Monitoring large structures over their lifetime is becoming increasingly important as robust monitoring will ensure safety and financial benefits. Offshore wind turbines (OWT) are in need of continuous structural health monitoring (SHM) methodologies for scour detection. Reliable detection of scour requires robust selection of structural features to adequately represent the integrity of the OWT. Successful identification of operational modal parameters of OWT is challenging because of the effect of environmental and operational variability (EOV). Therefore, mitigating the effect of EOV is crucial to guarantee that the variability in the extracted features is caused by structural degradation. This work presents a methodology based on stochastic subspace identification enhanced by a clustering technique to identify stable and robust structural modes over time. Gaussian process regression is used to mitigate EOV due to its flexibility and non-parametric nature. Data collected from a control OWT and a scoured OWT in operation over 8-months is used for verification. The results have demonstrated that scour affects OWTs by reducing their natural frequencies. The results also indicate that scour induces OWTs to react differently to EOV. The outcomes of this work provide the basis for real time monitoring of OWTs, thus facilitating a reliable identification of scour.