Forecasting Climate Trends with Stochastic Models: Comparative Analysis of TRAMO/SEATS-Based RegARIMA vs Seasonal ARIMA Models

Time series analysis plays a crucial role in understanding climate variability and supporting reliable forecasting efforts. In this study, the forecasting performance of TRAMO/SEATS-based RegARIMA and SARIMA models is evaluated using long-term meteorological data (1960–2020) from Beyşehir and Seydişehir stations in Konya, Turkey. The TRAMO (Time Series Regression with ARIMA Noise, Missing Observations, and Outliers) approach corrects for outliers, missing data, and deterministic components, while SEATS (Signal Extraction in ARIMA Time Series) extract stochastic components such as trend and seasonality. The seasonality structure of the data is assessed using the Webel-Ollech test, and stationarity is examined via the Augmented Dickey-Fuller test. Following seasonal adjustment with TRAMO/SEATS, more robust forecasting models are constructed. Comparative analyses reveal that TRAMO/SEATS-based RegARIMA models outperform SARIMA models, yielding lower Mean Square Error (MSE: 0.25–5.22) and Root Mean Square Error (RMSE: 0.50–2.29), alongside higher Nash-Sutcliffe Efficiency (NSE: 0.69–1.00). These findings demonstrate the superiority of the TRAMO/SEATS approach in modeling complex seasonal structures and improving forecast accuracy. This study provides a scientific basis for understanding regional climate dynamics and supports the development of effective climate change adaptation and risk mitigation strategies.