Five years calibrated observations from the University of Bonn X-band weather radar (BoXPol)

Polarimetric weather radars offer a wealth of new information compared to conventional technology, not only to enhance quantitative precipitation estimation, warnings, and short-term forecasts, but also to improve our understanding of precipitation generating processes and their representation in numerical weather prediction models. To support such research opportunities, this paper describes an open-access dataset between 2014–2019 collected by the polarimetric Doppler X-band weather radar in Bonn (BoXPol), western Germany. To complement this dataset, the technical radar characteristics, scanning strategy and the best-practice for radar data processing are detailed. In addition, an investigation of radar calibration is presented. Reflectivity measurements from the Dual-frequency Precipitation Radar operating on the core satellite of the Global Precipitation Mission are compared to those of BoXPol to provide absolute calibration offsets with the dataset. The Relative Calibration Adjustment technique is applied to identify stable calibration periods. The absolute calibration of differential reflectivity is determined using the vertical scan and provided with the BoxPol dataset. Measurement(s) Radar backscattering of precipitation Technology Type(s) Polarimetric Doppler X-band weather radar Measurement(s) Radar backscattering of precipitation Technology Type(s) Polarimetric Doppler X-band weather radar

the scientific exploitation of polarimetric radar data. Therefore, calibration offsets and best-practice processing scripts that utilise libraries from wradlib 26 and Py-ART 27 are provided with the BoXPol dataset.
Section Methods of this paper includes the technical description and scan strategy of the polarimetric X-band radar in Bonn (BoXPol), while Section Technical Validation outlines the calibration of horizontal reflectivity and differential reflectivity and the recommended data processing and correction algorithms for ground-based radar observations. An overview of the archive and the data formats is presented in Section Data Records.

Methods
The BoXPol weather radar is located in Bonn (50.7305° North and 7.0717° East), Germany, at 99.5 m above mean sea level (Fig. 1) on the rooftop of a 30 m building next to the department of meteorology of the University of Bonn. The hardware consists of a radome-less EEC DWSR-2001-X-SDP weather radar operating in Simultaneous Transmit and Receive of H and V channels (STAR) mode using an Enigma signal processor (Enigma 3 upgraded to Enigma 4 in April 2017). The random-phase magnetron system operates at a frequency of 9.3 GHz and employs a scanning strategy consisting of ten different plan position indicator (PPI) scans with elevation angles between 1° and 28°, a birdbath scan (90°) and a range-height indicator (RHI) scan within a 5 minutes scan schedule (approximately 30 s per scan). The technical characteristics are displayed in Tables 1,  2 summarizes elevation angle, maximal range, range resolution and the pulse repetition frequency (PRF) for all PPI scans with Enigma 3 and 4, respectively, i.e. before and after April 2017. The azimuthal resolution is 1° while the range resolution depends on the scan configuration (Table 2) and varies between 25 and 150 meters. The lowest PPI measured at 1° covers 150 kilometers range. A beam-blockage map and its derivation based on specific attenuation is provided in 28 .

