Enhancing Greenhouse Gas Emission Prediction and Monitoring using Machine Learning and Remote Sensing

Abstract

The escalating emissions of anthropogenic carbon dioxide (CO2) has been identified as the principal driving force on recent climate change and global warming. The surge in anthropogenic CO2, primarily from fossil fuel combustion and land-use changes, underscores the urgency for robust actions towards a sustainable low carbon future, as emphasized by international agreements like the Kyoto Protocol and the Paris Agreement. However, accurate predictions of CO2 and methane (CH4) emissions presents significant challenges in supporting effective climate mitigation strategies, due to incomplete reporting, uncertainties in bottom-up emission inventories, and difficulties in modelling efforts. The aim of this study was to enhance the monitoring and prediction capabilities of greenhouse gas (GHG) emissions using a data-driven approach integrating Machine Learning (ML) and satellite Remote Sensing techniques. A Random Forest (RF) model was employed to predict CO2 emissions within the highly industrialized and urbanized region of North-Western Europe. The model utilized various datasets including satellite observations of proxy specie NO2 from TROPOMI, meteorological data, and socio-economic variables. The RF model demonstrated high accuracy in predicting CO2 emissions within the area the model trained on, achieving an R² score of 0.989, a Root Mean Square Error (RMSE) of 1.9774e-08, and a Mean Absolute Error (MAE) of 9.9025e-09. While the RF model performed very well on the test set, it did not generalize well to a broader region (R²-score of -2,822), indicating the importance of understanding geographical differences and source apportionment of various pollutant emissions. This underscores the need for further refinement and maturation of the model to ensure it can be readily deployed as a reliable tool for global GHG monitoring and prediction, supporting more effective climate mitigation efforts.

Date of Award	9 Sept 2024
Original language	English
Supervisor	Lyana Curier (Supervisor) & Angelique Lansu (Co-assessor)