RAINDROP SIZE DISTRIBUTION MODELING FOR RADIO LINK DESIGN ALONG THE EASTERN COAST OF SOUTH AFRICA

Abstract

A study of the raindrop size distribution along the eastern coast of South Africa (Durban) is presented. The Biweight kernel estimator based on distometer measurement is used to determine the best estimate of the measured raindrop size probability distribution function (pdf). The best kernel estimator, which results in the lowest integral square error (ISE), is used to measure the closeness of the estimated lognormal and gamma pdf of raindrop size to the measured raindrop size distribution. It is established that the optimised lognormal pdf slightly outperforms the optimised gamma pdf in terms of the mean ISE and the RMSE values, with mean ISE values of 0.026 for lognormal and 0.04 for gamma distributions, respectively, and corresponding mean RMSE values of 0.073 and 0.081, respectively. The method-of-moments gamma and lognormal distributions are observed to be worse estimators of the measured pdf than the two optimized distributions. The N(D) distributions using the optimised lognormal and gamma distributions for the region are compared with those for difierent tropical regions, namely, India, Singapore, Nigeria, Indonesia, and Brazil. While the Indian lognormal N(D) model gives the highest peak for low raindrop sizes for all rain rates, Durban's gamma and lognormal models exhibit the widest raindrop size spread over all rain rates ranging from 1{120mm/h. Finally, the speciflc attenuation due to rain using the Durban models are compared against the ITU-R models and actual measurements over a 19.5GHz LOS link; the results indicate a need for further work involving both distrometer and radio link measurements for rain rates exceeding 30mm/h in the eastern coast of South Africa.