Abstract

In hydrological studies, continuous and long-term streamflow records are crucial for achieving reliable hydrological analysis and model development. However, obtaining such complete datasets remains challenging, particularly in developing countries. This study proposes a novel framework for daily streamflow imputation by coupling the spiking neural network (SNN) with the extra trees (ET) model. The SNN-predicted outputs were used to cluster flow regimes into normal- and high-flow using the gaussian mixture model (GMM) algorithm and subsequently served as augmented inputs for ET, which was trained separately within each regime. The coupled SNN-ET model was compared with stand-alone SNN and ET models using case studies from two streamflow stations in Thailand's southern basin. Two experimental objectives were examined: (1) comparing model training using observation-only data (scenario 1) with the integration of ERA5-Land reanalysis data (scenario 2), and (2) evaluating the robustness of the best-performing model under three artificial missing patterns (random, sequential, and combined) at missing proportions ranging from 5% to 90%. The results revealed that the coupled SNN-ET model outperformed the stand-alone models, particularly under scenario 2 with Kling-Gupta efficiency (KGE) more than 0.75. Moreover, it demonstrated strong robustness under all missing patterns, maintaining reliable predictive performance even with up to 80% missing data, especially under the combination pattern. This study highlights the importance of developing coupled models that integrate complex temporal networks with flexible ensemble methods to compensate for each other's limitations, thereby enhancing the accuracy of streamflow data imputation through the use of reanalysis datasets. The proposed framework presents a scalable and adaptable approach, specifically designed for broader application across data-scarce basins, providing a practical pathway to enhance hydrological data reliability and advance global water resource management.