of high-resolution convection allowing models (CAMs). The Weather Prediction Center
(WPC) is very interested in generating better ensemble visuals and verification of the
quantitative precipitation forecasts (QPF) from these models. Standard verification can
unfairly penalize the “blob-like” QPF features for a small displacement compared to a
missed forecast altogether. This talk will focus on tracking QPF objects through space
and time to quantify their object attributes and utilize them in a multitude of ways.
The High Resolution Rapid Refresh Version 2 (HRRRv2) and the Version 3
(HRRRv3) QPF is tracked for the 2017 and 2018 warm seasons. The HRRRv2 and
HRRRv3 models are initialized every hour and run at 3 km grid spacing out to forecast
hour 18. The HRRR one-hour QPF is verified against one-hour precipitation data from
the Stage IV analysis. Both model and observation QPF objects are tracked over the
Contiguous United States (CONUS) using the Model Evaluation Tools (METv6.0)
Method for Object-based Diagnostic Evaluation Time-Domain (MTD).
The tracker is used both quasi-operationally to create ensemble plots of object
attributes for WPC forecasters and retrospectively to assess biases in object attributes
over a verification period. This talk will focus more on the retrospective verification,
specifically the biases and error of object attributes such as displacement, area, intensity,
track density, object initiation, and dissipation. Before evaluating the entire warm season,
the MTD tracker performance is optimized for a few case studies to ensure that it
properly tracks the objects of interest.
The retrospective verification shows that the HRRRv2 and HRRRv3 has a wet
bias over most of CONUS with a northward displacement of precipitation in the model.
The HRRR predicts too few heavy precipitation objects (mostly in the later forecast
hours), and with the exception of the Central Plains region, overpredicts the size and
intensity of these objects. The largest northward displacement is over the Central and
Southern Plains for the more intense QPF regions. Future work will focus on how object-
oriented biases can be properly subset and displayed operationally to WPC forecasters.
![](images/flag.png)
![](images/blue_ribbon.png)