P76 Understanding ensemble behavior using the Method for Object based Diagnostic Evaluation (MODE)

Jensen, Tara, Tressa Fowler, Randy Bullock, John Halley Gotway, National Center for Atmospheric Research (NCAR), Ed Tollerud, Cooperative Institute for Research in the Atmosphere and National Oceanic and Atmospheric Administration (NOAA), Barbara Brown, NCAR, and Isidora Jankov, NOAA

Ensemble numerical weather prediction (NWP) is geared toward capturing the uncertainty in the numerical modeling process. There are many methods for running ensembles as well as for post-processing ensembles. Traditional statistics like the Brier Score, the Area Under the Receiver Operator Characteristic Curves (Area under ROC), the Reliability Diagram are well known traditional scoring metrics for ensemble-based probability scores. Rank histogram, also known as Talagrand Diagrams, and Spread-Skill relationships look the impact of the individual members. Traditional categorical and continuous statistics can be calculated for the ensemble mean. This talk breaks out of the traditional metric box and looks at the use of spatial metrics defined by the Method for Object-Based Diagnostic Evaluation (MODE) contained in the Model Evaluation Tools (MET) for diagnosis and evaluation of ensembles. It will use data sets from forecast experiments focused on severe weather and extreme precipitation events and explore the use of MODE on both ensemble means and individual members to better understand ensemble performance, such as over and under-dispersive behaviors. It will also demonstrate the use of MODE on probabilistic fields and explore how to interpret the results.