Speaker: Professor Victor O.K. Li
Chair of Information Engineering
Head of Department of Electrical and Electronic Engineering
The University of Hong Kong
Air quality has deteriorated rapidly in Hong Kong and China with NO2 and PM2.5 levels frequently exceeding WHO safety guidelines. While poor air quality has clear public health impacts, very few monitoring stations (e.g. only 13 monitoring stations in Hong Kong and 35 in Beijing) are available for measurements of major air pollutants, severely limiting evidence-based air quality decision-making, and leading to severe criticisms about the transparency and public relevance of the official Air Pollution Index. Since air pollution is highly location-dependent, and monitoring stations are costly and bulky, a citywide air quality monitoring system would be prohibitively expensive. Urban big data may be used to fill this gap. By analyzing the causality between human dynamics data (such as vehicular traffic and points of interest data) and measured air quality, one may estimate air quality at locations not covered by monitoring stations. However, processing the massive volume of data poses another challenge. To overcome this challenge, we note that most of these data are spatially and temporally correlated. Therefore, our approach is to exploit such spatio-temporal (S-T) correlation to process “part” of the data instead of “all” of the data. We detect, measure, quantify and visualize causalities between various urban dynamics data and air quality. Here causalities can be expressed in a probabilistic manner spatially and temporally. Furthermore, we exploit parallel computing by separating and allocating relatively independent data blocks to different computing resources, based on causality measures. In this way, time efficiency and scalability can be achieved. Our approach will be illustrated using data from Shenzhen, China.