Quantification of long-term partitioning of precipitation into evaporation and runoff is a fundamental pursuit in catchment hydrology. The Budyko framework provides a theoretical basis for this and estimates the evaporative fraction based on the aridity index. However, deviations from the global-average Budyko curve point to additional controls on precipitation partitioning beyond the aridity index. We hypothesized that root zone storage capacity (Sr,max), defined as maximum subsurface water volume accessible to vegetation roots, is a key driver of these deviations. The relationship between Sr,max and precipitation partitioning in the Budyko space was investigated globally across >5000 catchments. Sr,max was calculated using the memory method based on runoff observations and the water balance. The ω-parameter from Fu’s equation, which was used here to construct parametric Budyko curves, reflects deviations from the global-average Budyko curve and hence precipitation partitioning. Results revealed a globally stronger correlation (Spearman’s ρ= 0.68) of ω with Sr,max, than with other potential controls, indicating Sr,max as a dominant driver of precipitation partitioning. Further analysis based on Köppen–Geiger climatic zone classification revealed variations in the Sr,max–ω relationship, with the strongest correlations observed in cold (ρ= 0.87) and Mediterranean (ρ = 0.83) climates, followed by temperate (ρ = 0.76), tropical (ρ = 0.64) and arid climates (ρ = 0.61). Regional differences in Sr,max indicate that, at a given aridity, EA/P largely reflects vegetation adaptation to the seasonal interplay between water supply and atmospheric water demand. This study provides strong empirical evidence on a global scale for Sr,max as a governing factor in modulating catchment precipitation partitioning, as evident in the Budyko space. As a major implication our results provide a theoretical basis for the maximum values of Sr,max found in nature, as constrained by the water and energy limits of the Budyko framework.

Quantification of precipitation partitioning into evaporation and runoff is crucial for predicting future water availability. Within the widely used Budyko framework, which relates the long-term aridity index to the long-term evaporative index, curvilinear relationships between these indices (i.e. parametric Budyko curves) allow for the quantification of precipitation partitioning under prevailing climatic conditions. A common assumption is that movement along a specific Budyko curve with changes in the aridity index over time can be used as a predictor for catchment responses to changing climatic conditions. However, various studies have reported deviations around these curves, which raises questions about the usefulness of the method for future predictions. To investigate whether parametric Budyko curves still have predictive power, we quantified the global, regional, and local evolution of deviations of catchments from their parametric Budyko curves over multiple subsequent 20-year periods throughout the last century based on historical long-term water balance data from over 2000 river catchments worldwide. This process resulted in up to four 20-year distributions of annual deviations from the long-term mean parametric curve for each catchment. To use these distributions of deviations to predict future deviations, the temporal stability of these four distributions of deviations was evaluated between subsequent periods of time. On average, it was found that the majority (62 %) of study catchments did not significantly deviate from their expected parametric Budyko curves. Out of the remaining 38 % of catchments that deviated from their expected curves, the long-term magnitude of median deviations remains minor, with 70 % of catchments falling within the range of ±0.025 of the expected evaporative index. When these median deviations were expressed as relative changes in discharge, catchments in arid regions showed higher susceptibility to larger discharge shifts compared to those in humid regions. Furthermore, a significant majority of catchments, constituting around the same percentage, was found to have stable distributions of deviations across multiple time periods, making them well suited to statistically predict future deviations with high predictive power. These findings suggest that while trajectories of change in catchments do not strictly follow the expected long-term mean parametric Budyko curves, the deviations are minor and quantifiable. Consequently, taking into account these deviations, the parametric formulations of the Budyko framework remain a valuable tool for predicting future evaporation and runoff under changing climatic conditions within quantifiable margins of error.