A study on seasonal forecasting of air quality in East Asia using statistical-dynamical methods

In this study, we developed a statistical-dynamical model that predicts the concentration of particulate matter in Korea 2-3 months in advance using the correlation between meteorological and climate factors and presented its performance. Temperature and atmospheric circulation around the Arctic Ocean considering the predictive performance of the National Centers for Environmental Prediction (NCEP) climate forecast system version 2 (CFSv2), the concentration of particulate matter in winter in Korea, and the correlation between meteorological and climate factors, potential predictors such as sea surface temperature in the Bering Sea region and sea level pressure in the Atlantic region were discovered. Using this, a multiple linear regression model was constructed between the average concentration of particulate matter during the winter of that year and latent factors for the second half of October and the first half of November in NCEP CFSv2, respectively. Seasonal predictions were made for the concentration of particulate matter in winter for a total of 20 years from 2001 to 2020. The result of the winter predicted in the second half of October showed r=0.49. And the result predicted in the first half of November showed a predictive performance of r=0.45. Considering the linear trend of particulate matter reduction, which was strong during the study period, r=0.72 in the second half of October and r=0.71 in the first half of November. This is judged to be the result of maximizing climate prediction performance considering the relatively long time scale of seasonal forecasting. In addition, additional forecasting ability can be expected through improved predictability of climate prediction models such as multi-model ensemble technology. However, although the results of the dynamical model were reflected, there are still limitations of the statistical model, and additional research is needed, such as problems due to limitations in observational data.