Distinct dynamical and thermodynamic pathways compound to amplify extreme atmospheric rivers

Atmospheric rivers (ARs) making landfall over western North America are a primary driver of extreme precipitation and flood hazard, yet the upper bounds of their intensity remain poorly constrained by the short observational record. Here, we combine a differentiable global climate model (∼1.4◦ )with high-resolution dynamical downscaling (∼0.11◦ ) to construct physically plausible storylines of unprecedented AR events along the coast of British Columbia. By optimizing minimal perturbations to historical initial conditions, we generate events that maximize integrated vapor transport (IVT) at landfall and exceed the observational record under present-day climate conditions. Pseudo-global warming perturbations under SSP5-8.5 provide a second pathway to unprecedented intensity through end-of-century warming. The two 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 of the IVT budget, and when both are included simultaneously in numerical experiments, the combined effect is nearly additive, with differences generally below 15%. Yet, while relatively small, non-linear effects are always positive and the most extreme physically plausible ARs therefore arise from the compounding of both drivers. However, the two 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 30%. These findings demonstrate that bounding the upper tail of AR hazard requires accounting for both dynamical variability and thermodynamic change.