© 2020 EDP Sciences. All rights reserved. Multi-year droughts in Mediterranean climates may shift the water balance, that is, the partitioning rule of precipitation across runoff, evapotranspiration, and subsurface storage. Mechanisms causing these shifts remain largely unknown and are not well represented in hydrologic models. Focusing on measurements from the headwaters of California s Feather River, we found that also in these mixed rain snow Mediterranean basins a lower fraction of precipitation was partitioned to runoff during multi-year droughts compared to non-drought years. This shift in the precipitation runoff relationship was larger in the surfacerunoff-dominated than subsurface-flow-dominated headwaters (39% vs. 18% decline of runoff, respectively, for a representative precipitation amount). The predictive skill of the Precipitation Runoff Modeling System (PRMS) hydrologic model in these basins decreased during droughts, with evapotranspiration (ET) being the only water-balance component besides runoff for which the drop in predictive skill during drought vs. non-drought years was statistically significant. In particular, the model underestimated the response time required by ET to adjust to interannual climate variability, which we define as climate elasticity of ET. Differences between simulated and data-driven estimates of ET were well correlated with accompanying data-driven estimates of changes in sub-surface storage (1S, r D 0:78). This correlation points to shifts in precipitation runoff relationships being evidence of a hysteretic response of the water budget to climate elasticity of ET during and after multi-year droughts. This hysteresis is caused by carryover storage offsetting precipitation deficit during the initial drought period, followed by vegetation mortality when storage is depleted and subsequent post-drought vegetation expansion. Our results point to a general improvement in hydrologic predictions across drought and recovery cycles by including the climate elasticity of ET and better accounting for actual subsurface water storage in not only soil, but also deeper regolith that stores water accessible to roots. This can be done by explicitly parametrizing carryover storage and feedback mechanisms capturing vegetation response to atmospheric demand for moisture.