Three different Quantitative Precipitation Estimations (QPE) are produced in real-time on a 5-minute basis : the first one, referred to as CONV, is a conventional estimation that only relies on horizontal reflectivity converted into rainfall rate using Marshall-Palmer. It is the current state-of-the-art of the operational radar QPE at Météo France and stands therefore as the benchmark to beat in the conditional evaluation. The two other estimations, reffered to as POL1 and POL2, are obtained with Testud's ZPHI algorithm : POL1 only differ from CONV by the fact that it includes a correction for attenuation by rain. POL2 includes a correction for attenuation but also an adjustment of the N0* parameter.
The three radar QPE are compared on a hourly basis with a very dense gauge network. All rainfall episodes are included in the comparison. The originality of the validation approach lies in the fact that the results are stratified as a function of the severity of each possible source error : attenuation, N0* fluctuation, bright band contamination, partial beam blocking, ... Such a stratification, which is rendered possible by the size of the dataset, allows assessing the specific improvement of dual-polarization for each kind of errors.
Results are very promising, the most impressive improvement being the correction for attenuation in convective precipitation. The N0* adjustment has generally a positive impact. However, as the adjustment is only performed in the rain region, its impact is severely limited in the winter season because the radar beam is very rapidly in the melting layer.