Abstract:
This study uses linear inverse modeling (LIM) to examine the predictability of North
Atlantic sea surface temperatures (SSTs) in both observations and simulations of an unforced
coupled climate model. For observations, historical North Atlantic SSTs from 1919
to 2018, we assess how predictability is impacted by the way in which the climate change
signal is removed from the basin across three different observed SST datasets. In the climate
model simulations, which possess no externally forced signal, we examine the influence
of increasing the ocean resolution to an eddy resolving scale (0.1 degrees Lon/Lat) on the
predictability of North Atlantic SSTs.
Analysis of the observations ndfis that 1) SSTs in the Irminger Sea and Iceland Basin
have the highest predictability within the North Atlantic and 2) basin averaged SSTs are
predictable at a time scale of 3-5 years. The modes calculated based on the LIM supports
these predictability results: all of the least damped modes, regardless of detrending method
or dataset, indicate memory storage around the Irminger Sea and Iceland Basin with a
decay timescale of 4.5 to 5 years.
Neither the 3-5 year range of basin-wide predictability nor the enhanced Irminger Sea
and Iceland Basin predictability are present in the SSTs of the eddy-resolving climate model.
Examining the LIM eigenmodes for this dataset shows that the low Irminger Sea and Iceland
Basin predictbaility is primarily the result of variable subsurface heat storage and a shallow
mixed layer in these regions. It is worth noting that climate model's external forcing (
i.e. greenhouse gas and aerosol concentrations) is fixed and thus one possible caveat in
comparing the results of the simulations to those of observations is that neither detrending
method has completely removed the forced signal from observed SSTs. That said, the
results of our study, regardless of detrending method, show the Irminger Sea and Iceland
Basin to be highly predictable regions within the observed North Atlantic and the fact that
this is not the case in the simulation is likely due to model inadequacies.
