As for c but for calibrated simulations based on the emergent constraint shown in b and the RPS, with shading showing the 5–95% confidence interval diagnosed from the r.m.s.e. ( Methods ). All data are for boreal winter (October–March) 31-year rolling means. Significance is based on a two-sided t test in b and block bootstrapping ( Methods ) in the other panels. For the historical period, only ensemble members for historical + SSP2-4.5 (Table 1 ) are shown leading to discontinuities at the start of future projections for the other scenarios.
