Quantifying Accuracy and Dispersion of Ensemble Forecasts of Severe Convection Using an Object-Based Technique

Assessing the accuracy and spread characteristics of ensemble forecasts is critical to understanding deficiencies an ensemble system may possess in terms of forecast error and/or biases. Presented work will focus on the verification results a 50 member Ensemble Kalman Filter (EnKF) configuration using the WRF-ARW model with 4km grid spacing across portions of the U.S. Great Plains. The Data Assimilation Research Testbed Ensemble Adjustment Filter (DART-EAKF) is used to assimilate mesocale-resolving observations into the model prior state. Ensemble forecast accuracy and spread is assessed using both traditional and object-based techniques. The object-based technique utilizes a time tracking algorithm that objectifies areas of model and observed rainfall that can be used to calculate occurrence scores and spatio-temporal tendencies in the initiation of deep, moist convection in terms of surface rainfall. Initial verification results through a handful of high-impact severe weather events reveal appreciable skill in the ability to properly represent the pre-storm environment as well in the general evolution of the convective event. However, several issues are identified within the suite of forecast members for all events, including a cool temperature bias at low-levels that is believed to be reason for a slight late bias in storm initiation. Most notably, underdispersion in several state variables is seen at low levels and near the surface, while slight overdisersion tends to exist aloft. In the net, the ensemble of forecasts tends to produce too few rainfall objects for all cases, also suggesting a lack of ensemble spread. In light of initial findings, cases are rerun with varied physics parameterizations in attempt to force ensemble forecast spread. This procedure is performed on the outer domains of the model incrementally down to the domain that explicitly resolves convection. Presented work will focus on the impact to state variable and object-based convection both in terms of accuracy and ensemble spread.