Scour remains a significant threat to bridge infrastructure, as it erodes the sediment around pier foundations and undermines structural integrity. Conventional inspection techniques, such as visual surveys and impact vibration tests, are often limited, particularly during flood events where access constraints impede the timely acquisition of scour-related information. Scour assessment by identifying the rocking motion natural frequency of the pier through ambient vibration analysis is expected to address these issues enabling real-time assessment of natural frequency changes without disrupting bridge traffic or requiring physical inspection, while the accuracy is limited. The identified natural frequency over months (see Figure 1) shows significant variation before, during, and after the anti-scouring maintenance work, confirming the sensitivity of natural frequencies to structural changes. However, the red circles highlight instances where non-rocking motion modes were unintentionally identified, an issue this study seeks to minimize. This research thus aims to improve the accuracy of natural frequency identification by employing advanced modal identification techniques. The proposed methodology applies the Random Decrement Technique (RDT) to extract free vibration responses from ambient vibration data, filtering out random noise while taking into account specific amplitude conditions in sample picking. The specific amplitude conditions were investigated by examining the clarity of the Power Spectral Density (PSD). PSD analysis of ambient acceleration measured at a bridge pier demonstrates distinct frequency peaks between 12-14 Hz, consistent with impact test results, for data obtained at certain condition datasets (see Figure 1a) while other datasets highlight the need for improved frequency identification (see Figure 1b), where vibration modes, other than the rocking mode, show their distinct peaks. The sample picking considering amplitude conditions is meant to ensure that rocking-motion structural vibrations are well excited, improving the accuracy of natural frequency identification. Fast Bayesian FFT is then used to estimate natural frequencies probabilistically, accounting for uncertainties in noisy data environments. Together, these methods improve frequency identification accuracy. This research highlights the potential of ambient vibration-based monitoring for continuous, non invasive scour detection, offering a safer and more efficient alternative to traditional methods. By improving the accuracy of natural frequency identification, this approach can significantly enhance early detection of scour-induced damage, ultimately improving the resilience and safety of bridge infrastructure.