Sources of Errors in Rainfall Measurements by Polarimetric Radar: Variability of Drop Size Distributions, Observational Noise, and Variation of Relationships between R and Polarimetric Parameters

Abstract

Using a set of long-term disdrometric data and of actual radar measurements from the McGill S-band operational polarimetric radar, several sources of errors in rain measurement with polarimetric radar are explored in order to investigate their relative importance and the feasibility of a polarimetric technique for estimating R in the context of the McGill S-band operational radar that performs a full volume scan of 24 plan position indicators (PPIs) every 5 min. The sources of errors considered are the variability of drop size distributions (DSDs), observational noise, and systematic variation of the relationships between R and polarimetric parameters at different climate regimes. Additional polarimetric parameters dramatically reduce the effect of the DSD variability on rain estimates by radar. The effectiveness of various multiparameter relationships is investigated. The relationships from the literature that are derived from the DSD model and measured DSDs at a different climate regime differ from those derived from the disdrometric dataset herein. An application of these relationships to the Montreal dataset results in a bias (about 10%?20%) and the significant random error resulting from the DSD variability. These errors should be eliminated by using a relationship suitable for the local climate. Assuming a measurement noise as expected from a slow scanning polarimetric radar [?1 rotation per minute (rpm)] and a 10-min smoothing, the R ? (Zh, ZDR) relationship outperforms the conventional R ? Zh because of the combined effect of the DSD variability and measurement errors. In addition, the marginal measurement noise that is required to have the same accuracy of R ? Zh and R ? (Zh, ZDR) algorithms is obtained as a function of temporal smoothing. The quantified measurement noise of the McGill S-band fast scanning operational radar (?6 rpm) is significantly larger than that of a slow scanning radar, implying that a temporal averaging of ZDR of 1 h is needed to achieve some gain with R ? (Zh, ZDR).