Data-driven machine learning quantifies ozone transport in the Hangzhou Bay urban cluster

Abstract Severe ozone (O 3 ) pollution has always been a serious problem faced by areas with rapid economic development, and the regional O 3 transport between cities is a major cause of this problem. Therefore, we used a bidirectional long short-term memory (Bi-LSTM) model to quantitatively identify the regional O 3 transport in Hangzhou Bay, China. Combined with the meteorological removal method, we were able to model O 3 concentrations that were not affected by transport. The contribution of regional transport to Shanghai’s O 3 was quantified and validated using two different simulation schemes, which yielded highly consistent results of 18.41 μg/m 3 (24% contribution) and 20.52 μg/m 3 (27% contribution). According to the model simulation results, we found that approximately 24% of the O 3 pollution in Shanghai originates from other cities in the summer when the O 3 pollution is high. In addition, the regional O 3 transport was mainly concentrated during the high-value weather of O 3 pollution in Shanghai, and transport on non-pollution days was not apparent. Therefore, the regional O 3 transport from other cities is an important source of O 3 pollution in Shanghai. Overall, our study demonstrates the potential of machine-learning models coupled with meteorological removal for quantifying the inter-city influence of atmospheric pollutants.