Author(s): Emanuel Escobar; Alexandre Fabregat; Josep A. Ferre; Anton Vernet

Linked Author(s):

Keywords: Basin biophysical features; Machine learning; Precipitation; Streamflow

Abstract: The ability of a basin to convert rainfall into streamflow depends on interacting meteorological and biophysical factors including, for instance, soil, terrain, vegetation and land cover, that control infiltration, evapotranspiration, and groundwater recharge. Traditional hydrological models such as SWAT or HBV can reproduce basin responses, but their reliance on calibration, coarse aggregation of heterogeneous properties, and parameter compensation often obscures the causal influence of individual predictors. Large-Sample Hydrology (LSH) initiatives like CARAVAN have advanced comparative hydrology by providing standardized attributes, yet they are constrained by either limited catchment–stream consistency or relatively large basin scales. This study introduces a calibration-free, data-driven framework that leverages the NHDPlus database to isolate the role of biophysical attributes in rainfall–runoff conversion. NHDPlus provides fine-resolution, hydrologically consistent catchments (COMIDs), each uniquely associated with a local drainage unit, making it well suited for basin-scale water balance analysis. Using daily precipitation and streamflow records from up to 200 U.S. headwater catchments (2019–2023), combined with static attributes such as topography, soils, and land cover, we exploit cross-basin variability to quantify how individual predictors affect hydrological response. The approach provides robust, generalizable insights into how land use change, forestry, and climate warming influence water availability.

DOI:

Year: 2026