Fig 1 Mean March ice edge, 1979-2002
This shows the mean ice concentration for March in Davis Strait and the Labrador Sea using the average of 24 Marches (1979-2002) of passive microwave ice concentration data. Greenland is at upper right, Newfoundland at lower left. The ice edge is defined as the 30-39% ice concentration contour. The white pixels have 30-39% ice concentration, and the black dots defining the ice edge are spaced at 25 km. Initially I selected the ice edge points manually, then smoothed the resulting curve, then resampled along the smooth curve at 25 km increments to get the ice edge plotted in this figure.

Fig 2 March 1979 ice edge
This shows the ice concentration for March 1979, along with the ice edge (black dots on white pixels), and the 24-year mean March ice edge (heavy black line).

Fig 3 Measuring the deviation from the mean
The black dots are the mean March ice edge, with 25 km spacing. The red dots are the March 1979 ice edge. They were selected manually from the March 1979 ice concentration image. The deviation from the mean March ice edge is computed following the method of Shapiro et al. [2003]. For each 25-km interval along the mean ice edge, a perpendicular line segment is constructed. The intersection of this line segment with the March 1979 ice edge is found. The length of the segment, from the mean ice edge to the March 1979 ice edge, is the deviation. The deviation is a function of the arc length along the mean ice edge. The deviation is negative when it represents a retreat of the ice edge relative to the mean. The deviation is positive when it represents an advance of the ice edge relative to the mean. Three negative deviations (blue line segments) are shown for illustration. The arc length of the mean ice edge is measured from Greenland (0 km) to Newfoundland (~2500 km).

Fig 4 Deviation for 1979
This shows the deviation of the March 1979 ice edge as a function of the arc length along the mean ice edge.

Fig 5 All deviations, 1979-2002
This shows the deviations of all 24 March ice edges as a function of the arc length along the mean ice edge. The large positive deviations at a distance of 500-1000 km correspond to the years when the ice advanced farther than average down the west coast of Greenland. The large positive spike at 2300 km in some years is a lobe that extends east-southeast from Newfoundland. In other years the ice doesn't even reach Newfoundland. Notice that in the middle range of distance (1200-1800 km) there is little variability. This is the north-south portion of the ice edge off the coast of Labrador.

References
Shapiro, I., R. Colony, and T. Vinje, 2003: April sea ice extent in the Barents Sea, 1850-2001, Polar Research 22(1), 5-10.

In the above figures and captions, the perpendicular distance from a given ice edge to the mean ice edge is called the "deviation". The paper by Shapiro et al. [2003] refers to this quantity as the "anomaly". I think it would be best to use their terminology and change all of the above "deviations" to "anomalies", in order to maintain consistency and to avoid confusion with the term "standard deviation".

Here are some further details about the definition of the mean ice edge (see Fig 1 above).

• I chose the 30-39% ice concentration contour as the ice edge in order to conform with the definition used by Shapiro et al. [2003]. They say that "The 30% ice coverage has historical relevance: it is the maximum ice cover in which sailing vessels can easily manoeuvre".
• I manually selected 16 points along the 30-39% contour of the mean March ice concentration. I constructed a cubic spline interpolant to pass through these points. Then I sampled the spline at 25-km increments. The resulting mean ice edge consists of 99 points, or 98 segments.
• The anomaly is measured perpendicular to the mean ice edge. I construct the perpendicular bisector of each segment along the mean ice edge and find the distance to its intersection with the given ice edge. Thus the anomaly for each March ice edge consists of 98 values. In case the bisector does not intersect the given ice edge anywhere, a missing value is recorded. Out of 98 x 24 = 2352 potential values, there are 2227 actual anomaly values (94.7%) and 125 missing values (5.3%). The missing values tend to lie at the ends of the ice edge, near land.
• Shapiro et al. [2003] emphasize the importance of a smooth mean ice edge, and that the mean anomaly should be zero. The cubic spline interpolant gives the smoothest ice edge that still passes through all 16 manually selected points. The mean anomaly over all 2227 anomaly values is just 0.074 kilometers, quite close to zero. The standard deviation is 115 km, and the extremes are -418 km and +556 km. The standard deviation is comparable to that found by Shapiro et al. [2003] in the Barents Sea.

Principal Component Analysis
The anomalies can be put into a matrix, A, with 98 columns (space dimension) and 24 rows (time dimension). I form the covariance matrix A^TA (I'm leaving out some details here) and find the eigenvalues (variances) and eigenvectors (characteristic anomaly patterns). The largest eigenvalue accounts for 60% of the total variance, and the second largest eigenvalue accounts for 19%.
Fig 6 shows the standard deviation (over 24 Marches) of the anomaly as a function of the distance along the mean ice edge (black curve). The anomaly is measured in kilometers. The figure also shows the first and second eigenvectors. Each eigenvector or principal component has an associated time series that gives its evolving contribution to the anomaly pattern.
Fig 7 shows the correlation of each principal component time series with the winter (Dec-Mar) North Atlantic Oscillation (NAO) index [Hurrell, 1995]. The correlation of the first time series (0.57) is significantly greater than zero with more than 99% confidence according to a one-tailed test of Student's t distribution. This indicates that a large part of the March ice edge anomaly can be explained by the current winter's NAO pattern. A strong NAO brings colder temperatures and stronger northerly winds to Baffin Bay / Davis Strait, pushing the ice edge further south. I also looked at the correlation of the previous winter's NAO index with the first principal component time series - it is 0.43 - not so impressive. It seems clear that the current winter's NAO should affect the March ice edge, and indeed it does. I still don't understand the strong correlation we found previously between the March ice concentration and the one-year lagged winter NAO, but others have noticed it too. Partington et al. [2003] says, "the winter sub-polar seas respond clearly to the NAO after a lag of 1 year".

References
Hurrell, J. W., 1995: Decadal trends in the North Atlantic Oscillation: regional temperatures and precipitation, Science, 269, 676-679.
Partington, K., T. Flynn, D. Lamb, C. Bertoia, and K. Dedrick, 2003: Late twentieth century Northern Hemisphere sea-ice record from U.S. National Ice Center ice charts, J. Geophys. Res., 108, C11, 3343, doi:10.1029/2002JC001623.