Deployment of the C-band radar Poldirad on Barbados during EURECA

The German polarimetric C-band weather radar Poldirad (Polarization Diversity Radar) was deployed for the international field campaign EURECA (ElUcidating the RolE of Cloud-Circulation Coupling in ClimAte) on the island of Barbados where it was operated from February until August 2020. Focus of the installation was monitoring clouds and precipitation in the trade wind region east of Barbados. Different scanning modes were used with a temporal sequence of 5 minutes and a maximum range of 375 km. In addition to built-in quality control performed by the radar signal processor, it was found that 5 the copoloar correlation coefficient ρHV can be used to remove contamination of radar products by sea clutter. Radar images were available in real-time for all campaign participants and onboard of research aircraft. Examples of mesoscale precipitation patterns, rain rate accumulation, diurnal cycle, and vertical distribution are given to show the potential of the radar measurements for further studies on the life cycle of precipitating shallow cumulus clouds and other related aspects. Poldirad data from the EURECA campaign are available on the EURECA AERIS database: https://doi.org/10.25326/218 (Hagen et al., 2021a) 10 for raw data and https://doi.org/10.25326/217 (Hagen et al., 2021b) for gridded data. Copyright statement. TEXT

In this context the mission and the contribution of the dual-polarization Doppler weather radar Poldirad was: to provide an overview for research aircraft and vessels in the experimental area east of Barbados (see Fig. 1), to investigate the life cycle of the precipitation cells approaching Barbados and in particular those passing over the Barbados Cloud Observatory ( BCO Stevens et al., 2016), 55 to investigate the possibilities about cloud microphysics retrieval when combining C-band (wavelength 5.5 cm) measurements with profiles from the Ka-band (wavelength 8 mm) radar at BCO.
To support the first item, Poldirad images and data were transferred in real-time to Barbados Meteorological Service and incorporated into the Caribbean Radar Composite. Real-time images were available through the Atmosphere Planet software system 1 onboard the German HALO (Hight Altitude and LOng range research aircraft) and French ATR-42 research aircraft, 60 and for ground staff. Radar images were also of importance for the flight crew of the French remote controlled unmanned aerial vehicle Boréal which should avoid flying into precipitation.

Poldirad on Barbados
The C-band polarimetric weather radar Poldirad (Polarization Diversity Radar, Schroth et al. (1988), Fig. 2) was deployed on Barbados for the EUREC 4 A campaign. Due to unforeseeable long delays during shipping from Germany, custom handling and 65 local permissions, the radar was ready for operation on Feb. 5 th with a two week delay for the campaign. Originally, it was 1 http://www.atmosphere.aero/products-services/planet, last access: 27 April 2021. planned to operate the radar after the campaign with a reduced schedule until June 2020. However, due to the rapid outbreak of the SARS-CoV-2 virus and the restriction in traveling, operation was stopped in March 2020. A limited radar operation was resumed mid May 2020 until a major failure of the antenna drive at the beginning of August 2020.

Radar system 70
Poldirad was installed in 1986 at the DLR (Deutsches Zentrum für Luft und Raumfahrt -German Aerospace Center) site Oberpfaffenhofen (Germany) as one of the first polarimetric weather radars in Europe and is presently -after some major upgrades -up-to-date (Table 1). Its unique fast ferrite polarization network allows for any kind of polarization basis -even elliptical ones. The polarization basis can be independent for transmit and receive and can be changed from pulse to pulse.
During EUREC 4 A two modes were used: (i) the hybrid or STAR (Simultaneous Transmit And Receive) mode and (ii) the 75 alternate HV mode (Horizontal Vertical). In the hybrid mode a linear polarized pulse with an orientation of 45 • is transmitted and on reception the signal is received simultaneously with linear horizontal and vertical polarization. In the alternate HV mode the polarization on transmit is alternating from pulse to pulse from linear horizontal to linear vertical polarization. On reception always both, the linear horizontal and vertical polarization are recorded. Only the latter mode allows measuring the full back-scattering matrix and hence the depolarization signal (LDR: Linear Depolarization Ratio).

