacle in the DSD retrievals by accurately measuring the vertical air motion. Figure 1 shows the 94-GHz normalized backscattering cross section σb as a function of raindrop diameter at 20°C. At 94 GHz, the backscattering cross-section versus size function for raindrops with a diameter greater than 1 mm oscillates. The raindrop diameters for which these minima and maxima occur are well predicted by Mie theory [Mie, 1908]. The first minimum is well defined and occurs at a raindrop diameter of 1.7 mm. The vertical air velocity can then be deduced from the simple difference between the terminal velocity of a raindrop with diameter 1.7 mm and the value of the first minimum in the Doppler spectrum (Figure 2) observed at vertical incidence with the millimeter wave Doppler radar. Lhermitte  first mentioned this innovative technique in the context of stratiform rain observations. Recently, Firda et al.  and Kollias et al. [1999, 2001] used this approach to study the retrieval of precipitation and vertical air motion in stratiform rain and light convective rain using a vertically pointing 94-GHz Doppler radar.
 The use of millimeter radars for the retrieval of raindrop spectra offers more advantages besides the accurate decomposition of the observed Doppler velocity. Millimeter wave radars are designed as research tools, rather than weather warning and monitoring platforms. While their spatial coverage is no match for conventional radars and wind profilers, the short pulse width and the very narrow beam width beam results in small sampling volumes. As a result, the effects of turbulence and wind shear on the Doppler spectrum are minimized [Kollias et al., 2003].
 Apart from the advantages in the use of millimeter radars for precipitation studies, strong attenuation of the 94-GHz EM waves in heavy rain is a serious disadvantage. Lhermitte  reports that the one-way attenuation can reach 7–8 dB km−1 for a rainfall rate of 10 mm h−1. Hence, it is unlikely that the mm-wavelength radar can penetrate convective systems from the ground to the cloud top height in intense rainfall. In the region of cloud observed by the 94-GHz radar, the strong attenuation makes the accurate measurement of reflectivity values difficult. As a result, the retrieved DSD from the Doppler spectra are unscaled. However, reflectivity profiles from a collocated 915-MHz wind profiler (the approach taken here) or other long wavelength radars can be used to scale the retrieved DSD. In a later section, the relationship between attenuation and rainfall rate is evaluated.
2.2. Raindrop Distortion
 It is well known that the raindrop shape, especially that of large raindrops (D ≥ 2 mm), deviates significantly from sphericity [e.g., Beard and Chuang, 1987]. Large raindrops falling at terminal velocity exhibit an asymmetric shape with a flattened base. This deviation of the equilibrium shape from sphericity and the use of very short wavelength (λ = 3.2 mm) requires the use of a scattering model for nonspherical particles. The hydrostatic model proposed by Green  is used to describe the equilibrium shape of falling raindrops. This deviation from sphericity creates differences in the backscattering cross-section. Therefore, it is important to correct the Doppler spectra corresponding to spherical raindrops to more realistic oblate spheroid raindrops (Figure 1). The T-matrix [Mishchenko and Travis, 1994; Mishchenko, 2000] approach (or the extended boundary condition method), is used to solve the problem of scattering from nonspherical particles. The method uses the phase-scattering matrix that relates the intensity of the incident and scattered radiation.
 The difference between the Doppler spectra for oblate spheroids and spherical raindrops is shown in Figure 3. Up to 4 m s−1 the differences in the Doppler spectrum are negligible, but beyond this limit the deviation is substantial. The use of the Mie solutions for spherical particles can lead to an overestimate of the large raindrops concentration [Aydin and Lure, 1991]. In addition, the use of the Mie solutions for spherical particles will lead to an overestimate of downdraft intensity. Using the T-matrix method, the location of the first Mie minimum is at D = 1.71 mm rather than D = 1.67 mm for spherical particles. This difference causes a 7 cm s−1 shift in the location of the first minimum in the terminal fall speed (5.95 m s−1 instead of 5.88 m s−1).
3. Instrumentation and Data Processing
 During the summer and autumn of 1999 a combination of instruments that included the University of Miami 94-GHz Cloud Radar (MCR) and a 915-MHz wind profiler were used to observe convective and stratiform precipitating systems passing over Virginia Key, Miami, Florida.
