Iterative imputation of river monitoring data using Singular Value Decomposition

River monitoring datasets often contain gaps, inconsistencies, and variable data quality arising, for example, from differing measurement periods, equipment changes, and operational interruptions. These issues complicate the use of such monitoring records in studies that do not tolerate gaps in input data such as hydrological and groundwater modelling. This paper presents a robust and efficient imputation methodology that exploits the strong spatio-temporal correlations commonly found in hydrological systems to reconstruct missing or poor-quality river gauging station records. Our approach iteratively replaces missing or unreliable entries with values derived from a low-rank Singular Value Decomposition approximation. This process repeats until the solution stabilizes, ensuring the imputed data aligns with the dominant spatio-temporal patterns of the system. Synthetic data experiments demonstrate that the method consistently performs better than other imputation techniques, particularly when large fractions of data are missing (>20%). A real-world application of the workflow is presented to extract complete monthly average river monitoring datasets for the Condamine River (Queensland, Australia), where even river data for stations with more than 70% missing records were imputed. The study also derives a flow-stage converting formula that allows obtaining site-unique relationship. The results confirm that the proposed workflow provides a robust and computationally efficient tool for generating complete and consistent river datasets, which also opens the door to applications in other fields requiring reliable spatio-temporal inputs.