Seasonal Forecast Skill of Arctic Sea Ice in Two Versions of a Dynamical
Forecasting System and Comparisons with Potential Predictability
Estimates
Abstract
In this study, we assess pan-Arctic and regional seasonal sea ice
forecast skill in versions 1 and 2 of the Canadian Seasonal to
Inter-annual Prediction System (CanSIPSv1 and CanSIPSv2) dynamical
seasonal prediction systems. Each version applies a multi-model ensemble
approach using two coupled general circulation models. CanSIPSv2
features a new model formulation (where one of the underlying models,
CanCM3, was replaced with GEM-NEMO) and improved sea ice initialization.
We show that the modifications made in the development of CanSIPSv2
substantially enhance forecast skill. For example, the lead time for
skillful forecasts of detrended pan-Arctic September sea ice area
increases from three months in CanSIPSv1 to seven months in CanSIPSv2.
We also show that forecasts of detrended winter sea ice area are
improved, with CanSIPSv2 producing skillful forecasts for all considered
lead times (up to 11 months) for December, January, and February. We
find that improvements in pan-Arctic forecast skill are due primarily to
improved initialization methods.Further, a potential predictability
experiment is conducted for one of the two CANSIPSv2 models, CanCM4, in
order to establish – in conjunction with similar studies – the
potential to further increase forecast skill with improved models,
observations and initialization methods.