TTU WRF Ensemble Modeling Techniques and their Application to the Home Utility Management System
Manser, Russell P P
MetadataShow full item record
Convection-allowing model (CAM) ensemble forecasts provide quantitative probabilistic guidance of convective hazards that forecasters would otherwise qualitatively assess, such as updraft helicity. However, various ensemble strategies can be used to generate CAM probabilistic forecasts and it is still unclear how different configurations perform. This suggests that a focus on quality of initial conditions may be worth further study, especially in the frame of probabilistic guidance for convective hazards. Comparisons of CAM ensembles are challenging due to the computational resources required. Because of this, previous studies have been limited in the number of cases the authors consider, as well as the number and type of verification methods used to assess forecast quality. In this study, the current Texas Tech University (TTU) ensemble modeling system and three additional initial condition ensemble modeling systems are verified. Holding the model configuration constant, initial conditions are generated by downscaling GEFS member forecasts, applying random perturbations to GFS initial conditions, and re-centering EnKF perturbations on GFS initial conditions. These three strategies represent feasible systems that could become the TTU ensemble modeling system given its use in several public and private forecasting applications. Forecasts are initialized for 48 hours every 12 hours for six consecutive weeks from 0000 UTC 27 April 2016 to 1200 UTC 3 June 2016 and are verified using traditional, neighborhood, and object-based methods. This study finds that higher quality forecasts are produced by ensembles centered on GFS analyses. Ensembles initialized with random and EnKF perturbations added to GFS analyses have the most accurate surface and upper air forecasts. Adding random perturbations to GFS analyses tends to produce the most reliable forecasts. EnKF and Downscaled GEFS ensemble forecasts are less accurate and reliable than those with GFS analyses as their center. Object-based verification of precipitation forecasts shows small, but statistically significant differences in mean accuracy of object locations in favor of EnKF perturbations centered on GFS analyses. The most skillful probabilistic forecasts generally are produced by ensembles initialized with random and EnKF perturbations added to GFS analyses. However, flow-dependent ensembles tend to have more accurate precipitation forecasts for early lead times and larger neighborhoods. The EnKF ensemble has the least skillful forecasts of convective parameters with few exceptions. This work suggests that initial condition center is important for forecast accuracy and that random perturbations can provide sufficient ensemble spread. Random perturbations added to GFS analyses prove most promising as the next TTU WRF ensemble modeling system due to smaller computational expense and the ability to tune perturbations. Such an ensemble would provide value by making forecasts available at earlier lead times and improving upon ensemble spread and accuracy at later lead times.