The U.S. Department of Energy (DOE) Atmospheric Radiation Measurements (ARM) facility has been at the forefront of millimeter-wavelength radar development and operations since the late 1990s. The operational performance of the ARM cloud radar network is very high; however, the calibration of the historical record is not well established. Here, a well-characterized spaceborne 94 GHz cloud profiling radar (CloudSat) is used to characterize the calibration of the ARM cloud radars. The calibration extends from 2007 to 2017 and includes both fixed and mobile deployments. Collectively, over 43 years of ARM profiling cloud radar observations are compared to CloudSat and the calibration offsets are reported as a function of time using a sliding window of 6 months. The study also provides the calibration offsets for each operating mode of the ARM cloud radars. Overall, significant calibration offsets are found that exceed the uncertainty of the technique (1–2 dB). The findings of this study are critical to past, ongoing, and planned studies of cloud and precipitation and should assist the DOE ARM to build a legacy decadal ground-based cloud radar dataset for global climate model validation.
The first millimeter-wavelength cloud radars (MMCRs; Moran et al., 1998) of the U.S. Department of Energy Atmospheric Radiation Measurement (ARM) facility were installed at the Tropical Western Pacific (TWP), Manus, and Southern Great Plains (SGP) sites in 1996. Since then, the ARM facility has been at the forefront of short-wavelength radar development and operations for over 2 decades (Kollias et al., 2016). In the beginning, emphasis was placed on demonstrating high operational stability and developing standard hydrometeor location and spectral products (Clothiaux et al., 2001; Kollias et al., 2007b). The ARM facility MMCR calibration efforts were limited to subcomponent characterization (i.e., antenna gain), monitoring of the transmitted peak power, and infrequent detailed characterization of the radar receiver by injecting signal with known amplitude. In 2005, the ARM facility started the deployment of its mobile facilities and the gradual modernization of the MMCR receiver. This led to the development of the W-band ARM Cloud Radar (WACR). In 2009, the ARM facility embarked on a significant expansion of its radar facilities (Mather and Voyles, 2013). The expansion included the addition of scanning millimeter- and centimeter-wavelength radars with Doppler and polarimetric capabilities (Kollias et al., 2014a; North et al., 2017) and the development of the next-generation profiling cloud radar, the Ka-band ARM Zenith Radar (KAZR) and its upgraded second generation (KAZR2).
Part of the motivation for the ARM radar expansion was to improve cloud microphysical retrievals through the use of dual-wavelength ratios, that is, making use of the relative difference in radar scattering at different wavelengths. This difference signal is often only a few decibels and as one might expect, this requirement brought the characterization of the ARM radar calibration into focus. Early comparisons between collocated profiling ARM cloud radar indicated differences in reported radar reflectivity profiles. This hardly came as a surprise to those involved in radar characterization (Atlas, 2002). Soon after the National Aeronautics and Space Administration (NASA) Tropical Rainfall Measuring Mission (TRMM) spaceborne radar was in orbit, its remarkable stability made it a calibration standard and its comparison to the ground-based observations of the Weather Surveillance Radar-1998 Doppler (WSR-88D) network uncovered several issues with the calibration of the radars despite the mandate of the WSR-88D network on quantitative precipitation estimation and the implementation of routine calibration procedures (Bolen and Chandrasekar, 2000). On the other hand, establishing routine calibration procedures based on engineering measurements or natural targets for the ARM profiling cloud radars is a far more challenging task. The systems are only vertically pointing, thus making the use of corner reflectors or metal spheres difficult; are designed with sensitive receivers that can detect very low radar reflectivity targets but saturate in rain, thus making the use of disdrometers challenging (Gage et al., 2000); and operate in climate regimes that often have no or little precipitation and suffer from considerable gaseous and hydrometeor attenuation (Kollias et al., 2005, 2007a). Furthermore, the four different profiling cloud radars (MMCR, WACR, KAZR, and KAZR2) were deployed in different climatological regimes, for small periods of time (9–24 months mobile deployments) and often with no gaps between deployments, thus making it even more challenging to develop calibration standards. At present, the ARM facility employs a larger radar operation and engineering group and has set procedures for characterizing the ARM radars using a combination of subsystem calibrations, corner reflectors, and natural targets. However, these methods are still not fully operational today and certainly not applicable to the historic ARM profiling cloud radar dataset that spans over 2 decades.
