A machine learning approach for resource mapping analysis of greenhouse gas removal technologies

Abstract

In this study, machine learning (ML) was applied to investigate the suitability of a location to deploy five greenhouse gas removal (GGR) methods within a global context, based on a location's bio-geophysical and techno-economic characteristics. The GGR methods considered are forestation, enhanced weathering (EW), direct air carbon capture and storage (DACCS), bioenergy with carbon capture and storage (BECCS) and biochar. An unsupervised ML (hierarchical clustering) technique was applied to label the dataset. Seven supervised ML algorithms were applied in training and testing the labelled dataset with the k-Nearest neighbour (k-NN), Artificial Neural Network (ANN) and Random Forest algorithms having the highest performance accuracies of 96%, 98% and 100% respectively. A case study of Scotland's suitability to deploy these GGR methods was carried out with obtained results indicating a high correlation between the ML model results and information in the available literature. While the performance accuracy of the ML models was typically high (76 - 100%), an assessment of its decision-making logic (model interpretation) revealed some limitations regarding the impact of the various input variables on the outputs.