Predictability of a mesoscale circulation associated with the high-impact rainfall during the first night of the great Colorado flood of September 2013: Performance of the HRRR then and now

Northeastern Colorado experienced historic flooding in mid September 2013. While one characteristic of the event was the long duration of the rainfall, from 9 through 15 September, a key part of the flooding happened during the evening of 11 September and early morning hours of 12 September, when very heavy rains fell in and near the foothills of Boulder and Larimer Counties. Studies of this portion of the flood have noted the importance of a mesoscale circulation that stalled over far southeastern Boulder County, sending convective cells west-northwestward into the foothills on intensified east-southeast upslope flow. About half the total precipitation in the nearby foothills and Front Range for the entire event fell during this six to ten hour period on the evening into the night of the 11th, and while significant rains also fell on the following day, it is likely that the magnitude of damage would have been considerably lower without the initial burst of heavy precipitation. Operational models, even at short range, did not correctly resolve this mesoscale feature and consequently heavier precipitation that was predicted stayed east of the foothills on the plains, without a threat of any flood. Examination of the high-resolution (3 km horizontal grid resolution) convection-resolving HRRR (for High Resolution Rapid Refresh) model, which as of September 2013 was still run in experimental mode, showed that a circulation was successfully resolved at low levels, but it was displaced eastward on the plains. As a result, enhanced precipitation associated with this mesoscale feature in the HRRR was also displaced to the east of the foothills for runs initialized during the day on 11 September, and it was not until late in the day when the HRRR began to move the circulation and associated heavy rains into the foothills. Since September 2013 a number of important changes have been made to the RAP (the ESRL Rapid Refresh model, the parent model for the HRRR) and the HRRR, and the HRRR is now (as of October 2014) operationally run at NOAA/NCEP (National Center for Environmental Prediction) and distributed to National Weather Service (NWS) Weather Forecast Offices (WFOs) at hourly intervals with forecasts out to 15-h. In light of these developments, we have made several reruns for the Colorado flood event, and the focus of this study will be on forecasts for the critical hours during the evening/night of 11 September and a comparison of the HRRR real-time runs with the improved versions of the HRRR, with emphasis on whether improvements result in better predictability of the key mesoscale circulation noted above.

We will review observations to document the evolution of the mesoscale circulation, which developed during the day south of the Denver and drifted northwards before stalling. It appears to be much like the well-documented “Denver Cyclone”, though recent study by others suggests latent heaving feedback during the evening probably intensified the circulation, resulting in the enhanced upslope flow. The official National Weather Service (NWS) Assessment Report noted that the operational models were not particularly accurate in their details (although certainly gave a longer-range indication of a wet period), and as noted tended to focus precipitation away from the foothills as one got closer to the event. The type of circulation seen on 11 September is on a scale that might not be readily resolved by the larger scale models like the GFS or ECMWF. As noted above, while the higher-resolution HRRR model did forecast the circulation, it was displaced too far to the east. In this study we will document the real-time forecasts from the HRRR on 11 September and compare them to the observed flow and reflectivity. We will then compare the forecasts to those from the latest HRRR version, which performed considerably better and would have been of use to forecasters at the Boulder WFO. By contrasting the reruns using the latest configuration of the HRRR with other HRRR reruns for this case that have been made at various times since September 2013, we will be able to speculate on which changes are most important for the observed improved forecasts.