SELECTIVE ENSEMBLE MEAN TECHNIQUE FOR TROPICAL CYCLONE TRACK FORECASTS USING MULTI-MODEL ENSEMBLES

A selective ensemble mean technique for tropical cyclone (TC) track forecasts, which excludes from the ensemble those models that have large position errors at short lead times, was applied to a set of TC track forecasts produced by 11 operational global deterministic models. The position errors of the resulting selective ensemble mean TC track forecasts were verified for 91 TCs in the western North Pacific from 2010 to 2013 that reached an intensity classification of "tropical storm"or stronger. The TC position errors of the selective ensemble mean were smaller than those of a simple 11-member ensemble mean by 14.4%, 7.4% and 4.7% at forecast times of 24, 48 and 72 hours, respectively. However, the errors were larger than those of the best single-model-based deterministic forecasts, which were ECMWF forecasts. The correlation between TC position errors at short and long lead times was weak, which partially explains why the selective ensemble mean technique in this study had lesser skill than ECMWF forecasts. For operational forecasting, simple ensemble mean forecasts by ECMWF and NCEP generally provide the best forecast performance for verification samples from 2010 to 2013.