Observations collected during the 25 February 2020 deployment of the Vapor In-Cloud Profiling Radar at the Stony Brook Radar Observatory clearly demonstrate the potential of G-band radars for cloud and precipitation research, something that until now was only discussed in theory. The field experiment, which coordinated an X-, Ka-, W- and G-band radar, revealed that the Ka–G pairing can generate differential reflectivity signal several decibels larger than the traditional Ka–W pairing underpinning an increased sensitivity to smaller amounts of liquid and ice water mass and sizes. The observations also showed that G-band signals experience non-Rayleigh scattering in regions where Ka- and W-band signal do not, thus demonstrating the potential of G-band radars for sizing sub-millimeter ice crystals and droplets. Observed peculiar radar reflectivity patterns also suggest that G-band radars could be used to gain insight into the melting behavior of small ice crystals.
G-band signal interpretation is challenging, because attenuation and non-Rayleigh effects are typically intertwined. An ideal liquid-free period allowed us to use triple-frequency Ka–W–G observations to test existing ice scattering libraries, and the results raise questions on their comprehensiveness.
Overall, this work reinforces the importance of deploying radars (1) with sensitivity sufficient enough to detect small Rayleigh scatters at cloud top in order to derive estimates of path-integrated hydrometeor attenuation, a key constraint for microphysical retrievals; (2) with sensitivity sufficient enough to overcome liquid attenuation, to reveal the larger differential signals generated from using the G-band as part of a multifrequency deployment; and (3) capable of monitoring atmospheric gases to reduce related uncertainty.
Over the past 20 years, millimeter-wavelength radars have become the
instrument of choice for the study of cloud and precipitation. Today, radars
operating at 35 and 94 GHz frequencies are routinely operated at
ground-based observatories (e.g., U.S. Department of Energy Atmospheric
Radiation Measurement, ARM, user facilities; Stokes and Schwartz, 1994;
and the Aerosol, Clouds and Trace Gases Research Infrastructure, ACTRIS;
Pappalardo, 2018) and from a variety of ship-based and airborne
platforms (Kollias et al., 2007b). In space, the CloudSat 94 GHz cloud
profiling radar (CPR) has been operating since May 2006 (Stephens et al.,
2002) and the Earth Cloud Aerosol and Radiation Explorer (EarthCARE), the
first spaceborne Dopplerized cloud profiling radar, is expected to be
launched in 2023 (Illingworth et al., 2015). Reasons for the popular use
of millimeter-wavelength radars include the facts that this frequency range is much
more sensitive (in contrast to centimeter-wavelength radars) to cloud droplets and
small ice crystals and that it allows for the collection of observations at
excellent spatial resolution (
Fortunately, attenuation and scattering of radar signals are frequency
dependent such that they can be exploited to retrieve independent
information about particle mass, habit, or size, depending on the character
of scattering. For instance, the observations from two (or more) radar
frequencies within the same scattering regime, but different absorption
regime, can be combined to isolate differential attenuation signals useful
for the retrieval of liquid water content (Hogan et al., 2005; Zhu et
al., 2019). Alternatively, observations collected at two (or more) radar
frequencies experiencing similar signal absorption, but differential
scattering, can be combined to reveal information about ice crystal habit
and size (Kneifel et al., 2015). That being said, modern multifrequency
pairings are limited because (i) they rely on frequencies that experience
little differential attenuation in liquid clouds causing larger liquid water
content retrieval uncertainty and (ii) they do not produce differential
scattering signals for hydrometeors smaller than 800
In response to these limitations, the research community has expressed an interest in developing radars operating at higher frequencies in the so-called G-band (110–300 GHz; Battaglia et al., 2014). Compared to a Ka–W (35–94 GHz) frequency pair, a Ka–G frequency pair should experience measurable differential attenuation at smaller water mass amounts and non-Rayleigh scattering at smaller particle sizes (e.g., Battaglia et al., 2014; Hogan and Illingworth, 1999; Lhermitte, 1988). What is more, a Ka–G frequency pair is expected to always produce differential signals larger than that of traditional pairs, thus increasing the resilience to noise and precision of hydrometeor mass or size retrievals. Other applications of G-band and sub-millimeter-wavelength radars come from the presence of a water vapor absorption line at 183 GHz. By tuning the radar frequency between positions of higher and lower absorption near a water vapor line, (e.g., 183 or 325 GHz), G-band radars can be used to profile water vapor using the differential absorption radar (DAR) technique (Battaglia and Kollias, 2019; Lebsock et al., 2015; Roy et al., 2018; Cooper et al., 2018).
