Narrow regions of intense snowfall, also known as snowbands, present hazardous travel conditions due to rapid onset, high precipitation rates, and limited visibility. A promising approach to operational prediction of snowbands is the application of convection-allowing models such as the High-Resolution Rapid Refresh (HRRR). However, to the best of our knowledge, no attempt has been made to verify the skill of such high-resolution forecasts. We have adapted an objective snowband definition to allow for algorithmic detection of bands in both model-simulated and observed base radar reflectivity fields. Comparison between these fields will then enable us to assess band predictability by the HRRR.
Our automated identification method is based upon a thresholding procedure combined with Real-Time Mesoscale Analysis (RTMA) temperature data. Similar studies have utilized a set reflectivity threshold, such as 20, 25, or 30 dBZ, to identify precipitation features. In contrast, we define a snowband as a reflectivity region some level above the background reflectivity. Calibration with a manual classification yielded a “best” definition of 1.25 standard deviations above the background reflectivity. This distribution-based threshold allows for improved detection of bands of varying intensities and reduces biases associated with differences between model-simulated and observed reflectivity fields.
419 hourly reflectivity images containing bands were identified in the 2015-16, 16-17, and 17-18 winter seasons. The POD and FAR by the HRRR are approximately 35% and 65%, respectively. However, allowing for small timing discrepancies drastically improves performance, with a POD and FAR of approximately 65% and 40%, respectively. In order to perform an object-oriented verification, we plan to apply the Method for Object-based Diagnostic Evaluation (MODE) to directly compare the salient features of forecasted and observed bands such as location, shape, and orientation.