Probabilistic Prediction of North Atlantic Hurricane Track and Intensity

Probabilistic Prediction of North Atlantic Hurricane Track and Intensity

Christopher Dickson, Judith Curry

Climate Forecast Applications Network

The hurricane forecasts of track and intensity produced by the National Hurricane Center are deterministic, focused on a single best forecast plus a cone of track uncertainty. The desire for probabilistic forecasts of track and intensity is frustrated by the relatively small ensemble size of the NOAA GEFS/FVS3 forecast system of 21 members. The ECMWF forecast system provides a total of 52 ensemble members, but even the larger ECMWF ensemble is not sufficiently disperse to provide true probabilistic forecasts. Inclusion of multiple models with ensemble forecasts increases the size of the ensemble, but introduces additional issues that make determination of a true probability forecast far from straightforward. Climate Forecast Applications Network (CFAN) has approached the challenge of creating probabilistic forecasts of hurricane tracks and intensity by creating synthetic tracks from the forecasts thus expanding the size of the ensemble from a given forecast model. A distribution of historical track forecast error (along- and cross-track) is generated using the model reforecasts (hindcasts) and observed tracks. A Monte-Carlo resampling technique is used to create synthetic tracks by a random draw from the historical along-track and cross-track error distributions. In the current forecast configuration the Monte Carlo resampling technique produces a set of 1,500 synthetic forecast tracks for each tropical cyclone forecast. The pdf derived from the synthetic tracks provides a dynamic cone of uncertainty. The synthetic tracks are used to create probabilistic intensity forecasts, after correcting the forecast intensities against the model reforecasts and observed intensities using a quantile-to-quantile calibration. PDFs of track intensity and minimum pressure are created. From calibrated 2-D wind fields, probabilistic predictions of the storm Integrated Kinetic Energy and Cyclone Damage Potential are created. Examples and verification statistics are presented from the 2019 Atlantic hurricane season.