How BLUE is the Sky? Estimating air qualities in Beijing during the Blue Sky Day period (2008–2012) by Bayesian Multi-task LSTM.

• Air quality data in Beijing during the Blue Sky Day period are re-estimated. • A Bayesian multi-task learning model is constructed for air quality re-estimation. • Official air quality data, US Embassy PM 2.5 data, and proxy data are utilized. • Re-estimated air quality values are higher than the official values on average. • Re-estimated air quality data exhibit reduced statistical irregularity. For over three decades, air pollution has been a major environmental challenge in many of the fast-growing cities of the world, including Beijing, China. Given that any long-term exposure to high levels of air pollution has devastating health consequences, accurately monitoring and reporting air pollution information to the public is critical for ensuring public health and safety, while facilitating rigorous air pollution and health-related scientific research. Recent statistical research examining China's air quality data has posed questions on data accuracy, especially data reported during the Blue Sky Day (BSD) period (2000–2012), even though the quality of publicly available air quality data in China has improved substantially over the recent years (2013–2017). Until now, no attempts have been made to re-estimate the air quality data during the BSD period. In this study, we propose a multi-task machine-learning model to re-estimate the official air quality data during the recent BSD period, from 2008 to 2012, utilizing PM 2.5 data reported by the US Embassy in Beijing and proxy data covering Aerosol Optical Depth (AOD) and meteorology. Results have shown that average re-estimated daily air qualities are respectively 56% and 55% higher than the official ones, for air quality index (AQI) and AQI equivalent PM 2.5 , during the BSD period, from 2008 to 2012. Moreover, the re-estimated BSD air quality data exhibit reduced statistical discontinuity and irregularity. Our novel data re-estimation methodology can be used to provide more credible historical air quality data for evidence-based environmental and public health studies. [ABSTRACT FROM AUTHOR]