Abstract

The NASA dual-frequency, dual-polarization Doppler radar (D3R) was deployed as part of the GPM Iowa Flood Studies (IFloodS) ground validation field campaign from 1 May through 15 June 2013. The D3R participated in a multi-instrument targeted investigation of convective initiation and hydrological response in the midwestern United States. An overview of the D3R’s calibration and observations is presented. A method for attenuation correction of Ka-band observations using Ku-band results is introduced. Dual-frequency ratio estimates in stratiform rain and ice are presented and compared with theoretical values. Ku-band quantitative precipitation estimation results are validated against IFloodS ground instruments.

1. Introduction

The NASA dual-frequency, dual-polarization Doppler radar (D3R) is a deployable weather radar that enables synchronized, beam-aligned observations at Ku (13.91 GHz) and Ka band (35.56 GHz; Vega et al. 2014). The D3R was designed as a ground-based radar with operating frequencies analogous to the Global Precipitation Measurement (GPM) core satellite Dual-Frequency Precipitation Radar (DPR). As part of the ground validation mission for the GPM program, the D3R is a critical platform for comparison with the DPR because of its similar operational frequency and the ability to be deployed for targeted, long-term observation of meteorological events. The D3R provides high-temporal-resolution and high-spatial-resolution observations for a more detailed understanding of precipitation that will ultimately be used to refine algorithms for retrieval of microphysical properties (Vega et al. 2014).

The D3R is equipped with a solid-state transmitter system and coaligned, beam-matched antennas. The D3R allows for simultaneous dual-polarization observation of the same volume at two frequencies. The Ku band provides greater penetration into moderate and heavy convection. The Ka band has enhanced sensitivity to light rain and snow but is more susceptible to attenuation in moderate and heavy precipitation. The solid-state transmitter and receiver design supports reconfigurable transmitter waveforms and receiver filters, making the D3R an agile research platform for microphysical investigation.

The D3R was deployed collocated with NASA’s S-band polarimetric radar (NPOL) during the NASA Iowa Flood Studies (IFloodS) field campaign as part of the GPM ground validation. Both D3R and NPOL were deployed in Traer, Iowa, from 1 May to 15 June 2013 to observe precipitation in concert with an array of ground instruments, NEXRAD S-band radars, and the University of Iowa’s X-band radars in the area of the Cedar River and Iowa River basins. During the field campaign, severe flooding was observed in the surrounding areas and river basins. April and May of 2013 were the wettest on record, with the area experiencing unprecedented rainfall for the 2013 spring season in its 141 years of record (as of 2013; Hillaker 2013). The dataset collected by D3R and other instrumentation deployed for the IFloodS field experiment is, in part, being used to better understand and model the microphysical structure of precipitation and the hydrological response of the IFloodS region.

The IFloodS field campaign was the second operational D3R deployment and the first summer deployment of the D3R, demonstrating continuous operation in warm, humid conditions. The IFloodS experiment represents another milestone D3R event as it was the first deployment of the high-powered (40 W) Ka-band transmitter that replaced the temporary 1 W stand-in. The D3R operated continuously during the campaign, experiencing only one non-radar-related issue. This deployment demonstrated the D3R’s robustness and completed the operational validation of the radar over its anticipated operating temperature range [supplementing validation during winter operations in early 2012 as part of the NASA GPM Cold-Season Precipitation Experiment (GCPEx)]. The high-powered Ka-band solid-state transmitter was successfully demonstrated. The upgrade provides a 16-dB improvement in sensitivity compared to the 1 W transmitter. The upgrade increases the effective operational range of the Ka band and enhances the capability of the D3R for microphysical studies.

The D3R Ku-band radar’s rainfall mapping capability, also referred to as quantitative precipitation estimation (QPE), is examined. The IFloodS field campaign provides an ideal test bed for observation and cross validation with ground instruments. For comparison between ground instruments and the D3R observations, Autonomous Parsivel Unit (APU) disdrometers and rain gauges (Tokay et al. 2014) are considered here. For accurate microphysical investigation of hydrometeors, it is necessary to ensure that the radar estimates are well calibrated and corrected for attenuation effects. The D3R calibration results and preliminary attenuation-correction method for IFloodS are examined in section 3. The cross-validation results using dual-frequency observations and the other instruments deployed at IFloodS are compared in section 4. Quantitative precipitation estimates based on specific differential phase using Ku-band observations are presented in section 5. Finally, section 6 summarizes the D3R’s performance and results from the IFloodS ground validation field experiment.

2. D3R overview at IFloodS

During the field experiment, D3R collected over 45 000 scans in coordination with NPOL and the other deployed ground instruments as part of the IFloodS field campaign. There were observations of precipitation on 36 of the 46 days during the experiment. From these, there are no less than five severe convective thunderstorms, two squall lines, and four stratiform rain cases. Table 1 provides a qualitative summary of the D3R observations during the deployment.

Table 1.

Summary of D3R observations (by date) during the IFloodS field campaign.

Summary of D3R observations (by date) during the IFloodS field campaign.
Summary of D3R observations (by date) during the IFloodS field campaign.

