Development and application of a multi-task oriented deep learning model for quantifying drivers of air pollutant...

Quantitative assessment of the drivers behind the variation of six criteria pollutants, namely fine particulate matter (PM_2.5), ozone (O₃), nitrogen dioxide (NO₂), sulfur dioxide (SO₂), particulate matter (PM₁₀), and carbon monoxide (CO), in the warming climate will be critical for subsequent decision-making. Here, a novel hybrid model of multi-task oriented CNN-BiLSTM-Attention was proposed and performed in Taiyuan during 2015-2020 to synchronously and quickly quantify the impact of anthropogenic and meteorological factors on the six criteria pollutants variations. Empirical results revealed the residential and transportation sectors distinctly decreased SO₂ by 25 % and 22 % and CO by 12 % and 10 %. Gradual downward trends for PM_2.5, PM₁₀, and NO₂ were mainly ascribed to the stringent measures implemented in transportation and power sectors as part of the Blue Sky Defense War, which were further reinforced by the COVID-19 pandemic. Nevertheless, temperature-dependent adverse meteorological effects (27 %) and anthropogenic intervention (12 %) jointly increased O₃ by 39 %. The O₃-driven pollution events may be inevitable or even become more prominent under climate warming. The industrial (5 %) and transportation sectors (6 %) were mainly responsible for the anthropogenic-driven increase of O₃ and precursor NO₂, respectively. Synergistic reduction of precursors (VOCs and NOx) from industrial and transportation sectors requires coordination with climate actions to mitigate the temperature-dependent O₃-driven pollution, thereby improving regional air quality. Meanwhile, the proposed model is expected to be applied flexibly in various regions to quantify the drivers of the pollutant variations in a warming climate, with the potential to offer valuable insights for improving regional air quality in near future.