Short- and Medium-Range Prediction of Tropical and Transitioning Cyclone Tracks within the NCEP Global Ensemble Forecasting System

Abstract

The NCEP Global Ensemble Forecasting System (GEFS) is examined in its ability to predict tropical cyclone and extratropical transition (ET) positions. Forecast and observed tracks are compared in Atlantic and western North Pacific basins for 2006?08, and the accuracy and consistency of the ensemble are examined out to 8 days. Accuracy is quantified by the average absolute and along- and cross-track errors of the ensemble mean. Consistency is evaluated through the use of dispersion diagrams, missing rate error, and probability within spread. Homogeneous comparisons are made with the NCEP Global Forecasting System (GFS). The average absolute track error of the GEFS mean increases linearly at a rate of 50 n mi day?1 [where 1 nautical mile (n mi) = 1.852 km] at early lead times in the Atlantic, increasing to 150 n mi day?1 at 144 h (100 n mi day?1 when excluding ET tracks). This trend is 60 n mi day?1 at early lead times in the western North Pacific, increasing to 150 n mi day?1 at longer lead times (130 n mi day?1 when excluding ET tracks). At long lead times, forecasts illustrate left- and right-of-track biases in Atlantic and western North Pacific basins, respectively; bias is reduced (increased) in the Atlantic (western North Pacific) when excluding ET tracks. All forecasts were found to lag behind observed cyclones, on average. The GEFS has good dispersion characteristics in the Atlantic and is underdispersive in the western North Pacific. Homogeneous comparisons suggest that the ensemble mean has value relative to the GFS beyond 96 h in the Atlantic and less value in the western North Pacific; a larger sample size is needed before conclusions can be made.