Application of Machine Learning for Radio Refractivity Prediction across Climatic Zones in West Africa

Abstract

This research explores the use of machine learning (ML) models to predict surface radio refractivity in diverse climatic zones across West Africa. We evaluated three machine learning approaches for predicting surface radio refractivity across West Africa’s diverse climates, using data from five locations: Niamey, Bamako, Monrovia and Jos. Models used including; LightGBM, Random Forest, and Long Short-Term Memory (LSTM) trained on ERA5 reanalysis atmospheric dataset, optimized through grid search and early stopping to improve predictive accuracy. Result showed LightGBM performed best overall (MAE: 35.014, MSE:1918.844), though Random Forest achieved slightly lower in some regions but struggle with overall accuracy (MAE: 113.457, MSE: 59707.833). The analysis reveal wind speed and solar radiation as the key predictive factors across different climate zone. These models offer practical improvements over traditional methods for radio wave propagation, forecasting particularly valuable for telecommunication planning in regions with limited observational data.