The D3R was able to observe, unobstructed, from 90° clockwise to 230° in azimuth and from 0° to 90° in elevation. Observations were made along a line of ground instruments to the southeast that included an array of disdrometers and rain gauges. The D3R’s maximum range of 39.75 km allowed coordinated observations with a multitude of instrument types: the NPOL, four APUs, two rain gauges, and three 2D video disdrometers (2DVDs). The ground instruments mentioned were deployed along a radial at approximately 130.4° in azimuth from the radar site. This radial was selected for coordinated RHI scans between D3R and NPOL to maximize the opportunity for cross validation and microphysical investigation using the diverse array of instruments. A map of the location of D3R, NPOL, and the ground instruments discussed here is shown in Fig. 1. The geographic coordinates and the relative bearing and range from the D3R to the ground instruments are presented in Table 2.

Fig. 1.

This map shows the locations of a subset of instruments deployed during the IFloodS field campaign during May and June 2013 located in eastern Iowa. The location of NPOL and D3R is shown with the blue marker. The blue circle with a 40-km radius represents D3R’s max observational range. The red dots mark the locations of four disdrometers deployed within D3R’s domain.

Fig. 1.

This map shows the locations of a subset of instruments deployed during the IFloodS field campaign during May and June 2013 located in eastern Iowa. The location of NPOL and D3R is shown with the blue marker. The blue circle with a 40-km radius represents D3R’s max observational range. The red dots mark the locations of four disdrometers deployed within D3R’s domain.

Table 2.

Instrument locations within D3R’s observation domain. The range and bearings relative to D3R and NPOL are calculated using the World Geodetic System 1984 Earth model. APU01–APU04 are Autonomous Parsivel Unit disdrometers. SN25, SN35, and SN36 are 2DVDs. The gauges are tipping-bucket rain gauges.

Instrument locations within D3R’s observation domain. The range and bearings relative to D3R and NPOL are calculated using the World Geodetic System 1984 Earth model. APU01–APU04 are Autonomous Parsivel Unit disdrometers. SN25, SN35, and SN36 are 2DVDs. The gauges are tipping-bucket rain gauges.
Instrument locations within D3R’s observation domain. The range and bearings relative to D3R and NPOL are calculated using the World Geodetic System 1984 Earth model. APU01–APU04 are Autonomous Parsivel Unit disdrometers. SN25, SN35, and SN36 are 2DVDs. The gauges are tipping-bucket rain gauges.

3. D3R system calibration and attenuation correction

The calibration of radar systems, such as the D3R, is necessary for accurate retrievals (Vega et al. 2014). A variety of calibration methods are considered to provide consistent and redundant verification of the radar system parameters necessary for accurate and precise radar observations. These system parameters include transmitted power, receiver gain, antenna pointing angle, and antenna coalignment. The time scales in which the various calibration techniques are used depend on the rate of variation and the ability to track parameter drift during normal operation. The results specific to D3R’s deployment and operation in IFloodS are discussed, with a focus on demonstrating the accuracy of D3R’s meteorological parameter estimation.

a. System calibration

A full system characterization of the radar was performed using calibrated test equipment. This procedure is invasive and required breaking connections at reference planes in the radar system for injection and measurement of signals. Because of the complexity and invasive nature of this characterization, it is generally only performed during scheduled maintenance or when radar component changes occur (component changes may have slight performance deviations from unit to unit). The signal paths, for injection or monitoring, are broken at specific calibration planes, which allows for deterministic characterization of system performance between fixed points. After initial characterization is performed using the test equipment, the internal calibration hardware may be used to track slight variations in the system’s performance. The radar’s receiver provides a consistent monitor of the radar’s performance to estimate any parameter change in the transmitter and receiver paths from the calibration planes.

A metallic sphere is used as an overall, end-to-end radar system calibration target. The metal sphere provides a known radar cross section. The metal sphere is tethered to a balloon and released on a free-flight trajectory while being tracked and observed by the radar. The radar observations are compared against the expected radar cross section for valid observations. Deviation of the observations from the theoretical curve provides an overall radar calibration taking into account all components. Calibration of reflectivity Z with less than 1 dB absolute error is expected using this method. For the IFloodS experiment, a sphere calibration was performed using a 254-mm (10 in.) diameter metal sphere starting at 0000 UTC 14 June 2013. The sphere calibration was performed on a clear, dry, calm evening, and the sphere flight path was well clear of the sun. The sphere calibration results, in Fig. 2, show that the radar’s horizontal polarization reflectivity calibration Zh error is less than 0.5 dB for both frequencies.

Fig. 2.

D3R (top) Ku- and (bottom) Ka-band sphere calibration results from a metal sphere flight beginning at 0000 UTC 14 Jun 2013. The Ku-band results show a Zh calibration mean error of −0.002 dB with a std dev of 0.44 dB. For the Ka band, the Zh calibration mean error is −0.05 dB with a std dev of 0.51 dB. Ku- and Ka-band results show excellent reflectivity calibration.

Fig. 2.

D3R (top) Ku- and (bottom) Ka-band sphere calibration results from a metal sphere flight beginning at 0000 UTC 14 Jun 2013. The Ku-band results show a Zh calibration mean error of −0.002 dB with a std dev of 0.44 dB. For the Ka band, the Zh calibration mean error is −0.05 dB with a std dev of 0.51 dB. Ku- and Ka-band results show excellent reflectivity calibration.

