Brier skill scores for six statistical methods for 3 precipitation thresholds

BSS=1-BS/CS,     BS=1/nj(fk-ok)2

k=1

n

From Applequist et al. Wea. and Forecasting,2002

Logistic regression scores higher than linear regression

Based on 4 cold seasons

Any ideas why no higher thresholds were used in the study? 

The authors did not think they would have enough cases to be statistically meaningful!

A Junker editorial comment:  The models now change every year.

PC=principal component analysis

2.5 mm

1.25 mm

0.25 mm