Seasonal Variation and Machine Learning-Based Prediction of Atmospheric Refractivity over Southwestern Nigeria

Abstract

Atmospheric radio refractivity strongly influences radio wave propagation, link reliability and signal quality in tropical regions. In south-western Nigeria, pronounced wet-and-dry seasonal transitions lead to substantial variability in refractivity, yet data-driven predictive tools remain limited. This lack of reliable prediction poses a challenge for communication system planning and performance optimisation in the region. This study investigates the seasonal variation in surface atmospheric refractivity over south-western Nigeria and evaluates the effectiveness of machine learning techniques in predicting it. Four years of meteorological data from an automatic weather station were used to compute refractivity using the Smith–Weintraub formulation. Seasonal and monthly patterns were analysed using descriptive statistics, correlation analysis, and time-series techniques, including trend and anomaly detection. Supervised machine learning models, including Support Vector Regression, Random Forest, Gradient Boosting, and a Multi-Layer Perceptron, were trained to predict daily refractivity from meteorological inputs. Results show seasonal dependence, with higher refractivity during the wet season driven by increased humidity. Among the models, Support Vector Regression achieved the highest predictive accuracy (R² ? 0.9999, RMSE ? 0.066), followed by Random Forest. The findings demonstrate that machine learning is a reliable and effective approach for predicting atmospheric refractivity and capturing its seasonal variability in tropical environments.