A Comparative Analysis of Advanced Modeling Techniques for Global Methane Emission Forecasting Using SARIMA, LSTM, and GRU Models

Abstract

Forecast methods are an important aid to the early detection of future levels of pollutant amounts released from global pollutants. This research predicts changes in future global methane gas emissions using SARIMA, LSTM, and GRU models, and also compares the accuracy of these three prediction methods. In the study, a time series analysis was conducted by focusing on the monthly methane (CH4) gas emission amounts recorded between 1984 and 2024. Methane emission data measured between 1984 and 2024 were used as input in the development of the models. By comparing the prediction results and actual values, they were evaluated with performance criteria such as R², RMSE, MAE, and MAPE%. The results revealed that all three methods performed well in estimating global methane gas emissions. The SARIMA model shows the best performance, followed by the LSTM and GRU models. It was determined that the SARIMA model had the lowest error rate with 0.0020 MAPE, 0.0335 MAE, 0.0335 RMSE, and 0.9998 R² values. It has been revealed that estimated global methane emission values may be approximately 1.5 times higher than today's level by 2050.