Predicting Ground-based PM2.5 Concentration in Queensland, Australia

Abstract

Air quality degradation driven by rapid global development, industrialisation, and urbanisation puts human health at serious risk. Accurate air quality forecasting and early warning systems are critical to allow health authorities to provide warnings and health recommendations. In this paper, we use data-driven learning models using historical hourly, ground-based meteorological and air pollutant data to forecast particulate matter $(PM_{2.5})$ matter concentrations on a short-term basis (1-6 hour) across tropical, subtropical and arid climates. We found that the triple exponential smoothing (TES) model developed for the paper consistently outperformed two common benchmarking models, (the persistence and average models) in all three climates. In the tropical climate, the root mean square errors at lead times of 1, 3 and 6$\mu \mathrm{g}/\mathrm{m}^{3}$ hours for the TES model $(0.54\mu \mathrm{g}/\mathrm{m}^{3}, 0.78\mu \mathrm{g}/\mathrm{m}^{3}, 1.66\mu \mathrm{g}/\mathrm{m}^{3})$ were significantly smaller than the average $(2.28\mu \mathrm{g}/\mathrm{m}^{3}, 1, 69\mu \mathrm{g}/\mathrm{m}^{3}, 2.99\mu \mathrm{g}/\mathrm{m}^{3})$ and persistence models $(0.59\mu \mathrm{g}/\mathrm{m}^{3},0.93\mu \mathrm{g}/\mathrm{m}^{3}$, 1.75 $\mu \mathrm{g}/\mathrm{m}^{3})$. Similar results were mirrored in experiments using subtropical and arid data. We also establish the interdependence of meteorological parameters on atmospheric pollutants across the three climates.