All fasting durations were calculated as 24 days shorter than the summer period with sea ice extent <30%. Red lines are using PMW observations. Black and grey lines use the ensemble means of the CESM1 simulations with the RCP8.5 and RCP4.5 scenarios, respectively, with both having the same historical integrations through year 2005 (the RCP4.5 scenario was only run to 2080 due to computational limitations). The range of the ensemble of simulations is shown by shading plus and minus one standard deviation about the ensemble means. Numbers show the mean per-decade increase in the number of fasting days from 1979-2016, from PMW (red) and from the full range of CESM1 ensemble members (grey: RCP4.5; black: RCP8.5). Ice-free season lengths from CESM1 nearly always agree with or slightly underestimate the ice-free season lengths observed from PMW, again rendering our forecasts optimistic. Green background: SIE subpopulations; red background: DIE subpopulations; blue background: CIE subpopulations.