Regime-Dependent Verification and Calibration of a 10-Member Convection-Permitting Ensemble during the 2019 HWT SFE

Many statistical post processing techniques exist for the removal of global bias in convection-allowing model (CAM) ensembles. Recent studies have shown that these global post processing techniques have the ability to significantly improve objective model skill. However, it is also known that model error characteristics can exhibit sharp regional biases. Furthermore, CAM forecasts can exhibit unique error characteristics depending on the prevailing synoptic regime under which a forecast is generated. Incorporating this information into the existing post-processing technique framework, we generalize the cumulative distribution function (CDF) bias correction technique of Johnson and Wang (2014) to account for both regime and regional based bias.

Self-organizing maps are used to facilitate the clustering of meteorologically similar atmospheric regimes over a 9-year climatological period. Using these regimes, we classify 3-km grid spaced forecasts generated by the University of Oklahoma Multiscale data Assimilation and Predictability (MAP) group's experimental CAM ensemble run during the 2019 NOAA Hazardous Weather Testbed Spring Forecasting Experiment (HWT SFE). These forecasts are verified using a neighborhood maximum ensemble probability (NMEP) approach both before and after classification in order to establish a suitable baseline skill.

Bias correction was employed utilizing both a region-blind and regionally modified CDF approach performed with and without atmospheric regime segregation. Model performance was then re-verified and analyzed to gauge the benefits of regime segregated calibration versus the regime-blind approach, within both the region-blind and regionally modified context. These results will provide greater insight into the effectiveness of enhanced post-processing schemes in bolstering CAM performance.