Global climate models (GCMs) capture large-scale thermodynamic responses to anthropogenic forcing; however, their coarse spatial resolution limits their utility for impact-relevant hydroclimatic assessment. Regional climate models (RCMs) have been developed to address this limitation through dynamic downscaling, yet their hydroclimatic projections frequently diverge from those of their driving GCMs. Here, contrasting projections for Central Europe under a high-emission scenario are examined: GCMs indicate increasing aridity, whereas RCMs project an intensifying hydrological cycle, with increases in evapotranspiration (ET), precipitation (P), and runoff. An independent Budyko framework linking the aridity index to the evaporative index (EI = ET/P) is used to diagnose these contrasts. Across ensembles, EI remains relatively stable, revealing a robust, state-dependent bimodality in the ET response to vapour pressure deficit (VPD). RCMs occupy the wet branch with increasing ET under rising VPD, whereas GCMs span a broader range of hydroclimatic states, including regimes characterised by declining ET at higher VPD, consistent with an apparent feedforward response whereby enhanced atmospheric demand coincides with stomatal limitation. This bimodality implies that antecedent hydroclimatic model states continue to shape future projections despite standard statistical post-processing, challenging the assumption that higher spatial resolution alone guarantees physically consistent projections for impact and adaptation assessments.

The manuscript is currently under review at Communications Earth & Environment and is therefore provided in restricted form. The analysis code and full reproducibility workflow are available at: https://github.com/MilanFischer/GCM-RCM