Artificial Intelligence and Clean Air: Development of Novel Algorithms with Machine Learning and Deep Learning

Artificial Intelligence and Clean Air: Development of Novel Algorithms with Machine Learning and Deep Learning

Air pollution has detrimental impacts on the people, the environment, and the global economy; however, this issue is somewhat under-recognised in many nations, despite having an advanced healthcare infrastructure. Accurate, reliable, and real-time forecasting of air pollutants is critical to minimise adverse health outcomes, and improving the quality of human life. In comparison with other developed nations, Australia has a relatively higher standard of air quality. Despite this, approximately 5,000 annual deaths are attributable to air-borne issues (AAS 2021). The Environment Protection Agency (EPA-VIC 2016) forecasts that population growth, increased urbanisation, and uncontrolled fossil-fuel emissions may result in summer smog, getting significantly pronounced after 2030. Australia's arid nature with frequent dust-storms and bushfires could further exacerbate the particle-borne air pollutants. Therefore, scientifically verified, and robust artificial intelligence methods are essential for efficient air quality forecasting that can support the Australian Government's health and environment protection policies.

This doctoral research presents a novel study based on the development of computationally efficient artificial intelligence models that can provide early warnings through an air quality forecasting system. In the first objective, a novel hybrid artificial intelligence framework based on machine learning or an online sequential-extreme learning machine algorithm is developed to emulate hourly air quality variables i.e., fine particulates (PM2.5), coarse particulates (PM10), and the visibility reducing particles. These are associated with increased respiratory-induced mortality resulting in recurrent healthcare costs. The second objective further advances the first objective by introducing deep learning algorithms to develop a forecasting system for suspended particulate matter. As remote sensing data are also an important predictor for meteorological variables, the third objective has constructed a one-dimensional convolutional neural network (CNN) integrated with a one-directional fully gated recurrent unit (GRU) model using satellite and ground-based observations producing reliable estimates of PM10. The study areas considered are all major Australian air pollution hotspots, posing a growing hazard to public health. The results reflect that meteorological data, geographical information, and important statistical metrics have further improved the prediction accuracy in all the objectives.

By providing a high-resolution spatiotemporal forecasting system, the doctoral research results may provide scientific innovation and knowledge contributions facilitating further studies, enhancing the understanding and dynamic evolution of airborne pollutants. These findings could also facilitate the necessary pre-emptive actions during bushfires seasons, frequent dust storms in real-time warnings so our citizens can plan and minimise exposure risks. Another significance of the doctoral study is protecting vulnerable population groups (e.g., frail elderly, people with sickness, expectant mothers, and children) so the proposed models can be updated when new, higher-resolution satellites are launched.

In synopsis, the novel state-of-the-art artificial intelligence models in this doctoral research have the potential to provide significant economic and advisory benefits for the government, particularly in air quality and health policy development. The predictive system can also be of global importance, where air quality poses a serious health risk. Therefore, the outcomes of this doctoral research can be adopted in constructing a real-time forecasting system tailored for environment protection industries, public health, governments, and other stakeholders.

File reproduced in accordance with the copyright policy of the publisher/author.