Start
July 16, 2024 - 9:30 am
End
July 16, 2024 - 10:00 am
Categories
School of Meteorology (Defense)Thesis Defense
Identification and Verification of Mesoscale Convective Systems in the Models of the High-Resolution Ensemble Forecast System
Jeffrey Milne
Tuesday, July 16th , 2024
NWC 5930 / 9:30 am
Abstract: To better understand the characteristics of storms produced by forecast models and those seen in observations, contiguous and nearly contiguous radar objects with reflectivity above a convective threshold were identified and their shapes analyzed. The area, aspect ratio, orientation angle and maximum reflectivity within the object were calculated. For all models of the High Resolution Ensemble Forecast system, the modeled and observed storm objects had similar distributions in area-aspect ratio space. Modeled storm objects had a preferred band of maximum reflectivity that was not seen in the observed storm objects. The modeled storm objects also had a more north-south orientation than observed. Despite these differences, the shape characteristics of the modeled storm objects were still close enough to modeled storm objects to proceed with the creation of an MCS identification and tracking algorithm. The identification algorithm was based on one developed for observed storms
. It was tuned to match expert labeling. A novel tracking algorithm was developed based on continuous swaths of maximum reflectivity. Together, the identification and tracking algorithm produced a spatial distribution that was somewhat consistent with an observed climatology. Looking further into verification of the identification and tracking algorithm revealed an underforecast from all of the models. At both the diurnal and annual scale, the models generally did not match the shape of the observed distribution. Matching the forecast MCSs with observed MCSs further showed the low bias, though the forecasts did have very low FAR. The tuned parameters did not necessarily improve the forecast compared to the default parameters. The positional error between the modeled and observed MCSs showed that when the model produced an MCS, it was usually close to an observed MCS and oriented the same direction as the observed MCS.