For D3R, solar scans are performed primarily as a means of verifying antenna absolute alignment and coalignment of the Ku- and Ka-band antennas. The relative position of the solar radiation compared to the expected position of the sun is compared. When the centroid of the relative position is zero, the antenna positioner’s absolute calibration is perfect. Using the same technique, the antenna coalignment is verified with results presented in Fig. 3. For the IFloodS campaign, the coalignment is 0.01° in azimuth and 0.11° in elevation with an absolute alignment error less than 0.06° in azimuth and 0.15° in elevation. A more detailed discussion of the D3R antenna coalignment can be found in Vega et al. (2014).

Fig. 3.

The D3R antenna coalignment and positioner error verification results at 2247 UTC 18 May 2013. Coalignment of the antennas is within 0.01° in azimuth and 0.11° in elevation. The absolute position error is within 0.06° in azimuth and 0.15° in elevation.

Fig. 3.

The D3R antenna coalignment and positioner error verification results at 2247 UTC 18 May 2013. Coalignment of the antennas is within 0.01° in azimuth and 0.11° in elevation. The absolute position error is within 0.06° in azimuth and 0.15° in elevation.

Periodic differential reflectivity calibration is performed using vertical pointing observations (elevation of 90°) made through a full azimuth rotation and colloquially referred to as a “birdbath” scan. Birdbath calibrations are ideally performed when light rain is present over the radar. The full rotation during vertical pointing is necessary to nullify contribution of antenna side-lobe contamination from nearby ground clutter (Bringi and Chandrasekar 2001). Using this method, calibration of less than 0.2 dB can be attained. An example calibration scan performed at 0015 UTC 9 June 2013 is shown in Fig. 4. The melting layer can easily be identified in Fig. 4 by the simultaneous increase in reflectivity and decrease in copolar correlation for both Ku- and Ka-band observations. Range volume cells below the melting layer, containing observations of rain, are averaged over the full rotation and, if necessary, a bias is applied so = 0 dB. The bias is then saved and used for subsequent estimates of . The results of rain in Fig. 4 show a near-zero bias in that requires no adjustment.

Fig. 4.

Calibration results of from a birdbath scan performed at 0015 UTC 9 Jun 2013 in light stratiform rain. The radar parameters are averaged over a full azimuth rotation. The melting layer is identified by the drop in copolar correlation coincident with an increase in reflectivity at ~3 km. The for distances below the melting layer, representing rain observations, are adjusted to be 0 dB. From the results, the bias is ~0 dB for Ku and Ka bands.

Fig. 4.

Calibration results of from a birdbath scan performed at 0015 UTC 9 Jun 2013 in light stratiform rain. The radar parameters are averaged over a full azimuth rotation. The melting layer is identified by the drop in copolar correlation coincident with an increase in reflectivity at ~3 km. The for distances below the melting layer, representing rain observations, are adjusted to be 0 dB. From the results, the bias is ~0 dB for Ku and Ka bands.

Finally, a ray-by-ray calibration (a ray is a full integration period of N pulses) is performed using an internal calibration loop (Vega et al. 2014). This calibration loop provides a means to measure the transmitted power through the receiver path. Using the measurement of transmitted power, variations in the transmitted power and receiver gain can be corrected in the radar equation. Using the internal calibration, both Zh and can be corrected for transmitter power drift and receiver gain drift to maintain the accuracy demonstrated by both birdbath scans and sphere calibration.

Using this tiered calibration approach, the D3R system’s accuracy is maintained continuously during operation. The sphere calibration results show very good calibration of both the Ku- and Ka-band subsystems. Later discussions will compare meteorological estimates with other ground instruments in the IFloodS campaign to further substantiate the calibration quality of the D3R.

b. Attenuation correction

The attenuation correction of reflectivity and differential reflectivity observations is a challenging task in weather radars. However, dual-polarization-based attenuation-correction methods have been fairly successful and used widely in the recent years. Many operational systems have implemented differential phase–based attenuation correction (Bringi and Chandrasekar 2001). This can be further improved by more sophisticated attenuation-correction techniques such as the self-consistency approach from Gorgucci and Baldini (2007). The fundamental premise of this attenuation-correction technique is that horizontal and vertical polarization specific attenuation— and , respectively—can be estimated from specific differential phase for frequencies up to Ku band (Bringi and Chandrasekar 2001). We have introduced a preliminary attenuation correction for D3R observations at the Ku and Ka bands. A full analysis of attenuation correction is beyond the scope of this paper.

The drop size distributions (DSDs) observed by the four APUs deployed within D3R’s domain (see Fig. 1) are used to derive the theoretical relationship among , , and . The APU is an optical disdrometer based on single-particle extinction that measures both particle size and fall velocity (Tokay et al. 2014). Given the DSD data, assumed water temperature of 10°C, and drop shape ratio from Thurai et al. (2007), the dual-polarization radar measurements are simulated using scattering calculations (Waterman 1965). The simulated radar observations include and specific attenuation for each polarization (i.e., and ). During IFloodS, the APU temporal sampling resolution was configured for 1-min intervals. In this study, the DSD dataset from the APUs is divided into training and testing datasets. The training dataset is used to estimate a linear relationship between specific attenuation and specific differential phase. The training dataset includes 15 219 (1-min averaged) DSDs collected over nine precipitation days of the IFloodS campaign. The testing dataset, which is independent from the training dataset, is used to verify the accuracy of the model and includes 8631 (1-min averaged) DSDs. Linear regression is applied to the training dataset’s specific attenuation and to estimate the model’s coefficient (see Figs. 5a,b). The resulting relationship for the Ku-band horizontal polarization specific attenuation is

 
formula

and similarly, the Ku-band vertical polarization specific attenuation is estimated by

 
formula

where is in degrees per kilometer and the specific attenuation is in decibels per kilometer.

