Evaluation of two integrated techniques to estimate the rainfall rates from polarimetric radar measurements and extensive monitoring of azimuth-dependent ZDR and phiDP biases

This work presents a comparative evaluation between two so-called ‘integrated' algorithms to estimate the rainfall rate from polarimetric radar data. The two algorithms are ZPHI (Testud et al. 2000) on the one hand, which entirely relies on the horizontal reflectivity (ZH) and the differential phase (fDP), and ZZDR (Thompson 2007) on the other hand, which makes use of the horizontal reflectivity (ZH) and the differential reflectivity (ZDR). Both algorithms are said to be ‘integrated' in the sense that the Drop Size Distribution (DSD) characteristics, NW essentially, are retrieved over subdomains of the radar image and then used at the pixel scale to convert the reflectivity into rainfall rate. This approach reduces the noise inherent in the ZDR and fDP measurements when using operational radars with rapid antenna rotation rate and finite beam widths. Both algorithms have been applied to data collected by the French C-band operational polarimetric Trappes radar located near Paris. 12 major episodes have been selected and the radar-derived rainfall estimations are compared at the hourly time step against a dense rain gauge network. Both algorithms are only valid in rain and require, in a first step, a rigorous identification of non-rain echoes such as ground-clutter, clear-air, bright band, snow and hail.

As a first conclusion, we show that calibration issues, whether of ZH (case of ZPHI and ZZDR) or of ZDR (ZZDR only), are extremely important. We demonstrate that the radome induces azimuth-dependent biases both on ZH and ZDR. The ZDR azimuth-dependent calibration biases, retrieved by assuming that the intrinsic ZDR of 20 – 22 dBZ rain is 0.2 dB, are extensively analyzed using several tens of episodes from several polarimetric radars of the French operational network. The “reproduceability” of the curves is not always evident. Wet radome and / or rotary joint effects may make the quantitative use of ZDR more difficult than initially anticipated. The same result is found for the fDP0 curve, which also shows azimuth-dependent variations and a sometimes large event-to-event variability. This questions the static normalization of fDP profiles.

The analysis of the results reveals that the ZPHI algorithm is triggered (at C-band) for rain rates above 3 – 4 mm h-1, which can be considered as moderate to heavy rain according to (northern) European standards, yet the most critical ones in terms of hydrological consequences. In those cases, the attenuation correction together with the NW (equivalently N0*) adjustment significantly improve the quality of the rainfall estimation.

The ZZDR algorithm is triggered much more often than ZPHI and also improves the radar – rain gauge scores. The applicability of ZDR to heavy rain and attenuated areas depends critically on the performance of the attenuation correction. The sensitivity of the ZZDR algorithm is investigated by simulating a bias on ZDR and computing the resulting radar – rain gauge bias. A +0.3 dB bias on ZDR typically leads to a 30% bias of the radar rainfall estimation.