Abstract
Estimating low flows is essential for water-resources management, especially in semi-arid regions with sparse hydrometric monitoring and high drought vulnerability. This study advances low-flow regionalization in a Brazilian semi-arid basin by integrating spatial climate data, precipitation and actual evapotranspiration from TerraClimate, into regional regression models. Of fourteen compiled gauges, ten met data-quality criteria and were used for calibration and leave-one-station-out cross-validation (LOSO-CV). The streamflow-equivalent water balance computed from TerraClimate fields ( $$\:{\text{WBeq}}_{\text{TC\_TC}}$$ ) showed the best univariate predictive performance, outperforming models based on drainage area and on precipitation alone. Spatialization applied a WBeq > 0 domain-of-applicability mask and an empirical envelope check based on the observed station range. Uncertainty was quantified by residual bootstrap in log space, and scenario perturbations ( P - 10%, AET + 10%) indicated strong sensitivity of low-flow estimates in the driest reaches. Overall, the results show that freely available gridded climate datasets can improve low-flow regionalization in data-scarce basins and support water management in semi-arid environments.