PREreview of Indoor PM2.5 forecasting and the association with outdoor air pollution: a modelling study based on sensor data in Australia

This study aimed to predict hourly indoor PM2.5 concentrations and investigate their correlation with outdoor PM2.5 levels across 24 buildings in Australia. The researchers collected indoor air quality data from 91 sensors in eight Australian cities from 2019 to 2022. They developed an innovative three-stage deep ensemble machine learning (DEML) framework to predict hourly indoor PM2.5 concentrations. The DEML model consistently outperformed benchmark models, achieving high accuracy for most sensors. The study also found that outdoor PM2.5 concentrations significantly impacted indoor air quality, especially during bushfires. The findings highlight the importance of accurate indoor air quality prediction for developing location-specific early warning systems and informing effective interventions to protect public health.

Major issues

1. The study lacks detailed information on the specific building characteristics, such as building materials, surrounding environment, and ventilation system efficiency. These factors could potentially impact indoor PM2.5 variations and should be considered in the analysis.

2. The hourly outdoor PM2.5 data from monitoring stations were only available until December 2020, limiting the comparison between indoor and outdoor PM2.5 to this period. The authors should discuss the potential bias introduced using monitoring station data that may not accurately represent outdoor PM2.5 levels around the buildings.

3. The study was conducted mainly during the COVID-19 pandemic when human behaviors and activities were altered due to lockdowns and stay-at-home orders. The authors should further explore how these changes might have influenced indoor air quality and the relationship between indoor and outdoor PM2.5.

Minor issues

1. The authors should provide more details on the inter-comparison tests, calibrations, and uncertainty estimations performed on the sensors before installation to ensure data quality and reliability.

2. The study focuses on buildings in metropolitan areas of Australia. The authors should acknowledge that indoor and outdoor PM2.5 associations may differ in rural or outskirt areas due to distinct microclimate environments and ventilation conditions.

3. The manuscript would benefit from a more detailed description of the developed air quality monitoring dashboard, including its user interface, data visualization, and potential applications for building managers and occupants.

Competing interests

The author declares that they have no competing interests.