Fig. 5.

Simulation results at Ku band using APU-observed DSDs show the relationship between Ku-band specific attenuation and : (a) vs and (b) vs . (c),(d) Scattergrams of attenuations estimated using relations in Eqs. (1) and (2) vs attenuations directly computed using the testing DSD dataset, respectively.

Fig. 5.

Simulation results at Ku band using APU-observed DSDs show the relationship between Ku-band specific attenuation and : (a) vs and (b) vs . (c),(d) Scattergrams of attenuations estimated using relations in Eqs. (1) and (2) vs attenuations directly computed using the testing DSD dataset, respectively.

The testing dataset is used to evaluate the model error. The simulated specific attenuation is compared to the model’s estimated specific attenuation [Eqs. (1) and (2)] from the simulated Kdp. Figures 5c and 5d show scattergrams of the model’s estimated specific attenuation versus the intrinsic specific attenuation from simulation. To quantitatively evaluate the accuracy of Eqs. (1) and (2), the mean absolute percentage error (MAPE) is used, defined as

 
formula

where is the -based estimate of specific attenuation, A is the attenuation directly simulated from testing DSD data, and angle brackets denote the sample average. The resulting MAPE values for Eqs. (1) and (2) are 13.7% and 24.8%, respectively.

With the specific attenuation, the path-integrated attenuation (PIA) for the horizontal polarization is calculated by

 
formula

and the vertical polarization PIA is similarly calculated as

 
formula

where r is range, s is the propagation interval path length, and both and are in decibels. The attenuation-corrected horizontal reflectivity is estimated by

 
formula

and the attenuation-corrected differential reflectivity is estimated by

 
formula

The -based attenuation-correction methods assume Rayleigh scattering. Attenuation correction of is subject to overestimation because of deviation from the Rayleigh scattering assumption. The effects of Mie scattering reduce the accuracy of the attenuation correction for for regions of heavy rainfall. Mie scattering effects can result in a deviation of the actual backscatter and signal extinction cross sections from those predicted from the simple regressions in Eqs. (1) and (2). Similarly, the differential copolar backscatter phase shift cannot be neglected for bigger raindrops. The bias leads to a local overestimation of from the radar’s observations of differential phase .

For observation volumes with hydrometeors in the Mie scattering regime, the specific attenuation estimates are subject to bias. For Ku-band observations of convective precipitation with intrinsic reflectivity on the order of 45 dBZ or more, the estimation accuracy of specific attenuation may be reduced. For PIA estimation, the bias may not be significant given the overall natural distribution of reflectivity and expected reflectivity estimation accuracy (typically 1 dB). For differential PIA, this bias can exceed the expected accuracy for estimates of (typically 0.2 dB).

The dual-frequency radar configuration is designed to allow a broader class of precipitation to be observed. The Ka-band is intended for light rain and ice microphysical investigation, as attenuation effects are significant for moderate and heavy precipitation. Because of Mie scattering, the simple -based attenuation-correction method does not extend to Ka-band observations for rain. However, the specific attenuations at Ku- and Ka-bands are tightly connected. In this paper, the Ku-band is used to obtain an estimate of Ka-band specific attenuation. Figure 6a shows a scattergram of Ka- versus Ku-band specific attenuation from simulations of the training DSD dataset. For simplicity, the approximated Ka-band specific attenuation is a linearly scaled Ku-band specific attenuation given by

 
formula
Fig. 6.

(a) Scattergram of Ka- vs Ku-band attenuations based on simulations using training DSD dataset. (b) Scattergram of estimated attenuation using Eq. (8) vs theoretical attenuation at Ka band using the testing DSD dataset.

Fig. 6.

(a) Scattergram of Ka- vs Ku-band attenuations based on simulations using training DSD dataset. (b) Scattergram of estimated attenuation using Eq. (8) vs theoretical attenuation at Ka band using the testing DSD dataset.

Similarly, we use the testing DSD dataset to evaluate the accuracy of the coefficient in Eq. (8). Figure 6b shows a scatterplot of the Ka-band specific attenuation estimate from Eq. (8) versus the intrinsic Ka-band specific attenuation simulation from the testing DSD data. Evaluating the testing dataset using Eq. (8) yields an MAPE of 41.1% for Ku-band less than 4° km−1. This relationship is determined to be more accurate for the regions where specific attenuation is small; this indicates its viability for use in light and stratiform rain observations during the IFloodS campaign.

The attenuation-correction relationships derived here applied to the D3R observations. These results are compared to other instruments in the IFloodS domain in section 4. The relations in Eqs. (1), (2), and (8) are derived particularly for application to observations during the NASA IFloodS field experiment.

