Tracking and verification of tropical cyclone development in global ensemble prediction systems: evaluations during recent field programs

This study evaluated two global ensemble forecast systems from the National Centers for Environmental Prediction (NCEP) and the Naval Research Laboratory (NRL) in their performance of predicting the genesis and evolution of tropical cyclones. The skill of the ensembles was assessed using cases studies from recent field programs. For each model, the case studies consisted of five named tropical cyclones and two unnamed nondeveloping tropical systems. The overall skill of ensemble forecasts, ensemble means and spreads, and probabilities were verified. Various tracking methods with different variables and vertical levels were compared for effectiveness. A manual tracking method with three variables (vorticity maxima, geopotential height and circulation) at the 850 hPa level was adopted to track tropical cyclones in ensemble forecasts for both their pregenesis and postgenesis phases. In the NCEP ensemble forecasts, the overall forecast skill of ensemble tracks and intensity in the storm's pregenesis phase was relatively low. The forecast uncertainty generally decreased with the reduction of forecast lead time, while the short-range forecasts tended to be more accurate. In contrast, the skill of ensemble track forecasts significantly improved for matured storms, possibly because of the implementation of the NCEP storm relocation scheme. Although the high resolution deterministic GFS forecasts were equal or better than the ensemble forecasts in almost half of the cases, the ensemble forecasts added value in predicting the probability of the tropical cyclone genesis. For example, the ensemble spread could be a good indication of the forecast track errors. Two versions of the NOG APS ensemble are compared: with and without the addition of stochastic convection. The forecast skill of tropical cyclone tracks and development from the NOG APS ensemble without stochastic convection was relatively low, although it provided additional information over the deterministic control. Ensemble track spreads were small and underdispersive in the control ensemble. Including stochastic convection enlarged the ensemble track spread and enhanced genesis predictability. The ensemble spread also showed a better correlation to the forecast track errors, especially in the postgenesis phase. However, further evaluation showed that weaker tropical systems could be over-intensified in the ensemble with stochastic convection.