Distinct dynamical and thermodynamic pathways compound to amplify storylines of extreme atmospheric rivers in British Columbia

Atmospheric rivers (ARs) over western North America drive extreme precipitation and flood hazard, yet their intensity upper bounds remain poorly constrained by the short observational record. We combine a differentiable global climate model (~1.4°) with high-resolution dynamical downscaling (~0.11°) to construct physically plausible storylines of five unprecedented AR events in British Columbia. By optimizing minimal perturbations to historical initial conditions, we generate events that maximize integrated vapor transport (IVT) and exceed the observational record under present-day conditions. Pseudo-global warming perturbations under SSP5-8.5 provide a second pathway through end-of-century warming. Both approaches amplify AR intensity through distinct mechanisms: the optimization primarily modifies the wind field, while the pseudo-global warming signal primarily increases atmospheric moisture. These contributions act on largely independent components, and when applied simultaneously, the combined effect is nearly additive, with differences generally below 15%. Non-linear effects, though small, are always positive, and within these storylines the most extreme physically plausible ARs arise from compounding both drivers. The pathways differ in precipitation efficiency: the dynamical amplification largely preserves the conversion of moisture transport to precipitation, whereas the thermodynamic amplification reduces it by up to 29%. Bounding the upper tail of AR hazard may therefore require accounting for both dynamical variability and thermodynamic change.