<p>Vibration-based structural health monitoring methodologies significantly rely on signal processing techniques to accurately analyze and interpret complex measurement data from instruments installed on the structure. In this paper, the multivariate variational mode decomposition technique (MVMD) is exploited for the condition assessment of the structure by utilizing the vibration data to provide reliable system identification parameters. Presently, limited studies have explored the applicability of the MVMD and various practical challenges that revolve around full-scale civil structures. Therefore, further investigation is required to evaluate the robustness of MVMD for mode decomposition and identification in the presence of non-stationary excitation, narrowband frequency, limited sensor measurements, and full-scale structures. In this paper, an optimization strategy is proposed to identify suitable signal decomposition parameters of MVMD, which are then used to isolate modal responses. The implementation of the proposed methodology is validated by a series of experiments conducted on the lab-scale experimental model subjected to various earthquakes and on the pedestrian bridge subjected to narrowband excitation. Moreover, the performance of the proposed method is compared in the presence of a limited number of sensor measurements. The average relative error in frequency estimation remains below 1.5% for the lab-scale model and below 0.5% for the pedestrian bridge. In addition, the accuracy of the extracted mode shapes is validated through both modal assurance criterion and coordinate modal assurance criterion analyses, which confirm strong global and pointwise correlations between the identified and exact mode shapes. These findings demonstrate the robustness of MVMD for real-world applications and contribute to creating a robust and user-friendly framework for optimal parameter selection and subsequent modal decomposition.</p>