HKU duo want to look at air quality ‘down to the individual street’ with HK$50 million project
New platform relies on deep learning technology to produce precise pollution data by crunching government information
Live in Kowloon but want to check on the air quality in Tai O before your hike tomorrow morning? Two University of Hong Kong academics hope to make that possible, as they work on a HK$50 million big data research project.
Harnessing deep learning technology, the duo is engineering a new air quality monitoring platform that can crunch information from government air quality monitoring stations to produce the equivalent effect of 110,000 vastly more precise individual stations spread across the city.
Deep learning, put simply, involves feeding software large amounts of data, which it “learns” and uses to make decisions about other data, similar to how neural networks in the human brain work.
The aim, according to Professor Victor Li On-kwok and Dr Jacqueline Lam Chi-kei of HKU’s electrical and electronic engineering department, will be to build an air quality mobile app on this framework that will offer more location-specific forecasts and personalised advice to the public on how to reduce exposure to bad air.
“The key words here are ‘personalised and smart, timely and interactive’. Telling me how bad the air is after I breath it is useless – I want to know right now,” Li said. “If in the afternoon I want to go to Mong Kok, I want to know the forecast for the air there this afternoon.”
The project overcame intense competition to win HK$45 million from the Research Grants Council’s theme-based funding scheme earlier this year. It is the first interdisciplinary clean energy and environment project to be funded under the theme of “emerging research and innovations”.
Because of the huge volume of data it will be dealing with, the team has struck an agreement with Microsoft to host the platform on its cloud service for free.
Using a type of regression analysis called Granger causality testing on data from the Environmental Protection Department’s 16 roadside and ambient monitoring stations, the models can be trained to estimate pollution concentrations within one square kilometre “grids” across the city with about 82 per cent accuracy.
“Hong Kong’s land mass is about 1,100 square kilometres, so we’ve been able to divide it into 1,100 little boxes. But we feel even this is too large. We hope to narrow each grid down to 100 metres by 100 metres, meaning 110,000 grids around the city,” Li said. The assumption will be that air in the spaces within these grids is the same.
Li says the information provided will be greater than what the government’s network provides.
Government monitoring stations are situated at just 13 general and three roadside locations around the city. Areas without stations must reference data from the ones closest to them. Li and Lam claim their platform will fill in the gaps in the data.
“It will allow us to look at air quality down to the individual street,” said Lam, a co-director of the HKU-Cambridge Clean Energy and Environment Research Platform. “Research has shown that air pollution can be drastically different on two sides of a hill or even on two adjacent streets.”
The two hope to fine tune the data over the next five years and discover ways to boost the accuracy rate to “above 90 per cent”.
But therein lies a problem.
“If we have only 16 stations [to draw data from], perhaps it is impossible to get above 90 per cent. We’ll need more mathematical modelling to find out,” Li said, adding that the research team would consider using mobile monitoring stations and other technologies to collect more data.
The end product will be a free public app that will allow users to input their own personal and health data and receive tailor-made advice on how to avoid bad air.
“We hope to create an app in the future that is just as convenient and easy to understand as the [Hong Kong] Observatory’s,” Li said.
“In terms of academic impact, it will be a new way of measuring things ... but at the end of the day, we as engineers also hope to push for changes that can help society and the wider community.”
High-precision data on air pollution could help the government in city planning, for instance. The data mined could also be useful for the commercial sector
“We won’t rule out that entrepreneurs may find the data very useful and, based on our platform, develop new apps of their own. In these circumstances, we will license the technology to them,” he said.
“Once we license it, we will have some revenue, but we are not allowed to keep it, we must invest it back in the project, continue to improve it and make it more accessible to the public”.