COMPARISON OF PARAMETRIC, NON-PARAMETRIC AND MACHINE LEARNING TECHNIQUES FOR TEMPERATURE TREND IDENTIFICATION

Abstract

Understanding local climate trends is critical for effective adaptation planning, particularly in rapidly urbanizing regions like West Africa. This study examines temperature changes in Kano, Nigeria, from 1993 to 2023 using a multi-method approach. We combined parametric techniques (linear regression, ARIMA), non-parametric tests (Mann-Kendall, Sen’s Slope), and machine learning models (Random Forest, SVM) to analyze monthly maximum temperatures, with careful attention to seasonal effects. Our analysis identified a gradual warming trend of 0.16°C per decade—significantly slower than the 0.3°C regional average for West Africa. More strikingly, seasonal patterns accounted for over 80% of temperature variability, with April temperatures consistently peaking 10-15°C above January lows. The research highlights how simple models with categorical month variables (achieving R² = 0.88) can outperform both basic linear approaches and complex algorithms in tropical climate studies. These findings have important practical implications: they suggest that local factors may be moderating temperature rise in Kano, and they underscore the need for climate adaptation strategies that prioritize managing seasonal extremes rather than just long-term warming trends. The study demonstrates the value of combining multiple analytical methods to produce robust, actionable insights for urban climate planning in tropical regions.