4. D3R cross validation with other instruments

The IFloodS field campaign presents a unique opportunity for multi-instrument comparison and cross validation of observations. By comparing observations of precipitation at multiple frequencies, and using ground observations from APU disdrometers, the integrity of the D3R observations will be further established. The microphysical features can also be evaluated in greater detail. As part of the cross validation, both Ku- and Ka-band observations will be compared, and NPOL observations will be considered. In addition, the APU-observed DSDs are used to simulate radar observations at the appropriate frequencies, which are then compared with real radar measurements.

a. Self-consistent attenuation correction and system calibration validation

A self-consistent validation of the radar parameters illustrates that the attenuation correction improves the D3R observations (Scarchilli et al. 1996). The self-consistent relation between reflectivity and differential reflectivity is examined for the D3R Ku-band and NPOL observations. The radar observations are compared to the intrinsic relationship expected using simulated radar parameters from APU-observed DSDs from the field campaign. The self-consistency comparison in Fig. 7 shows the two-dimensional distribution of versus ; the DSD-based, simulated radar moments are shown as black markers and the attenuation-corrected radar observations (at 1.4° elevation angle) are presented as a two-dimensional color density histogram. The radar observation and simulated results are in good agreement in the distribution range, providing validation that the results, after attenuation correction, are microphysically consistent.

Fig. 7.

(a) A histogram showing vs of D3R Ku-band observations at 0250 UTC 25 May 2013, after attenuation correction, shown as a color map with Ku-band-simulated results using APU-observed DSDs as black markers. (b) From the same observation time, a histogram of the NPOL observations shown as a color map with S-band-simulated results using APU-observed DSDs as black markers.

Fig. 7.

(a) A histogram showing vs of D3R Ku-band observations at 0250 UTC 25 May 2013, after attenuation correction, shown as a color map with Ku-band-simulated results using APU-observed DSDs as black markers. (b) From the same observation time, a histogram of the NPOL observations shown as a color map with S-band-simulated results using APU-observed DSDs as black markers.

b. Comparison of D3R Ku band and NPOL

The dual-polarization radar observations of the D3R Ku band and NPOL are compared using a plan position indicator (PPI) scan for 1.4° elevation at 2248 UTC 29 May 2013. The results are presented in Fig. 8. The figure compares a D3R Ku band and an NPOL PPI scan of heavy convection embedded in stratiform rain observed at the same elevation and same time. Four radar measurements are considered: , , , and copolar correlation . Attenuation correction is applied to the observations of reflectivity and differential reflectivity presented here.

Fig. 8.

Radar observations from PPI scans for 1.4° elevation at 2248 UTC 29 May 2013. (left) NPOL observations after attenuation correction. (right) D3R Ku-band attenuation-corrected observations. Note that the extinction of the D3R Ku-band signal results in a mismatch of observations in the southeastern quadrant. The NPOL range extends to 100 km while D3R range limit is 39.75 km.

Fig. 8.

Radar observations from PPI scans for 1.4° elevation at 2248 UTC 29 May 2013. (left) NPOL observations after attenuation correction. (right) D3R Ku-band attenuation-corrected observations. Note that the extinction of the D3R Ku-band signal results in a mismatch of observations in the southeastern quadrant. The NPOL range extends to 100 km while D3R range limit is 39.75 km.

A threshold of is used as a data filter to select observations of precipitation with sufficient signal-to-noise ratio (SNR). From inspection of Fig. 8, the NPOL and the D3R Ku-band results show good agreement in reflectivity. Differential reflectivity agrees well for areas of light and moderate precipitation. The D3R Ku-band observations are higher than expected along radials for ranges after propagation through areas of intense precipitation. This is because of overcompensation for differential attenuation by the linear attenuation-correction model that is not adequate for correction of in heavy precipitation at Ku band. The effects of heavy precipitation are evident in the differential phase observations and appear as a large gradient. For areas along the southeastern line of convection, observations have a reduced SNR because of signal attenuation. The effect of signal attenuation is evident by decreased and high gradient in . The D3R observations in the extreme southeast do not contain data because of signal extinction caused by the line of heavy precipitation.

A detailed ray profile comparison of the observed D3R Ku-band and NPOL reflectivities is shown in Fig. 9. The figure shows a ray plot comparison between the two radars after attenuation correction. Figure 9 also shows simulated S-band observations using the APUs in D3R’s observation domain. The radial selected is at azimuth 128.4°, which is in line with the APU disdrometers and at an elevation of 1.4°. The S-band simulated reflectivity, using DSD observations from APU02, APU03, and APU04 at the same time period, are shown as black markers in Fig. 9. The two radars and three APUs show excellent agreement in the reflectivity estimates. For this observation, the root-mean-square error (RMSE) between the D3R and NPOL observations is 2.8 dB for ranges after 13 km with observations having a copolar correlation greater than 0.9.

Fig. 9.

Comparison of the D3R Ku-band, NPOL, and S-band simulated reflectivity using APU-observed DSDs. All observations are at 2248 UTC 29 May 2013. The presented NPOL and D3R reflectivity are from the same radial. Note that signal extinction results in a mismatch between NPOL and D3R, followed by a loss of data after approximately the 35-km range.

Fig. 9.