Luckily, NASA's CloudSat mission, a 94 GHz spaceborne cloud profiling radar
(CPR) was launched in April 2006 (Tanelli et al., 2008) on a circular
sun-synchronous polar orbit providing coverage from 82
In Sect. 2, the ARM facility cloud radars are presented and the Protat et al. (2011) methodology is revised and improved. Section 3 presents the results from the application of the calibration procedure to almost the entire record of ARM profiling cloud radar observations at the fixed and mobile sites from 2007 to the end of 2017 (a total of 43.5 years of radar observations). Finally, Sect. 4 presents a summary of our findings and their implications. The application of the technique to such a diverse set of radar systems and locations is expected to demonstrate the applicability of this approach to existing profiling radar networks such as the ARM facility and the future European research infrastructure network for the observations of Aerosol, Clouds and Trace gases (ACTRIS).
Here, the ARM and CloudSat CPR measurements and the methodology used for the comparison between the ground-based and space-based observations are described.
The record of ARM profiling radar observations compared to CloudSat is
detailed in Table 1. In total, ARM cloud radar observations from 14
different locations (Fig. 1) with four different radar systems (MMCR, WACR,
KAZR, KAZR2) for a total of 43 years and 8 months are analyzed.
At a couple of sites, the calibration record starts as early as the launch of
CloudSat (mid-2006) and at several sites stops at the end of 2017. For much
of the record analyzed here, the WACR was the primary profiling cloud radar
of the first ARM Mobile Facility (AMF) and as such has been deployed in
different climatological locations. A marine version of the WACR (M-WACR)
with smaller antenna and a ship-motion stabilizer has been the primary radar
for marine deployments of the second AMF (AMF2). The WACR does not use pulse
compression and operates only in co-polarization and cross-polarization
modes. The single operating mode of the WACR combined with the fact that it
uses the same frequency as the CloudSat CPR makes their comparison
relatively straightforward. The MMCR used a complicated operating mode
sequence (Moran et al., 1998; Kollias et al., 2007b) in order to
meet the requirement of detecting all radiatively important clouds with
radar reflectivity above
Location of the fixed and mobile ARM profiling cloud radars calibrated using the CloudSat Cloud Profiling Radar.
Information on the sites (name and location), the type of radar, and the start and end dates of the calibration period.
Therefore, as a prelude to comparing CloudSat and ARM, we begin with a
comparison of reflectivity values between ARM radar modes. As will become
clear later, changes in the intramode reflectivity differences are often,
though not always, indicative of changes in overall calibration. A detailed
comparison between the reported radar reflectivities from all the radar
systems with more than one operating mode was conducted (Fig. 2). The
difference between mode 3 and mode 2 is reported for the MMCR systems and
the difference between the GE and MD modes is reported for the KAZR and
KAZR2 systems. The difference (dB) in the measured radar reflectivity
between two modes is estimated at heights where both modes provide
observations (e.g., the MMCR mode 2 does not provide data below 3.6 km) with
high signal-to-noise ratio (SNR > 0 dB), and at ranges where the
averaged profiles were correlated to filter our ranges where big
discrepancies due to radar artifacts were present. At each height, the
average reflectivity profile of each mode (in linear units) is computed
using a 1-month running window. The mean of the differences in the averaged
radar reflectivity profiles between the two modes is computed and shown as a
function time in Fig. 2. Overall, the mode reflectivity differences are
small (
The difference (dB) in the radar reflectivity reported between different ARM modes. For KAZR/KAZR2 systems the GE–MD difference and for MMCR systems the mode 3–mode 2 differences are reported.
The comparisons between the ARM radars (MMCR, KAZR, and KAZR2) and the
CloudSat CPR are performed independently for two modes for the MMCR (2 and
3) and two modes for the KAZR and KAZR2 (MD and GE). The approach is similar
to that described in Protat et al. (2011). The technique consists in a
statistical comparison of the mean vertical profiles of non-precipitating
ice cloud radar reflectivities from the ground-based and spaceborne radar
observations. One of the improvements introduced in this study is that the
averaging of the radar reflectivity value at each height is performed in
linear space (
Several factors need to be taken into account to achieve an objective statistical comparison between ground-based and space-based observations: frequency of each radar, sensitivity, viewing geometry, attenuation correction, etc. The approximations to deal with all these factors introduce errors that are difficult to assess. The necessary steps required to find the calibration offset for each radar are described here, following the algorithm flow outlined in Fig. 3.