Data Records
The archive dataset consists of daily netCDF files (Conventions CF-1.7 (https://github.com/cf-convention/ cf-conventions) and following Cf/Radial-2.1 (no standard yet)) for each of the ten PPI scans (birdbath scan and RHI will be included in later versions) and includes the following polarimetric variables: reflectivity at horizontal polarization (Z H ), reflectivity at vertical polarization (Z V ), differential reflectivity (Z DR ), cross-correlation coefficient (ρ hv ), total differential phase (Φ DP ), uncorrected horizontal/vertical reflectivity factor (T H , T V ), horizontal/vertical radial velocities (V H , V V ) and horizontal/vertical spectral width of radial velocity (W H , W V ). Note that Z H and Z V are corrected for clutter, speckle, interference and second/third trip echoes by the radar processor. In relation to these corrections, a clutter map (CMAP) is also available since April 2017. Calibration offsets (see Fig. 2 and Table 3), however, need to be applied by the user. The data is archived by the DKRZ (German Climate Computing Centre 29 ) technical Validation Processing of ground based radar data. Accurate absolute calibration of radar data requires a thorough preprocessing. Even though raw data is provided, the algorithms we applied before the calibration are outlined in the following as an optional guideline. First, the BoXPol polarimetric moments are filtered for erroneous observations by excluding reflectivities Z H lower than −20 dBZ and higher than 80 dBZ, differential reflectivities Z DR lower than −6 dB and higher than 7 dB, differential phase texture SD(Φ DP ) higher than 20° 30 and cross-correlation coefficient ρ hv lower than 0.6 to remove non-meteorological signals. The SD(Φ DP ) is the spatial variability of Φ DP , expressed as the root mean square difference in a region of three pixels in range and azimuthal direction. This variable is e.g. used in 30 to distinguish between precipitating and non-precipitating echoes. We follow 31 to process raw Φ DP with linear programming to provide improved estimates of Φ DP , in the following referred to as processed Φ DP , and to derive specific differential phase K DP (Py-ART 27 ,). K DP values lower than −4° km −1 and  www.nature.com/scientificdata www.nature.com/scientificdata/ higher than 15° km −1 are excluded from the dataset. In the ensuing step processed Φ DP is used for attenuation correction using the ZPHI method 32 . The correction is only applied to the liquid region below the freezing level determined with the ERA5 geopotential height and dry bulb temperature profiles on pressure level dataset 33 following 34 . Linear interpolation was applied to get the geopotential height exactly at the 0 °C level. Based on the 3 second resolution Digital Elevation Model (DEM) from NASA's Shuttle Radar Topography Mission 35 , the method from 36 , implemented in the wradlib library 26 , is applied to determine partial beam-blockage (PBB). Areas showing PBB >10% are excluded to improve the accuracy of calibration retrievals. For example Fig. 1 illustrates affected areas for the PPI at the lowest (1°) elevation angle 37 .   Table 3. Calibration offsets for BoXPol's horizontal reflectivity (Z H ) and differential reflectivity (Z DR ) in the selected periods.
www.nature.com/scientificdata www.nature.com/scientificdata/ Calibration of horizontal reflectivity. We applied the relative calibration adjustment (RCA) technique to determine stable calibration periods and also volume matching with a satellite-based precipitation radar of the Global Precipitation Mission Core-satellite (GPM 38,39 ) for the absolute calibration. In contrast to conventional calibration methods [40][41][42][43] , these two calibration techniques do not require any changes with the operational scan strategy or extra hardware installations. Furthermore 39 , demonstrated that the use of the self-consistency technique for calibration purposes as described in 44 requires additional local disdrometer measurements to determine the relationship between K DP /Z H and Z DR . Without this assumption the difference between characteristic drop size distributions in the mid-latitudes used in 45 and the tropical case in 39 led to 2 dB difference in calibration. The Dual-frequency Precipitation Radar (DPR) observations are well-calibrated using internal and external calibration 38 with an accuracy within ±1 dB 46 and the satellite-based measurements are freely available (https:// storm.pps.eosdis.nasa.gov/storm/). The RCA technique can be applied continuously even in absence of precipitation.
Relative calibration. The RCA method exploits statistics generated from local stable clutter 39,47 to detect changes in calibration offsets. Radar pixels within 20 km range of the lowest scan are identified as stable clutter if the uncorrected reflectivity is 50 dBZ or higher in at least 50% of the daily measurements. The 95th percentile of all reflectivity samples within the persistent clutter bins is then used to estimate the relative calibration for that day. Application of RCA to the BoXPol data set reveals four significant changes in calibration across the period with GPM overpasses, namely on 2014-06-01, 2015-04-25, 2016-06-24 and 2017-05-19 (Fig. 2). Indeed, radar hardware changes, operational changes or radar services occurred on these dates, which confirms the reliability of the method. For each stable period identified between two subsequent changes in the RCA time series (Fig. 2,  top), the GPM radar measurements (more details on the GPM measurements are provided in section ' Absolute Calibration') are used to determine the respective mean absolute calibration values (Fig. 2, center). The RCA time series is not sufficiently stable to provide relative calibration based on the mean GPM offset for each period, as recommended by 39 . Rather, the RCA time series shows strong seasonal variability, with increased values during warmer and decreased values during cooler months. Therefore we use the RCA time series only to select stable periods to use for calibration with GPM and do not apply the RCA analysis for calibration. The overall mean and standard deviation of the daily stable clutter pixels used for the relative calibration is also indicated in Fig. 2 (top). We hypothesize these seasonal variations are the result of the annual temperature cycle, however, similar findings have not been documented before and further investigations are suggested to corroborate this connection.
Absolute calibration. Due to lower attenuation compared to K α -band, this study exploits the K u -band (13.6 GHz) measurements of the DPR on board of the Global Precipitation Mission (GPM) for calibration of the ground-based radar BoXPol. The K u system has a footprint of 5 km, 125 m vertical resolution and 245 km swath width 48,49 . GPM overpasses for Germany occur approximately twice per day and we selected all overflights in the period from 8 August 2014 (first rain event in BoXPol area after GPM launch) to 8 April 2019 with more than 1% of the BoXPol region covered with precipitation. This region is defined between 51.4° and 49.4° north and 9.0° and 5.8° east. The GPM data (version 5, file specification 2AKu) are freely available. Specific GPM parameters required for the calibration technique are the quality index (dataQuality), zenith angle (localZenithAngle), precipitation flag (flagPrecip), bright band height (heightBB), bright band width (widthBB), bight band quality (qual-ityBB), precipitation type (typePrecip), precipitation type quality (qualityTypePrecip) and attenuation corrected reflectivity (zFactorCorrected). For more detailed information about specific GPM parameters we refer to 38 .
In this technique, the K u -band radar bins of the space-borne radar (SR, Fig. 3b) are geometrically matched with the radar beams of the ground-radar (GR, Fig. 3a) to enable the comparison of identical volumes. Therefore all BoXPol bins located in a DPR footprint and all DPR bins from the same footprint located vertically within the BoXPol radar beam are identified and averaged (matched). The averaged reflectivities of the GR bins corresponding to the DPR footprints are shown in Fig. 3c and the averaged reflectivities of the SR bins corresponding to a vertically GR beam width are shown in Fig. 3d. The generated matched volumes are used for the calibration (see Fig. 3c/d, more details on the matching are shown in Fig. 2 of 38 ). The Z H offset is calculated by subtracting the SR reflectivity from the GR reflectivity (Fig. 3e) followed by averaging over all matched samples identified in one overpass (Fig. 3f). In order to take differences between the frequencies into account, the K u -band reflectivity (Z H (K u )) is first converted to X-band (Z H (X)) following mainly the S-K u band conversion introduced by 50 . We performed T-matrix scattering simulations 51 for rain, dry snow and dry hail to simulate the reflectivities at K u and X-band. Drop size distributions, particle orientation, the complex dielectric constant and the aspect ratio are simulated as in 50 . We calculated the aspect ratio for snow following 52 and for hail following 53 and the dielectric constant is calculated according to 54 . Thus, to convert the SR measured at K u -band to X-band the following equation is applied: The last term is the dual-frequency ratio with the specific coefficients for the frequency conversion c i in rain, dry snow and dry hail are provided in Table 4. The overall accuracy of the frequency conversion is 0.23 dB for rain, 0.42 dB for dry snow and 0.20 dB for dry hail. Note that the transformation for hail is only used if the DPR has detected convective precipitation above the bright band.
GPM overpasses containing at least 10 valid precipitation samples (indicated by the flagPrecip fields in GPM files) within 20 to 150 km range from the ground radar site have been selected for the comparison with the www.nature.com/scientificdata www.nature.com/scientificdata/ BoXPol dataset. Searching for the closest radar volume in time for each GPM overpass a maximum time difference of 2.5 min between the BoXPol volume start time and the GPM overpass time was allowed. Following 38 , we verified the sensitivity of the GR-SR difference to the GR (Fig. 4, left) and SR (Fig. 4, right) reflectivities for all matched volumes and all overpasses to identify the reflectivity thresholds for the calibration. Only reflectivities between 19 dBZ and 25 dBZ have been taken into account. The upper threshold mitigated the impact of   Table 4. Coefficients for relationship to convert reflectivities at K u -band to X-band. The coefficients are given for rain, dry snow and dry hail. www.nature.com/scientificdata www.nature.com/scientificdata/ uncertainties in the attenuation correction of DPR measurements and the lower threshold is due to SR sensitivity. Matched samples with standard deviations greater than 4 dBZ or observations contaminated with bright band are removed. The minimum number of pairs in one matched volume is set to 20. With these constraints, 85 valid DPR overpasses for liquid and solid precipitation in the BoXPol region have been identified for the dataset. Absolute Z H offsets for periods identified as stable with the RCA are provided in Table 3 with standard deviations ranging between 1.66 dB and 2.15 dB. Similar deviations can be found in 39 and 37 . The overall mean calibration offset and standard deviation are 0.03 and 1.73 dB (see also Fig. 2, center). Note that the calculated reflectivity calibration offset in Fig. 2 (center) have to be subtracted from the ground-based radar reflectivity.
Calibration of differential reflectivity. Calibration time series for differential reflectivity Z DR are determined using the measurements in light precipitation with the birdbath scan at 90 deg elevation 55,56 . Since the mean canting angle of small raindrops is close to 0°, they appear spherical seen from below, which implies that both the Z V and Z H are expected to be equivalent and deviations from Z DR = 0 dB can be exploited for calibration. According to 57 this can be applied to all hydrometeor classes. For our study we consider all regions except the melting layer. Azimuthal averaging has been performed to reduce noise and measurement within the first 600 meters have been excluded due to possible clutter contamination. To avoid any biases introduced by strong precipitation events and melting particles, only samples with Z H < 30 dBZ and ρ HV > 0.99 have been included in the analysis. All data 250 meters above and below the freezing level are also removed. Here, the freezing level height derived from the ERA5 reanalysis is used again. To exclude turbulence, only fall velocities below 1 ms −1 are allowed. The median of remaining data points between the 10 and 90 percentile provides the Z DR offset ( 57-59 , Fig. 2, bottom). Note that the calculated Z DR calibration offset illustrated in Fig. 2 and listed in Table 3 have to be subtracted from the measurements. The overall standard deviation of the Z DR offset is 0.26 dB. The standard deviations in the specific periods are within the required uncertainty range between 0.1 dB and 0.2 dB 57 . The daily standard deviations also satisfy this condition with the exception of a few specific days (red colored points in Fig. 2 bottom).

Code availability
The described Z H calibration and correction procedures are available in the python packages wradlib 26 , Py-ART 27 , cluttercal (https://github.com/vlouf/cluttercal https://github.com/vlouf/cluttercal) and gpmmatch (https://github. com/vlouf/gpmmatch) and demonstration scripts for data visualization, processing and absolute calibration are provided as part of the data repository and the published relative calibration codes are also available on github (cluttercal).