80
In contrast to most other C-band weather radar systems, Poldirad is equipped with an offset antenna and the antenna is not sheltered by a radome. With an offset design the antenna feed does not cause blockage of the antenna beam and cannot create additional cross-polar signal making it better suitable for depolarization measurements. During the development of Poldirad it was decided to design the radar without a protective radome, since it was not appraisable how much a radome (especially a wet one) will influence polarization purity. For the mobile deployment of Poldirad the antenna can be installed on a 3 m high tower. Radar control electronics are housed in one 20 ft container while another 20 ft office container is used for radar operation (Fig. 2). Two further 20 ft containers are used for transportation of the equipment.

4@300
The sensitivity or lowest observable reflectivity, named minimum detectable signal (MDS) Z m is a function of range r Z m (r) = Z 0 + 20 log 10 r + K a r (1) 90 with Z 0 as MDS in dBz at a range of 1 km, r range in kilometres, and K a the two-way atmospheric attenuation in dB per kilometre. Z 0 is −47.5 dBz for the long range mode and −39 dBz for the short range mode (Fig. 3). Poldirad's signal processor uses K a = 0.019 dB km −1 as an estimation of the two-way atmospheric attenuation at C-band. See also the discussion in Sect.
4.3 for the impact of MDS on long range measurements. The actual sensitivity is slightly varying within ca. 0.5 dB since the transmit power is varying during the course of the scanning cycle in dependency of the transmitter temperature. Note that, due 95 to internal settings in the signal processor, the lowest recorded reflectivity value is −31.5 dBz. This limitation is only relevant for the first 6.2 km in long-range mode and 2.4 km in short range mode, respectively.

Location
The main mission of Poldirad during EUREC 4 A was the provision of precipitation information in the trade wind region east of Barbados. Therefore the radar location needed an unobscured view towards the experimental area. Logistical limitations were

Data quality
Several aspects are related to the quality of radar data: (i) calibration; (ii) antenna pointing direction; (iii) sanity checks; and (iv) polarization purity. Most important is the calibration of the radar system. An absolute calibration can only be done using a target with well known backscatter properties lifted into the air. Normally a metal sphere with a diameter which is large compared to the wavelength is used for that purpose. This is hardly achievable in the field. The architecture of Poldirad is that 125 the transmitter is located in the radar cabinet at ground whereas the polarization network and the receivers are mounted at the rotating antenna. The transmit power is measured for each pulse by the receiver. During the deployment at Barbados the signal path from the antenna feed horn to the receiver was not changed. For these reasons the calibration parameters determined during the latest system upgrade are kept for the campaign. In order to account for temporal variations of the receiver chain, a noise sample is taken before each scan. Additionally an offset calibration is performed using a single frequency signal with 130 defined power injected at the receiver frontend.
Verifying the antenna pointing direction and the alignment of the pedestal is frequently done by using the sun as radiative source (e.g. Huuskonen and Holleman, 2007;Altube et al., 2015). The initial alignment was done with dedicated solar scans (Reimann and Hagen, 2016, cf. Appendix A). Solar hits during the regular scan sequence were used to verify the antenna pointing. Measurements after sun rise and as long as the sun was within the volume PPI or RHI scan revealed that elevation 135 error was less than 0.1 • and that azimuth error was about 0.5 • to the left.
The radar signal processor (Selex GDRX © ) performs quality checks, ground clutter filtering, and second-trip echo removal.
Data with a low signal quality index (SQI) i.e. with too large standard deviation are disregarded and no further radar moments are calculated. A ground clutter filter operating in frequency domain is applied by default to Poldirad radar parameters. Secondtrip echoes are considered if reflectivity is above noise level, but no phase coherence can be achieved within the sampling 140 interval.
The ferrite polarization network of Poldirad allows for a high flexibility in defining the polarization base for transmission and reception (Schroth et al., 1988). However, the required settings for power division and phase delay need calibration.
This was done by Reimann (2013) and it was not possible to repeat this procedures in the field. A simple way to check the polarization configuration is the assumption that for light rain the differential reflectivity Z DR should be close to zero (e.g. 145 Gourley et al., 2006) and LDR much below −30 dB. Occasionally setting the polarization fails, which becomes visible in data where reflectivity is much less compared to the previous or following scan. These data should be omitted.
During the warm-up phase of the magnetron (about one hour after switching on the transmitter), the nominal frequency is not yet reached and the receiver may fail to lock to the stabilized local oscillator (STALO) frequency. This occurs mainly for the long range PPI scans and becomes visible in the data when reflectivity and other products are empty and the raw data files 150 are much smaller than the others. These data should not be used for further processing.

