Relating tropical cyclone track forecast error distributions with measurements of forecast uncertainty

Abstract

Tropical cyclone (TC) track forecasts will always contain uncertainty. This thesis relates ranges (bins) of uncertainty measurements with historical TC track forecast errors, to provide statistically distinct error distributions for use with the Monte Carlo (MC) method. T-test and Kolmogorov-Smirnov tests are used to confirm distinctness among error distributions associated with the bins of either European Center for Medium-Range Weather Forecasts (ECMWF) ensemble spread or TVCN Goerss Predicted Consensus Error (GPCE). The statistical tests indicate that distinct error distributions (consisting of official TC forecast error, ECMWF ensemble mean [EMN] error, or TVCN error) exist when using four bins of uncertainty (of either uncertainty measurement). Furthermore, error distributions of ECMWF EMN error are distinct with five bins of ECMWF ensemble spread. Along- and cross-track official errors could not be directly related to either measurement of uncertainty at even three bins. These results suggest that the National Hurricane Center test and evaluate the use of four bins of uncertainty for operational use with the MC method to further improve its Wind Speed Probability products and overall TC track forecasts. TC forecasters should also exploit the more impressive relationship established using five bins ECMWF ensemble spread with ECMWF EMN error.