Comparison of the D3R Ku-band, NPOL, and S-band simulated reflectivity using APU-observed DSDs. All observations are at 2248 UTC 29 May 2013. The presented NPOL and D3R reflectivity are from the same radial. Note that signal extinction results in a mismatch between NPOL and D3R, followed by a loss of data after approximately the 35-km range.

c. Dual-frequency ratio

Dual-frequency ratio (DFR) is the ratio of reflectivities at two radar frequencies. Dual-frequency ratio is an important observed parameter that aids in describing the microphysical characteristics of the radar volume. For the GPM DPR, DFR is an important parameter used in algorithms for microphysical retrieval and hydrometeor classification. The D3R design inherently provides temporal and spatial synchronization between the Ku and Ka bands. The simultaneous, aligned observations allow for direct calculation of DFR. The measured dual-frequency ratio is the difference between the observed reflectivity at two frequencies without attenuation correction; and DFR differ only in attenuation correction. For the D3R,

 
formula

The D3R Ku- and Ka-band observations of a convective cell with stratiform and ice aloft is shown in Fig. 10. The convective cell was observed at 0405 UTC 21 May 2013, and the Ku-band reflectivity, Ka-band reflectivity, and are shown in Fig. 10. It should be noted that no attenuation correction has been applied to the observations presented, and only data with are shown as a means to filter low SNR observations. In this case, attenuation due to precipitation does not significantly impact reflectivity observations for ranges near the radar where only light precipitation ( < 10 dBZ) is observed. Similarly, regions of ice do not have significant attenuation effects, as attenuation from ice is less pronounced than attenuation from water particles.

Fig. 10.

RHI observations at 0405 UTC 21 May 2013 of a convective cell with lofted ice. No attenuation correction has been applied and data with a copolar correlation coefficient greater than 0.8 are shown. The measured (a) Ku- and (b) Ka-band reflectivity is shown. (c) The D3R (Ku minus Ka band) results. The values at or above 13 dB are shown as the 13-dB value.

Fig. 10.

RHI observations at 0405 UTC 21 May 2013 of a convective cell with lofted ice. No attenuation correction has been applied and data with a copolar correlation coefficient greater than 0.8 are shown. The measured (a) Ku- and (b) Ka-band reflectivity is shown. (c) The D3R (Ku minus Ka band) results. The values at or above 13 dB are shown as the 13-dB value.

From the observation in Fig. 10, two radial reflectivity profiles are presented in Fig. 11. The observations have been averaged for longer duration (576 ms vs 64 ms in Fig. 10) to reduce the effects of measurement fluctuations. The presented reflectivity observations are echoes from ice crystals in the observed convective cell in Fig. 10. Figures 11a and 11c consider a radial covering 15.1°–16.2° for reflectivity and DFR, respectively. Figures 11b and 11d consider a radial covering 30.1°–31.1° for reflectivity and , respectively. The two different elevations observe different projections of the echo’s polarimetric signature onto the radar as well as different ice crystal densities. The observed values in Figs. 11c and 11d are consistent with the reported simulation values of DFR for Ku- minus Ka-band reflectivity observations of ice crystal aggregates (Tyynelä and Chandrasekar 2014).

Fig. 11.

Two radials representing the observations of lofted ice in Fig. 10. The (a),(b) observed Ku- and Ka-band reflectivity (without attenuation correction) and (c),(d) difference between the reflectivities . The observed is consistent with simulations of DFR for aggregates of ice crystals (Tyynelä and Chandrasekar 2014). The radials are integrated over 576 ms (1152 pulses) to estimate reflectivity and only observations with are shown.

Fig. 11.

Two radials representing the observations of lofted ice in Fig. 10. The (a),(b) observed Ku- and Ka-band reflectivity (without attenuation correction) and (c),(d) difference between the reflectivities . The observed is consistent with simulations of DFR for aggregates of ice crystals (Tyynelä and Chandrasekar 2014). The radials are integrated over 576 ms (1152 pulses) to estimate reflectivity and only observations with are shown.

Next, the dual-frequency ratio of rain will be considered from an RHI observation of stratiform rain at 1530 UTC 27 May 2013. The attenuation-corrected reflectivity profiles at Ku and Ka bands are presented in Figs. 12a and 12b, respectively. Examination of the reflectivity plots shows signal extinction as a result of signal attenuation in rain, which is apparent in the Ka-band observations when compared to the Ku-band results.

Fig. 12.

Reflectivity observations at 1530 UTC 27 May 2013 of an RHI scan through stratiform rain for (a) Ku and (b) Ka band after attenuation correction.

Fig. 12.

Reflectivity observations at 1530 UTC 27 May 2013 of an RHI scan through stratiform rain for (a) Ku and (b) Ka band after attenuation correction.

Figure 13 presents simulations of radar observations using APU-observed DSDs at 0° in elevation. The effects of Mie scattering result in a nonlinear relationship between Ku- and Ka-band reflectivity. The nonlinear behavior is characterized in greater detail by the dual-frequency ratio versus Ku-band reflectivity in Fig. 13 (middle). The median raindrop equivalent volume diameter is defined such that drops less than contribute to half the total rainwater content W (Bringi and Chandrasekar 2001) and can be estimated from the following relationships:

 
formula

where D is the raindrop equivolume diameter and is the density of water. With the relationship in Fig. 13 (right), can be estimated directly from DFR observations (Meneghini et al. 1997).

Fig. 13.

The intrinsic (left) Ka- vs Ku-band reflectivity, (middle) DFR vs Ku-band reflectivity, and (right) DFR vs . All results are from the simulation of APU-observed DSDs. The vertical black lines extend one std dev from the mean. The black line with red markers represents the mean. Histogram values with two or less observations are omitted for clarity but are used for estimates of the mean and errors.

Fig. 13.

The intrinsic (left) Ka- vs Ku-band reflectivity, (middle) DFR vs Ku-band reflectivity, and (right) DFR vs . All results are from the simulation of APU-observed DSDs. The vertical black lines extend one std dev from the mean. The black line with red markers represents the mean. Histogram values with two or less observations are omitted for clarity but are used for estimates of the mean and errors.

