THESIS
2021
1 online resource (xviii, 186 pages) : illustrations (some color), color maps
Abstract
Low-cost air quality sensors for air pollution monitoring have shown huge potential in
enhancing the spatial and temporal resolution of much needed pollution data at a lower cost,
greater flexibility in use with less maintenance than air quality monitoring stations. Air quality
sensors provide opportunities to extend a range of existing air pollution monitoring capabilities
and provide ideas for new monitoring applications. The widespread data sources combined with
sensors improve people’s understanding of air quality research areas. However, severe gaps
existed between the guarantee of sensor data quality and derivation of meaningful information
from sensor applications in air quality monitoring. This study provides avenues from sensor
calibration to applications in urban air quality m...[
Read more ]
Low-cost air quality sensors for air pollution monitoring have shown huge potential in
enhancing the spatial and temporal resolution of much needed pollution data at a lower cost,
greater flexibility in use with less maintenance than air quality monitoring stations. Air quality
sensors provide opportunities to extend a range of existing air pollution monitoring capabilities
and provide ideas for new monitoring applications. The widespread data sources combined with
sensors improve people’s understanding of air quality research areas. However, severe gaps
existed between the guarantee of sensor data quality and derivation of meaningful information
from sensor applications in air quality monitoring. This study provides avenues from sensor
calibration to applications in urban air quality monitoring. Three steps were divided as follows:
Firstly, the air quality sensors were tested in the laboratory-controlled environment to
understand the sensor performance and influence factors. The ambient factor influence was
identified and quantified to build correction algorithms. Based on the understanding of the
sensor working principle, a principle-based correction algorithm was built. The new algorithm
proves the sensor sensitivity variation trend during different seasons. Meanwhile, compared
with commonly used correction models, it shows the best performance and spatial
representativeness in the sensor network.
Secondly, a mobile sensor network was deployed in Hong Kong to monitor traffic-related air pollutants (TRAP) along bus routes. To estimate local and background contributions, a robust
baseline extraction algorithm was developed and evaluated. The result indicates NO and NO
2 are locally dominated air pollutants and varied within roads. Background concentrations
primarily arose from CO and PM
2.5 and decreased during summertime. The regional transport
pollution is the primary contributor during high pollution episodes. For the traffic induced local
concentration, the most polluted road segments cluster at tunnel entrances and congested points.
Some of these polluted locations were observed in Low Emission Zones and suggest limitations
to the existing control strategies. These finding gives insights in the importance of regional
cooperation to control background air pollution combined with local control strategies to
improve roadside air quality in Hong Kong.
Finally, based on the mobile measurement, we build machine learning based land use regression (LUR) models to estimate the street-level pollutions distribution in complex urban environments. In addition to traditional geographic variables, two innovative variables from
Google maps were integrated as candidate predictors: street view image and traffic congestion
index. These two variables can be used to represent the pollutant dispersion status and traffic
activities which is more relevant to vehicle emissions. The result indicates the integrated
machine learning models perform well in transportation dominated local urban environments.
The traffic-related predictor variables were identified based on the feature importance
evaluation that indicates the traffic and transport infrastructure are major concerns both for
individual exposure and urban planning. Furthermore, the estimated pollution maps can help
identify environmental hotspots that need to be improved by taking appropriate urban planning
measures and management strategies.
Keywords: Air quality, traffic-related pollutants, air quality sensor, correction algorithm, traffic
congestion, machine learning, land use, crowdsourcing data
Post a Comment