Improving Real-time Forecast Accuracy of Convective Morphology Through Use of an Ensemble

Testing is being performed to determine the best way to provide real-time forecast guidance on expected convective morphology through the use a convection-allowing Weather Research and Forecasting (WRF) model ensemble. A total of 34 convective systems occurring over the Upper Midwest from the 2015 and early 2016 warm seasons have been simulated. These events are classified based on the classification scheme used in Gallus et al (Weather and Forecasting, 2008) which allows for nine morphologies: individual cells, cluster of cells, broken lines, linear system with trailing stratiform precipitation, linear system with leading stratiform precipitation, linear system with parallel stratiform precipitation, linear system with no stratiform precipitation, bow echo and nonlinear system. Four WRF ensemble members are being used to study the best way of formulating guidance on expected morphology during the 12-hour period over which the members are integrated. The members are constructed by incorporating NAM, GFS and RAP model data for initial and lateral boundary conditions. Three of the four members use the same model data for both initial and lateral boundary conditions with the fourth mixing NAM initial conditions with GFS lateral boundary conditions. These four members also use different combinations of microphysics and planetary boundary layer schemes. A classification of morphology is made every hour in each ensemble member as long as convection is occurring. The forecast morphology is based on the ensemble average which is determined by the majority classification for each hour. For hours in which there are an equal number of convective modes, a variety of methods are being studied to determine the most accurate prediction. Factors such as initial RAP analysis background conditions and classifications made in the previous or subsequent hours when such a disagreement is not present are being take into consideration to determine the ensemble average. An objective score is being calculated for each event to determine the accuracy of the simulated morphologies. Preliminary results are consistent with prior work showing that linear systems are not being simulated very well overall. RAP analyses of the model initialization times are being used to determine whether a bias correction based on prior results that suggest linear systems are underestimated and some cellular modes are overpredicted should be implemented to improve convective morphology forecasts