From observations presented in Fig. 12, a direct comparison of the D3R’s estimated reflectivities along an aligned radial is shown in Fig. 14. The radial is at approximately 1.5° in elevation with an integration period of 320 ms. Observations from 5 to 16 km in range showed good agreement between Ku- and Ka-band reflectivity estimates. At 15 km, simulated Ku-band reflectivity from APU02-observed DSDs is presented and is also in good agreement with observed reflectivities. Using the results of Fig. 13 (right) and the DFR estimates of Fig. 14b, the for the stratiform rain observations along the radial are between 1.3 and 2.0 mm, which is consistent with the estimate of from the APU02 disdrometer of 0.954 mm for the observation period.

Fig. 14.

(a) Ku- and Ka-band reflectivity along a radial covering 1.0°–1.5° elevation from Fig. 12. Five consecutive rays were averaged to produce the results. (b) The DFR (Ku- minus Ka-band reflectivity) for the same observations.

Fig. 14.

(a) Ku- and Ka-band reflectivity along a radial covering 1.0°–1.5° elevation from Fig. 12. Five consecutive rays were averaged to produce the results. (b) The DFR (Ku- minus Ka-band reflectivity) for the same observations.

5. Ku-band QPE

The D3R Ku-band provides an effective means to quantitatively estimate precipitation rate and total rainfall accumulation, especially in light rain. It has been shown that various rainfall algorithms can be derived with respect to dual-polarization radar measurements via the DSD information (Bringi and Chandrasekar 2001). However, the choice of rainfall relations at Ku band gets complicated since and must be corrected for attenuation before being used for rainfall estimation. Therefore, we consider only for rainfall estimation with the D3R. At Ku band, is the only estimator not affected by signal attenuation due to propagation through precipitation. In addition, is not sensitive to hail contamination or absolute calibration errors of the radar system (Bringi and Chandrasekar 2001).

A Ku-band relation is developed based on disdrometer observations collected during the IFloodS field experiment in Chen and Chandrasekar (2015). The relationship is given by

 
formula

The rainfall rate R is measured in millimeters per hour and has units of degrees per kilometer. As with the estimation of specific attenuation, the methodology proposed in Wang and Chandrasekar (2009) is implemented for the D3R to estimate .

Using Eq. (11), the rainfall rate is estimated every 1 min based on the scanning strategy for the D3R. To evaluate the performance of the D3R rainfall product, 5-min rainfall observations from ground instruments are used as a reference for cross comparison. The ground instruments considered are two pairs of collocated APUs and tipping-bucket rainfall gauges. For comparison between the instruments, the D3R’s rainfall is presented as 5-min rainfall accumulations of the 1-min estimated rainfall rates. To evaluate the D3R QPE performance in a quantitative manner, the normalized standard error (NSE) is defined as

 
formula

where is the radar estimate of rainfall, is the APU measurement of rainfall, and angle brackets denote averaging over the observation time period.

Figure 15 compares the 5-min rainfall accumulation observations on 28 May 2015. The D3R Ku-band estimates at two locations during the same time period are considered: the location of the APU02 and APU04 disdrometers. The observation period on 28 May 2013 is of a widespread stratiform event. It can be seen from Fig. 15 that the radar measurements agree very well with the observations from the APUs and rain gauge measurements. For the 28 May 2013 sample event shown in Fig. 15, it is concluded that the NSE for the 5-min rainfall accumulations is 36.9%. The NSE statistic is computed based on observations at both the APU02 and APU04 locations.

Fig. 15.

The 5-min rainfall accumulation estimates from 28 May 2013 using D3R Ku-band, APU02 and APU04 disdrometers, and rain gauge collocated with disdrometers. The observations are of widespread stratiform rain.

Fig. 15.

The 5-min rainfall accumulation estimates from 28 May 2013 using D3R Ku-band, APU02 and APU04 disdrometers, and rain gauge collocated with disdrometers. The observations are of widespread stratiform rain.

It should be mentioned that rain gauge observations are not used here for quantitative evaluation because the tipping-bucket gauges suffer from significant errors in light rain cases. The rainfall accumulation resolution of the gauge is 0.254 mm (0.01 in.). For light rain or high temporal sampling, the gauge resolution is not adequate and highlights the limited applicability of rainfall gauge data for use in light to moderate rainfall rates with high-temporal-resolution models. The rain gauge error during light rain at high temporal resolution is evident from review of Fig. 15.

Figure 16 shows example 5-min rainfall comparisons on 29 May 2013, characterizing a strong convective storm. It can be seen from Fig. 16 that the radar measurements agree with the observations from the APUs and rain gauge measurements very well for this convective case. The NSE for the 5-min rainfall accumulation is 49.4% using observations from both APU02 and APU04. Similarly, the NSE between radar rainfall accumulations and rain gauge measurements (collocated with APUs) is computed for this convective case as 45.6%.

Fig. 16.

The 5-min rainfall accumulation estimates from 29 May 2013 using D3R Ku-band, APU02 and APU04 disdrometers, and rain gauge collocated with disdrometers. The observations are of convective rain.

Fig. 16.

The 5-min rainfall accumulation estimates from 29 May 2013 using D3R Ku-band, APU02 and APU04 disdrometers, and rain gauge collocated with disdrometers. The observations are of convective rain.

