414 Development of Localized Aviation MOS Program for Main Airports in South Korea

Tuesday, 30 January 2024
Hall E (The Baltimore Convention Center)
Jeonghoe Kim, Seoul National University, Seoul, South korea; and J. H. Kim

To support flight operations near airports, aviation meteorological offices issue the terminal aerodrome forecasts (TAFs) for several weather phenomena including wind, temperature, visibility, and ceiling. To produce TAFs, weather forecasters can utilize model output statistics (MOS), which is a collection of statistical models that predict various meteorological variables observed at the specific sites and airports (i.e., predictands) from numerical weather prediction (NWP) outputs (i.e., predictors). In this study, statistical models forecasting local weather at the main airports in South Korea (Incheon, Gimpo, Gimhae, and Jeju) were implemented by using NWP outputs and in-situ observations as the predictors. We follow the concept of the Localized Aviation MOS Program (LAMP), which was originally developed by the National Weather Service/Meteorological Development Lab (NWS/MDL). For the model training and validation, multi-year data of (1) Local Data Assimilation Prediction System (LDAPS) which is the Korea Meteorological Administration’s (KMA’s) operational regional NWP model, and (2) in-situ weather observations at the airports were utilized with pre-processing procedures. Two types of statistical models were developed: (1) regression models that generate continuous predictand values, and (2) classification models that select the best categorical expression of the predictands. First, a statistical forecast system was developed by using multilinear regression with forward feature selection, which is the method used in the KMA’s current operational MOS, as a baseline model. Then, statistical forecast systems using both linear (LASSO regression) and nonlinear (random forest regression and classification) models were developed and compared with the baseline model using scoring metrics such as bias, mean absolute error (MAE), and area under the ROC curve (AUC). It was found that the model incorporating in-situ observations as predictors increases its forecast performance for the short-term forecast (< 6-hr) compared to the forecasts from the traditional MOS system. A sample of automated TAF was produced from the newly developed statistical model, which can be used as potential guidance for aviation weather forecasters. Acknowledgement: This work was funded by the Korea Meteorological Administration Research and Development Program under Grant KMI2022-00310.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner