Pollen forecasting systems can provide information for coping with respiratory allergies. They estimate daily pollen production, dispersal, deposition, and removal based on daily weather conditions to predict daily pollen concentrations and provide allergy warnings. As of 2023, the Korea Meteorological Administration (KMA) provides 2-day forecast of allergenic pollens. However, unlike these models, long-term analysis of annual observations of tree pollen reveal annual variations. Therefore, in this study, we aimed to develop annual prediction models for allergenic tree pollens based on long-term multi-site pollen and meteorological data. Daily pollen concentrations were observed using Hirst-type volumetric spore traps at nine sites in Korea from 1998 to 2021, and daily weather data from the closest KMA stations were utilized. Models were developed to predict the seasonal pollen integral of seven tree species based on monthly mean temperature, wind speed, and total precipitation using three variable selection methods: 1) the t-test based key variable screening followed by linear regression with stepwise procedure (TM), 2) direct linear regression with stepwise procedure from the full variable model (FM), and 3) LASSO regression from the full variable model (LM). Data obtained during 1998-2017 and 2018=2021 were utilized for model development and validation, respectively. The root mean squared error, mean absolute error, mean error, and coefficient of determination (R²) revealed that the TM models were best suited for actual forecasting, even though R² in the TM model was lower than those of the FM and LM models. The annual variation model in this study can be integrated with the daily pollen forecast model by controlling the annual pollen potential, and the accuracy of the daily forecast can be improved accordingly.
Read full abstract