Session 6R.1 Evaluation of the NCAR Auto-Nowcaster during the NWS Ft. Worth Operational Demonstration

Wednesday, 26 October 2005: 10:30 AM
Alvarado ABCD (Hotel Albuquerque at Old Town)
Eric J. Nelson, NCAR, Boulder, CO; and S. J. Fano, R. Roberts, W. Bunting, T. Saxen, C. Mueller, H. Cai, A. Crook, D. Megenhardt, and J. Pinto

Presentation PDF (471.1 kB)

The NCAR Auto-Nowcaster (ANC) is an automated thunderstorm nowcasting system that produces routine short-term nowcasts for thunderstorm initiation, growth and dissipation. The ANC ingests a multitude of meteorological observations including automated surface weather observations, profiler retrievals, radiosonde observations, satellite imagery and data from 7 National Weather Service (NWS) WSR-88D radar sites. These data as well as operational Rapid Update Cycle (RUC) model output, forecaster input, feature detection algorithms, and a boundary layer numerical model (VDRAS) and its adjoint are used to produce predictor fields. The predictor fields are inserted into a fuzzy logic engine to produce a final thunderstorm nowcast.

The ANC has been deployed to the National Weather Service forecast office in Fort Worth Texas as part of the NWS Man In The Loop (MITL) demonstration. The MITL effort seeks to increase the role of the human in automated short term forecasting for the public and the aviation community. As part of the demonstration, the person working the short term forecast desk in the forecast office has been given additional responsibilities. He/she is asked to enter significant convergence boundaries (fronts, gust fronts, drylines, etc) using tools found on the ANC display. Unlike the automated boundary detection algorithm in the ANC which use only radar data to find shear zones and reflectivity fine lines on a single radar site, the human forecaster can quickly integrate radar data along with other operational data sets to identify features that may extend over multiple radar coverage areas. Human-inserted boundaries provide valuable information in areas where radar coverage is limited and feature detection algorithms are ineffective. The human-entered features are combined with automated feature detection algorithms and ingested into the fuzzy logic engine for the production of the final nowcast product.

The quantitative performance of ANC short term forecasts with and without human forecaster input will be discussed using validation statistics collected during real-time operations on a variety of intense convective events. Qualitative feedback provided by a real-time event log and post event surveys will be presented documenting the impact of ANC output on real-time NWS operations.

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