Algorithm flowchart of the calibration of the ARM cloud radars using CloudSat observations.
The CloudSat overpasses are predicted using the two-line element (TLE) set files that encode all necessary information to define the latitude and longitude of the satellite over the Earth's surface at any given time. Using these files, the satellite position is computed with high resolution in time and the distances to each ARM radar location are used to define the overpass. Only CloudSat data passing in a radius between 100 and 300 km around the ARM radar location are extracted. Knowing the orbits of the overpasses, the CloudSat respective files are read. In this present study, the data from the fourth and fifth releases (R04 and R05) of the 2B-GEOPROF product are used to extract the CPR reflectivity, height, DEM elevation, CPR cloud mask, gaseous attenuation, and data quality flags. In addition, the height of the freezing level is extracted from the 2C-PRECIP-COLUMN product. Figure 4a shows the probability density function (pdf) of the freezing level height at the North Slope of Alaska (NSA).
Example of the CloudSat–ARM radar calibration at the NSA for
mode 2 of the MMCR for the period January 2008 to December 2009.
All CPR observations within 500 m from Earth's surface are removed to avoid
residual surface clutter contaminations. In addition, all CPR detections at
very low signal-to-noise ratio (SNR) conditions (CPR cloud mask < 20) and
poor data quality points (data quality
The gaseous attenuation correction reported in the CloudSat files is added
to the reflectivity profile. The CPR reflectivity is normalized for the
differences in the values used for the dielectric constant ( The profiles are carefully separated into two groups: precipitating and
non-precipitating ice clouds. Ice clouds are assumed at heights above the
freezing level while liquid particles are assumed below. An ARM column is
considered to be precipitating if at least 10 % of the range gates below
the freezing level report echoes higher than The performance of the most sensitive radar is degraded to match the minimum
detectable signal (MDS) of the least sensitive radar. Due to the large
distance between CloudSat and the Earth's troposphere, the CloudSat MDS is
practically constant around If the ARM radar operates at 35 GHz, the radar reflectivity is converted to
94 GHz radar reflectivity using Eq. (3). Using all available columns within the selected time window (6 months), a
reflectivity frequency by altitude diagram (CFAD) is constructed for each
radar (Fig. 4b, c). This diagram will be used to generate the mean vertical
reflectivity profile used in the final comparison (Fig. 4e). Steps 1 and 3 are repeated for all possible calibration offsets, from Each CFAD constructed with the previous methodology is representative of one
averaged profile. As we have The final calibration result is found by computing the root-mean-square
error (RMSE) between the profiles of each radar for each calibration and at
heights with enough data points (at least 3 % of the total sample size).
The calibration offset representative of the profiles with the least RMSE
will be the final calibration result (Fig. 4f). The probability density function (pdf) of cloud top heights (Fig. 4d) is
also used for verification purposes, assuming that occurrences of the
highest clouds should be similar when the ground and spaceborne radars have
equal sensitivity (Protat et al., 2010).
The most important factor in determining our ability to perform a good
comparison is the number of available CloudSat profiles. Several temporal
windows were considered, and the decision was made to use a time window of 6
months throughout this study. In addition to the length of the time window,
the impact of the maximum distance of the CloudSat observations from the ARM
site (we tested values from 100 to 300 km) was investigated. In particular,
we examined the sensitivity of the estimated calibration offset to the
selected maximum distance of the CloudSat observations. Using different
distance values from 100 to 300 km every 25 km at different sites, we
investigated the behavior of the estimated calibration offsets. Our analysis
indicated that a maximum distance of 200 km was optimum for most ARM
locations and was therefore selected as a fixed value throughout the study.
Figure 5 shows the number of CloudSat profiles with suitable measurements
(non-precipitating ice) with a 6-month window for all the ARM fixed and
mobile sites as a function of time. As expected, there is strong seasonal
variability that is dictated by the seasonal cloud type and atmospheric
temperature profile variability. Of particular interest is the availability
of suitable CloudSat profiles at the NSA. There is a significant decrease in
the number of available CloudSat profiles during the period when the ARM
facility transitioned from the MMCR to the KAZR radar system. The reduction in
the number of CloudSat profiles is not related to the changes of the ARM
radar system (these two systems have similar MDS) nor to
significant changes in the cloud climatology at the NSA. The transition from
the MMCR to the KAZR system coincided with the battery anomaly that occurred
on CloudSat in 2011 and resulted in CloudSat operating since then only
during daylight conditions, thus effectively halving the possible number of
CloudSat columns (Stephens et al., 2018). The daylight-only operations
of CloudSat challenged our ability to collect a good size sample of column,
especially at very high latitudes (e.g., ARM West Antarctic Radiation
Experiment (AWR) during the Southern Hemisphere winter).
