Deep Learning–Based Greenhouse Gas Emissions Prediction with Sensitivity and Uncertainty Analysis for Environmental Decision-Making

Baye Yemataw
College of Computing, Department of Information Systems, Debre Berhan, Debre Berhan University, Ethiopia
College of Engineering and Technology, Department of Computer Science, Kabridahar, University of Kabridahar, Ethiopia

Betegiorgis Mamo
College of Social Science, Department of English and Literature, Debre Birhan, Debre Berhan University, Ethiopia

Binyam Eshetu
College of Computing, Department of Information Systems, Debre Berhan, Debre Berhan University, Ethiopia

Dagmawi Legesse
College of Computing, Department of Data Science, Debre Berhan University, Ethiopia

Gezahign Tibebu
College of Engineering, Department of Mechanical, Debre Berhan University, Ethiopia

Abstract

Greenhouse gas emissions currently endanger the ecosystem by causing climate change and global warming, as a result of the burning of coal, fossil fuels, industrial processes, and deforestation. Therefore, tracking past, current, and future greenhouse gas emission levels is indispensable for mitigating their adverse effects on the global environment. The present study evaluated the 2011–2023 U.S. Environmental Protection Agency Greenhouse Gas Reporting Program data using a multilayered Gated Recurrent Units model, providing a deep learning approach to forecasting non-linear, multi-year industrial emission time series. By automatically identifying complex data patterns, the model outperformed deep learning approaches to forecasting greenhouse gas emissions. The model showed a high predictive accuracy (R² of 0.940, MAE = 0.023, RMSE = 0.048), considerably robustified by the integration of the sensitivity and uncertainty analyses. The results indicated that the optimal model for 2028 consists of a five-year projection plus the 2023 actual CO₂e levels. The model also confirmed that CO₂e levels were on a downward trajectory, showing a decline in projected emissions for 2028 when compared to the 2023 actual emissions of 3,757,462,133 metric tons. Implications for future study were forwarded. 

Keywords: greenhouse gas emissions, CO2e, deep learning prediction, patterns, insights