Improving convective mode and streamflow forecasts through the use of convection-allowing ensembles
During the warm season (March – September), high-intensity convection events that can produce intense winds, large hail, tornadoes, and flash floods frequently occur in the U.S. Upper Midwest. Accurately forecasting convection is therefore vital for the development of warnings for these high-intensity events. Despite the improvements in the ability to forecast convection in recent years, mainly due to the enhancements in technology and computational resources, forecasting the timing, location, and intensity of convection remains a difficult task. The use of ensembles to predict various characteristics of convection have become increasingly more common to further improve these forecasts. It can be argued that a deterministic ensemble mean forecast typically performs better than an individual member and the probabilistic forecasts from ensembles have an advantage over deterministic because they limit over- and underforecasting while accounting for a level of uncertainty. Forecasting two characteristics of convection using ensembles is the focus of this work: convective mode evolution forecasts and quantitative precipitation forecasts (QPFs) for use in streamflow forecasting.
First, a small 4-member Weather and Research Forecasting (WRF) model ensemble with diverse members was used to test the ability of such an ensemble to predict convective evolution. Because this ensemble is likely smaller than one used operationally, the 10-member National Center for Atmospheric Research (NCAR) ensemble was also tested for a select number of cases. Both deterministic and probabilistic forecasts were created and verified with the observed radar reflectivity. Overall results indicated that a deterministic statistical mode forecast typically performed better than the individual members. Probabilistic forecasts showed that the ensemble performed better at predicting broader convective groups and for longer time periods over individual convective modes and shorter time periods.
Second, a systematic shifting method to mitigate spatial displacement errors in QPF for use in the Sacramento Soil Moisture Accounting – Heat Transfer (SAC-HT) hydrologic forecast model was tested. Streamflow forecasts using systematically shifted QPF from the 9-member High-Resolution Rapid Refresh Ensemble (HRRRE) were compared to streamflow forecasts using the HRRRE raw QPF output. Each HRRRE member was shifted in the four cardinal and four intermediate directions to create an 81-member ensemble. The shifted QPF streamflow ensemble showed an overall improvement in the ability to capture flooding events over the raw QPF streamflow ensemble. Despite the improved ability for the shifted QPF ensemble to capture an event, forecast probabilities of an event remained relatively low.