Advances in Quantitative Precipitation Estimation with Polarimetric C-Band Radar Networks

Abstract

Areal quantitative precipitation estimation (QPE) with high spatial and temporal resolution is crucial to feed nowcasting and hydrological models. Ground-based weather radars are the best-suited tool for this purpose. Despite the upgrades of the German C-band radar network to polarimetry, the operational QPE products offered by the German Meteorological Service (DWD, Deutscher Wetterdienst) are still based mainly on the traditional radar variable, reflectivity Z, leading to large uncertainties. To enhance QPE quality from DWD radars, this study presents several novel polarimetric QPE algorithms at C-band. These include i) hybrid rainfall estimators based on specific attenuation A and specific differential phase KDP, denoted as R(A, KDP), ii) a polarimetric vertical profile of reflectivity (PVPR) correction method for Z biases within and above the melting layer (ML), and iii) a KDP-based snowfall estimator. Additionally, a warm-rain event is investigated to develop approaches that mitigate the impacts of vertical precipitation gradients below the ML on QPE.<br /> The S-band R(A) algorithm has been proven to be a powerful tool for rainfall estimation and has recently been operational in the U.S.A. The attenuation parameter α required for the A calculation, however, is susceptible to the variability of drop size distributions (DSD), which is more accentuated at C-band compared to S-band due to stronger resonance effects. This limits the applicability of R(A) to C-band radars. To overcome this challenge, this study derives DSD-dependent α values based on the slope of differential reflectivity ZDR against Z. Furthermore, in cases of heavy rain where Z > 40 dBZ, the R(A) relationship at C-band becomes relatively sensitive to the DSD variability and therefore R(KDP) is used. The resulting hybrid algorithms R(A, KDP) surpass in quality the DWD real-time QPE product, particularly for convective rain where the normalized root-mean-square error (NRMSE) is reduced by 13% and the normalized mean bias (NMB) by 16%.<br /> On 14 July 2021, western Europe experienced severe floods caused by intense stratiform precipitation, for which the above rainfall algorithms underestimated surface rain with an NMB of -30% owing to the increased rain rates below the observing altitudes of the operational radars. Since most existing vertical profile of reflectivity (VPR) correction techniques focus on data within and above the ML, this study proposes a vertical profile (VP) correction method that considers vertical precipitation gradients near the surface for both Z and KDP. This correction method involves i) projecting radar observations from the bottom of the ML down to 700 m taking range-defined quasi-vertical profiles (RD-QVP) as references, and ii) deriving rainfall relationships from vertically-pointing radar data in the lower few hundred meters, in order to better capture warm-rain processes. Moreover, observations from a local X-band radar are utilized to fill gaps in the operational radar network. Results show that the VP correction method reduces the NRMSE and NMB values of the estimates by more than 20%, and with the additional use of the gap-filling radar, the R(A, KDP) retrievals even outperform the hourly gauge-adjusted QPE product from DWD.<br /> When the radar beam intersects and exceeds the height of the ML, significant errors in surface precipitation estimates are induced due to Z biases caused by partially frozen and clumped hydrometeors within the ML and beam-broadening effects above it. Current VPR correction approaches mitigating the biases rely solely on Z and do not explore the potential of polarimetry. Here, the - as far as is known - first PVPR correction method for C-band is developed. This method reconstructs the intrinsic VPR and estimates Z biases within and above the ML based on the statistics of polarimetric profiles. For pure and uniform stratiform rain, the rainfall retrieval based on the corrected Z has an up to 20% lower NRMSE value than the hydrometeor-type-specific rainfall retrieval suggested by Giangrande and Ryzhkov (2008); and for estimates above the ML, the PVPR-corrected retrieval is more accurate than the polarimetric snowfall retrieval proposed in this study.<br /> The PVPR correction method is rendered inapplicable when snowfall reaches the ground. Additionally, the accuracy of snowfall estimation using Z is restricted due to the high diversity of snowflakes. To solve this issue, Bukovčić et al. (2020) introduced a generalized polarimetric snowfall estimator using KDP and demonstrated its efficacy with S-band radar observations. This estimator considers changes in the shape and orientation of snowflakes or ice crystals by incorporating an assumed aspect ratio ar and width of the canting angle distribution σ in a power-law snowfall relationship. Our adjustment of the algorithm to C-band yields better estimates than conventional Z-based retrievals, and it additionally mirrors the typically higher ice concentration within the dendritic growth layer (DGL) at higher levels.<br /> The novel QPE algorithms exhibit promising results and hold potential for real-time applications to C-band radar networks throughout Europe. These advancements profit from the strong cooperation with DWD in the program “Near-Realtime Quantitative Precipitation Estimation and Prediction (RealPEP)”. Currently, the algorithms are implemented in DWD’s operational platform “Polarimetric Radar Algorithms (POLARA)” in order to assess their stability, feasibility, and robustness for online applications. In the near future, optimization of the algorithms will continue based on evaluations using long-term databases in POLARA.