Are snow accumulation forecasts generally overdone?

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 19 January 2010: 9:00 AM
B217 (GWCC)
Bruce Rose, The Weather Channel, Atlanta, GA; and J. Koval and E. Floehr

Presentation PDF (1.8 MB)

Amongst weather forecasters in the public and commercial domains, there is anecdotal evidence that short-term weather predictions (6-48 hours ahead) of snowfall accumulation are generally overdone or overpredicted in amount and incidence. Moreover, it is thought these biased predictions are evident across the entire winter season, and occur irrespective of snowfall intensity or amount, geographic region, or forecast provider. Yet there is little or no published documentation of this phenomenon to be found in the literature. While there may be some precautionary or preparedness value in providing worse-case scenarios in winter storm prediction, a persistent or large bias of day to day snowfall prediction would seem less than optimal in serving typical consumers of weather information. Consistently biased or dire snowfall predictions can lead to the same negative consequences associated with other forecasts that exhibit high false-alarms and/or low confirmation; thus, the perceived problem is one worthy of closer study.

Therefore, we propose to compare snowfall forecasts for first-order stations (those providing at least once-per-day manual snowfall amount measurements) across the Conterminous United States and Alaska (about 200 stations) against observations for the upcoming winter season, here defined as November 1, 2009 through March 31, 2010. Each morning during this timeframe we will extract consecutive 6-hour duration snowfall and melted quantitative precipitation (QPF) forecasts through the upcoming 48 hours, from The Weather Channel's Global Forecast Center digital forecast feed (TWC), the National Weather Service's National Digital Forecast Database (NDFD), and the National Center for Environmental Prediction (NCEP) operational NAM/WRF forecasting model. These three realtime forecasts will be published onto a publically-available data service such as Google base™ and the individual observations or verifications will later be amended to that same published record - once ground truth can be collected, quality-controlled and compared to the predictions.

This seasonal study should present an interesting comparison of basic skill in snowfall (and QPF) prediction amongst the three forecasts, as well as lending objective insight to support or contradict the assertion that broadly available weather forecasts in the public domain (TWC, NAM/WRF and NDFD) tend to overpredict snowfall. It will also be instructive to know if NAM/WRF output contains a positive snowfall or QPF bias that might influence the human-prepared forecasts - which rely partially on guidance supplied by that same model.

Because of the timing of this contemporary work, only partial results can be presented. However, the complete methodology of the season-long study, along with details of the on-line aspects of the data management will be fully described.

Supplementary URL: http://snow.forecastwatch.com