Synthetic storm technique and machine learning algorithms for rain attenuation characterization across selected locations in tropical Nigeria

Abstract

One of the major problems of wireless communication systems is rain attenuation, especially in tropical and sub-tropical areas of Nigeria. Interactions between the water molecules and electromagnetic radiation leads to the scattering and absorption of radio signals, and a reduction in their amplitude and strength. This study employs a hybrid approach of the Synthetic Storm Technique (SST) and the Machine Learning (ML) models for rain attenuation characterization and prediction across eight (8) selected locations in Nigeria. Ten (10) years rain rate statistics, at 30 min integration time, were obtained from the archive of the Global Precipitation Mission (GPM) of National Aeronautics and Space Administration (NASA), and converted to 1 min integration time for the period between January 2010 and December 2019. ITU-R P.618–13 and Synthetic Storm Technique (SST) models were used for the computation of rain specific attenuation and rain attenuation timeseries for each station. The rain attenuation timeseries were grouped into two categories A and B, and fed into four (4) supervised machine learning algorithms for model development and implementation, respectively. Findings demonstrate that the hybrid SST-ML model is capable of recreating predicted rain attenuation across various climatic regions, and the best performing ML model is the RF with minimal RMSE values of 0.015388 and 0.799156 in Ilorin and Sokoto, respectively, and R² values approaching unity across many of the stations. The study underscores the need for satellite services operators to embrace site-specific, machine-learning-based fade margin analysis for Nigeria to achieve 99.99% link availability at Ku-band frequencies and above.