3.1. Vertically Pointing Radars
 A single-antenna Doppler radar operating at 94 GHz [Albrecht et al., 1999], is the principal source of observations made in this study. The radar was operated with a 10 kHz Pulse Repetition Frequency (PRF) to give a unambiguous Doppler velocity window of ±8 m s−1. Doppler spectra measurements were based on the integration of 10,000 samples (1-s dwell time). The recorded Doppler spectra had 512 points with 3.2 cm s−1 velocity resolution. In the radar observations reported here, the vertical resolution is 30 m, thus providing a fine vertical resolution mapping of the cloud structure and boundaries. The antenna beam width is 0.24°, so that the radar horizontal resolution is about 8 m at 2 km range. For a 1-s dwell time and a typical cloud motion of 10 m s−1, the effective beam cross section is approximately doubled. In the applications described in this paper, a 1-s signal sampling is followed by a 2 to 3.5 s FFT processing so that a new vertical profile is obtained every 3 to 4.5 s.
 The 915-MHz wind profiler collocated with the cloud radar was operated in vertical incidence mode (beam width of about 9°) to give a temporal resolution of 20 s. The range gate spacing was 210 m and the Doppler spectra were recorded in all cases. The 64-point recorded Doppler spectra had 32.8 cm s−1 velocity resolution over a range of ±10.5 m s−1. The 915-MHz and 94-GHz radars were collocated within 10 m to provide overlapping sample volumes of the overlying vertical atmospheric column.
3.2. Supplementary Observations
 In addition to the vertically pointing radars, a tipping-bucket rain gauge and a surface meteorology station were available to complement the radar observations. The resolution of the tipping bucket is 0.254 mm and provides a time series of the rainfall rate during the observing period. The rainfall rate record is essential for rain intensity determination and comparison with the retrieved rainfall rates from the radars. Since both the 94-GHz radar and the wind profiler were vertical pointing, data from the Miami WSR-88D (KAMX) weather radar, operating at λ = 10 cm and located 28.4 km southwest of the MCR site were used for monitoring the evolution and horizontal structure of the observed precipitating systems.
 The WSR-88D radar completes volume scans in multiple elevations within 5–6 min. However, due to it's coarse spatial and temporal resolution, the reflectivity values at the location of the site are of less significance than the vertical profiles of reflectivity deduced by the vertical pointing radars. In this study the WSR-88D provides monitoring capability and continuous sampling coverage of large areas to map the horizontal structure, motion, and evolution of the precipitating systems.
3.3. Data Postprocessing
 The basic data sets used in this study are the vertical profiles of Doppler spectra from the 915-MHz wind profiler and the 94-GHz cloud radar. Initial processing included de-aliasing of the Doppler spectra due to frequency folding and noise-thresholding. Aliasing occurs if the Observed Doppler shift frequencies (velocities) exceed the Nyquist frequency (velocity).
 Once the de-aliased Doppler spectrum is recovered, it is centered and a noise thresholding technique is applied that uses the edges of the Doppler spectrum [Lhermitte and Kollias, 1999].
 The next step is the spectral peak detection. The peak detection methods for the wind profiler and the cloud radar Doppler spectra are different. In the case of the wind profiler, the method searches for possible bimodality arising from the coexistence of Bragg and precipitation spectral peaks. Two spectral peaks are often observed by the profiler under stratiform conditions, where the intensities from clear air echoes and light precipitation are of the same order of magnitude. In most cases there is no clear air echo in the 915-MHz profiler spectra above 2 km [Rajopadhyaya et al., 1999]. The spectral peak detection algorithm is based on fitting a high order polynomial (up to 12th order) to the Doppler spectra. The high order in the polynomial fit is essential to capture the bimodality of the observed spectrum [Sato et al., 1990]. The first and second derivatives of the polynomial fit are calculated, and the local maxima and minima are located.
 The 94-GHz radar spectra are treated in a slightly different manner. In addition to the identification of the local maxima and minima, it is necessary to identify the spectral peaks created by the modulation of the Doppler spectrum by the Mie backscattering function (Figure 2). This procedure is more complex than that applied to the 915-MHz spectra, since one, two or three peaks can be detected depending on the turbulence intensity and the shape of the DSD. If only one peak is detected, then no retrieval of the vertical air motion and the DSD is computed. In this case, either the drop size distribution does not contain large enough raindrops to give returns beyond the first Mie minima (Dmax ≤ 1.7 mm), or the attenuation of the signal is so strong that radar receiver noise overwhelms the atmospheric return. For cases where there are two or more peaks, the relative spacing of the peaks is used to identify which are real peaks generated by the scattering mechanisms, and which are artifacts created by processes related to the noisy nature of the Doppler spectrum. Once the peaks are correctly identified, the displacement of the first Mie minimum is used to estimate the mean air motion. The observed Doppler spectrum is shifted to zero air velocity conditions. A nonlinear least squares fitting procedure, similar to the one used in the retrieval of the DSD in the wind profiler is applied. As in the case of the wind profiler, there are limits to the applicability of the technique. It is applicable as long as the signature of the Mie oscillation is apparent in the observed spectrum. Throughout this study, a conservative approach was pursued, applying the method only in spectra with well defined Mie maxima and minima.
