Improving Temporal and Spatial Coverage of UK Black Carbon Measurements by Applying a Machine Learning Approach

Abstract

The WHO Global Air Quality Guidelines suggest that continuous and systematic monitoring of black carbon (BC) should be implemented due to BC from both transport and residential wood burning posing adverse health effects. Beyond PM₁₀ and PM_2.5, BC serves as a strong indicator for the study of health risks related to primary combustion sources. In the UK, only 14 monitoring stations report BC concentrations, contrasting with the extensive network measuring NO_x, SO₂, O₃, CO, PM₁₀ and PM_2.5 and meteorological parameters (more than 170 stations across the country). The sparse spatial and temporal coverage of BC data caused by this limitation constrains scientists’ research and policy makers’ decisions. This study applies machine learning algorithms to address this challenge: first, filling gaps in missing BC data at 9 urban and 4 rural background sites in the UK between 2009 and 2020, and then further estimating hourly concentrations of BC at 7 sites without observations (1 urban and 6 rural background sites). The result is a new data set with greater temporal and spatial coverage of BC data, providing a resource for further research on BC impacts such as epidemiological studies on health outcomes. Although the data are specific to the UK, the proposed methodology can potentially be applied to other countries.