Objective Consensus Typhoon Track Forecasting Using Multimodel Superensemble

This study applies the multimodel superensemble, a objective and weighted consensus technique to typhoon track forecasting in the western North Pacific to provide good guidance to forecasters. This consensus is constructed using linear regression analysis. Each forecasted track (latitude, longitude) per forecast period (1-5-day) is calculated by forecasted results of selected models and their regression coefficients trained in more than the past 2-years. We used the KMA (Korea Meteorological Administration) typhoon analysis data as the reference value and various models typhoon forecast data over the western North Pacific during 2011-2013 (up to typhoon 1330 HAIYAN). For 3-day forecast, multimodel superensemble shows better performance approximately 3% in the cross-validation (2011-2012) and 7% in the real-time forecast (2013) than non-weighted average consensus. However the track error of multimodel superensemble is very large for 5-day forecast. Also performance characteristics (biases) are examined. Fig. 1. Period mean track error of the multimodel superensemble (blue) and non-weighted average consensus (green) each forecast period (1-5-day) in the (a) cross validation (2011-2012) (b) real-time forecast (2013).