 Signal attenuation has no effect on the Doppler spectrum shape and dynamic range. Consequently, the retrieved DSD shape from the 94-GHz Doppler spectra is not affected by the signal attenuation and an accurate reflectivity measurement is needed in order to scale properly the DSD.
4.1. High Reflectivity Core
 The data for this case were collected on 7 and 8 September 1999, at the Rosenstiel School of Marine and Atmospheric Science (RSMAS) site. Figure 4a shows a time-height cross-section of reflectivity from the 915-MHz wind profiler. The convective system sampled, formed inland from the site, and was observed for almost 3 hours as a substantial trailing stratiform area moved overhead after the passage of a convective core. The rainfall rate measured by the rain gauge collocated with the radars ranged between 30 and 80 mm h−1during the passage of the high reflectivity core.
 The arrival of the storm was indicated by a sudden 15 K drop of the equivalent potential temperature, θe (Figure 4b). A wind shift and the lower θe values occurred 30 min ahead of the storm, before any rain was detected at the site. The drop in θe was associated with storm outflow caused by downdrafts. The lower elevation scan data from the Miami WSR-88D (KAMX) verified that the observed precipitating system was part of a convective line triggered by outflows of other convective regions further inland and propagating in a southeast direction. The first cloud detected by the profiler (Figure 4a) was a shelf cloud at 2325 UTC, nearly coincident with the θe drop. Convective precipitation was detected at 2345 UTC as illustrated by the vertically oriented high reflectivity area and the high rainfall rates. The convective part of the storm lasted almost 30 min, and after a short transition period, stratiform precipitation was observed for more than 2 hours (0045–0245 UTC) with the 915-MHz wind profiler. A bright band signature was observed during the stratiform rain (Figure 4a) at an altitude of 4.5 km. Rainfall rates during this period were very low with values ≤1 mm h−1, suggesting strong evaporation as indicated by the weak reflectivity values at low levels. Bragg scattering is evident at these levels, creating a noisy image during the last hour of 915-MHz observations.
 The analysis focuses on the convective core, since the stratiform part of the system was very weak and associated with few ≥1.7 mm drops (e.g., no retrieval). As a result, the Mie minimum was detectable only in a few locations, making the analysis difficult. The reflectivity values observed by the wind profiler (Figure 5a) were verified by the WSR-88D. There are two high reflectivity cores (dBZ ≥ 50) at levels between 1.5 and 4.0 km. The presence of such high values is likely related to the existence of large raindrops at these levels. Due to strong attenuation, the cloud radar could not penetrate higher than 1.5 km in the convective part of the cloud. However, DSD and w analysis was possible in the domain shown in Figure 5a. The retrieved vertical motion field within this domain is shown in Figure 5b. The vertical resolution of the cloud radar for this case was 60 m, and the temporal resolution was 4.8 s, providing 300 profiles of Doppler spectra to define the structure of the air motion field. Each Doppler spectrum was analyzed independently without any input from nearby Doppler spectra in the time-height domain. The results (Figure 5b) show a narrow updraft at the approaching side of the convective core. The interior of the convective system at these levels is dominated by two weak downdrafts separated by a narrow, weak updraft area. The magnitude of the updraft reaches 4 m s−1 and the main downdraft approaches −3 m s−1. In general, the retrieval of the vertical air motion under these conditions (R between 30–80 mm h−1) at different levels is coherent in time and space.
 After retrieving the vertical air motion, the Doppler spectrum is used to retrieve DSDs, which illustrate the different nature of the two downdrafts. The cloud radar signal is severely attenuated, and the reflectivity measurements are not reliable. The retrieval method captures the DSD shape. However, scaling of the raindrop size distribution requires accurate measurements of reflectivity. Since there is little attenuation of the wind profiler radiation by precipitation particles, the wind profiler reflectivity values are used to scale the raindrop spectra retrieved by the 94-GHz radar. To compensate for the lower temporal resolution of the wind profiler, we assume a homogeneous reflectivity field within a 20 s time interval.
 A reflectivity cross section at 1.3 km from the wind profiler and the corresponding surface rainfall rates are shown in Figure 6a. Initially, a high reflectivity region (2348–2354 UTC) is observed that corresponds to the area of high reflectivity (dBZ ≥ 50) observed within the convective rain core (Figure 5a). Later, a plateau of reflectivity is observed (≈45 dBZ) until the end of the convective core where a sharp decrease of the radar reflectivity is observed. The corresponding surface rainfall rate shows a bimodal structure with time within the convective core. A relative minimum is observed during the transition from the high reflectivity region to the reflectivity plateau. The maximum rainfall rate is observed during the reflectivity plateau (R = 60–80 mm h−1).
Figure 6b shows the medium volume diameter (Do) and the vertical air motion at the same height retrieved from the 94-GHz Doppler spectra. Do can be estimated by the retrieved DSD shapes without reflectivity scaling. Despite the differences in the sampling volume, the reflectivity from the wind profiler and the Do retrievals from the cloud radar Doppler spectra show consistent variability. The observed high reflectivity region (2348–2354 UTC), overlaps with high Do values. After a very sharp transition area, lower reflectivity values (≈45 dBZ) are observed by the wind profiler. During the same period (2355–0003 UTC), the Do values are slightly lower (Do ≤ 2 mm). Figure 6b, also shows the vertical air motion retrieved by the cloud radar. A weak downdraft area is adjacent to the strong updraft observed at the leading edge of the convective core. The transition between the updraft and the downdraft is very sharp and is followed by a sharp increase in the Do values.
 Most of the rain falls within the wider and stronger downdraft (2355–0003 UTC). This observation is supported by the tipping bucket rainfall data. In the area of the main downdraft, the raindrops have smaller sizes and the maximum observed sizes did not exceed 3 mm in diameter. In addition to the variability of Do induced by the drafts, small-scale variability in Do is observed. Other processes, and especially turbulence in the interior of the convective core, may contribute to this variability.
4.2. Hurricane Irene
 On 15 October 1999, Hurricane Irene made landfall in Southwest Florida. Figure 7a shows the lowest elevation scan (0.5°) mapping the reflectivity of the hurricane from the Miami WSR-88D. At 1440 UTC, the center of the hurricane was located in the lower Florida Keys and a rainband passed over the radar site (Virginia Key, Miami). The WSR-88D data indicate reflectivity values of 40–45 dBZ within the hurricane rainband. Figure 7b, shows a time-height section through the rainband as observed by the vertical beam of the 915-MHz profiler. The observations also show a bright-band radar signature at a height of 4.7 km.
 During the same period, and despite the intense rainfall and sustained horizontal wind of 20–25 m s−1, the 94-GHz cloud radar was used to collect high-resolution Doppler spectra profiles. During the high reflectivity periods the cloud radar signal was totally attenuated within the lowest 2 km, while during the stratiform period the radar signal reached the melting layer. Figure 8 show a time series of microphysical retrievals from the 94-GHz Doppler spectra from the lowest radar gate (200 m).
 The retrievals from the MCR show exceptional correlation with the rainfall rate collected from the rain gauge. In Figure 8a, the retrieved Do time series at the lowest gate (200 m) and the rainfall rate are shown. During the convective reflectivity core, high rainfall rates were observed. The burst-type variability of the surface rainfall data during that convective event was consistent with the variability of the wind profiler reflectivity, showing four reflectivity maxima within the convective core (≥35 dBZ). The retrieved values of Do vary with the fine scale variability of the reflectivity within the core. Figure 8b shows a comparison between the rainfall rate calculated using the retrieved N(D) from the mm-wavelength Doppler spectra scaled using the 915-MHz reflectivity and the observed surface rainfall rate. This comparison is encouraging since the retrieved rainfall reasonably tracks the gauge record. The retrieval performs better during the stratiform periods, where the assumption of homogeneity in the reflectivity field is valid. In the convective core, the fine-scale variability is resolved, but there are differences between the retrieved and the actual rainfall measurements. In general, however, the retrieved rainfall rates capture the observed small-scale variability.
4.3. 94-GHz Attenuation
 In the introduction, we discuss the prohibiting effect of attenuation on the use of a 94-GHz Doppler radar for precipitation studies. In an effort to quantify and evaluate the signal attenuation at 94 GHz, attenuation data obtained during heavy rainfall rates observed during Hurricane Irene are used. The attenuation was estimated from the observed signal decrease (in dBm) in the lowest 300 m above the radar near field (200 m). Assuming homogeneous conditions for the precipitation targets in the 200–500 m layer, the signal drop was attributed to attenuation. In Figure 9, the attenuation A (dB km−1) of the radar signal is plotted with the observed surface rainfall rate R (mm h−1). In addition to the attenuation versus surface rainfall observations, modeling results of signal attenuation at 94 GHz for a variety of exponential raindrop size distribution (N = Noe−ΛD) are shown. For modeling purposes three different No values were selected. These are the Marshall-Palmer (MP, No = 0.08 cm−4) [Marshall and Palmer, 1948], the Joss Thunderstorm (JT, No = 0.014 cm−4) and Joss Drizzle (JD, No = 0.3 cm−4) [Joss and Gori, 1978] for an ambient temperature of 20°C. The dashed line shows the best regression fit A = 0.89R0.827. As expected, there is strong correlation between the signal attenuation at 94 GHz and rainfall rate. The correlation suggests that signal attenuation at 94 GHz, observed in a shallow precipitation layer can be used for rainfall measurements.
4.4. Comparison of Vertical Air Motion Retrievals
 Wind profilers are widely used for quantitative measurements of air motion and precipitation. An interesting application is the collocation of a 915 MHz wind profiler with the MCR. This configuration was implemented during our observations. Comparison of the vertical air motion retrievals can be performed only in stratiform rain, since the Bragg scattering is overwhelmed by the precipitation return in convective rain. A comparison is shown in Figure 10. The stratiform rain data were collected during the high reflectivity core case (0130–0200 UTC). The MCR data are more dense and coherent and a general trend from a weak updraft to a weak downdraft is observed. The wind profiler air motion measurements follow the same trend, but overestimate the magnitude of the vertical air motion. Such a comparison can greatly enhance our understanding of wind profilers data and their accuracy. Similar comparisons can be made for the estimation of the turbulence intensity and drop size distribution. This is a very important dimension of the cloud radar, since there is already an extensive database of wind profilers data. Such a comparison tests the assumptions required for retrievals with wind profilers and other radars. The high sampling cloud radar can document the homogeneity of the sampling volume of the wind profiler. In addition, the low-level retrieval of the cloud radar can be used to correct the wind profiler data and with assumptions, extrapolate these measurements to higher altitudes where the wind profiler is able to penetrate.
5. Discussion and Conclusions
 In this paper, the potential of using a 94-GHz radar for precipitation studies is demonstrated. Emphasis is given to convective rain observations. The observations clearly demonstrated that a 94-GHz radar, combined with a lower frequency radar, is a very useful tool for looking at microphysics and kinematics associated with both convective and stratiform rain. The time-height retrieval of vertical air motion and DSDs, and the superior sampling (temporal and spatial resolution) relative to other remote sensors, makes the cloud radar a unique instrument for resolving small-scale variability in the interior of convective cores when attenuation is not a major problem. The complexity of the convective structures implies that we need horizontal winds as well as vertical profiles of w and DSDs to understand these structures. Such an observing platform can be the basis for future precipitation research, especially at small scales.
 Interesting vertical draft structures were observed for the two cases studied. The spatial resolution of the observed updraft and downdraft structures goes well beyond any other previous observations of the lower part of convective updrafts. The two-dimensional view provided by the radar adds one more dimension to that provided by aircraft penetrations. In the first case studied, strong evidences of drop sorting effects due to the kinematics of the interior of the convective core were found. The size-sorting of raindrops in space due to convective updrafts was also documented by Kollias et al.  using the same retrieval technique. The elevated high reflectivity cores (Figure 5a) are consistent with the suspension of raindrops at high levels until they reach terminal velocities large enough to overcome the upward motion. The vertical air motion retrievals (Figure 5b) verify the presence of a strong updraft. The results in Figure 6b show very large raindrops (Do ≥ 2.4 mm) for a 3-minute interval inside the area of the first weak downdraft. Although we are limited to a two-dimensional snapshot of the air motion field and we cannot easily explain the existence of the large raindrops next to the main updraft, the observations provide a physical model of the role of convective updrafts in the precipitation process. Szumowski et al.  observed a similar behavior using combined radar and aircraft observations. If the main updraft is tilted in three-dimensions or is weakening, the large drops will escape and fall rapidly from the higher levels of the cloud. Actually, this is evident in the high reflectivity tail of the first high reflectivity core (Figure 5a). Since the large raindrops are observed at lower altitudes, they somehow avoid collisional breakup with small raindrops during their fall from higher altitudes. A tilted updraft can create this type of drop separation due to the different terminal fall velocities of raindrops of different sizes. Thus, the large drops must be falling through regions with small concentrations of small raindrops [Rauber et al., 1991]. In addition, the region where the large raindrops are observed is adjacent to the strong updraft. Thus, recirculation and further growth of some of the raindrops is a plausible mechanism for the generation of large raindrops and high reflectivity values. The vertical structure from the observations provides a more complete picture than possible from in situ measurements and facilitates the interpretation of the data. While more observations of this type are required, the observations underline the need for higher spatial resolution precipitation models with explicit microphysics so the effects of updrafts on raindrop spectra can be simulated.