The quantitative precipitation estimation results demonstrate D3R as an excellent platform for rain-rate estimation in light rain mainly because of the increased sensitivity of at Ku band over lower frequencies such as S band. The high-temporal-resolution QPE results presented demonstrate the capability of Ku-band observations for monitoring fast-moving precipitation with a high degree of accuracy.

6. Summary

The D3R was deployed as part of a focused multi-instrument ground validation field experiment to support the GPM program. D3R is a well-calibrated, accurate instrument for microphysical investigation of meteorological phenomena. The collocated D3R and NPOL enabled coordinated multifrequency observations of precipitation at fine spatial and temporal resolution. The coordinated observations in IFloodS extend across multiple, diverse, and densely deployed ground instruments for in situ microphysical observations and measurement of rain rates. This experiment provided the conclusive end-to-end system validation for the D3R as a deployable radar system.

The D3R’s various methods of calibration and verification were presented for the IFloodS field campaign. All results show the D3R system operates in a state with excellent calibration, which was confirmed by self-consistent validation derived from microphysical relationships and comparison with a variety of instruments deployed in the D3R’s observational domain. Using simulation results from APU disdrometer DSD observations, attenuation-correction relations specific to the region were determined for the D3R. The attenuation-correction performance was similarly verified against other instruments. The D3R shows very good agreement with both ground-based disdrometers as well as the collocated NPOL.

From the D3R observations, examination of the sample microphysical characteristics of the IFloodS domain is presented. Dual-frequency ratio observations of ice and rain were shown for simultaneous Ku- and Ka-band observations. The observations of ice were consistent with theoretical values for aggregate ice crystals. The DFR observation in light, stratiform precipitation yielded estimates of the median raindrop diameter and verified using an APU disdrometer’s estimate of . Finally, high temporal sampling of rainfall using the Ku-band observations was demonstrated for a stratiform rain case and a convective rain case. The normalized standard error results from the D3R comparison with APUs demonstrate that good performance is attainable using Ku-band observations for light rain observations with high temporal update rates.

Acknowledgments

The authors acknowledge the support of the GPM program, David Wolff and David Marks for NPOL support, and Mathew Schwaller and Walter Petersen for deployment and logistics support. In addition, the authors acknowledge all the participants of the IFloodS field campaign.

REFERENCES

REFERENCES
Bringi
,
V. N.
, and
V.
Chandrasekar
,
2001
:
Polarimetric Doppler Weather Radar: Principles and Applications.
Cambridge University Press, 664 pp.
Chen
,
H.
, and
V.
Chandrasekar
,
2015
:
Estimation of light rainfall using Ku-band dual-polarization radar
.
IEEE Trans. Geosci. Remote Sens.
, 53, 5197–5208, doi:.
Gorgucci
,
E.
, and
L.
Baldini
,
2007
:
Attenuation and differential attenuation correction of C-band radar observations using a fully self-consistent methodology
.
IEEE Geosci. Remote Sens. Lett.
,
4
,
326
330
, doi:.
Hillaker
,
H. J.
,
2013
: Preliminary Iowa weather summary–2013. Iowa Dept. of Agriculture, accessed 2 February 2015. [Available online at www.iowaagriculture.gov/climatology/weatherSummaries/2013/pas2013.pdf.]
Meneghini
,
R.
,
H.
Kumagai
,
J. R.
Wang
,
T.
Iguchi
, and
T.
Kozu
,
1997
:
Microphysical retrievals over stratiform rain using measurements from an airborne dual-wavelength radar-radiometer
.
IEEE Trans. Geosci. Remote Sens.
,
35
,
487
506
, doi:.
Scarchilli
,
G.
,
E.
Gorgucci
,
V.
Chandrasekar
, and
A.
Dobaie
,
1996
:
Self-consistency of polarization diversity measurement of rainfall
.
IEEE Trans. Geosci. Remote Sens.
,
34
,
22
26
, doi:.
Thurai
,
M.
,
G. J.
Huang
,
V. N.
Bringi
,
W. L.
Randeu
, and
M.
Schönhuber
,
2007
:
Drop shapes, model comparisons, and calculations of polarimetric radar parameters in rain
.
J. Atmos. Oceanic Technol.
,
24
,
1019
1032
, doi:.
Tokay
,
A.
,
D. B.
Wolff
, and
W. A.
Petersen
,
2014
:
Evaluation of the new version of the laser-optical disdrometer, OTT Parsivel2
.
J. Atmos. Oceanic Technol.
,
31
,
1276
1288
, doi:.
Tyynelä
,
J.
, and
V.
Chandrasekar
,
2014
:
Characterizing falling snow using multifrequency dual-polarization measurements
.
J. Geophys. Res. Atmos.
,
119
,
8268
8283
, doi:.
Vega
,
M. A.
,
V.
Chandrasekar
,
J.
Carswell
,
R. M.
Beauchamp
,
M. R.
Schwaller
, and
C. M.
Nguyen
,
2014
:
Salient features of the dual-frequency, dual-polarized, Doppler radar for remote sensing of precipitation
.
Radio Sci.
, 49, 1087–1105, doi:.
Wang
,
Y.
, and
V.
Chandrasekar
,
2009
:
Algorithm for estimation of the specific differential phase
.
J. Atmos. Oceanic Technol.
,
26
,
2565
2578
, doi:.
Waterman
,
P. C.
,
1965
:
Matrix formulation of electromagnetic scattering
.
Proc. IEEE
,
53
,
805
812
, doi:.

Footnotes