Predicting A Monotonic Trend of The Surface Air Temperature Using Generalized Additive Model (Gam) Over North-Eastern Region of Nigeria

Abstract

Given the growing global concern over climate change, understanding how temperature patterns 
evolve over time is increasingly important. Analyzing long-term temperature data can provide 
valuable insights into these trends, helping researchers and policymakers assess climatic shifts 
and their potential impacts. Monthly temperature records spanning several decades offer a 
robust foundation for identifying such patterns, revealing both seasonal fluctuations and 
broader climatic trends. Traditionally, linear regression models have been widely employed to 
analyze temperature data due to their simplicity and interpretability. While these models can 
effectively capture annual temperature variations, they have significant limitations in 
representing long-term trends, especially when those trends are non-linear. Climate systems 
often exhibit complex, non-stationary behavior that linear models fail to adequately capture, 
necessitating more flexible modeling approaches. Generalized Additive Models (GAMs) provide a 
powerful alternative by fitting smooth, non-linear functions to the data, allowing for a more 
nuanced representation of temperature patterns. Unlike rigid linear assumptions, GAMs adapt to 
the underlying structure of the data, making them particularly useful for detecting seasonal 
variations and gradual climatic shifts over extended periods. This study explores how GAMs can 
enhance the modeling of temperature trends by applying them to a 41-year dataset (1981–2022) 
obtained from the Nigerian Meteorological Agency (NiMet). The analysis demonstrates that 
GAMs offer superior flexibility in capturing both seasonal cycles and long-term warming trends. 
Graphical representations of smoothed model estimates reveal a clear rise in regional 
temperatures over the past four decades, aligning with broader observations of global warming. 
While GAMs are highly effective in identifying non-linear relationships, their performance 
depends on the careful selection of smoothing functions. The choice of basis functions and 
smoothing parameters must align with both the data characteristics and the modeling objectives 
to ensure accurate and interpretable results. By leveraging GAMs, this study underscores their 
utility in climate research, providing a more refined tool for analyzing temperature trends in a 
warming world.