Heavy Rainfall Forecasts from Two Very Different Ensembles

Operational, convection-allowing ensemble guidance at the National Centers for Environmental Prediction (NCEP) consists of the High-Resolution Ensemble Forecast version 2 (HREFv2), an 8-member ensemble of opportunity that leverages both mixed physics and time-lagging, at horizontal grid spacings near 3 km. A multi-agency effort is underway with the goal of achieving performance similar to or better than HREFv2 using a system with a single dynamic core and single set of physical parameterizations, offering advantages such as a more physically based representation of uncertainty and simpler code management. One such system currently in an experimental status at the Earth System Research Laboratory is the High-Resolution Rapid-Refresh Ensemble (HRRRE).

Although a future single-core ensemble system will leverage the FV3 dynamic core, HRRRE provides a real-time, ARW-based proving ground for testing various techniques to improve ensemble performance. In addition to initial condition perturbations that target random error, HRRRE has been using a stochastic parameter perturbation approach to address model error since late spring 2019.

Here we compare forecasts of heavy rainfall from the HREFv2 and HRRRE ensembles over the eastern United States during Jun-Aug 2019. Various metrics are used to evaluate model biases, spatial representation of precipitation, and the reliability and sharpness of probabilistic forecasts. These early results indicate generally useful guidance from both systems, in spite of a number of deficiencies inherent in forecasts of localized heavy rainfall. By several measures, HRRRE achieves performance comparable to that of HREF, suggesting potential for accurate representation of atmospheric uncertainty using approaches other than a multi-core, multi-physics ensemble.