Toward an object‐based assessment of high‐resolution forecasts of long‐lived convective precipitation in the central U.S.

Abstract Forecast models have seen vast improvements in recent years, via both increased resolutions and the ability to assimilate observational data, particularly that which has been affected by clouds and precipitation. The High‐Resolution Rapid Refresh (HRRR) model is an hourly updated, 3 km model designed for forecasting convective precipitation recently deployed for operational use over the U.S. that initializes latent heating profiles as a function of assimilated radar reflectivity. An object‐oriented verification process was developed to validate experimental HRRR convective precipitation forecasts during the 2013 warm season using the NCEP Stage IV multisensor precipitation product. A database of 467 convective precipitation features that were observed during the forecast assimilation period and their corresponding HRRR forecast precipitation features was created. This database was used to evaluate model performance over the entire forecast period, and to relate that performance to model processes, especially those related to precipitation production. Generally, HRRR precipitation is located within 30 km of the observed throughout the forecast period. Validation statistics are best at forecast hour 3, with median biases in mean, maximum, and total rainfall and raining area near 0%. Earlier in the forecast, median biases in the mean and maximum rain rate exceed 30%, with bias values often exceeding 150%. The median bias in areal extent at the beginning of the forecast is near −40%. This low areal bias and POD values <0.6 appear to be related to the model's ability to produce deep convection relative to atmospheric moisture content and concentration of rainfall in convective cores.