Explaining spatial variation and catchment characteristics thresholds in streamflow response

Understanding hydrological processes, considering the interactions among system state, structure, and climate, and the thresholds that regulate catchment behavior, remains a challenge, mainly across scales where multiple processes interact. Here, we move beyond average feature rankings to identify where and when 16 climate and catchment characteristics change their effect on streamflow across 735 Brazilian catchments. Using random forest models together with global and local feature attribution methods, we identify nonlinear thresholds behavior across the flow-duration cure. Model performance supported diagnostic interpretation, with the main streamflow signatures (Q 5 , Q mean , and Q 95 ) showing R 2 and NSE values of 0.63–0.81 and KGE values of 0.70–0.85, while the rarest conditions showed lower but still informative skill. Although these relationships are not causal, provide valuable insights into hydrologic functioning based on the attribute–streamflow relationships identified within the model. Modeled relations indicate that climate exerts the strongest large-scale control: an aridity threshold near 1.23–1.37 marks where drying begins to reduce streamflows, though this effect varies regionally; aridity can enhance streamflow in parts of the Amazon and southern Brazil. Low streamflows were particularly sensitive to precipitation seasonality (SI), which showed a coherent threshold near SI = 0.40, as well as to subsurface porosity, slope, elevation, and vegetation cover. Forest cover showed limited relevance for higher streamflows, but low streamflows decline when forest cover exceeds 25%, and a weaker 32–40% threshold appears for Q mean and higher streamflows. Sand content shows a dual behavior with > 56% boosts low streamflows, whereas values above 45% attenuated high streamflows. Regionally, for the rarest high-streamflow (Q 99 ), soil properties can override climate; under shallow bedrock they become the dominant amplifiers. Overall, at large scales, Brazilian streamflow regimes seem organized by a set of physically interpretable thresholds that distinguish climate-dominated regimes from those governed by catchment state and structure, with variation across space and the flow-duration curve.