Akademik Veri Yönetim Sistemi
3rd International Conference on Optimization and Data Science in Industrial Engineering, ODSIE 2025, Hybrid, Istanbul, Türkiye, 20 - 22 Kasım 2025, cilt.2854 CCIS, ss.440-455, (Tam Metin Bildiri)
Global climate change is reflected in persistent shifts in temperature anomalies, which create challenges for detection and modeling that need to be addressed. A key methodological gap arises in the reliable analysis of long-term anomaly series under strong nonstationarity, since traditional approaches often depend on assumptions of normality, constant variance, or independence. To address this, we develop a framework that integrates robust quality-control charts, ARIMA/SARIMA models, and Hidden Markov Models, which are free from distribution. Robust CUSUM and EWMA charts are used to identify structural drifts while minimizing false alarms, and they prove effective in detecting sustained warming signals. Time series modeling shows that SARIMA improves on ARIMA, reducing the root mean square error (RMSE) from 0.0970 to 0.0964 (−0.6%) and the mean absolute error (MAE) from 0.0753 to 0.0747 (−0.8%), with corresponding gains in log-likelihood and information criteria. A Hidden Markov Model with seven Gaussian states, selected based on the minimum BIC value (−2975), captures distinct climate regimes, which allows regime-wise modeling that reduces residual bias and improves interpretability. These findings show that nonparametric methods and regime partitioning strengthen robustness in long climate anomaly series. The contribution of this study is to provide a strong methodological basis for analyzing long term warming trends and to establish a foundation for future climate risk assessment and adaptation planning.