The number of CloudSat profiles found within a
A total of 653 ARM–CloudSat comparisons were performed using a running 6-month time window. The relationship between the minimum RMSE value achieved in a particular ARM–CloudSat comparison and the corresponding number of CloudSat columns is shown in Fig. 6. As expected, the RMSE value decreases with the number of samples. The analysis of the entire ARM–CloudSat comparison record suggests that when the number of CloudSat columns is less than 500, the comparison is difficult to perform. In addition to the value of RMSE and the number of CloudSat columns, the goodness of the fit between the ARM and CloudSat cloud top height pdf's is evaluated when the minimum RMSE is achieved. Out of the possible 653 calibration coefficients, 616 were accepted, i.e., a 94.3 % success rate.
The relationship between the minimum RMSE (dB) achieved in a particular ARM–CloudSat comparison and the corresponding number of CloudSat columns.
First, the results of the ARM–CloudSat comparison at the two sites that
feature the most recently acquired profiling cloud radar systems of the ARM
facility are discussed. The two KAZR2 systems are located at critical
climatological locations (ENA and OLI) and are the primary sources of cloud
observations. The OLI KAZR2 is compared against the CloudSat CPR for the
period September 2015 to December 2017. Figure 7a shows the calibration offset (dB) we need
to add to the MD mode observation to minimize the RMSE with the CloudSat
observations. If the calibration offset is positive, this suggests that the
MD mode underestimates the radar reflectivity compared to CloudSat. Although
a 6-month running time window is used, considerable temporal variability is
observed, especially at the beginning of the period. At the beginning of the
period,
Figure 7b shows the calibration offset for both KAZR2 operating modes (MD and GE) using the ARM–CloudSat comparison methodology applied to the recorded radar reflectivities of each mode. Overall, the calibration offsets closely follow each other throughout the observing period. During the first 6 months, the calibration offset for the MD is about 1 dB higher, suggesting that the MD reported on average 1–2 dB lower radar reflectivities than the GE mode. This relationship is reversed around April 2016 and until the end of the observing period; the calibration offset for the MD mode is now 1–2 dB lower than that estimated for the GE mode. Noticeably, the reversal in relationship of the calibration offsets coincides with the period that we argued earlier coincides with changes in the radar configuration around April 2016. During that period, the number of fast Fourier transform (FFT) points in the recorded radar Doppler spectra changed from 256 to 512 and the calibration was updated (Joseph Hardin, ARM radar engineer, personal communication, 2018).
As discussed in Sect. 2.1 the ARM MD and GE mode observations can be used to estimate their relative offset. Figure 7c shows the difference (MD–GE) in decibels of the two KAZR2 operating modes (black line). On the same plot, the difference (MD–GE) in decibels as seen from CloudSat is also reported (circles). Overall, a very good agreement is found between the two estimates of the radar reflectivity offset between the two KAZR2 modes. This suggests that the ARM–CloudSat comparison can provide high-quality information regarding the absolute and relative calibration offsets between radar modes.
The ENA KAZR2 calibration offset as reported by the CloudSat–ARM comparison (star symbol) and by the KAZR2 and Parsivel disdrometer comparison (line). The colors indicate the RMSE of the CloudSat–ARM comparison for different radar calibration offsets.
