Forecasting upper atmospheric scalars advection using deep learning: an $$O_3$$ O 3 experiment

Weather forecast based on extrapolation methods is gathering a lot of attention due to the advance of artificial intelligence. Recent works on deep neural networks (CNN, RNN, LSTM, etc.) are enabling the development of spatiotemporal prediction models based on the analysis of historical time-series, images, and satellite data. In this paper, we focus on the use of deep learning for the forecast of stratospheric Ozone ( $$O_3$$ ), especially in the cases of exchanges between the polar vortex and mid-latitudes known as Ozone Secondary Events (OSE). Secondary effects of the Antarctic Ozone Hole are regularly observed above populated zones on South America, south of Africa, and New Zealand, resulting in abrupt reductions in the total ozone column of more than 10% and a consequent increase in UV radiation in densely populated areas. We study different OSE events from the literature, comparing real data with predictions from our model. We obtained interesting results and insights that may lead to accurate and fast prediction models to forecast stratospheric Ozone and the occurrence of OSE.