Air quality and health impact of 2019–20 Black Summer megafires and COVID-19 lockdown in Melbourne and Sydney, Australia

Highlights

• Air pollution in Australia was modelled using meteorological normalisation.

• Accounting for meteorology is necessary for attributing air quality impacts.

• Air quality improvement during COVID-19 lockdown highly regionally dependent.

• PM_2.5 and O₃ from Black Summer smoke likely caused 187 excess deaths.

Abstract

Poor air quality is an emerging problem in Australia primarily due to ozone pollution events and lengthening and more severe wildfire seasons. A significant deterioration in air quality was experienced in Australia’s most populous cities, Melbourne and Sydney, as a result of fires during the so-called Black Summer which ran from November 2019 through to February 2020. Following this period, social, mobility and economic restrictions to curb the spread of the COVID-19 pandemic were implemented in Australia. We quantify the air quality impact of these contrasting periods in the south-eastern states of Victoria and New South Wales (NSW) using a meteorological normalisation approach. A Random Forest (RF) machine learning algorithm was used to compute baseline time series’ of nitrogen dioxide (NO₂), ozone (O₃), carbon monoxide CO and particulate matter with diameter < 2.5 μm (PM_2.5), based on a 19 year, detrended training dataset. Across Victorian sites, large increases in CO (188%), PM_2.5 (322%) and ozone (22%) were observed over the RF prediction in January 2020. In NSW, smaller pollutant increases above the RF prediction were seen (CO 58%, PM_2.5 80%, ozone 19%). This can be partly explained by the RF predictions being high compared to the mean of previous months, due to high temperatures and strong wind speeds, highlighting the importance of meteorological normalisation in attributing pollution changes to specific events. From the daily observation-RF prediction differences we estimated 249.8 (95% CI: 156.6–343.) excess deaths and 3490.0 (95% CI 1325.9–5653.5) additional hospitalisations were likely as a result of PM_2.5 and O₃ exposure in Victoria and NSW. During April 2019, when COVID-19 restrictions were in place, on average NO₂ decreased by 21.5 and 8% in Victoria and NSW respectively. O₃ and PM_2.5 remained effectively unchanged in Victoria on average but increased by 20 and 24% in NSW respectively, supporting the suggestion that community mobility reduced more in Victoria than NSW. Overall the air quality change during the COVID-19 lockdown had a negligible impact on the calculated health outcomes.

Introduction

The austral Spring 2019 - Autumn 2020 period in south-eastern Australia was marked by two significant events impacting air quality. In the Spring and Summer (November 2019–January 2020), severe bushfires along much of Australia’s east coast shrouded the country’s most densely populated regions in smoke (Jalaludin et al., 2020; Borchers-Arriagada et al., 2020). The magnitude of the fires was unprecedented with 21% of Australia’s temperate broadleaf forest being burnt (Boer et al., 2020). In the Autumn, Australia along with much of the rest of the world implemented widespread economic, mobility and social restrictions in response to the COVID-19 pandemic, resulting in reduced traffic and air travel.

Both events resulted in significant changes to emissions of primary pollutants such as particulate matter and nitrogen oxides: an increase in the case of smoke conditions (Jalaludin et al., 2020) and a decrease in response to the COVID-19 restrictions (e.g. Keller et al., 2020). Changes in primary pollutant levels are likely to also result in changes to secondary pollutants such as ozone both during wildfires (Buysse et al., 2019) and the COVID-19 lockdowns (Kroll et al., 2020). Because tropospheric ozone, nitrogen oxides and particulate matter are associated with poor health outcomes (World Health Organization 2005), it is crucial for public health policy that the impact of such events is understood. This a priority for Australia with recent studies showing the large and detrimental health and economic cost of smoke exposure (Johnston et al., 2020; Borchers-Arriagada et al., 2020) that can only be expected to grow as climate change increases the frequency and severity of bushfires (Sharples et al., 2016; Pitman et al., 2007). In the case of the COVID-19 shutdown, changes in air pollution seen in this period may offer an insight into a ‘greener’, lower emissions future scenario. The shutdown may also provide the opportunity to assess the health benefits associated with decreased anthropogenic activity.

Quantifying the magnitude of changes in pollutant concentrations as a direct result of specific events is complicated by their dependence on meteorological variables (e.g., temperature, wind speed and wind direction, solar radiation), seasonal patterns and also by long term trends in these meteorological variables and emissions. As a result, the technique of ‘meteorological normalisation’ was developed to control for the influence of background meteorology and regular emissions cycles on air quality time series’ and allow robust interpretation of emissions changes (Grange et al., 2018; Grange and Carslaw, 2019). The increase in accessibility of machine-learning (ML) algorithms for statistical modelling has led to their use in meteorological normalisation, including for analysis of emissions changes during the recent COVID-19 lockdown periods (Keller et al., 2020; Petetin et al., 2020). In practice, meteorological normalisation involves training a statistical model to predict a dependent variable (e.g., pollutant concentration) on the basis of independent variables (e.g., meteorological parameters). The statistical model used may be a parametric (e.g., Libiseller and Grimvall, 2003) or ML technique. Throughout an event of interest, if the model explains a high amount of variance in the predicted air quality variable, pollutant levels predicted on the basis of the trained model and known meteorology provide a baseline ‘business-as-usual’ scenario against which to compare the actual observations. In this work, the Random Forest (RF) machine learning algorithm (Breiman, 2001) is trained to provide the meteorologically normalised baseline air quality time series’.