The second KAZR2 system has been operated at the ENA since the fall of 2015. Figure 8 shows two calibration offset (dB) values for the KAZR 2 MD (white
symbols). Contrary to the OLI site, the ENA site cloud and temperature
climatologies do not favor the collection of a large number of suitable
CloudSat columns for calibration (Fig. 5). During the first 9 months of
operation (October 2015–July 2016) the calibration offset was very small (
The ARM TWP Darwin, Manus, and Nauru sites are located deep in the tropics
and featured MMCR systems until the first quarter of 2011. Only at two sites
(Darwin and Manus), the MMCR systems replaced by KAZR systems. All TWP
sites terminated operations in 2014 (Long et al., 2016). The
calibration offsets for the period 2007 to 2014 at the TWP sites are shown
in Fig. 9. The calibration offset record is not continuous since the number
of CloudSat columns is affected by the significant inter- and intraseasonal
cloud and precipitation variability driven by large-scale features at
different temporal–spatial scales such as El Niño–Southern Oscillation, the
Madden–Julian oscillation, and the movement of the intertropical convergence
zone (ITCZ). The operational record of the TWP systems is also intermittent
due to the logistical challenges associated with the physical presence of
ARM engineers at these sites: delays associated with the delivery of
hardware components at the TWP sites and poor communications for instrument
monitoring, especially at Manus and Nauru (Long et al., 2016).
Overall, the calibration offsets are within
The calibration offset for the MMCR mode 2 and KAZR GE mode at the
Tropical Western Pacific (TWP) sites of
The ARM NSA and SGP sites are the two longest operating sites of the ARM
facility (Sisterson et al., 2016; Verlinde et al.,
2016). The NSA represents a typical Arctic environment with very low
temperatures while the SGP has been the observational centerpiece and anchor
of the ARM facility since 1992. The calibration offsets for the period 2008
to 2017 at these two sites along with the ARM intramode differences are
shown in Fig. 10. The NSA MMCR 2 significantly overestimates the radar
reflectivity, and a calibration offset between
The calibration offset for the MMCR mode 2 and KAZR GE mode at
The radar calibration offset we should add to the reported ARM cloud radar reflectivities in order to minimize their differences with those reported by the CloudSat CPR at the ARM Mobile Facility (AMF) sites. The size of the circles indicates the ratio of the sample size of the CloudSat columns for any given calibration offset estimate relative to the maximum sample size of CloudSat columns observed during the same period by the same mode. Circles correspond to 94 GHz (WACR) calibration offsets and squares correspond to 35 GHz (KAZR) calibration offsets.
The ARM Mobile Facility (AMF) is a portable atmospheric observatory equipped
with a sophisticated suite of instruments designed to collect essential data
from cloudy and clear atmospheres in important but under-sampled climatic
regions. As such, the AMF deployments are often the only source for
ground-based observations of clouds and precipitation at some of the AMF
deployments (Miller et al., 2016). Here, we report the calibration
offsets for five deployments of the first ARM Mobile Facility (AMF1) and two
deployments of the second ARM Mobile Facility (AMF2). The results are shown
in Fig. 11. The AMF1 deployments are Niger, west Africa (NIM); Black
Forest, Germany (FKB); Graciosa island, Azores (GRW); Cape Cod,
Massachusetts (PVC); and Manacaparu, Brazil (MAO) and the AMF2 deployments
are Hyytiälä, Finland (TMP) and McMurdo Station, Antarctica (AWR).
The AMF deployments are typically 1-year deployments, except for the GRW
and MAO deployments that lasted for 2 years. At the AMF1 deployments the
main profiling cloud radar system was the WACR and at the AMF2 deployments a
KAZR. The short duration of the mobile deployments coupled with the time
needed to relocate the AMFs to their next field location makes the AMF
calibration offset record sparse. At NIM, the AMF deployment was over 13
months long but the WACR was deployed for only 8 months and two WACR
calibration offsets are estimated (
The AMF2 was established later than the AMF1; thus, its deployment record is
shorter. The AMF2 deployment in Hyytiälä, Finland (TMP), has been
considered the first successful deployment of triple-frequency radar
observations by the ARM facility (Kneifel et al., 2015) with well-calibrated radar systems. The ARM–CloudSat comparison confirms that the
KAZR MD mode was well calibrated during the TMP deployment and the
calibration offsets are
The DOE ARM facility has been at the forefront of the development and application of profiling and scanning millimeter-wavelength radars for over 20 years. The long record of ARM cloud radar observations represents a unique dataset that provides a bottom-up, high-resolution view of clouds and precipitation at a number of locations around the globe. The characterization of a decade-long cloud radar record from multiple locations is a necessary step for the development of unbiased statistics on cloud occurrence and the estimation of microphysical parameters using retrieval techniques. Once the characterization and reprocessing of the ARM radar observations is completed, the decade-long record and its added-value products can be used as observational targets for global climate model evaluation studies using suitable forward operators (Zhang et al., 2018; Lamer et al., 2018).
