says
"These data are a compilation of data from many sources integrated into a single gridded product by John Walsh and Bill Chapman, University of Illinois. The sources of data for each grid cell have changed over the years from infrequent land/sea observations, to observationally derived charts, to satellite data for the most recent decades. Temporal and spatial gaps within observed data are filled with a climatology or other statistically derived data."
Your reference says:
"The missing data have been reconstructed using statistical (regression-like) models relating atmospheric processes (SLP gradients and SAT) to ice extent (Kovalev and Nikolaev 1976; Yulin 1990)."
The choice is between hockey stick ice from a temperature-based model of theoretical ice extent, or non-hockey-stick real world measurements of the actual ice extent.
Seems like both groups had to resort to using models to fill data gaps.
By the way, what are the units on those figures?
The paper explaining the chart is in a PDF link on this page: http://www.frontier.iarc.uaf.edu:8080/~igor/research/ice/index.php and it says:
ice-extent anomalies ( 1000 km 2 )
Yes, both used models, but the Univ of Alaska folks just filled in a few missing years. The Univ of Illinois doesn't have any explanation of measurements before 1953 except the cryptic reference to the Walsh models.