Thursday, 16 January 2025: 9:15 AM
354 (New Orleans Ernest N. Morial Convention Center)
Handout (2.4 MB)
Wind forecasting at the Hong Kong International Airport (HKIA) during tropical cyclones (TCs) is critical for aviation safety and operational efficiency. Traditional forecasting methods based on climatology, statistical regression and analogue often struggle with accuracy and reliability. This study introduces XGBoost wind prediction models using observation and forecast dataset as features, including recent METAR reports, TC track and intensity forecast, and post-processed numerical weather prediction multi-model output, namely the objective consensus forecast (OCF). Data preparation involved trigonometric transformations to handle periodic parameters and avoid nonsense out-of-range predictions.
Model performances were evaluated for models with different forecast ranges: immediate (T+0 hour), short-term (T+1 to T+12 or T+24 hours), and extended (T+1 to T+36 hours). The XGBoost models demonstrated improved accuracy over baseline models (OCF), with lower mean square error (MSE), lower mean absolute error (MAE), and higher R-squared values. Notably, the models showed better performance in predicting wind speed, particularly in the immediate and short-term ranges.
The XGBoost model's ability to accurately predict wind conditions at HKIA during TCs can significantly enhance decision-making processes for supporting aviation operations. These models provide basis to construct an interactive tool for aviation forecasters to obtain guidance on hourly wind forecasts under alternative scenarios and access forecast uncertainties by adjusting the TC forecast tracks, offering better preparedness and response strategies in severe weather scenarios.
This presentation will detail the data processing techniques, model training procedures, and performance evaluation metrics of the XGBoost wind prediction model. It will also highlight case studies demonstrating the model's practical applications for more accurate airport wind forecasts contributing to aviation safety and efficiency during TCs.
Model performances were evaluated for models with different forecast ranges: immediate (T+0 hour), short-term (T+1 to T+12 or T+24 hours), and extended (T+1 to T+36 hours). The XGBoost models demonstrated improved accuracy over baseline models (OCF), with lower mean square error (MSE), lower mean absolute error (MAE), and higher R-squared values. Notably, the models showed better performance in predicting wind speed, particularly in the immediate and short-term ranges.
The XGBoost model's ability to accurately predict wind conditions at HKIA during TCs can significantly enhance decision-making processes for supporting aviation operations. These models provide basis to construct an interactive tool for aviation forecasters to obtain guidance on hourly wind forecasts under alternative scenarios and access forecast uncertainties by adjusting the TC forecast tracks, offering better preparedness and response strategies in severe weather scenarios.
This presentation will detail the data processing techniques, model training procedures, and performance evaluation metrics of the XGBoost wind prediction model. It will also highlight case studies demonstrating the model's practical applications for more accurate airport wind forecasts contributing to aviation safety and efficiency during TCs.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner