On the relative benefits of a multi-centre grand ensemble for tropical cyclone track prediction in the western North Pacific

Authors


Abstract

In this study, ensemble tropical cyclone (TC) track predictions were examined using the THORPEX Interactive Grand Global Ensemble (TIGGE) dataset. The main goal was to investigate the relative benefits of a multi-centre grand ensemble (MCGE) over a single-model ensemble (SME) from both deterministic and probabilistic perspectives. Here, the SME was composed of all ensemble members of the ensemble prediction system (EPS) at a certain numerical weather prediction (NWP) centre, while the MCGE was composed of all ensemble members of all or selected SMEs. Nine NWP centres participating in the TIGGE project were considered, and 58 TCs in the western North Pacific from 2008 to 2010 were verified.

In the verification of TC strike probability, the Brier skill score of the MCGE was larger than that of the best SME, which was the ECMWF EPS, on a medium-range time-scale, although this was not true on a short- to medium-range scale. In addition, the reliability was improved by the MCGE, especially in the high-probability range. Moreover, the MCGE reduced the number of missed events by about one tenth compared with the best SME. In the verification of confidence information, the MCGE was successful in extracting confidence information across all prediction times from 1 to 5 days. The relative benefits of the MCGE over the SME were seen in cases where the ensemble spread was extremely small. In such cases, the position errors of the MCGE were generally smaller than those of the best SME, indicating that, when multiple SMEs simultaneously predict a low uncertainty, the confidence level increases and the probability of a large position error decreases. In the verification of deterministic track predictions, the ensemble mean TC track predictions by the combination of the ECMWF, JMA, and UKMO EPSs were found on average to be slightly more accurate for 5-day predictions than those of the best SME, although the differences in the errors were not statistically significant. Copyright © 2012 Royal Meteorological Society

Ancillary