Surprisingly, the first G-band radar for meteorological applications
was only built in the late 1980s; McIntosh et al. (1988) designed a
215 GHz non-Dopplerized high-power extended interaction klystron transmitter
radar system and demonstrated that it was capable of making backscatter
measurements from terrain targets at ranges of several kilometers under
normal atmospheric conditions. Mead et al. (1989) attempted to use the
system to characterize clouds and fog and realized that it did not possess
sufficient sensitivity to detect clouds and light precipitation. Thirty
years past before we saw the development of the next generation of G-band
radars. In 2018, thanks to significant technological advancements in radar
front ends, mixers and low-power wide-bandwidth solid-state G-band sources,
the Jet Propulsion Laboratory (JPL) developed a highly sensitive
non-Dopplerized frequency-modulated continuous-wave (FMCW) G-band radar
tunable from 167 to 174.8 GHz (i.e., DAR; Cooper et al., 2020; Roy et
al., 2018). The system, named Vapor In-Cloud Profiling Radar (VIPR), was
deployed during 7 d in April 2019 at the ARM Southern Great Plains
facility. This first deployment aimed to evaluate VIPR's ability to exploit differential absorption
signatures to retrieved in-cloud humidity profiles (Roy et al., 2020).
VIPR's retrievals were evaluated against coincident measurements from ARM
water vapor sensors, with the primary comparison coming from frequently
launched radiosondes. Furthermore, VIPR's integrated water vapor measurement
capabilities in clear air columns were investigated by comparing with both
radiosonde and Raman lidar profiles. These comparisons highlighted VIPR's
ability to profile in-cloud water vapor with high resolution (
VIPR was deployed again in February 2020 at the Stony Brook Radar Observatory (SBRO) to demonstrate the capability of G-band radars for characterizing rain, ice crystals and snow. There, it collected observations alongside three radars operating respectively at 9.4, 35 and 94 GHz, thus providing first light multifrequency radar observations including the G-band. Here, we present the results of the quadruple-frequency radar field experiment that sampled a frontal system accompanied by prefrontal cirrus clouds followed by ice transitioning into light warm rain. The presented work demonstrates the value of using a G-band radar as part of a multifrequency radar observatory and underlines some important lessons learned and requirements needed for taking full advantage of G-band radar observations for cloud and precipitation microphysical studies.
The SBRO is a fenced-in facility located on the edge of Stony Brook
University's commuter parking lot located on Long Island, New York state,
USA (40
This section provides details specific to the operation of these systems during the deployment of the G-band VIPR radar on 25 February 2020, beginning with pictures of the instrument layout during the field deployment (Fig. 1). Like the pictures illustrate, all systems were installed in very close proximity in order to facilitate multifrequency retrievals.
VIPR is a first-of-its-kind solid-state G-band differential absorption radar (DAR). Its technical specifications are described in detail in Cooper et al. (2020) and Roy et al. (2020).
