C3 - Predictability of tropical and hybrid cyclones over the North Atlantic Ocean
During Phase 1, this project focused on the predictability of extratropical storms that underwent Tropical Transition (TT) into cyclones with tropical traits over the Mediterranean Sea and the subtropical North Atlantic Ocean. For these hybrid cyclones, novel, cyclone-relative metrics were introduced to better describe changes in the forecast ensemble with lead times. Sudden changes in the ensemble statistics with lead time, termed "forecast jumps", were a common feature of eight storms exhibiting tropical characteristics over the Mediterranean Sea and the Bay of Biscay. The concept of a "predictability barrier" is introduced. These barriers start to disintegrate when the model's initial data contain sufficient information on the variety of factors influencing the life cycles of storms undergoing TT. Such changes and associated predictability regimes were also found in the case of the TT of Hurricane Chris over the North Atlantic. In this case, the changes were attributable to characteristics in the upper-level development of the potential vorticity (PV) fields and the relative location of the storm to PV streamers.
In Phase 2, we will build on our results from Phase 1 and substantially expand the scope to study the predictability of a much larger number of tropical and hybrid cyclones over the North Atlantic Ocean in ensemble forecasts. The hybrid systems comprise subtropical cyclones and cyclones that undergo tropical and extratropical transitions, respectively. The forecast periods will be split into less than 10 days, for which the predictability of individual cyclones will be considered and into lead times from one week to one month, for which the predictability of bulk cyclone properties such as cyclone numbers or accumulated cyclone energy (ACE) will be in the focus.
We will follow a statistical-dynamical approach for forecast lead times of more than one week. A neural network will be developed, validated, and compared with a pre-existing multilinear regression model. Relevant predictors for the neural network will be, for example, wave-filtered Hovmoeller diagrams in the tropical wave channel and simplified representations of Rossby wave breaking. Later these predictors will be amended by geometry-based, low-dimensional descriptions of wave-related objects. A generalized concept from computational geometry to describe the latter will be developed and efficiently applied to analysis and large ensemble data sets to infer robust statistics of object characteristics (e.g., origins and splits of PV features). For forecasts up to 10 days, particularly rapid changes in the evolution of ensembles with lead time will be related to "wave objects" and other cyclone-related features with the aim to understand the meteorological causes of "typical" forecast jumps over a large number of cases. The data sets used will be the "Sub-seasonal to Seasonal" (S2S, > 1 week) ensemble (re-)forecasts and post-2016 operational ensemble forecasts (< 10 days) of the European Centre for Medium-Range Weather forecast (ECMWF) ensemble data. From the proposed research, a better understanding of what limits tropical and hybrid cyclone predictability at different lead times will be achieved.