Experimental study and modelling of rain drop size distribution in southern England

Modelling raindrop size distribution (DSD) is important for understanding, predicting and mitigating rain-induced attenuation in satellite signals in the millimetre band. Early studies suggested that the DSD followed an exponential distribution. Later studies showed lognormal and gamma distributions yielded better rainfall predictions when combined with radar data. However, in all these studies, goodness-of-fit of the models was only done by eye, sum of squared errors or accuracy of predicted rainfall. This work helps fill this gap by calculating chi-squared goodness-of-fit for distributions fitted using standard moment calculations, on 3 years´ data from Chilbolton´s observatory in southern England. Our results suggest that although the method of moments for fitting lognormal and gamma may produce reasonable rain rate predictions, the distributions produced do not actually match the DSDs, as neither lognormal nor gamma distributions fit the DSDs particularly well (we would expect around 95% non-reject rates). Our results equally suggest that the lognormal fits better than the gamma, and that the fits seem worse at higher rain rates.