When it was deployed at the SBRO, VIPR transmitted 300 mW of power at 167
and 174.8 GHz, and it was operated in frequency-modulated continuous-wave
(FMCW) mode with a chirp bandwidth of 10 MHz and corresponding range
resolution of 15 m. With a single-pulse coherent integration time of 1 ms,
VIPR realizes a noise-equivalent reflectivity of
Around noon on 24 February 2020 (1 d before the official field deployment)
VIPR was installed near but outside a large shipping container. That day,
VIPR was mostly operated off zenith for calibration purposes (details in
Sect. 3.2). On the official deployment day of 25 February 2020, VIPR continued
operating next to the large shipping container but this time in vertically
pointing mode. Following the onset of rain that day, VIPR's transmitter had
to be turned off on a number of occasions to wipe water droplets off of the
radar antenna (gaps seen in Fig. 8c). In some instances, we noted that
strong radar returns from close-range rain caused an increase in the system
noise floor of up to 20 dB stemming from broadband phase noise in the
transmitted signal (Cooper et al., 2020). At 20:41 UTC, following the
onset of heavier surface rain, VIPR was moved inside the adjacent container
and pointed 40
ROGER, named after late radar pioneer Roger Lhermitte, is a refurbished
version of the W-band (94.8 GHz) radar previously integrated into the Center
for Interdisciplinary Remotely-Piloted Aircraft Studies Twin Otter aircraft
(Mead et al., 2003). ROGER is a single-polarization 0.3
During VIPR's deployment, ROGER was set to operate with a range gate spacing
of 30 m and collected a full radar Doppler spectrum every 4 s, achieving a
sensitivity of roughly
KASPR is a mechanically scanning 0.3
For most of the VIPR deployment, until 21:00 UTC to be exact, KASPR was
operating vertically pointing with 15 m range resolution and 13.6 km maximum
range. It only transmitted horizontally polarized waves and collected a full co-polar
and cross-polar radar Doppler spectrum every 1 s, achieving a sensitivity
of roughly
SKYLER is a dual-polarization X-band low-power phased-array radar with an
antenna beamwidth of 1.98
During VIPR's deployment, SKYLER was only operated between 18:00–24:00 UTC.
SKYLER was mounted on a rotation table installed on a mobile truck's flatbed
oriented facing upward to enable the collection of vertically pointing
observations. SKYLER was set to operate with a 2
Because SKYLER's receiver blanking parameters were incorrectly set, its reflectivity observations collected below 1.25 km are biased low (hatching in Fig. 8a). Knowing that this bias could be corrected for, we elected to display these observations but only performed quantitative retrievals using SKYLER observations collected above 1.25 km.
One of the SBRO Parsivel
The US National Weather Service (NWS) performs balloon-borne radiosonde measurements twice a day (00:00 and 12:00 UTC) from the Brookhaven National Laboratory campus in Upton, NY, 22 km east of the SBRO location. On 25 February 2020, SBRO staff and Stony Brook University students also launched two GRAW DFM-90 radiosondes at 01:46 and 15:44 UTC directly from the SBRO.
A Stream Line XR Doppler lidar and a Lufft CHM 15k backscatter lidar were also operated during the field experiment. The Doppler lidar was set to operate at 60 m range resolution and 1 s temporal resolution, providing estimates of air motion in the subcloud layer (not analyzed as part of the current study), while the backscatter lidar was set to operate with a 15 m range resolution and 15 s temporal resolution for monitoring the location of liquid layers.
Before they can be used to gain insight into atmospheric liquid and/or ice, high-frequency radar measurements must be post-processed to remove signal attenuation caused by atmospheric gases. Also, and especially in the context of multifrequency analysis, radar signals should be calibrated to improve the accuracy of quantitative retrievals. This section describes the steps used to postprocess and calibrate the radar observations collected by the VIPR, ROGER, KASPR and SKYLER radar and how these corrected observations are combined to conduct a multifrequency analysis.
When thermodynamic information is available, radio-wave propagation models
can be used to estimate radar signal attenuation by atmospheric gases. Here
we use the MPM93 model, an updated version of the millimeter-wave
propagation model described by Liebe (1985) and Liebe et al. (1993) to
compute two-way gas attenuation of X-, Ka-, W- and G-band signals for the
conditions that occurred at 12:00 and 15:44 UTC on 25 February
2020 when two radiosondes were launched. Figure 2a and b show the profiles
of temperature, dew-point temperature and humidity recorded at the US NWS site
22 km east of SBRO at 12:00 UTC and at the SBRO at 15:44 UTC. The two-way
gas attenuation profiles depicted in Fig. 2c confirm that millimeter radar
signals, particularly at G-band, experience non-negligible gas attenuation.
For this particular mid-latitudinal winter case, we estimate two-way gas
attenuation at 11 km to reach
From sounding observations collected on 25 February 2020 at 12:00 UTC at the US NWS Upton site (22 km east of SBRO; solid lines) and at 15:44 UTC from the SBRO (dashed lines); profile of
Since the following analysis focuses on quantifying hydrometer properties, we correct all radar signals for two-way gas attenuation using the profiles derived above. The profiles estimated using the 12:00 UTC sounding are used to correct radar measurements collected before 13:52 UTC, while the ones estimated using the 15:44 UTC sounding are used to correct the rest of the radar measurements. The variability between the consecutive profiles can be used to get a sense of the uncertainty associated with using only two soundings to correct the daylong radar dataset.
