| Summary: | The spatial and temporal complexity of tropical cyclones (TCs) raises a number of scientific questions regarding their genesis, movement, intensification, and variability. In this dissertation, the principal goal is to determine the current state of predictability for each of these processes. To quantify the current extent of tropical cyclone predictability, we assess probabilistic forecasts from the most advanced global numerical weather prediction system to date, the ECMWF Variable Resolution Ensemble Prediction System (VarEPS). Using a new false alarm clustering technique to maximize the utility of the VarEPS, the ensemble system is shown to provide well-calibrated probabilistic forecasts for TC genesis through a lead-time of one week, and pregenesis track forecasts with similar skill compared to the VarEPS's postgenesis track forecasts. To quantify the predictability of TCs on intraseasonal time scales, forecasts from the ECMWF Monthly Forecast System (ECMFS) are examined for the North Atlantic Ocean. From this assessment, dynamically based forecasts from the ECMFS provide forecast skill exceeding climatology out to weeks three and four for portions of the southern Gulf of Mexico, western Caribbean and the Main Development Region. Forecast skill in these regions is traced to the model's ability to capture correctly the variability in deep-layer vertical wind shear, the relative frequency of easterly waves moving through these regions, and the intraseasonal modulation of the Madden-Julian Oscillation. On interannual time scales, the predictability of TCs is examined by considering their relationship with tropical Atlantic easterly waves. First, a set of easterly wave climatologies for the CFS-R, ERA-Interim, ERA-40, and NCEP/NCAR Reanalysis are developed using a new easterly wave-tracking algorithm. From the reanalysis-derived climatologies, a moderately positive and statistically significant relationship is seen with tropical Atlantic TCs. In relation to large-scale climate modes, the Atlantic Multidecadal ...
|
|---|