Convective System QPF Displacement Errors in HRRRE and Potential Use for Shifting QPF Fields to Improve Flood Forecasting

For small watersheds, displacement errors in weather model quantitative precipitation forecasts (QPF) can result in very large errors in forecast streamflow. As a result, most streamflow forecasts are made after precipitation has begun when observed precipitation can be used as the input to the model. However, more lead time could be provided for flood warnings in these small basins if QPF could be used instead of observed rainfall. One way to improve the QPF skill so that it can be used in hydrological models would be to adjust for displacement errors. The current study explores displacement errors for heavy convective rainfall events in the Midwest to determine whether systematic biases are present and if displacement error for the primary period of heavy rainfall is correlated well with displacement error at initiation time.

To explore these questions, the nine-member High-Resolution Rapid Refresh Ensemble (HRRRE) is examined for a sample of 28 cases, four from 2017, and 24 from 2018. For each, we isolated the synoptic or mesoscale event which caused heavy rainfall and computed a center of mass for each of the nine HRRRE member rainfall regions. These centers of mass were then compared to the center of mass of Quantitative Precipitation Estimate data taken from the North Central River Forecast Center.

Our results show that the HRRRE generally has a westward bias in QPF for the 28 heavy rainfall events, and the bias is significant with an average westward displacement varying between approximately 10 and 28 kilometers for the nine members. However, in the north-south direction, although displacements are common, there is not a systematic displacement as a roughly equal number of cases are displaced north and south, with average displacements varying from approximately 9 kilometers northward to 4 kilometers southward for the nine members.

Even though a significant displacement toward the west was found in the full sample of cases, there were some events where most or all of the nine HRRRE members showed displacements to the east. This suggests that there are likely better ways to anticipate the QPF displacement errors. Work is underway to examine the displacement during the first hour of precipitation in the HRRRE members to see how well they correlate with the displacement of the QPF center of mass over a much longer period during the event. If a correlation exists, QPF fields could be shifted after the first hour of observed and modeled rainfall and input into hydrologic models, likely yielding much more accurate forecasts of streamflow than are possible from existing ensemble guidance. We also plan to explore whether other environmental parameters such as wind speed and direction, along with temperature and humidity could better anticipate the displacement errors in HRRRE QPF fields. In addition, we have obtained QPF from the High Resolution Ensemble Forecast (HREF) system for the same sample of cases and will perform a similar analysis of displacement errors for that dataset. The HREF performed better than the HRRRE during the 2018 NOAA Spring Forecast Experiment, so it will be important to document differences in the displacement climatologies.

This project is part of a larger effort to improve probabilistic streamflow forecasting in flash and river flooding situations, and is designed to supplement that work. The HRRRE QPF for each of the nine members, along with eight 0.5 and 1.0 degree shifts in the cardinal and intermediate directions are fed into a hydrological model. If better estimates of expected displacement errors can be identified, these shifts can be weighted accordingly to better predict the probabilities of streamflow exceedances, and possibly improve flood forecasts for small basins.