Evaluation of Extra-Tropical Cyclones Using a Feature-Relative Method

Many of the current verification metrics used in operations are grid point relative (e.g., regional), and averaging over a fixed geographic region will smooth many important processes and significant weather features around the cyclone. Some verification metrics are gathered for cyclones, but they are usually limited to central pressure, displacement, and cyclone density errors. In order to better understand operational model and ensemble performance and processes that may be leading to any systematic errors of significant weather around the cyclone, a cyclone-relative verification approach is needed. Knowledge of the performance of cyclone-relative features, such as jets, precipitation bands, and fronts, can also help develop confidence metrics and tools for forecasters about the skill in predicting those features at various lead times from 0 to 10 days.

A feature-relative approach was added to the unified modeling framework built on the Model Evaluation Tools (MET). MET is a state-of-the-science verification package supported to the community through the Developmental Testbed Center. Standard measures, such as continuous statistics may be computed in this cyclone relative framework. The errors associated with the cyclones can then related to the various important physical processes (moisture flux, stability, strength of upper-level PV anomaly and jet, surface fluxes, and precipitation), and the results will be separated by different stages of the cyclone (genesis, mature, and decay), cyclone intensity, and large-scale flow regime. This approach will be transferred to the U.S. National Oceanic and Atmospheric Administration (NOAA) National Center for Environmental Prediction (NCEP) to help model developers and forecasters better understand the origin of cyclone biases (e.g., hypothesized under-deepening of surface cyclones in the GFS in the medium range), and the ensemble performance for the significant weather around the cyclone. This presentation will describe the feature-relative methodology and provide examples from the prototype system.