Sea clutter identification
The backscatter from sea surface can considerably contaminate radar observations if a weather radar is located close to the coast. Reflectivity from sea clutter was observed up to 35 dBz for Poldirad scans at 0 • elevation and up to 15 dBz for 0.5 • elevation ( Fig. 6 top). Ground clutter from land surface is removed using Doppler filters for stationary targets. However, since 155 sea surface is in motion, ground clutter filters will fail and additional filters have to be applied. Various algorithms do exist to identify hydrometeors using polarimetric radar data (e.g. Vivekanandan et al., 1999;Park et al., 2009). The separation of meteorological targets from non-meteorological targets is a further aspect of some classification techniques (e.g. Giuli et al., 1991;Berenguer et al., 2006), and more recently by Kilambi et al. (2018) or Overeem et al. (2020). Besides some other properties observed in polarimetric parameters, like the texture of differential reflectivity Z DR or 160 differential propagation phase φ DP , the copolar correlation coefficient ρ HV is one of the most promising parameters to identify sea clutter. Meteorological echoes have a high correlation between reflectivities from horizontal and vertical polarization. Ryzhkov et al. (2002) proposed to use a threshold of ρ HV = 0.7 to discriminate between meteorological echoes and sea clutter. Figure 6 shows the observed reflectivity factor, copolar correlation coefficient and the filtered reflectivity factor using a threshold of 0.7 for ρ HV . Some speckles are not removed and also some weak echoes at the edges of precipitation cells are removed. The latter is caused by non-uniform beam filling while scanning across the edges of precipitation. For further processing of Poldirad data this threshold will be applied. Figure 7 shows the radar operation during February 2020 (blue bars) together with rain events (any non-zero 10 second rain intensity) recorded at BCO (green ticks) and HALO flight times (red). The flight on Feb. 18 th was the return flight to Germany.

170
The radar is designed for research application and is not suited for unattended 24/7 operation. During the campaign a 24-hour operation was envisaged. Failures and limited personal resources caused gaps in a continuous operation during EUREC 4 A and afterwards.
Further measurements, only during daytime, were performed from mid-May until beginning of August. During the nearby passage of the Tropical Storms Gonzalo (July 23 rd to 25 th ) and Isaias (July 28 th to 29 th ), Poldirad was operated continuously 175 also during nighttime.

Data format
Poldirad raw data are stored in HDF5 (Hierarchical Data Format version 5) format following the ODIM 2.0.1 specifications (Michelson et al., 2010). This data structure was developed by the weather radar operators group of European meteorological services to ease the international exchange of weather radar data. Some additions and modifications to the ODIM specification 180 had been necessary, since Poldirad offers more flexibility and options than operational weather radar systems. Each surveillance PPI, volume PPI, or RHI scan is stored as one file with an approximate size of 2 to 12 MBytes. Each file consists of general meta data, meta data for each sweep or sub-scan (named as dataset1 to datasetn) as well as for each radar product (named as data1 to datan). Details on the structure and the attributes (meta data) can be found in Michelson et al. (2010).
Depending on the scanning mode up to 22 radar products are defined. The available radar products are listed in Appendix B.

Gridded dataset
To ease the usage of Poldirad data within the EUREC 4 A community, a 2-dimensional gridded dataset was generated from the long range surveillance scans. The sector scans were interpolated on a 1 by 1 km grid with a size of 400x400 km 2 . Interpolation was done using the griddata routine from the Python SciPy package. For each grid point reflectivity factor Z, rainfall rate R, longitudinal (x) distance from radar, meridional (y) distance from radar, longitude, latitude, and approximate height above 190 sea level is provided. Sea clutter was removed according to the procedure outlined in Sect. 3.5.
Conversion from reflectivity to rainfall rate was done by the commonly used empirical z-R relation z = 200R 1.6 (Marshall et al., 1955) with R in mm per hour and z in mm 6 m −3 . There are only a few z-R relations attributed to shallow trade wind showers, e.g. Stout and Mueller (1968)

Examples from Poldirad measurements
In this section we will show some exemplary measurements from the Poldirad observations during the EUREC 4 A campaign. range only precipitation can be observed. Therefore, the term precipitation or rain patterns or cells will be used in the following when describing radar echoes.   On 11 Feb. 2020 precipitation systems were larger and widespread (Fig. 8 bottom) and did reach heights of 4 to 5 km. The precipitation patterns had a structure with multiple cores of heavier precipitation. Cloud patterns on that day were classified as Flowers. With the limited view of the volume scan the whole Flower pattern is not visible.

