Machine learning estimates for G20 subnational urban GHG emissions from 2000-2020.

Area of Science:

• Environmental science
• Climate change modeling
• Data science

Background:

• Subnational greenhouse gas (GHG) emissions data are crucial for climate action but are often scarce and inconsistent.
• Limited data availability hinders tracking progress and identifying mitigation opportunities in cities and regions.
• Existing methods face challenges with spatial relevance and methodological variations.

Purpose of the Study:

• To develop a machine learning (ML) framework for estimating annual Scope 1 and 2 carbon dioxide-equivalent (CO2-eq) emissions.
• To provide globally consistent, administratively-aligned emissions data for subnational jurisdictions in G20 countries (2000-2020).
• To support data-informed policy decisions for urban and regional decarbonization.

Main Methods:

• Developed an ML framework integrating geospatial, socioeconomic, and environmental data.
• Incorporated self-reported inventories where available.
• Aligned model predictions with subnational administrative boundaries for enhanced spatial relevance.

Main Results:

• The ML model successfully estimated annual Scope 1 and 2 CO2-eq emissions from 2000 to 2020 for G20 subnational jurisdictions.
• The approach demonstrated improved spatial relevance and predictive performance over traditional methods.
• The resulting dataset captures locally specific emission drivers, even in data-poor contexts.

Conclusions:

• The developed ML framework provides a reliable and comparable dataset for subnational GHG emissions.
• This resource serves as a baseline for assessing climate progress and guiding mitigation strategies.
• The findings support targeted policy decisions for effective urban and regional decarbonization efforts.