Winter snowfall prediction in the United States using multiple discriminant analysis

This study seeks to determine the skill of multiple discriminant analysis for predicting seasonal snowfall. Winter total snowfall amount and frequency of snowfall events are examined for 440 stations in the United States from 1930 to 2006. The independent variables used to create the forecast include ocean-atmosphere teleconnection patterns [such as the Pacific Decadal Oscillation (PDO) and El Niño Southern Oscillation (ENSO)], large-scale atmospheric patterns [such as the Arctic Oscillation (AO), North Atlantic Oscillation (NAO) and Pacific/North American (PNA)], land cover (such as Arctic sea ice extent and Eurasian snow cover extent), and temperature. Based on a jackknife analysis, forecasts are correct 20-80% of the time for categories of 'below normal', 'near normal', and 'above normal'. When broader categories are used of 'normal or below', 'near normal', and 'normal or above' the forecasts are correct as much as 90% of the time at some stations. The Central United States, Ohio River Valley, Great Lakes, and Upper Midwest regions show the highest level of skill. Results not only confirm relationships previously documented between atmospheric phenomena and US snowfall (such as with the PNA, NAO, and ENSO), but also expand our understanding of factors that influence decadal-scale snowfall variation (such as Arctic sea ice extent and Eurasian snow cover extent).