On 24 February 2020 (1 d before the official field experiment), VIPR's calibration was verified using the methodology described in Roy et al. (2020); the exercise required hanging a small calibration sphere between two light posts roughly 200 m from the SBRO. KASPR's calibration is similarly checked twice a year by SBRO staff using a corner reflector located 300 m away from the SBRO.
SKYLER, ROGER and KASPR measurements are also sporadically calibrated using
Parsivel
Based on measurements from the Parsivel
For this analysis, Parsivel
Ideally, multifrequency analysis would be performed using perfectly time-matched and volume-matched observations in order to be able to attribute any signal differential to the properties of the hydrometeor population. Unfortunately, previous work has shown that perfectly matching radar observations is extremely challenging even for radars installed on the same pedestal (Kollias et al., 2014). Observation volume differences unavoidably occur as a result of using different radar frequencies, which require the use of different transmitting configurations such as pulse width, pulse repetition frequency and number of samples for integration. Temporal and vertical averaging of radar data on a common grid has been used in an attempt to alleviate radar observation mismatching.
Here we co-gridded the post-processed radar observations from all four
radars on a joint 15 m, 4 s resolution grid. The gridded observations are
subsequently averaged in time in 60 s increments to reduce noise. The
denoised radar observations are used to estimate the dual-wavelength ratio
(
On 25 February 2020, following the movement of a surface trough and associated low-pressure system, a stationary front established itself over the SBRO. The four profiling radar systems and the two lidar systems operating at the time observed the transition from prefrontal cirrus to rain associated with this system. The following sections discuss key findings attributable to the deployment of a G-band radar as part of a multifrequency radar deployment in these two weather regimes.
The radar and lidar observations displayed in Fig. 4 reveal that a deck of
prefrontal cirrus clouds, whose top extended near 9–10 km, advected over the
observatory between 07:00 and 10:00 UTC. Observations from KASPR, ROGER and
VIPR show that the thickness of the cirrus layer varied over time between
Time–height of radar reflectivity measured by
Differences in radar reflectivity measured by the Ka- (Fig. 4a), W- (Fig. 4b) and G-band (Fig. 4c) radars are a direct result of differences in signal attenuation and scattering, which can be best visualized in dual-wavelength ratio (DWR) space. Figure 5 shows DWR estimated using the traditional Ka–W pair (panel a), the W–G pair (panel b) and the Ka–G pair (panel c). These first light DWR observations involving G-band confirm all the advantages predicted by scattering theory.
Time–height of dual-wavelength ratio from
Profiles taken at 08:00 UTC during the ice cloud period,
Focusing below
Interpreting and performing retrievals from DWR observations always requires
considering the interplay of signal attenuation and non-Rayleigh scattering
(Tridon et al., 2013). Observations collected during the period around
08:42 UTC highlight this important limitation of DWR analysis targeting the
characterization of ice crystals. The lack of convergence at 0 dB in the
profile extracted at 08:42 UTC suggests the presence of considerable water
condensate (liquid and/or ice) mass in the atmospheric column (Fig. 6d).
Backscatter lidar observations do allude to the presence of liquid layers
(of unknown depth) over that period (Fig. 4e). Tridon et al. (2020)
suggest that if DWR reaches a constant value with height (a.k.a. a Rayleigh
plateau), the DWR of this plateau can be used to estimate integrated water
condensate mass within the layer. In this particular profile, the Ka–W
pair reached a clear Rayleigh plateau at 5 km showing a 1 dB DWR loss to
hydrometeor attenuation. We argue that both the Ka–G pair and the W–G pair
also reached a Rayleigh plateau near 6.8 km showing in the neighborhood of
3.5 dB DWR loss to hydrometeor attenuation. This signal could be qualified
as being the first quantitative hydrometeor mass signal recorded at G-band.