The use of CloudSat as a global calibrator for cloud radars was first proposed by Protat et al. (2011). Here, the Protat et al. (2011) technique is revised, improved, and automated and the entire record of CloudSat observations (2007–2017) is used to provide a calibration reference for over 43 years of ARM profiling cloud radar observations at fixed and mobile sites. Four generations of ARM cloud radar systems, operating at two different radar frequencies (35 and 94 GHz) are evaluated. All the radar systems (with the exception of the AMF1 WACR) operate using a sequence of modes with different capabilities in order to provide a uniform radar sensitivity and performance throughout the troposphere. The offsets in the reported radar reflectivity by these different modes for each radar are documented as a function of time. Abrupt changes in the offset magnitude and sign are found to correlate well with changes in the radar calibration as deduced by the statistical comparison with CloudSat. Thus, changes in the reflectivity offset between the modes should be monitored and used to identify periods where the calibration stability is suspect and moving forward perhaps trigger more prompt additional external calibration evaluations. Furthermore, the geographical location, the seasonal variability of the clouds and precipitation occurrence, and the operational status of the CloudSat CPR significantly affect the number of samples available within a 6-month time window to perform the ARM–CloudSat comparison. When the number of CloudSat columns is fewer than 500–1000, the comparison is difficult to perform. Out of the possible 653 calibration coefficients, 616 were accepted, i.e., a 94.3 % success rate.
The analysis demonstrates that both historic (i.e., MMCR) and recent ARM radar operations (i.e., KAZR2) require considerable adjustments before they can be used in a quantitative way. The analysis from Protat et al. (2011) and the experience gained in this study using the technique in a much larger dataset suggest that the accuracy of the CloudSat-based calibration of ground-based cloud radar systems is accurate within 1–2 dB. In most cases, the observed calibration offsets exceeded this uncertainty value, suggesting that the ARM profiling radar record contains considerable calibration biases. The reported calibration biases are expected to have a large impact on routine ARM microphysical data products such as the Continuous Baseline Microphysical Retrieval (MICROBASE) value-added product (Zhao et al., 2012). In addition, cloud retrieval techniques and associated products are impacted by the reported calibration offsets (Shupe et al., 2015; Dong et al., 2014). For reference, a 3 dB calibration offset is equivalent to a factor of 2 bias in hydrometeor content or number concentration retrievals. As part of the outcome of this study, the estimated calibration offsets, the RMSEs, and the number of samples as a function of time for each radar system evaluated here have been provided to the ARM facility. The ARM facility is currently considering reprocessing of the ARM radar record with these new calibration offsets. Furthermore, the gradual temporal change in the observed calibration offsets and the correlation of large swings in the calibration offset with periods when the ARM radar hardware and/or software was not operating in an optimal way suggest that the use of CloudSat can provide reliable information that can be used to characterize the calibration of ground-based radar systems.
Planned and future spaceborne radar systems such as the Earth Clouds
Aerosols and Radiation Explorer (EarthCARE; Illingworth et al., 2015;
Kollias et al., 2018) or future spaceborne radar concepts (Tanelli
et al., 2018) will provide similar spaceborne radar measurements to evaluate
large profiling cloud radar networks (e.g., ARM, ACTRIS) in the future. A
project website that describes the ARM–CloudSat comparison at all the ARM
sites and radar systems is now available to the entire user community:
Finally, there is merit in extending the presented analysis to other satellite measurements. For example, NASA's Global Precipitation Mission (GPM) Dual-Frequency Precipitation Radar (DPR) observations could be used in a similar manner to evaluate the calibration of the ARM facility centimeter-wavelength radars (Lamer et al., 2019). In addition to radar calibration, the statistical comparison between cloud and precipitation properties such as cloud base height, cloud thickness, precipitation occurrence and intensity, and liquid water path measured at the ARM facility and those derived by research satellites such as NASA'S A-Train constellation (Stephens et al., 2018) should be considered. The ARM facility provides a bottom-up view of clouds and precipitation with superior vertical resolution, especially in the boundary layer. Statistically significant differences with the top-down view provided by the A-Train satellites should be considered when conducting cloud-scale process studies using global satellite datasets.
The code used for the ARM–CloudSat comparison can be made available upon request. The ARM data were obtained from the Atmospheric Radiation Measurement (ARM) user facility, a U.S. Department of Energy (DOE) Office of Science user facility managed by the office of Biological and Environmental Research (BER). The CloudSat observations are available at the CloudSat Data Processing Center.
