Seismic surveys in coastal and shallow waters have gained increasing importance due to the growing demand for marine infrastructure development, coastal ground stability assessment, and environmental monitoring. However, in coastal waters with depths of approximately 5–30 m, limitations such as restricted vessel operation, short offsets, and a lack of low-frequency components make it difficult to obtain reliable long wavelength velocity models. This study proposes a strategy to regenerate secondary bubble signals from an air-gun source—typically treated as components to be removed—as effective low-frequency information in shallow marine seismic data obtained using a portable air-gun/streamer system. Using deterministic deconvolution with the source wavelet, the primary reflections were separated from the raw data, and the residual between them was used to construct the secondary bubble–dominant dataset. The long wavelength velocity structure was then inverted using direct envelope reflection-based full waveform inversion (DE-RFWI). Application to field data acquired in the Yeongil Bay area of Pohang demonstrates that the proposed approach provides a more structurally consistent, long-wavelength velocity model compared with primary-based FWI, confirming that reliable velocity models can be obtained even under short-offset and low frequency-deficient conditions. This study offers a practical workflow that overcomes the inherent limitations of portable shallow marine seismic systems and enhances subsurface property modeling in coastal and transition zone environments.