Development and Implementation of GEFS Based Probabilities for the Tropical Cyclone Logistical Guidance for Genesis (TCLOGG)

This presentation discusses the development and implementation of probabilistic genesis forecasts based on output from the Global Ensemble Forecast System (GEFS). These forecasts are the latest addition to the Tropical Cyclone Logistical Guidance for Genesis (TCLOGG). The developmental dataset was created following the methodology of Halperin et al. (2017) using the GEFS Reforecast (GEFSR) V12 dataset, spanning from 2000 to 2019. This dataset was created using the five GEFSR ensemble members initialized at 00 UTC each day.

To generate probabilistic output for tropical cyclone disturbances, an ensemble-based machine learning classification model (XGBoost) was employed to develop statistical models for the 0 to 168 hour forecast period across multiple basins: the Eastern Pacific (EP), North Atlantic (AL), West Pacific (WP), Indian Ocean (IO), and Southern Hemisphere (SH). The final model configurations were tested retrospectively over the 2020-2023 seasons on the operational control member. The experimental ML statistical models demonstrated well-calibrated TC genesis forecast probabilities globally.