Modeling flow persistence in headwater basins using climatic and morphometric data and machine learning

Study Region Southern Brazil, specifically the state of Paraná, characterized by distinct geomorphological units and subtropical to temperate climatic conditions. Study Focus This study investigates the controls of flow persistence (perennial vs. intermittent conditions) in first-order headwater basins using climatic and morphometric variables within a machine-learning framework. Rather than focusing on the hydrogeological origin of springs, the analysis targets the surface expression of flow regimes in headwater channels. A Random Forest (RF) classifier was applied to 164 basins with field-verified flow conditions using 152 predictors derived from ERA5-Land reanalysis and terrain metrics, including the Roughness Concentration Index (RCI). Model performance was evaluated through 1000 iterations and 10-fold cross-validation and optimized through systematic variable reduction to identify a parsimonious and physically interpretable set of predictors. New Hydrological Insights for the Region Results show that local morphometric properties, particularly terrain roughness (RCI), combined with hydroclimatic variables (runoff, precipitation, and vegetation indices), exert stronger control on flow persistence than basin area alone. The RF model achieved 78.2% accuracy (AUC = 0.84), outperforming traditional mapping approaches, while maintaining stable performance with only eight variables. These findings indicate that flow persistence in subtropical headwaters emerges from nonlinear interactions between terrain structure and climatic forcing, providing a scalable framework for hydrological assessment and water-resource management in data-scarce regions.