About WPC's Probabilistic Winter Precipitation Forecast (PWPF) Products
The operational WPC Winter Weather Desk (WWD) creates 24-h forecasts of snowfall and freezing rain accumulations for each of three consecutive 24-h periods (days) extending 72 hours into the future. These products are shared with the NWS Weather Forecast Offices (WFO) in a collaborative process resulting in refinement of the accumulation forecasts. After the 24-h snowfall and freezing rain accumulation forecasts are finalized, the WWD issues its public products: a limited suite of probabilistic winter weather forecasts. These probabilistic forecasts are computed based on the deterministic accumulation forecasts combined with ensemble information (see below). Prior to the 2013-14 season, the probabilistic forecasts were manually edited by the WWD forecaster. For the 2013-14 season and onward, the limited suite of probabilistic forecasts is usually not edited.
The probabilistic forecasts found here on the WPC PWPF page are also based on the deterministic WWD accumulation forecasts and are generated automatically using an ensemble of model forecasts along with the WWD forecasts. The automatic nature of this product generation allows an extensive set of displays of probabilities for snowfall or freezing rain exceeding a number of thresholds and accumulations of snowfall or freezing rain for various percentile levels. The percentile amounts and probabilities for 24-hour intervals are generated at six-hour increments through 72 hours.
The automatic processing also allows the generation of probabilistic winter precipitation forecasts for 48-h intervals based on 48-h accumulations obtained by adding two 24-h accumulations together. The same method used to compute the 24-h probabilistic products is applied to the 48-h intervals ending at 48 through 72 hours after the initial time. As with the 24-hour forecasts, the 48-h forecasts are produced at six-hour intervals. Finally, a single set of probabilistic forecasts are created for the entire 72-hour period.
A multi-model ensemble is utilized to create a distribution of values around the WPC accumulation at each grid point. The typical constituency of this ensemble is as follows:
Ensemble data Processing
To prepare the raw ensemble data for use by PWPF the following steps are taken:
PRISM-based downscaling is applied to the quantitative precipitation forecasts (QPF) from each member. The snow and ice forecasts for each member are computed from the downscaled QPF.
At each grid point, the precipitation type determination for the NCEP models is the dominant type algorithm (Manikin 2005). Precipitation type for non-NCEP models is determined by applying a simple decision tree algorithm using surface temperature, and temperatures on the 925-hPa, 850-hPa, and 700-hPa mandatory isobaric levels.
A unique snow level is also determined for each member based on the height of the 0.5 C Wet Bulb surface. At each grid point, a top down search is conducted to find where the wet bulb temperature first exceeds 0.5 C. It is then found where the snow level intersects the 5-km topography from the Realtime Mesoscale Analysis (RTMA). Grid points above the snow level are flagged as snow in the final precipiation type mask for each member.
Each member's snowfall is based on the QPF, the precipiation type, and a snow-to-liquid ratio (SLR). The SLR from the National Blend of Models (NBM), multiplied by an adjustment factor, is used as the SLR for each ensemble member.
For each member, the freezing rain accumulation is based on the QPF, the precipiation type, and an ice-to-liquid ratio (ILR). The ILR is based on a simplified version of the Freezing Rain Accumulation Model (FRAM; Sanders and Barjenbruch 2016) which uses precipiation rate as the sole predictor.
A binormal (Toth and Szentimrey 1990) probability distribution or density function (PDF), which allows skewness, is utilized for the PWPF. The fitting of the binormal distribution is a method of moments approach. The WPC forecast is the mode of the distribution. The placement of the WPC forecast in the ensemble order statistics determines the skewness of the distribution. The variance of the distribution is matched to the variance of the ensemble. The WPC deterministic forecast is included as an additional member of the ensemble for the computation of the variance. This fit is done at each grid point; so, the probability density function (PDF) varies from grid point to grid point.
The PWPF forecasts provide information in the following formats:
Probabilities of exceeding a threshold show filled contour levels of probability that the 24-hour, 48-hour, or 72-hour accumulation of winter precipitation will equal or exceed the given threshold. As an example, consider the 6-inch threshold for snowfall. If a point of interest falls within the 40% contour on the probability map, then the chance of snowfall exceeding 6 inches is 40% or greater. As the threshold values increase, the probabilities of exceeding them decrease.
Percentile accumulations for 24-, 48-, or 72-hour intervals show filled contours of snowfall or freezing rain amounts for which the probability of observing that amount or less is given by the percentile level. For example, if the 75th percentile map shows six inches of snow at a location, then the probability of getting up to six inches of snow is 75% at that point. Conversely, there is only a 25% probability of snowfall exceeding six inches at the location in this example. Percentile accumulations increase as the percentile level increases. To illustrate this point, take the previous example, but instead of the 75th precentile map consider the 10th percentile map showing two inches of snow at the location. In this case, the probability of getting up to but no more than two inches of snow is just 10%. The probability of getting more than two inches is 90%; so, a significant accumulation of snow is likely.
For more information on creation of the PWPFs and how to navigate the web page, please see this informational video.
Manikin, G. S., 2005: An overview of precipitation type forecasting using NAM and SREF data. Preprints, 21st Conf. on Wea. Analysis & Forecasting / 17th Conf. on Numerical Weather Prediction, Washington, DC, Amer. Meteor. Soc., 8A.6.
Roebber, P. J., M. R. Butt, S. J. Reinke, T. J. Grafenauer, 2007: Real-time forecasting of snowfall using a neural network. Wea. Forecasting, 22, 676-684.
Sanders, K.J. and B.L. Barjenbruch, 2016: Analysis of Ice-to-Liquid Ratios during Freezing Rain and the Development of an Ice Accumulation Model. Wea. Forecasting, 31, 1041-1060.
Toth, Z., and T. Szentimrey, 1990: The binormal distribution: A distribution for representing asymmetrical but normal-like weather elements. J. Climate, 3, 128-136.