Using 00z NCEP (including 20 ensemble members), let's begin by defining the EOFs describing the greatest amount of variance in the ensemble forecasts:
Alright what is this telling us?
First of all, note the asymmetry in the mean/spread ... there is a maximum in the spread over E PA and NJ ... to the west-northwest of the mean low.
EOF1 -- explains 46.6% of the variance -- describes the depth of the low for the cluster of solutions to the NW of the mean ... in general the average SLP across the domain of the low ... so a positive EOF1 is related to lower SLP.
EOF2 -- explains 21.5% of the variance -- describes the west-east position of the low ... positive EOF2 is west of the mean, negative is east of the mean.
EOF3 -- explains 18.2% of the variance -- describes the north-south position of the low, but also the amplitude of the system ... notice the maximum is colocated with the mean low ... a negative EOF3 is associated with a stronger low off the delmarva peninsula with higher pressures to the north and northwest.
Perhaps as you can tell, a negative EOF3 pattern is actually quite similar to the SLP pattern we might look for in an East Coast winter storm threat. So although it explains less of the variance in the ensemble forecast, I'm personally most interested in this EOF.
So now let's use this forecast diagnostic and find the correlation to a variable in the ensemble forecast, such as 500mb height.
Here we can see that the greatest correlation (sensitivity of the EOF3 to the 500mb height) is associated with the trough moving through the Northeast and Quebec over the next 48 hours. Lower heights over the northwest Atlantic over the next two days correlates with a negative EOF3 ... and a SLP pattern characteristic of Northeast snowstorms.