Combining Radar and NWP to Build a Snow Rate AI Algorithm

Snow rate is an important diagnostic parameter for decision support, as the necessity and effectiveness of mitigation strategies can strongly depend on how quickly snow accumulates. However, it is also generally a parameter that is not diagnosed operationally. This talk will present the initial results of an effort at CIWRO to build an AI-based snow rate algorithm to address this gap.

Creating a tool for snow rate is difficult due to the intrinsic challenges with both the detection and verification of snow rate. The limitations of the truth dataset used for this work, ASOS snow intensity, will be discussed, including the limited ability to map ASOS intensity to radar precipitation variables (reflectivity or Multi-Radar Multi-Sensor QPE). To improve on the use of reflectivity or QPE alone, mosaicked dual-polarization fields from MRMS are added to the radar inputs of the AI. As even these radar fields are also not perfectly discriminative, relevant NWP fields, such as vertical velocity, depth of the dendritic growth zone, and column maximum temperature, were additionally added to the training data to provide environmental parameters to the AI algorithm. The performance of the AI using the radar and environmental data will be demonstrated against a simple reflectivity threshold. additionally, the implications of these results for the future development of a snow rate algorithm will be discussed.