Daily accumulated precipitation and daily cycle
With long range surveillance scans about every 5 minutes daily rainfall accumulations can be generated and e.g. the diurnal 220 cycle of precipitation can be studied.
Figure 9 (top) shows the daily accumulated rainfall pattern for Feb. 9 th 2020 derived from the gridded datasets. The average daily precipitation amount within the HALO circle was 0.18 mm. Traces of individual cells can be located throughout the day.
Beyond approximate 250 km range less cells are observed. Only a few cells were observed further out. This can be attributed to the vertical extent of the majority of the cells and height of the radar beam in dependence with range (cf. Fig. 5 bottom). In 225 general, stronger precipitation is observed closer to the radar where the radar beam intersects within the more intense part of the precipitation cores of the cells (see Fig 10). In Fig. 9 (bottom) the diurnal cycle of precipitation fraction within the HALO circle flight pattern (red circle in Fig. 9 top) is shown. For that day, only a weak diurnal cycle of precipitation can be identified.
On Feb. 11 th the situation was different (Fig. 11): more precipitation was observed in the area. The average daily precipitation amount was 0.95 mm within the HALO circle. The cells were more widespread and more frequent. Vertical extent of the 230 precipitation was up to approximately 4 km (Fig. 12) and thus lower than on Feb. 5 th . The rain fraction within the HALO circle shows a pronounced daily cycle. Up to 20 % were observed in the morning hours and almost no precipitation during daytime.
This agrees well with the simulations and long-term observations by Vial et al. (2019).

Long-range reflectivity observations
As shown above, long-range reflectivity observations are limited by a number of factors: (i) earth-curvature and the height of  probably most of the high reflectivity precipitation cores can be observed up to that range. At further ranges the radar beam is only partial filled with echoes since only the upper part of the cells reach partly into the radar beam. This results in reduced reflectivity values since the radar equation always assumes a completely filled measurement volume. For that day no more precipitation echoes were recorded beyond 320 km.

Vertical reflectivity distribution
Contoured frequency by altitude diagrams (CFAD) are often used to investigate vertical structures of radar observables from precipitation systems (Yuter and Houze, 1995) or to compare measurements with numerical simulations. Here we use the frequent RHI scans towards BCO or other directions to derive reflectivity CFADs. In contrast to CFADs derived from vertical pointing radars, we have to consider the range dependent minimum detectable signal, as well as the inability to take measure-250 ments close to the surface at far ranges, the limited altitude range of RHI scans for short ranges, and the broadening of the radar beam with range (cf. Fig. 5).  also seen in Fig. 12 cells did to not exceed 5 km in altitude. The height of the 0 • C isotherm was derived from the soundings at BCO on those days (Stephan et al., 2021).
On February 15 th elevated moisture transport (Villiger et al., 2021) lead to a widespread stratiform cloud layer in the altitude range 4 to 9 km (Fig. 15). The cloud layer is visible in all three CFAD range intervals. Low level clouds are visible mainly in the short range interval. Beam broadening at far ranges overestimate the vertical extent in both, the RHI and CFADs, respectively.
260 Figure 16 shows the CFAD from the HALO cloud radar (Mech et al., 2014;Ewald et al., 2019;Konow et al., 2021) while HALO was flying on its standard circle (Fig. 1) on an altitude of 11.2 km between 21:50 and 23:30 UTC. The airborne cloud radar has a high vertical resolution (ca. 31 m) and a much higher sensitivity (ca. −38 dBz at an range of 5 km) compared to Poldirad (ca. 2 dBz at an range of 120 km). The structure of the CFADs show good agreement in the vertical extent of the since the HALO radar has been calibrated by independent means (Ewald et al., 2019;Konow et al., 2021).

