Short-Term Ensemble Streamflow Prediction Using Spatially Shifted QPF Informed by Displacement Errors

Abstract

Abstract Errors associated with the location of precipitation in QPFs present challenges when used for hydrologic prediction, particularly in small watersheds. This work builds on a past study that systematically shifted QPFs prior to inputting them into a hydrologic model to generate streamflow ensembles. In the original study, which used static, predetermined shifting distances, flood detection improved, but false alarms increased due to large ensemble spread. The present research tests a more informed approach by randomly selecting shift directions and distances based on the distribution of displacement errors from a sample of QPFs. Precipitation forecasts were taken from the High-Resolution Rapid Refresh Ensemble (HRRRE), and streamflow predictions were generated using the Weather Research and Forecasting hydrological modeling system, version 5.1.1, in a National Water Model 2.0 configuration. A 63-member streamflow ensemble was generated using the 9 original HRRRE and 54 shifted HRRRE members. Two ensemble updating schemes were tested in which ensemble member weights were adjusted using precipitation location and QPF displacement present at convective initiation. The ensembles using QPF shifted based on climatological spatial errors showed higher probabilistic forecasting skill, while having comparable dichotomous forecasting skill to the original HRRRE ensemble. Other methods of selecting nine ensemble members from the full 63-member suite did not show significant improvement. Flood peak timing showed frequent errors, with average timing errors around five hours early. Larger watersheds tended to have better skill metric scores than smaller basins, with increased skill added by the shifting of QPF.