871 Studying error growth in the tropical cyclone genesis environment with mesoscale high-resolution ensemble forecasts

Thursday, 10 January 2013
Exhibit Hall 3 (Austin Convention Center)
Levi S. Thatcher, University of Utah, Salt Lake City, UT; and Z. Pu

While many studies have used global models to study tropical cyclone genesis in ensemble-based forecasts, regional models have rarely been similarly employed. In light of this and the open questions related to ensemble creation, in this study we use the mesoscale community Weather Research and Forecasting (WRF) model at 5 km resolution to study error growth, structure, and characteristics in the vicinity of tropical cyclone (TC) genesis. The first numerical experiment is performed for the forecast of the genesis of Hurricane Ernesto (2006).

First, a control ensemble is created with the initial and boundary conditions (ICs and BCs) derived from the National Centers for Environmental Prediction (NCEP) Global Ensemble Forecast System (GEFS) with 14 members. A series of mesoscale regional ensemble forecasts is then created using a simple breeding vector method. The first type of bred ensemble is created by starting with GEFS ICs and BCs and then breeding potential temperature, water vapor, and/or the wind component for a predetermined amount of time. The second type of bred ensemble is created by performing the same breeding cycle for the same fields, after which the final perturbations are placed into the mesoscale field derived from the NCEP Global Forecast System (GFS) final analysis (FNL). The behavior of error structure and growth in these mesoscale ensembles is compared along with that of the control ensemble. The E-dimension, which is based around empirical orthogonal functions; perturbation versus error correlation analysis (PECA); and Fourier transforms are used to evaluate the results. The first two metrics are not only studied over time, but also in terms of distance from Ernesto's center.

It is found that the mesoscale ensembles created using breeding schemes produce notably higher E-dimension at the beginning of the simulations than does the control ensemble. This holds for E-dimension based on various fields and for the breeding of theta, vapor, and u/v. It is also found that E-dimension within 200 km of Ernesto was often 50% higher than that at large radius. The higher E-dimension at smaller radii held for all ensembles, but fluctuated greatly over time, ensemble type, and bred variable chosen. More results will be presented in the conference.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner