First row shows projected fast durations till the end of the century, estimated from all thirty ensemble members of CESM1 simulations of ice-free season lengths. Horizontal lines are as in Fig. 3, showing recruitment impact thresholds, assuming masses that are 20% lower (light shade), the same (medium-light shade), 20% higher (medium-dark shade), or 40% higher (dark shade) than in WH89-96. For each ensemble member, a threshold is defined to be crossed as the first occasion when three of the next five years exceed a fasting impact threshold, and we consider the mean across all thirty ensemble members to estimate years of first impact on polar bears. Recruitment declines are expected at a threshold crossing if the subpopulation’s fast-initiating body masses fall below the corresponding value. For example, recruitment declines would be expected in 2032 if the population’s G(M0,L0) is 20% or more below the G(M0,L0)(WH89-96) distribution in that year (vertical arrows and timeline of risk in second row). Third row: minimum convex polygons of the G(M0,L0)-distributions for the -20% (light shade), 0% (medium-light shade), +20% (medium-dark shade), and +40% (dark shade) body mass scenarios, showing for which G(M0,L0)-distributions recruitment declines would be expected (thick boundaries) or not (thin boundaries) at each threshold crossing. Risk increases with darker colors, both because higher body conditions are required to sustain increasingly longer fasts (contrast the four panels in third row), and because high body conditions become increasingly unlikely with longer fasts.