Nature 506, 89–92 (2014); doi:10.1038/nature12872

In this Letter, Figs 2 and 3 contained several minor errors, which have now been corrected. In Fig. 2c, we did not include the possible pathway from deciduous and freezing unexposed to evergreen and freezing exposed. This omission slightly alters the relative likelihood of the different pathways out of the evergreen and freezing unexposed state (<2%), but the interpretation is the same. In Fig. 2d, we also note that the arrow leading from large conduits and freezing unexposed to large conduits and freezing exposed and the arrow leading from large conduits and freezing exposed to small conduits and freezing exposed were switched when generating the figure. In general, the scale of the circles (persistence times) and arrows (transition rates) in Figs 2 and 3 were also found to be confusing. We have now corrected Figs 2 and 3 online such that the scale matches a discrete binning of the persistence times and transitions rates for each character state combination. We thank E. Edwards for bringing these issues to our attention. Finally, in Extended Data Table 3, we note an incorrect transition rate was provided for the transition from woody unexposed to woody exposed for the Superrosidae; the transition rate should be 0.01, not 0.001, and this has also now been corrected online.