Modeling and predicting city-level CO₂ emissions using open access data and machine learning.

Area of Science:

• Environmental Science
• Data Science
• Urban Planning

Background:

• Urban areas are major contributors to greenhouse gas (GHG) emissions, particularly carbon dioxide (CO2).
• Quantifying city-level CO2 emissions is challenging due to data limitations, especially in underdeveloped regions.
• Accurate emission data is crucial for effective climate change mitigation strategies.

Purpose of the Study:

• To develop and validate a machine learning model for estimating and predicting city-level CO2 emissions in China.
• To identify key socioeconomic and environmental variables influencing urban CO2 emissions.
• To provide a tool for generating carbon footprint maps in data-scarce areas.

Main Methods:

• Utilized open-access socioeconomic, geospatial, and meteorological data.
• Employed Recursive Feature Elimination and Boruta for feature selection, identifying 18 critical variables.
• Developed prediction models using machine learning algorithms, notably XGBoost.

Main Results:

• Statistical indicators of urban environmental pollution (e.g., SO2 and dust emissions per capita) were the most significant predictors.
• XGBoost models achieved high accuracy (R² > 0.98) with low relative error (~0.8%).
• An S-shaped relationship was observed between per capita CO2 emissions and economic growth, contrary to the expected U-shaped curve.

Conclusions:

• Easily accessible data combined with machine learning can accurately predict city-level CO2 emissions.
• The developed approach is valuable for regions with limited energy statistics.
• Findings support policymakers in designing targeted carbon emission reduction goals and strategies.