지자체 탄소중립계획 지원을 위한 지역 수준 탄소배출모델 개발 및 활용성 평가

• Published in: 한국기후변화학회지 Vol. 15; no. 51; pp. 691 - 712
• Main Authors: 정유정(Jeong, Yujeong), 송철호(Song, Cholho), 조현우(Jo, Hyeon-Woo), 고영진(Ko, YoungJin), 이우균(Lee, Woo-Kyun)
• Format: Journal Article
• Language: Korean
• Published: 한국기후변화학회 01.10.2024
• Subjects: 학제간연구; Carbon Emission Map; Local Government; Ridge Regression; Carbon Emission Model; Carbon Neutrality; Machine Learning; Carbon Neutrality Plan
• ISSN: 2093-5919; 2586-2782
• DOI: 10.15531/KSCCR.2024.15.5.691

Achieving carbon neutrality by 2050 necessitates significant contributions from local governments. To accelerate the creation and execution of local government carbon neutral plans, ongoing research focuses on spatial carbon emission data. However, in the Republic of Korea, policy-related use of spatial carbon emission data remains limited. This study aims to develop a spatial regression model for carbon emissions using machine learning-based ridge regression and regional greenhouse gas inventories to aid basic local governments in their decision-making regarding carbon neutrality and regional carbon mitigation. Input data were created by disaggregating subdivision land cover maps and facility-level national statistics to the local government level. Group K-fold cross-validation and area-based scaling were applied to improve generalization of the model. Two local carbon emission prediction models (the ‘LCE model’) were developed based on regional GHG inventory in SiDo-level (the ‘Level-1 model’) and LCE model based on Regional GHG inventory in SiGunGu-level (the ‘Level-2 model’). The Level-2 model exhibited higher accuracy, with R2 values of 0.84 and 0.66 at SiGunGu-level for 2019 and 2020, respectively, and at 0.93 and 0.76 at SiDo-level. The carbon emission maps generated by the Level-2 model exhibited higher accuracy than the Level-1 model. This study highlights the cost effectiveness of machine learning-based spatial regression models for carbon emissions compared to IPCC (Intergovernmental Panel on Climate Change) methods and fuel-based models. The machine learning-based methodology and its detailed emission maps are expected to provide timely scientific evidence for developing and assessing carbon neutrality plans at SiGunGu-level, delivering granular information on carbon emissions down to the DongRi-level KCI Citation Count: 0