Probabilistic Snowfall Forecasts

Product/Service Description Document (PDD)


TheWeather Forecast Office (WFO) in Caribou, Maine (CAR), will be producing a experimental probabilistic snowfall forecasts whenever a winter storm warning or winter weather advisory for snow is issued.  The methodology is based on the WFO Tulsa, Oklahoma (TSA) probabilistic QPF forecast process, but modified slightly as outlined below.  The primary modification is that our method assumes that where PoP for QPF is the probability of getting 0.01’ (one-hundredth of an inch) for snowfall we define PoP as the probability of 0.1”  (one-tenth of an inch).



Part I - Mission Connection

NOAA’s Strategic Plan states that we are to “Improve the reliability, lead-time, and effectiveness of weather and water information and services that predict changes in environmental conditions.” To support that effort, The National Weather Service Strategic Plan for 2005 – 2010: Working Together to Save Lives, establishes a mission goal to “develop new capabilities and move into a new direction of forecasts... and expand beyond traditional weather and water products....”

( )

In response to the NOAA and NWS goals, WFO CAR has modified a method developed by WFO TSA that provides probabilistic quantitative storm total snowfall forecasts on an as needed basis, in the form of probability of exceedance (POE) forecasts. These POEs will provide our clients and customers with more probablistic snowfall forecast during high impact winter weather situations, so they can use this informatoin in their decision-making processes.


a. Product Description

Forecasts for the probability to exceed particular snowfalll amount are being produced from other weather parameters already produced through the Gridded Forecast Editor (GFE) developed by the Global Systems Division (formerly Forecast System Laboratory, Boulder, CO). Specifically, the probability of exceeding a certain snowfall amount are calculated from data obtained from the 12hr PoP (Probability of Precipitation) and the Storm Total Snowfall gridded forecast. This is done by using the exponential distribution density function outlined below. The probabilities to exceed 0.1, 0.5, 1, 2, 3, 4, 5, 7, 10,12,15, 20 inches are provided for forecast zone . These forecasts are provided as a table (see Figure below).

b. Purpose and Intended Use

The purpose of the POE forecast is to provide customers and partners with probabilities of reaching a certain snowfall amount. By providing these probabities for 12 different snowfall amounts, the customer can get a better understanding of reaching various amounts of snowfall that are important to them. It has been shown (Amburn, 2006) that forecasters have sufficient knowledge to allow them to inform users of their probability to exceed certain meteoroloigcal threshold value. Over time, it is expected that customers will be able to use the POEs to help mitigate economic losses, maximize economic gains, and help protect life and property.

c. Audience

The audience for the POE forecasts includes NWS internal users, federal, state, and local emergency management agencies, mitigation and risk management specialists in all sectors of society, the media, and all others in need of information on the probability to exceed certain snowfall amounts. This product will be available to all who have an Internet connection.

d. Presentation Format

The POE forecasts are available on the Internet and will consist of locally produced tables. Each POE table displays the probability to exceed a threshold snowfall amount of 0.1, 0.5, 1, 2, 3, 4, 5, 7, 10,12,15, 20 inches. These probabilies are provided on a Forecast Zone basis. Although, there are plans to move this fuctionality over to a "Point and Click" methodology.


e. Feedback Method

Comments regarding the POE forecasts should be sent to the feedback email address on the Web page containing the product, and to the generic NWS Headquarters Customer Survey link:


Part II – Technical Description

a. Format and Science Basis

As noted above, WFO CAR, will be producing a experimental probabilistic storm total snowfall forecasts whenever a winter storm warning or winter weather advisory for snow is issued anywhere in the Caribou areas of responsibility.  The methodology is based on the WFO TSA probabilistic QPF forecast process.  The only modification is that our method is that PoP is assumed as a proxy for the probability receiving 0.1”  (one-tenth of an inch) of snowfall in a 12 hour period. The WFO TSA method assumed PoP is the normal NWS definition.

It has been established that the gamma distribution, specifically a special case of the gamma distribution called the exponential distribution, provides a reasonable approximation for the distribution of rainfall amounts (Wilks, 1995). An ESSA Technical Memorandum (Jorgensen and Klein, 1969, hereafter J&K69), based on 15 years of data, showed similar results. Finally, a recent study (Amburn, 2006), based on 10 years of data in eastern Oklahoma and northwest Arkansas again confirmed that the distribution of rainfall amounts for a given event were well represented by the exponential distribution. A quick review of snowfall amounts in the Caribou CWA further reveal that snowfall also follows a similar gamma distribution.

The Amburn study used the exponential distribution probability density function (PDF) to approximate the probability to exceed (POE) any selected rainfall amount (x). The unconditional POE(x) for any point and event is given by:

uPOE(x) = PoP * e (-x/μ) (1

PoP is the National Weather Service probability of precipitation, x is the selected exceedance value, and μ is the National Weather Service quantitative precipitation forecast (QPF). When the QPF is substituted for μ, it effectively changes the shape of the exponential PDF, therefore raising or lowering the probability to exceed selected rainfall amounts.


When calculating the probability of storm total snowfall, this equation is used with sum adjustment. For example, μ is defined as the storm total snowfall amount, x is the snowfall exceedance amount. PoP is the same as in the Amburn study equation, except we assume it means something slightly different (as stated above).

b. Product Availability

The Storm Total Snowfall probability forecasts will be available on the WFO CAR Internet site whenever a winter storm warning or winter weather advisory is issued for snow. Updates will be available anytime there are changes in the PoP or snowfall forecasts as long as the warning/advisory is in effect.

c. Additional Information

1. The product was created by Todd Lericos (SOO) at WFO Caribou, Maine This forecast will automatically be produced GFE in AWIPS when a WSW with the proper VTEC headings are detected.

2. No special software is necessary to display the forecasts. They are available via the Internet as both graphic and alphanumeric products.

Part III - References:

Amburn, Steven A., and J.M. Frederick, 2006: An Experiment in Probabilistic Quantitative Precipitation Forecasting, , 18th Conference on Probability and Statistics, 86th Annual Meeting of the American Meteorological Society, Atlanta, GA, January 2006.

Jorgensen, D.L. and W.H. Klein, 1969: Conditional Probabilities of Precipitation Amounts in the Conterminous United States. ESSA Tech. Memo, WBTM TDL 18, U.S. Dept. Commerce, 89 pp. [NTIS PB 183144]

Wilks, Daniel S., 1995: Statistical Methods in Atmospheric Sciences, Academic Press, San Diego, CA, 467 pp.