Within the U.S. National Weather Service (NWS), the Weather Prediction Center (WPC, part of NCEP – the National Centers for Environmental Prediction) is responsible for the issuance of deterministic and probabilistic ice accretion guidance twice daily out to three days over the contiguous United States. To generate the guidance, a blend of outputs from U.S. and international forecast models is post-processed using a modified version of the Freezing Rain Accumulation Model (FRAM) (Sanders and Barjenbruch 2016, http://journals.ametsoc.org/doi/abs/10.1175/WAF-D-15-0118.1). WPC forecasters assign weights to each model and then get an initial freezing rain accretion output from the resulting blend using the FRAM. Forecasters can then manually edit these forecast grids before issuing the guidance.
This guidance has historically been difficult to validate due to a lack of reliable ice accretion observations. Starting in 2013, several hundred surface observing stations within the Automated Surface Observing System (ASOS) network began reporting hourly ice accretion amounts as determined by a sensor used to detect freezing precipitation. These sensors consist of a vibrating probe whose frequency has been found to decrease linearly with increasing ice mass (see Ryerson and Ramsay 2007: http://journals.ametsoc.org/doi/pdf/10.1175/JAM2535.1). This provides the first network of routine, objective measurements of ice accretion in the United States.
I spent the summer of 2017 at the WPC and helped develop a verification technique using this new observational dataset. Using hourly surface observations from 562 stations with ice accretion reporting capabilities, we verified WPC’s ice accretion guidance for the winters of 2015-2016 and 2016-2017. By summing hourly accretion observations over each 24-hour forecast period at each station, we obtain an objective measurement against which we can fairly compare the forecasts. In general, ice accretion observations appear spatially and temporally coherent, lending confidence to the use of this dataset for verification purposes. Figure 1 shows an example from a severe freezing rain event over the Southern Plains of the United States in January 2017. The observed 24-hour ice accretion totals tend to be very close to the forecasts for this event (shown in contours for several thresholds). Our analysis showed that, overall, WPC freezing rain guidance performs quite well. However, we identified a low bias for the high ice accretion cases - the WPC tends to under-forecast more extreme amounts. This information will be used by WPC developers to test and improve their techniques for the post-processing of model output which generates the initial field of ice accretion forecasts. Through this improved verification method, we hope to improve WPC’s ice accretion guidance in the upcoming winter seasons.