Because ice and snow attenuation considerably increase when moving from the
W- to the G-band reaching one-way values of 0.9, 2.5, and 8.7 dB m
For observations collected
The DWR profile shown in Fig. 6b taken from observations collected at 08:00 UTC shows a contrasting situation where G-band signals can be directly used
for ice microphysical retrievals. In that profile DWR is seen to converge to
0 dB such that differential signal across the column can be interpreted from
resulting exclusively from non-Rayleigh scattering. Under such conditions
DWR can be related to ice crystal size given the proper ice scattering
library. Kneifel et al. (2015) initially proposed using DWR
The first period (07:45–08:12 UTC) depicted in Fig. 7a corresponds to the
period that presented a high-DWR slanted feature (referring back to Fig. 5)
and a thin liquid layer (referring back to the lidar backscatter
observations of Fig. 4). Plotting the radar observations in DWR–DWR space
can help determine if the amount of liquid attenuation caused by this thin
liquid layer is significant thus preventing us from inferring particle habit
directly from the gas attenuation corrected and calibrated radar
measurements. To be exact, a clustering of the DWR–DWR observations
collected in the upper part of the cloud (between 5.750–7.00 km) near the 0,0
point would indicate an absence of
signal attenuation. For this particular period, a 0.5 dB offset is seen in the contours on Fig. 7a,
suggesting that a slight adjustment should be made to the observed DWR
before they can be interpreted in terms of differential scattering and used
to infer particle habit. Even with this slight adjustment, we find that the
scattering calculation results only partially match the DWR–DWR signatures
observed leaving a noticeable gap in the high (
The second period (08:12–09:12 UTC) depicted in Fig. 7b corresponds to the
period containing non-negligible attenuation by water condensates. This
period also presents a broad high-DWR area between 2 and 5.5 km altitude
(referring back to Fig. 5). The offset from 0,0 DWR
The radar observations displayed in Fig. 8 show the light surface rain
episode that occurred following the frontal passage between 18:00 and 18:30 UTC. Observations from KASPR allow us to establish that the cloud sustaining
the rain extended up to 8 km. The bright band observed by all radars,
although notably different, is suggestive of a transition from ice particle
to liquid water near 2 km. This idea is substantiated by radiosonde reports
that place the 0
Time–height of radar reflectivity measured by
Difference in radar reflectivity measured by the X- (Fig. 8a), Ka- (Fig. 8b), W- (Fig. 8c) and G-band (Fig. 8d) radars during the period and specifically at 18:07 UTC (Fig. 9a) are a direct result of difference in signal attenuation and scattering.
Profiles taken at 18:07 UTC during the rain period;
Differential signal scattering explains the progressive reduction in the
overall radar reflectivity factor measured by the X-band SKYLER, Ka-band
KASPR, W-band ROGER and G-band VIPR.
During this light rain period, we expect the 3.2 cm wavelength X-band signal to experience Rayleigh scattering; especially considering the range of particle diameters measured by the disdrometer (Fig. 3a). In the Rayleigh
scattering regime, radar backscattering cross section (
Although G-band signals should allow for sizing smaller raindrops since
they experience non-Rayleigh scattering at smaller droplets sizes (compared
to longer wavelengths), one must remember that G-band signals also
experience non-negligible liquid attenuation. Theoretical calculations
suggest that extinction coefficients at 94 and 220 GHz rapidly increase for
particles with size up to
Like we saw in ice clouds, large DWR
Several gaps in cloud and precipitation remote sensing still exist especially at mid and high latitudes
(Battaglia et al., 2020). Radars at frequencies above 100 GHz are now
technologically feasible as proved by the VIPR system recently built by
JPL. This work presents multifrequency (X-, Ka-, W- and G-band) radar
observations from a field experiment at the Stony Brook Radar Observatory
(SBRO). Albeit short, the field experiment provided a long-sought-after
first light demonstration of the potential of multiwavelength radar
observations that include G-band for the characterization of ice crystals,
snow and rain. Besides confirming expectations derived from scattering
theory, the field experiment revealed a number of considerations relevant to
the deployment of G-band systems.
