Frequency-Resolved Seismic Source Localization in Fluvial Settings Using Dense Arrays

Monitoring dynamic surface processes such as sediment transport and turbulence during high-flow events remains a major challenge in fluvial geomorphology. Seismic methods provide a non-invasive alternative, but robust source localisation is challenged by the distributed nature of fluvial sources, heterogeneous shallow structures, and the broadband character of the signals they produce. Because turbulence and bedload dominate different parts of the spectrum, we require broadband analysis. We establish a generalised framework for frequency-resolved seismic source localisation using dense arrays and matched field processing. We introduce a hybrid processing strategy that exploits the array differently across frequencies: full-network matched field processing at low frequencies, where coherence spans the entire aperture, and sub-array averaging at higher frequencies, where coherence is confined to local scales. We apply the framework to a field case, where we retrieve frequency-dependent source regions across the active channel and separate low-frequency turbulent noise from higher-frequency bedload impacts. We conduct synthetic tests to quantify localisation uncertainty as a function of frequency, sensor density, and signal-to-noise ratio. Across the 2-40 Hz range, we find that localisation uncertainty varies from a few metres at low frequencies to more than 100 m at high frequencies, reflecting the expected loss of resolution at shorter wavelengths. By quantifying these trade-offs, we provide practical guidance for future deployments, including sensor spacing and array geometry required to achieve a target resolution.