Evaluating the Skillfulness of the Hurricane Analysis and Forecast System (HAFS) Forecasts for Tropical Cyclone Precipitation using an Object-Based Methodology

Tropical cyclones (TCs) are destructive, natural occurring phenomena that can cause the loss of lives, extensive structural damage, and negative economic impacts. A major hazard associated with these tropical systems is rainfall, which can result in flood conditions, contributing to the death and destruction. The role rainfall plays in the severity of the TC aftermath emphasizes the importance for models to produce reliable precipitation forecasts. Hurricane model precipitation forecasts can be improved through precipitation verification as the model weaknesses are identified. In this study, the Hurricane Analysis and Forecast System (HAFS), an experimental NOAA hurricane model, is evaluated for its skillfulness in forecasting TC precipitation. An object-based verification method is used as it is demonstrated to more accurately represent the model skill compared to traditional point-based verification methods. A 600 km search radius is implemented to capture the TC rainfall and the objects are defined by 2, 5, and 10 mm/hr rain rate thresholds. The 2 mm/hr threshold is chosen to predominantly represent stratiform precipitation, and the 5 and 10 mm/hr thresholds are used as approximate thresholds between stratiform and convective precipitation. Shape metrics such as area, closure, dispersion, and fragmentation, are calculated for the forecast and observed objects and compared using a Mann Whitney U test. The evaluation showed that model precipitation characteristics were consistent with storms that are too intense due to forecast precipitation being too central and enclosed around the TC center at the 2 mm/hr threshold, and too cohesive at the 10 mm/hr threshold. Changes in the model skill with lead time were also investigated. The model spin-up negatively impacted the model skill up to six hours at the 2 mm/hr threshold and up to three hours at the 5 mm/hr threshold, and the skill was not affected by the spin-up at the 10 mm/hr threshold. This indicates that the model took longer to realistically depict stratiform precipitation compared to convective precipitation. The model skill also worsened after 48 hours at the 2 and 10 mm/hr thresholds when the precipitation tended to be too cohesive. Future work will apply the object-based verification method to evaluate the TC precipitation forecasts of the Basin-Scale Hurricane Weather Research and Forecasting (HWRF-B) model.