Using Machine Learned Road Weather Observations to Verify A Gridded Snow Rate Product

How quickly snow accumulates has substantial impacts on aviation, impacting deicing
operations, runway clearing, and more. Departments of Transportation request real-time snow
rates to target snow clearing operations on the most impacted roadways. However, creating
hourly or subhourly snow rate fields is a difficult task. Snow accumulation is typically observed
with a daily cadence, making verification of an hourly or subhourly product difficult, if not
impossible. This project explores the utility of road impacts in evaluating a snow rate dataset in
the absence of direct observations of snow rate.

Here, observations of road conditions are used to investigate the decision support potential of
derived snow rates. These road condition observations are created by a machine learning
algorithm developed at the University of Albany. This ML algorithm classifies the road as dry,
wet, snow-covered, or severely snow-covered based on road camera imagery. These
classifications are compared to derived snow rates at the time of the observation from a
prototype Multi-Radar Multi Sensor (MRMS) gridded snow rate product. For this project, hourly
snow rates are calculated using MRMS hourly Quantitative Precipitation Estimates (QPEs) and
a variety of Snow-to-Liquid Ratios (SLRs). In addition to the snow rate at the time of the
observation, the impacts of the prior hours’ snow rates and the probability of subfreezing road
surfaces will be presented.