A comparative analysis of forecasting models for air temperature: Classical time series, machine learning, and functional approaches

Accurate temperature forecasting is essential for tackling current environmental, economic, and public health issues influenced by global climate changes. This study presents a new method for high-resolution temperature prediction by combining functional time series (FTS) decomposition with improved modeling techniques. The hourly air temperature (AT) data is broken down from Istanbul, Turkey, into deterministic and random parts to better track complex temporal patterns and short-term changes. Using a functional autoregressive (FAR) model together with standard methods including autoregressive integrated moving average (ARIMA), vector autoregressive (VAR), artificial neural networks (ANN), and autoregressive neural networks (ARNN), one-step-ahead forecasting is conducted for an entire year. Our results clearly show that the FAR model stands out. It significantly outperforms all other methods on key accuracy metrics like mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE). This results in the smallest forecast errors and the highest predictive consistency. This research supports more accurate decision-making in areas such as agriculture, energy management, and public safety. It makes a valuable contribution to societal resilience given climate uncertainty. The results demonstrate the clear superiority of the FAR model, which achieved the lowest error rates with a MAE of 10.39, a MAPE of 7.83%, and a RMSE of 9.49. This results in the smallest forecast errors and the highest predictive consistency. The findings highlight how important functional data analysis is in weather modeling and provide a strong framework for improving temperature forecasts. This research supports more accurate decision-making in areas such as agriculture, energy management, and public safety. It makes a valuable contribution to societal resilience given climate uncertainty.