Enhancing ozone nowcasting over East Asia using a data-to-data translation approach with observations from a geostationary environment monitoring spectrometer

Seong-Taek Oh,Jeong-Eun Park,Jaehoon Jeong,Sungwook Hong
DOI: https://doi.org/10.1016/j.apr.2024.102054
IF: 4.831
2024-01-21
Atmospheric Pollution Research
Abstract:The degradation of air quality caused by excessive anthropogenic ozone (O 3 ) concentrations negatively affects ecosystems and the atmosphere. To monitor this issue, the Geostationary Environment Monitoring Spectrometer (GEMS) aboard the Geostationary Korea Multi-Purpose Satellite (GK)-2B provides ozone products over Asia-centered circular regions based on the ozone algorithm derived from representative polar-orbit atmospheric environmental satellites. In this study, we developed a few-hour GEMS O 3 nowcasting model using data-to-data (D2D) translation with a conditional generative adversarial network. This model is based on hourly GEMS O 3 time-series products and can be used to predict ozone concentrations. The D2D model underwent training and testing employing paired input and output datasets of GEMS O 3 , with data collected from March 22, 2020, to June 21, 2020, and from March 22, 2021, to June 18, 2021, respectively. The resulting D2D ozone nowcasting model was used to determine ozone concentrations in time zones where GEMS O 3 products were unavailable. Test results of the D2D model demonstrated excellent statistical scores, including a bias of 2.162 Dobson Units (DU, where 1 DU corresponds to 2.687 × 10 16 molecules/cm 2 ), root-mean-square error (RMSE) of 5.606 DU, a correlation coefficient (CC) of 0.994 for 1-h prediction, a bias of 1.421 DU, RMSE of 5.903 DU, and CC of 0.992 for 2-h prediction, and a bias of 1.169 DU, RMSE of 6.797 DU, and CC of 0.988 for 3-h prediction. Despite the dataset pairing and number limitations, the D2D prediction model accurately forecasted GEMS O 3 within 3 h in East Asia.
environmental sciences
What problem does this paper attempt to address?