The observations clearly demonstrate that G-band radars can be made
sensitive enough to probe clouds and light precipitation in spite
of the strong water vapor attenuation occurring at this frequency. The large
sensitivity of G-band radars can in part be explained by improvements in
radar gain with increased frequency; all else equal, for a fixed aperture
size, radar sensitivity improves by 24 dB going from 10 to 170 GHz. Since G-band signals are especially prone to attenuation by water vapor, we
recommend that G-band radars targeting the characterization of clouds and
precipitation should have differential absorption capabilities in order to
avoid confounding effects due to water vapor attenuation. This could be
achieved through the use of interlaced pulses whose frequency would range
around a water vapor absorption line. The exact frequency range should
ideally be tuned to the specific water vapor condition like those proposed in
Roy et al. (2020), Cooper et al. (2020), and Battaglia and Kollias (2019). The observations presented here reinforce the idea that the sensitivity of
all the radar systems involved in future multiwavelength radar studies
should be sufficient to allow the detection of the Rayleigh plateau near the
top of ice clouds (or near the base if using an airborne system); that is
necessary to ensure that we have a robust estimation of the differential
(dual-wavelength) path-integrated liquid attenuation (Tridon et al.,
2020). For rain studies as well, G-band radar sensitivity should be large
enough to allow signals to penetrate through the rain shaft despite
attenuation by liquid water reaching several decibels. Nominally radar systems
should be capable of detecting unattenuated reflectivity as weak as The observations collected during this experiment confirm that the Ka–G pair
generates the strongest differential reflectivity signal, with observed
values of DWR reaching up to 13 dB in ice regions, which is 4 dB larger than
traditionally Ka–W pairs. The increased differential signal should allow for
increased retrieval confidence, especially in low-liquid-water-content
regions and/or for small particle sizes. The steep DWR In the absence of Ka–W differential signals, observations of non-Rayleigh
differential scattering signals at Ka–G and W–G demonstrates the potential
of G-band radars for sizing smaller ice particles. An ideal case observed during the field experiment allowed us to investigate
ice crystal habit. DWR–DWR observed by the Ka–W–G trio were compared to
estimates made using several scattering libraries. The scattering libraries
tested could only provide a partial explanation of the scattering properties
of the ice crystals observed with gaps in the high ( The observations collected during a melting event suggest that G-band radars
can detect radar bright bands. The character of this bright band is likely
indicative of the melting behavior of smaller ice crystals. In rain, the G-band radar reflectivity values are several orders of
magnitude lower than those measured by the W-band, Ka-band and X-band radar
systems creating measurable DWR signal. Interpreting these differential
signals may be challenging, because they result from both differential
scattering and attenuation. In large particle regimes where W- and G-band
signals experience similar attenuation by liquid, DWR
Longer datasets with similar measurement capabilities are needed to fully
assess the potential and challenges associated with using non-Dopplerized
G-band radar observations for the study of clouds and precipitation systems.
Such observations can in turn be used to raise the technology and science
readiness levels of spaceborne G-band systems. G-band radar signals coming from an above-cloud vantage point
should suffer from less signal attenuation than ground-based systems; that is
because water vapor and rain are typically concentrated in the lowest part
of the atmosphere. The reduced signal attenuation for airborne and spaceborne G-band radars should drive a less stringent
sensitivity requirement (
The datasets collected at the SBRO during the field experiment are available at
KL, MO, AB, RJR, KBC and PK were actively involved in the field experiment; KL operated the Doppler lidar, MO operated the KASPR, and RJR and KBC together operated VIPR. MO and RD performed initial data exploration work. PK finalized the data analysis. AB's and PK's inputs were instrumental in interpreting the radar signatures observed. KL lead the writing of the final version of this article. All members of the team reviewed and added to this final version.
The authors declare that they have no conflict of interest.
This research also used the Advanced Leicester Information and Computational Environment (ALICE) High Performance Computing Facility at the University of Leicester, UK. We would like to thank the Brookhaven National Laboratory staff and Stony Brook University students who assisted during the field experiment; special thanks go to Edward Luke for operating SKYLER and to Zeen Zhu, Samantha Nebylitsa, Jacob Segall and Kristofer Tuftedal for launching radiosondes from the SBRO.
This research has been supported by the Brookhaven National Laboratory (LDRD, grant no. 20-002 EE/EBNN), the National Science Foundation (grant no. 1841246), the National Aeronautics and Space Administration (grant no. 80NM0018D0004) and the UK-Centre for Earth Observation Instrumentation under the Grace project.
This paper was edited by Stefan Kneifel and reviewed by two anonymous referees.