The use of surface-based measurements of the raindrop PSD using impact or
optical disdrometers to calibrate profiling and scanning precipitation
radars is not new. This technique has been widely used in the past for
calibrating profiling centimeter-wavelength Doppler radars (Gage et al., 2000;
Tridon et al., 2013). In the case of centimeter-wavelength radars, wet radome
or antenna attenuation is negligible, the systems are configured to have
sufficient dynamic range to detect intense precipitation returns without
receiver saturation, and the Rayleigh scattering approximation is valid in
most cases. At millimeter-wavelength radars, several factors need to be considered:
the wet radome can induce considerable attenuation, at high rain rates the
Rayleigh scattering approximation is not valid, and receiver saturation
occurs at lower rain rates. Here, we use the Parsivel2 disdrometer (OTT
Hydromet GmbH) measurements. The disdrometer provides 1 min averaged
raindrop PSDs. From the Parsivel2 files, the variable
“equivalent_radar_reflectivity”, which is the
radar reflectivity calculated by the ARM ingest, is used. All 1 min Parsivel
measurements where raindrops with a diameter > 4.5 mm are detected
are filtered out to avoid the impact of false detection of large raindrops
in the Parsivel2–KAZR2 comparison. The Parsivel2 time assigned to each
data point indicates the beginning of a 1 min period of averaging. Using
this time, 1 min averages of the KAZR2 reflectivities in linear units are
estimated. Next, the KAZR2 radar reflectivities are corrected for path
attenuation induced by the hydrometeor. The relationship
Figure A1a shows the time series of the calibration offset between the Parsivel2 and the KAZR2 for different KAZR2 range gates. In general, the calibration offset is positive, thus implying that the KAZR2 underestimates the radar reflectivity. However, the calibration offset varies a lot with the range gate. The KAZR2 is a pulsed radar; thus after each pulse transmission the receiver protection circuit (T/R switch network) needs to switch from transmit (closed receiver) to receive (open receiver) mode. The switch takes several hundreds of nanoseconds; thus, the KAZR2 returns from the first range gates (3 to 7) report lower radar reflectivity values, resulting in higher radar calibration offset values. Our analysis identified range gate 8 (240 m) as the closest range gate to the surface that is unaffected by the KAZR2 T/R switch network. Above range gate 8, the calibration offset continues to decrease, highlighting the impact of the evaporation on modifying the raindrop PSD. The scatter plots between the KAZR2 radar reflectivity at range gate 8 and the Parsivel2 radar reflectivities during the two extensive periods are shown in Fig. A1b, c. These two periods match the periods used to estimate calibration offsets using the ARM–CloudSat comparison technique (Fig. 8). The ARM–CloudSat comparison indicated calibration offsets of 0.3 and 5.2 dB and the ARM–Parsivel2 comparison indicated calibration offsets of 0.57 and 3.91 dB.
Disdrometers certainly have the potential to monitor the calibration of profiling cloud radars and this topic warrants additional analysis using comprehensive datasets from different cloud radar systems and for different climatological conditions. For example, frequency-modulated continuous-wave (FMCW) radars (Küchler et al., 2017) do not have T/R switch networks, but careful analysis is required to ensure proper alignment of the two antennas or correction for the antenna parallax problem (Sekelsky and Clothiaux, 2002). Furthermore, careful analysis is required to avoid using radar returns that saturate the radar receiver, especially at short ranges, and to account for non-Rayleigh scattering in the case of 94 GHz radar systems. This careful analysis is beyond the scope of this study.
PK designed the ARM–CloudSat comparison project, wrote the paper, and prepared most of the figures. BPT carried out all the coding for the ARM–CloudSat comparison and provided edits to the paper. AP provided his original code that was used in a previous publication to conduct a similar study and provided assistance and comments during the coding phase.
The authors declare that they have no conflict of interest.
Pavlos Kollias and Bernat Puigdomènech Treserras were supported by the U.S. DOE ARM and ASR radar science project. The authors would like to thank Jim Mather, Bradley Isom, and Joseph Hardin for reviewing the paper and providing valuable feedback.
This research has been supported by the U.S. Department of Energy, Office of Science (grant no. DE-SC0012704).
This paper was edited by Mark Kulie and reviewed by Roger Marchand and two anonymous referees.