SeaQC-X: Transferability of a machine learning-based sea level quality control framework

The operational reliability of tide gauge networks depends on robust and broadly applicable quality control (QC) procedures. While machine learning (ML) has shown promise in improving data QC, the cross-regime transferability of these models across diverse coastal environments remains a challenge. This study tests a ML-based QC approach, originally developed for the macrotidal regime of Greenland, applying it to data from nine tide gauge stations along the Danish coast using a strategy of domain adaptation via fine-tuning. The results indicate that the framework effectively adapts to Danish conditions, supporting the viability of generalising ML methods across diverse oceanographic settings. The ML approach performs comparably to standard operational tools, showing particular strength in identifying subtle anomalies missed by traditional methods. Limitations of the method include the model’s high sensitivity, which can lead to over-flagging of valid data. These findings suggest a path forward for tide gauge QC procedures in which ML tools augment, rather than replace, existing statistical tests in a synergistic framework, creating more robust and comprehensive operational QC systems.