Using numerical weather model outputs to forecast wind gusts during typhoons

Numerical weather prediction (NWP) models are commonly used to predict mean wind speeds. However, methods for predicting gust wind speeds in NWP models are still lacking due to limitations of computer resources. This study develops two models to provide gust wind speed forecasts during a typhoon event (1) a linear regression-based model and (2) a micro-genetic algorithm (MGA)-based model. This study also proposes a successively accumulated regression process that can adaptively update the parameters in the models and thereby improve the forecasts. The performance was evaluated by comparing observations from gauge stations in Taiwan during six typhoons that occurred between 2015 and 2016. The results of the models were compared to a statistical model from a previous study. The results showed that the successively accumulated regression process improves the mean error of the linear regression-based model to 5.23 m/s, compared with 7.08 m/s from the previous study. Meanwhile, the MGA-based model provides the best performance, with a 6.26 m/s time-averaged root mean squared errors for lead times of 7-54 h. This result is superior to the 9.37 m/s obtained with the statistics-based model and the 8.22 m/s obtained with the linear regression-based model.