A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Environmental Engineering and Management

Thesis (M. Eng.) - Asian Institute of Technology, 2025

Methane (CH₄) emissions from rice cultivation significantly contribute to agricultural greenhouse gas emissions. Conventional estimation practices in developing countries often rely on generalized emission factors recommended by the Intergovernmental Panel on Climate Change (IPCC), due to the lack of sufficient ground-based monitoring and measurement. These estimations may not reflect the specific agricultural practices and environmental conditions of local regions. This study aims to localize methane emission factors for rice cultivation in Ayutthaya, Thailand, using a machine learning-based approach. A Gradient Boosting regression model was developed and trained on a global literature dataset containing known CH₄ emissions and associated agricultural variables. Key features influencing CH₄ emissions; pH, water regime, crop duration, organic amendment, N amount, and soil properties were selected based on feature importance analysis and SHAP (SHapley Additive exPlanations) values. The trained model was then applied to a Thailand-specific secondary dataset to predict methane emission factors. Results show that the predicted average CH₄ emission factor for the Thailand dataset is 0.8518 kg CH₄/ha/day, which is significantly lower than the IPCC Tier 1 default value of 1.30 kg CH₄/ha/day. This suggests that the IPCC default value may substantially overestimate methane emissions in certain local contexts, highlighting the importance of developing localized emission factors for improved regional GHG assessments. This research provides a practical framework for applying machine learning to improve emission factor localization. The findings offer valuable insights for policymakers and researchers in developing tailored mitigation strategies for rice agriculture in Southeast Asia.