A comparative analysis of advanced modeling techniques for global methane emission forecasting using SARIMA, LSTM, and GRU models
Clean Technologies and Environmental Policy, cilt.28, sa.7, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 28 Sayı: 7
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s10098-026-03538-0
- Dergi Adı: Clean Technologies and Environmental Policy
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, IBZ Online, ABI/INFORM, Compendex, Environment Index, Greenfile, INSPEC, Public Affairs Index, Natural Science Collection (ProQuest), Social Science Premium Collection (ProQuest), Materials Science & Engineering Collection (ProQuest), Pharma Collection (ProQuest), Technology Collection (ProQuest)
- Anahtar Kelimeler: Artificial neural networks, GRU, LSTM, Methane emissions estimate, SARIMA, Time series analysis
- Çukurova Üniversitesi Adresli: Evet
Özet
Greenhouse gases are among the issues that need to be controlled in the fight against climate change due to their heat-trapping capacity and the resulting temperature variations. To determine the variability levels of greenhouse gases, forecasting methods such as time series analysis are currently needed. When using these forecasting methods, their compatibility and accuracy with existing data must be tested. In this research, using a time series analysis based on monthly methane (CH4) gas emissions between 1984 and 2024, changes in global methane (CH4) gas emissions until 2050 were predicted using “Seasonal Autoregressive Integrated Moving Average Model (SARIMA),” “Long Short-Term Memory Neural Network (LSTM),” and “Gated Recurrent Unit (GRU)” models. The traditional forecasting method, the SARIMA model, was compared with deep learning-based LSTM and GRU models, and their performances were evaluated. Model performance was evaluated using RMSE, MAE, MAPE, R2, and KGE metrics to reveal different aspects of prediction accuracy. Additionally, the Diebold–Mariano test was used to statistically evaluate the differences in prediction performance between models in the test dataset. According to the findings obtained from the testing process, the RMSE values for the SARIMA, LSTM, and GRU models were found to be 1.0542, 2.8941, and 3.4210; the MAE values were 0.8124, 2.2715, and 2.7548; the %MAPE values were 0.0425, 0.1189, and 0.1441; the R2 values were 0.9963, 0.9720, and 0.9379; and the KGE values were 0.9912, 0.9410, and 0.9254. The results show that the proposed models can determine the temporal behavior of global monthly CH4 emissions with high accuracy, and model comparisons indicate that the SARIMA model has the best prediction performance. If current trends continue, the predictions suggest that global CH4 emissions could increase by approximately 13% from current levels by 2050.