Data availability
All data are currently considered as preliminary since calibration and data quality are based on best knowledge (cf. Sect
Additional details are given in Sect. 3.7. The data are as they have been generated in real-time by the radar signal processor. A description of the radar parameters is given in Appendix B. Data processing and quality control is performed by the implemented receiver and signal processor. Using dual-polarization correlation coefficient ρ HV with a threshold of 0.7 proved to be a suited parameter to eliminate contamination of radar data by 310 sea clutter. A preliminary comparison with observations from the HALO cloud radar shows that the calibration of Poldirad is reliable, details will be subject for further detailed studies. Processed data are available as HDF5 files (Hagen et al., 2021a), as well as gridded reflectivity and rain rate fields (Hagen et al., 2021b).
Poldirad adds a valuable contribution to the multifaceted data gathered during the EUREC 4 A campaign. Preliminary analyses of reflectivity distribution show that Poldirad could capture rain cells up to a range of 250 to 300 km. The typical vertical extent 315 of shallow rain cells is in the range of 3 to 6 km, their size is in the order of 10 to 20 km. On Feb. 11 th a pronounced diurnal cycle as described by Vial et al. (2019) of precipitation was observed. On other days either no or only a weak diurnal cycle was observed, or the radar observations didn't cover the full day (cf. Fig. 7). The high temporal and spacial frequency of long For transportation Poldirad's antenna has to be disassembled into four panels (faintly visible in Fig. 2). Reassembling is performed with high accurateness on the ground before lifting the antenna to the pedestal. Even though the supporting structure is very rigid, it can not be guaranteed, that the antenna will be mounted with the very same shape like it had on a factory antenna range in December 1984. The shape of the parabolic reflector defines the shape of the radar beam. Initial measurements show 325 a nearly circular beam pattern with a half-power beamwidth of approximately 1 • (see Fig. 3 in Schroth et al., 1988). Detailed measurements of the shape of the radar beam were performed in 2012 by Reimann and Hagen (2016) using the sun and an external signal source.
Dedicated solar scans were repeated after the set up of the radar on Barbados. Figure A1 shows the measurements on Feb.
3 rd , 2020 similar to Fig. 4 in Reimann and Hagen (2016). The top image shows the individual measurement points during the 330 scan, azimuth and elevation angles have been corrected for high elevation angles as described by Reimann and Hagen (2016).
Maximum received power was −103.2 dBm. Using a two-dimensional least-square fit to Gaussian shape of the main lobe  While the central part of the radar beam is nearly circular, the outer part has an elliptical shape which is much wider in azimuth. The reasons causing this deformation are not clear. The implication of the deformation is that observed precipitation patterns appear to be stretched in azimuthal direction. Figure A2 shows a simulated circular rain cell centered at a range of 100 km and an azimuth of 60 • which is observed by a radar. Only the main lobe, no sidelobes have been considered in the radar simulation. Range effects, i.e. non-rectangular pulse shape are ignored. The left image shows the "true" reflectivity pattern, the 345 central image shows the broadening as it would be observed by a radar having an azimuthal beamwidth of 1 • (like most weather Figure A2. Simulated effect of broadening of a circular rain cell when observed by a radar. Left: simulated rain cell; center: observed with a radar with beamwidth 1 • ; right: as center image, but with Poldirad beam. Range rings every 5 km, radials every 2 • . radars). The right image shows a gaussian approximation 2 of the Poldirad beam (c.f. Fig. A1 lower left). The consequence of the broad radar beam is that the size of rain cells will be overestimated, while their maximum rain rate will be underestimated. Table A1 gives an estimation for the effect on maximum reflectivity, maximum rain rate, area where reflectivity is greater 7 dBz (rain rate greater 0.1 mm h −1 ), and mean rain rate of the cell. Appendix B: Radar products Table B1 gives a short description of radar product identifiers used in the HDF5 files. The two different polarization modes require different signal processing and thus different radar products are generated. Physical details of radar products and their interpretation can be found in text books such as Bringi and Chandrasekar (2001).
Author contributions. MH, FE, and SG are the lead investigators for the deployment, LO is caring for all technical aspects, DF and MF 355 provided all administrative, logistical, and scientific support on the island of Barbados. MG, JH, JR, EM, and ET supported the technical installation, JR additionally cared for the radar software refinements. GM, CK, QL, RCL, AD, DG, and KH additionally supported the measurements during EUREC 4 A and after the campaign.