Reconstructing Sea Ice Conditions in the Arctic and Sub-Arctic Prior to Human Observations - Arctic Sea Ice Decline: Observations, Projections, Mechanisms, and Implications - page 29

Figure 1. Map showing the location of surface sediment samples used to develop the reference dinocyst database. The
dark gray squares correspond to Arctic sites, north of 66°N (see dashed line of the circle). The triangles and circles cor-
respond to Atlantic and Pacific sites, respectively. The database includes assemblages for 584 sites with seasonal sea ice,
about half of them corresponding to more than 6 months per year of sea ice. No sites under perennial sea ice are included
inasmuch as the assemblages are barren in such conditions because of extremely low productivity. Isobaths correspond
to 200, 1000, and 2000 m.

of the sediment may represent fluxes during a time range
varying from about 10 to 1000 years depending upon bio-
logical mixing and accumulation rates. On the other side, the
instrumental observations used as the “modern” reference
only span the last few decades, at the most. There is there-
fore a discrepancy between the time intervals represented
by the proxy, here the dinocyst assemblages, and the refer-
ence hydrographical values. Thus, when developing transfer
functions based on calibration or similarity techniques, one
uses an assumption concerning the correspondence between
surface sediment samples and the modern hydrography that
is potentially biased. In areas of relatively stable conditions
over the last hundreds of years, this is not critical. However,
in areas marked by a large interannual or interdecadal varia-
bility, this might be a problem and would yield noisy records.
Moreover, in areas characterized by long-term trends and
low sedimentation rates, we may expect some discrepancy

since the surface sediment sample would represent an older
interval with a different state or mean than the one defined
from the “modern” hydrographical values.
Despite fundamental limitations inherent to transfer func-
tions as exposed above, the accuracy of the MAT applied to
dinocyst assemblages permits reasonably adequate estimates
of sea ice cover. The accuracy of the sea ice reconstructions
has been evaluated from the leaving-one-out approach by
estimating the modern sea ice cover from dinocyst assem-
blages (Figure 3). The coefficient of correlation (r 2 ) between
observed and estimated values is 0.914, and the error of pre-
diction establishes at ±1.1 months/year of sea ice on aver-
age (Figures 3a and 3b). Such accuracy seems appropriate
given the interannual variability that characterizes the sea
ice cover extent in the domain of seasonal sea ice. At the
584 sites of the database characterized by sea ice occurrence,
the standard deviation (±1σ) around the 1954-2000 means