Prediction of Short-Term Ultrafine Particle Exposures Using Real-Time Street-Level Images Paired with Air Quality Measurements

Abstract

Within-city ultrafine particle (UFP) concentrations vary sharply since they are influenced by various factors. We developed prediction models for short-term UFP exposures using street-level images collected by a camera installed on a vehicle rooftop, paired with air quality measurements conducted during a large-scale mobile monitoring campaign in Toronto, Canada. Convolutional neural network models were trained to extract traffic and built environment features from images. These features, along with regional air quality and meteorology data were used to predict short-term UFP concentration as a continuous and categorical variable. A gradient boost model for UFP as a continuous variable achieved R 2 = 0.66 and RMSE = 9391.8#/cm3 (mean values for 10-fold cross-validation). The model predicting categorical UFP achieved accuracies for “Low” and “High” UFP of 77 and 70%, respectively. The presence of trucks and other traffic parameters were associated with higher UFPs, and the spatial distribution of elevated short-term UFP followed the distribution of single-unit trucks. This study demonstrates that pictures captured on urban streets, associated with regional air quality and meteorology, can adequately predict short-term UFP exposure. Capturing the spatial distribution of high-frequency short-term UFP spikes in urban areas provides crucial information for the management of near-road air pollution hot spots.

Dynamic Traffic Data in Machine-Learning Air Quality Mapping Improves Environmental Justice Assessment

“…Low-cost sensors can be flexibly and widely deployed at a substantially lower cost than regulatory stations, adding more inner-city information where limited monitoring exists . Mobile monitoring is increasingly used to detect fine-scale pollution patterns such as the gradients in traffic and point source impacts (e.g., highway, local industry), as well as dispersion and reaction dynamics within the built environment (e.g., street canyons). , Future research can deploy low-cost sensors or conduct full-coverage mobile monitoring in grids with the most racial/ethnic or income diversity to add details about hyperlocal exposure inequality. Also, more detailed parameters or refined measurement designs are essential for the development of the machine-learning model to reveal the complex interactions between fine-scale air quality gradients and demographic patterns, regional transport, multiple pollution sources, and the built environment.…”

“…Machine learning (ML)-powered geo-statistical models can improve the reliability and efficiency in leveraging real-world data. , They are always coupled with measurements such as satellite data, , ground observations, , and mobile monitoring data to depict air quality or exposures disparities from regional (∼10 km), , local (∼1 km), ,− to hyperlocal (∼100 m) , scale. However, most of these methods are insufficient to address the dynamic and fine-scale traffic-induced air pollution patterns as they typically consider static geo-features (e.g., road length) in their frameworks .…”

Societal Co-benefits of Zero-Emission Vehicles in the Freight Industry

“…Three electrification scenarios were modeled including the electrification of all heavy-duty trucks (Electric_HDT), which are primarily circulating on highways (8% of all on-road VKT in the GTHA), electrification of medium-duty trucks (Electric_MDT) currently diesel-fueled with movements along highways and major roads (4% of all on-road VKT in the GTHA), and electrification of light-duty trucks (Electric_LDT) currently gasoline-fueled and predominant on local roads and in residential neighborhoods (9% of all on-road VKT in the GTHA). We evaluated each scenario separately because they have different spatial patterns as indicated in a previous study conducted in the city of Toronto Figure S7 and Table S8 present the proportion of VKT associated with trucks in the GTHA and the total VKTs for the different truck categories.…”

A Machine Learning Approach for High-Resolution Modeling and Apportionment of Black Carbon Concentrations in an Impacted Community

“…27 Neural networks have been used in air quality modeling to downscale modeled ozone concentrations 28 and to predict ultrafine particle exposures from street-level images. 29 Various statistical and machine learning techniques have also been used to infer particulate matter concentrations from satellite measurements of aerosol optical depth 30−33 and predict nitrogen dioxide hotspots. 34 Examples of using machine learning to predict spatially comprehensive concentrations exist at monthly 35 and annual 36 time-scales, but often studies involve training a model on measured data to predict concentrations at discrete observation locations.…”