Historical genomes reveal scale-dependent predictability of climate adaptation

Predicting evolution remains a central challenge in biology. Contemporary spatial patterns are increasingly used as space-for-time proxies to forecast evolutionary responses to environmental change, yet the reliability of such predictions--and whether it varies among biological scales--remains unclear. Using historical and contemporary genomes of the invasive weed Ambrosia artemisiifolia--spanning the species' native range, invasions on two continents and nearly two centuries--we show that genomically predicted flowering-time clines remained stable in the native range while introduced populations re-evolved them. Haploblocks--likely structural variants--were likewise temporally stable in the native range, and two showed striking parallel evolution across all three ranges. Climate-associated SNPs with the strongest contemporary clines showed the greatest temporal change, with limited and variable parallelism among introduced ranges. Together, these results suggest adaptive evolution is partly predictable even when individual genomic trajectories remain flexible and contingent, with predictability emerging most clearly at the level of polygenic traits and large structural variants.