New Jersey Agricultural Experiment Station

• Main Authors: Robert P. Harnack, Michael W. Cammarata A
• Other Authors: The Pennsylvania State University CiteSeerX Archives
• Format: Text
• Language: English
• Subjects: Pacific; Sion; North Atlantic
• Online Access: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.667.5966; http://nwas.org/digest/papers/1987/Vol12-Issue2-May1987/Pg15-Harnack.pdf

A set of winter temperature forecast experiments were conducted using mllitiple discriminant analysis with a "jackknife " procedllre. Nine separate experiments were condllcted using variolls predictor fields inc/uding North Pacific sea sUI/ace temperatllre (SST). tropical Pacific SST, North Atlantic SST, and Northern Hemisphere 700 mb heights defined for variolls periods (November, October, fall. and slimmer). For most trials the data sample consisted of the years 1950-78. Principal conclllsions were: 1. Overall skill for most winter forecast trials were sta-tistically significant at the 95 % confidence level lIsing the binomial distriblltion. Overall percent correct skill ranged ji-om 59 % (summer heights) to 71 % (No vember All SST)for two-class models and 39 % (slimmer heights) to 54% (November All SST) for three-class models. 2. A direct comparison between discriminant and regres-sion analyses, lIsing November Pacific SST predictors. win-ter sub area temperatllres as predictands and a 1950-1978 data sample. showed that discriminant analysis gave sllpe-rior resllits. Overall. the two-class skill was 67 % versus 59% and the three-class skill was 48 % vel'SIlS 39%. 3. Overall skill decreased only slightly as forecast lead time increased. The skill for October and slimmer SST pre-dictor models (68 % and 67 % respectively for two-class) were as good as operational short lead-time winter fore-casts issued now. 1.