Session 17.6 Severe weather forecasts from an ensemble of human-perturbed simulations using an adjoint model

Friday, 8 October 2004: 11:45 AM
Victor Homar, NOAA/NSSL, Norman, OK; and D. J. Stensrud and J. J. Levit

Presentation PDF (280.6 kB)

The NOAA/Storm Prediction Center and National Severe Storms Laboratory conduct a yearly Spring Program experiment aimed at improving severe weather forecasts. The Spring Program 2003 (SP03) focused on the use of ensemble forecasting systems in assisting forecasters to produce short range severe weather outlooks. One of the sub-experiments of SP03 consisted of analyzing the benefits of having forecasters involved in the generation of a short-range ensemble forecasting system. The forecaster was asked to identify structures of interest in the control deterministic run and, using an adjoint model, a set of perturbations were obtained from the sensitivity fields to generate a 32-member MM5 ensemble. The simulations were run on a Linux supercomputer located at the Oklahoma Supercomputer Center for Education and Research (OSCER) at the University of Oklahoma. This talk will present details of the ensemble generation process and the verification scores of the system, focusing primarily on severe weather and precipitation. Results from this ensemble, using human involvement in the process, will be compared against other probabilistic forecasts including the SREF (multimodel and automatic breeding member generation) from the National Centers for Environmental Prediction and experimental SP03 day 2 outlooks. Severe weather probabilistic forecasts are verified using the National Weather Service reports database. Forecasts of 6-hourly and 24-hourly accumulated precipitation are verified against NCEP/CPC Stage IV analysis data.

Preliminary results confirm that the adjoint-generated ensemble mainly focuses the spread on areas highlighted by the human forecaster. This provides a set of possible outcomes linked to the features of interest of the day. Although simple postprocessing techniques such as bias correction are necessary, the skill of probabilistic forecasts from this ensemble are comparable to the other evaluated forecasts. Detected problems and suggestions to improve the system are also discussed.

- Indicates paper has been withdrawn from meeting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