The 2019-20 Australian spring and summer saw bushfires of unprecedented magnitude burn more than 12.6 million hectares in the states of Queensland, New South Wales (NSW) and Victoria (Jalaludin et al., 2020). These fires were fuelled by record warm and dry weather conditions throughout 2019 across Australia (Vardoulakis et al., 2020). As well as the devastating physical, psychological and economic impact of the fires on areas directly burnt, there was also a large health burden associated with smoke exposure in Australia’s most populous regions along the east coast (e.g., Walter et al., 2020; Borchers Arriagada et al., 2020). Borchers-Arriagada et al. (2020) estimated that 417 deaths were attributable to PM_2.5 exposure in smoke in Victoria, NSW and Queensland and Johnston et al. (2020) showed that the health impacts of the Black Summer fires totalled as much as $1.95 billion. We build on these works by demonstrating that the RF technique can be used to provide a meteorologically normalised air quality baseline against which to compare health outcomes resulting from smoke exposure, for both PM_2.5 and ozone.

We also show in this paper that the RF technique can be used to examine air quality and associated health outcomes from emission decreases. Sudden decreases in primary pollutant emissions were experienced throughout much of the world during economic and mobility restrictions to curb the spread of the COVID-19 pandemic. In the early months of 2020, Australia responded to the pandemic by partially closing national borders, quarantining returned travellers and imposing stay-at-home orders. Lockdown measures were introduced across Australia, restricting non-essential services and advising the public to stay at home as much as possible, on March 23rd, 2020 (Greene, 2020). Data collected by Google from its mobile platforms indicated a significant decrease in community mobility in Australia from the middle of March and throughout April 2020 (Supplementary Information Fig. 1) (Google LLC, 2020). A number of papers have highlighted changes in air pollution metrics in different parts of the world as a result of COVID-19 mobility restrictions. Reduced levels of primary pollutants, such as NO₂ and PM_2.5 have been observed in urban areas across, among other places, China (Chen et al., 2020), India (Sharma et al., 2020), the United States (Berman and Ebisu, 2020), Brazil (Nakada and Urban, 2020) and Europe (Menut et al., 2020). Some studies showed a corresponding increase in ozone as a result of reduced nitrogen oxides (e.g. Sharma et al., 2020; Siciliano et al., 2020). However, methods for estimating pollution changes as a result of the COVID-19 restrictions vary considerably, including spatial analysis of air pollution from satellite data products (e.g. Dutheil et al., 2020; Chen et al., 2020), comparison with historical (monthly or seasonal) means (e.g. Sharma et al., 2020; Berman and Ebisu, 2020; Nakada and Urban, 2020), comparison to chemical transport model simulations (e.g. Menut et al., 2020) and comparison with machine learning meteorologically normalised baseline pollutant time series computed using machine learning algorithms (e.g. Achebak et al., 2020; Petetin et al., 2020).

Air quality in Australia’s major cities is generally considered to be good, with declining levels of anthropogenic primary pollutants including nitrogen oxides and carbon monoxide over recent decades (Environment Protection Authority Victoria 2013). However, the impetus for improved air quality reporting is provided by the fact that air quality standards currently lag behind other parts of the world (Paton-Walsh et al., 2019; Schofield et al., 2020) and that Australia is the highest per capita emitter of nitrogen oxides in the Organisation for Economic Co-operation and Development (as of 2017, latest available data, Organization for Economic Cooperation and Development (2017)). Moreover, new air quality challenges are coming to the fore in Australia’s cities as they expand and as climate change makes extreme weather events more likely (Paton-Walsh et al., 2019). These include ozone smog events related to extreme temperatures (e.g. Utembe et al., 2018; Ryan et al., 2020a, b), dust storms related to drought and strong winds (e.g. Johnston et al., 2011) and elevated aerosol levels due to wild fires (e.g. Keywood et al., 2015). To our knowledge no previous study has examined the air quality resulting from COVID-19 specifically in Australia, and how it might compare to the air quality impact of the Black Summer fires.

The aim of this paper is to quantify the air quality impact resulting from emissions changes during bushfire smoke events and the COVID-19 lockdown period in Australia. We focus on Australia’s two largest states, Victoria and New South Wales and the pollutants NO₂, O₃, CO, and PM_2.5. To quantify the change in pollutant concentrations, we first train a Random Forest using 19 years of historical air quality and meteorological data before using this model to predict a baseline scenario between November 2019–May 2020, incorporating the bushfire period and the start of the COVID-19 shutdown.