Statistical–Dynamical Seasonal Forecast of North Atlantic and U.S. Landfalling Tropical Cyclones Using the High-Resolution GFDL FLOR Coupled Model

Abstract

Abstract Retrospective seasonal forecasts of North Atlantic tropical cyclone (TC) activity over the period 1980–2014 are conducted using a GFDL high-resolution coupled climate model [Forecast-Oriented Low Ocean Resolution (FLOR)]. The focus is on basin-total TC and U.S. landfall frequency. The correlations between observed and model predicted basin-total TC counts range from 0.4 to 0.6 depending on the month of the initial forecast. The correlation values for U.S. landfalling activity based on individual TCs tracked from the model are smaller and between 0.1 and 0.4. Given the limited skill from the model, statistical methods are used to complement the dynamical seasonal TC prediction from the FLOR model. Observed and predicted TC tracks were classified into four groups using fuzzy c-mean clustering to evaluate the model’s predictability in observed classification of TC tracks. Analyses revealed that the FLOR model has the highest skill in predicting TC frequency for the cluster of TCs that tracks through the Caribbean and the Gulf of Mexico. New hybrid models are developed to improve the prediction of observed basin-total TC and landfall TC frequencies. These models use large-scale climate predictors from the FLOR model as predictors for generalized linear models. The hybrid models show considerable improvements in the skill in predicting the basin-total TC frequencies relative to the dynamical model. The new hybrid model shows correlation coefficients as high as 0.75 for basinwide TC counts from the first two lead months and retains values around 0.50 even at the 6-month lead forecast. The hybrid model also shows comparable or higher skill in forecasting U.S. landfalling TCs relative to the dynamical predictions. The correlation coefficient is about 0.5 for the 2–5-month lead times.