Parameter regionalization is a critical challenge in hydrological modeling, impacting the ability to generalize models to ungauged areas. Traditional approaches often struggle with equifinality, raising doubts about the reliability of parameter sets they predict for ungauged regions. A more promising approach is to create more reliable targets for hydrological models’ calibration in ungauged regions through regionalizing hydrologic signatures, such as the flow duration curve, regime curve, and recession curve. This study presents a deep learning-based methodology for hydrological signature regionalization for HUC12 watersheds across the continental United States. Leveraging state-of-the-art deep learning techniques including deep transfer learning and meta-learning, our approach addresses the critical challenge of accurately predicting hydrological signatures in ungauged regions. It begins by extracting key features from watershed characteristics data. These features are then used to cluster watersheds with similar properties. Next, a deep learning model is trained at a continental scale. Finally, the model undergoes re-training for each regional cluster through transfer learning. This last step enables the model to adapt to more localized hydrological behaviors, improving its predictive accuracy for specific regions. To ensure this methodology can adapt to any hydrologic signature of importance, we implement a meta-learning approach. The findings demonstrate significant improvements in the accuracy and robustness of hydrological signature predictions compared to traditional methods, providing a scalable and practical solution for large-scale hydrological modeling and water resource management.