 In addition to highlighting the interaction between convective drafts and raindrops, the observations demonstrated that convective cores have internal variability. DSD and vertical air velocity variability are not highly correlated at one level (even though in the vertical each field is coherent), and exhibits fine-scale structure (10s of m) that has been unresolved by other measurement tools. The small-scale variability observed, indicates that other remote sensors like wind profilers with inherently large spatial filters will be ineffective in describing this variability. In addition, a comparison of the Z-R time series within the convective core (Figure 6a) exemplifies the difficulty of predicting rainfall rates in convective cores using reflectivity data.
 The DSD shapes and subsequently the Do retrievals are independent of reflectivity measurements. Consequently, the fact that Do variability retrieved by the cloud radar is correlated with reflectivities from the wind profiler, is strong support for the validity of the retrievals. The data within one of Hurricane Irene's rainbands underline another important application of using a 94-GHz radar for precipitation studies. Due to its sampling volume (3000 m3, which is significantly larger than the sampling volume of aircraft microphysical probes) and the strong signature of the Mie scattering oscillations on the Doppler spectra, the 94-GHz radar is an excellent tool for the detection of large raindrops (D ≥ 3 mm) in precipitation. The presence of a single large raindrop within the radar sampling volume will create a strong third Mie peak signature (see Figure 2) in the Doppler spectrum.
 An important issue for the application of 94-GHz radar in convective rain is signal attenuation. If it were not for attenuation, 94-GHz radar would be the ideal tool for precipitation studies. Under convective conditions, however, the radar signal experiences severe attenuation. The attenuation as a function of rainfall rate observed during Hurricane Irene, are within the theoretical limits estimated using different exponential distributions. At such high radar frequencies, scattering contributes as much to the attenuation as the signal absorption.
 Despite this serious disadvantage of short wavelength radars under precipitating conditions, the information contained from the Doppler spectrum (vertical air motion and DSDs) makes a 94-GHz Doppler radar a valuable tool for precipitation research. In particular, during low to moderate stratiform rain conditions (R ≤ 3 mm h−1), the 94-GHz Doppler radar retrieval technique is applicable from the ground to the melting layer (4–4.5 km in the tropics). Using its current configuration the MCR can retrieve the air motion and DSD in stratiform rain with 60 m vertical resolution and 3 s temporal resolution [Kollias et al., 2003]. Under the same conditions the wind profiler retrieval technique is applicable at the lower 2–2.5 km. At higher rainfall rates (R ≥ 10–20 mm h−1), a surface based 94-GHz Doppler radar with peak power 1 kW and 1 m antenna can penetrate the lower 2 km of the convective precipitation [Kollias et al., 2001]. The 915-MHz wind profiler retrieval technique is not applicable under convective rain since the Bragg scattering return is overwhelm by the Rayleigh scattering from the raindrops. Thus, the 94-GHz Doppler radar is the only remote sensing tool that can retrieve the vertical air motion and DSD in convective rain. Collocated with a lower frequency Doppler radar, 94-GHz radars can overcome to a great extent the uncertainties related to the retrievals of vertical air motion and DSDs [Kollias and Albrecht, 2002]. With simplicity and a minimum set of assumptions, this type of research radar can provide important, fundamental details of the precipitation processes.
 We are grateful to the technical assistance provided by Tom Snowdon during the collection of the data used in this study. This work was supported by NSF grant ATM9730119 and DOE grant DEFG0297ER62337.