Jeffrey Chang
3rd Place
Jeffrey Chang is currently a PhD Candidate focusing on Urban Climate and Air Pollution Modeling at The University of Hong Kong.
He is also a co-founder of GeogSTEM, a startup company integrating the concept of citizen science, IoT applications and environmental models for education advancement and extreme microclimate resilience for the local communities in Hong Kong. His research interests include numerical modelling, urban meteorological forecasts, electronic sensor development and citizen science applications, which the interdisciplinary studies aims to open-up new opportunities for citizen collaboration in urban sustainability in our future cities.
Laudatio for Jeffrey Chang
“This project presents a study carried out in collaboration with the Urban Climate and Air Pollution Laboratory of the University of Hong Kong. It aimed at developing an algorithm to rapidly identify the cross-sectional and vertical profile of wind-field and NOx concentration in urban canyons. The model combined ENVI-met microclimatic outputs with GIS data and used Mong Kok compact downtown area in Hong Kong for the study.
The problem of the potential health effect of pollutant hotspots within high traffic volume in dense urban areas to the general public is well stated. Overall, the methodology is clear indicating scientific rigour and sound results. This is an interesting model with clear relevance of outputs, such as pollutant hotspots. The model shows great potential to help policymakers, regulating bodies and governments to identify intervention actions for improving the health and quality of life of pedestrians and inhabitants of buildings of selected urban areas.”
Paula Cadima
Director of the March & MSc Sustainable Environmental Design at the Architectural Association School of Architecture (AA), PhD Supervisor, Member of the AA Academic Board, Academic Committee and Internal Assessment Committee.Architect
Machine learning pollution hotspots in urban street canyon
Why Hong Kong?
Hong Kong is a densely populated city with urban architecture since 1840s, with the strategic development from the British colonial government to now, under the sovereignty of China. The different urban planning styles have shaped Hong Kong to be special. Modern cyber cities and easily accessible transportation are commonly known as advantages for living, while the street canyon effect and the busy traffic give rise to a public health concern: “The roadside air pollution problem”.
The study was separated into two parts, the simulation of 3-dimensional NOx exposures in Mong Kok’s street canyon, and the integration of ENVI-met models with machine learning techniques. Since wind attributes, traffic flow and vehicle compositions are the contributing factors in the pollution profile throughout the street canyons, the first part of the study compares and investigates the pollution hotspots associated with different windspeed and wind direction boundary conditions by running multiple simulations with ENVI-met.
Computational fluid dynamic (CFD) models like ENVI-met offer a high-resolution spatiotemporal simulation in urban canopy, while computational resources and time consumption are always the research limitations and concern.
Therefore, in the second part of this study, the geospatial datasets including Aspect Ratio, Sky View Factor, Wind Effect, Wind Exposition Index, and Distance from Primary Roads were internalized with Python-based postprocessing model, to develop a Support Vector Regression with machine learning techniques. The model aims to couple ENVI-met and geospatial data with machine learning to provide a quick simulation of real-time roadside pollution for the public to stay alert.
Results
By using ENVI-met for the simulations, the 3-dimensional potential pollutant hotspots are clearly identified in Mong Kok (Hong Kong), and the Support Vector Machine learning model also shown good accuracy and consistency with the original model outputs.
There were in total 48 combinations of model runs being simulated with ENVI-met. The multiple simulations had illustrated the accumulation of NOx is more significant and they can propagate higher with Easterly Wind conditions. Regarding the building geometry in the selected domain, the model output shows there is a boundary height limit (~10 meters high) for the NOx pollutant propagation when the incoming wind is parallel to the street canyons. It is also notable that the building height difference between windward and leeward sides of buildings can amplifies the street canyon vortex effect and favors the pollutant accumulations inside deep canyon.
For the sensitivity test with road emission contributions, the result had remarked an important message that the downwind areas from Primary Roads (the two-way main road for traffic between the city centers) are vulnerable to the direct pollution impacts with wind effects
The simulations also suggested that if people are living at a level above 10 meters (i.e. 4th Floor or above), their pollution exposure will be about halved when compared to the ground-level.
Besides, as model postprocessing with coupling of ENVI-met and geospatial data was adapted in this study, it is stimulating that the model had shown a good accuracy and consistency with 90% simulation time reduction when comparing to the original runs. The model performs between with Southerly and Westerly wind scenarios, and may under-estimate high pollution concentration events (i.e. Easterly wind scenario), while the 3D pollutant pattern from machine learned model outputs are highly comparable with ENVI-met’s one.
Comparison and quantification through ENVI-met software indicates how the natural wind forcing and anthropogenic roadside emissions contribute to the variation in pollutant hotspots over the street canyon, and the experiment further highlights the potentials in coupling CFD models with machine learning techniques to provide a quick-response and fine-scale pollution forecasts in urban megacities.