LSTMAR-Net: A Deep Learning Model for Improving Time Resolution of AIRS Satellite Arctic Tropospheric Ozone Data

Arctic tropospheric ozone is a critical chemical component of the Earth’s atmosphere with diverse impacts on regional ecological environments and climate systems. Stratosphere-troposphere exchange (STE) is an important factor in the increase in Arctic tropospheric ozone. In this study, a long short-term memory with attention and residual network (LSTMAR-Net) model was developed to improve the time resolution of atmospheric infrared sounder (AIRS) satellite Level 3 ( $L3$ ) product tropospheric ozone profile data from multiple stations in the Arctic. The model performed well on the test set. The Pearson correlation coefficients for the upper troposphere at the Eureka station are mostly above 0.70, with root mean squared error (RMSE) and mean absolute error (MAE) below 0.027 ppmv. Similarly, in the upper troposphere at the Alert station, the Pearson correlation coefficients are mostly above 0.60, with RMSE and MAE values below 0.028 ppmv. The model successfully simulated ozone changes in the STE that were not detectable by AIRS data. The model also simulated the horizontal transport processes of ozone and successfully explained the evolution of ozone concentration changes.