Machine Learning Models for Ultrafine Particles in Copenhagen, Denmark

Abstract

Long-term mean ambient particle size (PS) in the ultrafine particle (UFP) range (< 100 nm) varies over space within cities, with locations near UFP sources having smaller PS. Spatial models for PS and lung deposited surface area (LDSA) within urban areas are limited. We collected particle number concentration (PNC), LDSA, and PS data over one-year monitoring campaign from May 2021 to May 2022 across 27 locations and estimated annual mean in Copenhagen, Denmark, and obtained additionally annual mean PNC data from 5 curbside monitors within the city. We developed 94 predictor variables (majority at 1 m spatial resolution (90%)), and machine learning models (random forest and bagged tree) were developed for PNC, LDSA, and PS. The annual mean PNC, LDSA, and PS were, respectively, 5,416 pt/cm3, 12.0 µm2/cm3, and 46.1 nm. The cross-validation R2 values (10-fold repeated 10-times) were 0.70, 0.67, and 0.60 for PNC, LDSA, and PS, respectively. Traffic-related variables, such as streets below/above specific speed-limits, and length of major roads within buffers of 100–150 m, amongst others, were strong predictors. External validation with high-quality data is warranted to ensure good performance of these models. These UFP predictions may assist urban planners, environmental justice studies, or epidemiologists conducting population-based studies.