Introduction

Microseisms are a type of continuous seismic noise generated by oceanic processes, and they can be classified into two main categories: single frequency (SF) microseisms and double frequency (DF) microseisms (Hasselmann, 1963;Longuet-Higgins, 1950;Tanimoto & Anderson, 2023). SF microseisms typically occur at frequencies between 0.05 and 0.1 Hz and are generated by the direct interaction of ocean waves with the seafloor (Hasselmann, 1963). In contrast, DF microseisms-more commonly observed-typically occur at frequencies between ∼0.1 and ∼0.35 Hz, and are generated by the nonlinear interaction of ocean waves, often involving the interference of opposing wave trains (Ardhuin et al., 2011;Longuet-Higgins, 1950;Tanimoto, 2007). It is also commonly observed that the DF microseisms peak splits into two distinct peaks (Bromirski et al., 2005(Bromirski et al., , 2013;;Lin et al., 2017;Stephen et al., 2003;Xiao, Xue, Pan, & Gao, 2018, Xiao, Xue, Yang et al., 2018), corresponding to long-period double frequency (LPDF) microseisms at ∼0.1-0.2 Hz and short-period double frequency (SPDF) microseisms at ∼0.2-0.35 Hz. Frequencies above 0.35 Hz are referred to as the Holu spectrum (McCreery et al., 1993;Stephen et al., 2003), and are primarily generated by local wind-driven ocean waves (Gal et al., 2015;Tanimoto & Anderson, 2023;Xiao et al., 2022;Zhang et al., 2009). Ambient noise cross-correlation functions (CCFs) method have been widely applied to understand subsurface structures (Shapiro & Campillo, 2004). These techniques have proven invaluable in various applications such as seismic tomography (Sabra et al., 2005;Shapiro et al., 2005), monitoring volcanic activity (Brenguier et al., 2008), assessing aquifer water levels (Mao et al., 2022), and investigating mantle structures through body waves (Poli et al., 2012). However, the source locations where microseisms are generated remain unclear (Bromirski et al., 2013). There has been a long-standing debate about whether the DF microseisms received on land mainly originate from deep oceanic regions or from shallow coastal waters. Some researchers believe that standing waves, formed by the interaction of incoming and reflected ocean waves at coastlines, are the main source (Barnes et al., 2015;Bromirski, 2001Bromirski, , 2023;;Bromirski & Duennebier, 2002;Bromirski et al., 2005Bromirski et al., , 2013;;Essen et al., 2003;Friedrich et al., 1998;Gal et al., 2015;Gerstoft & Tanimoto, 2007;Juretzek & Hadziioannou, 2016;Schulte-Pelkum et al., 2004;Tanimoto, 2007;Ying et al., 2014). Another perspective suggests that the open ocean is the primary source of DF microseisms (Beucler et al., 2015;Cessaro, 1994;Chevrot et al., 2007;Hillers et al., 2012;Kedar et al., 2008;Landès et al., 2010;Obrebski et al., 2012;Stehly et al., 2006;Tian & Ritzwoller, 2015). These microseisms can result from interactions between ocean waves generated by different storms or from opposite sides of a large, fast-moving storm system (Ardhuin et al., 2011(Ardhuin et al., , 2012;;Bromirski et al., 2005;Lin et al., 2017;Liu et al., 2016;Nishida & Takagi, 2016, 2022;Sufri et al., 2014;Xiao, Xue, Pan, & Gao, 2018, Xiao, Xue, Yang et al., 2018, 2021;Zhang et al., 2010).

Beamforming and polarization analysis, the primary techniques for microseisms surface wave analysis, can determine the direction of incoming waves but not the source-to-station distance (Friedrich et al., 1998;Gal et al., 2014Gal et al., , 2015;;Gerstoft & Tanimoto, 2007;Juretzek & Hadziioannou, 2016;Koper & Hawley, 2010;Sufri et al., 2014;Tanimoto et al., 2006). As a result, locating microseisms using land-based stations remains challenging, since multiple potential sources along the inferred azimuth-particularly in coastal regions-cannot be readily distinguished. Due to the previous lack of seafloor seismometer data, this challenge has been difficult to overcome. More recently, the increased deployment of ocean bottom seismometers (OBS) has provided valuable insights that help illuminate this issue. For example, Tian and Ritzwoller (2015) used seafloor seismic data combined with CCFs techniques to infer the direction of noise sources directly within the ocean.

Moreover, the recent development of Distributed Acoustic Sensing (DAS) on the seafloor has brought new insights and observations to this problem. DAS technology uses existing or dedicated fiber-optic cables to detect and measure acoustic signals with high spatial and temporal resolution, offering dense and continuous sensing capabilities previously unavailable with OBSs (Lindsey et al., 2019;Lindsey & Martin, 2021;Sladen et al., 2019;Williams et al., 2019). This enables more detailed insights into the location and characteristics of microseisms sources (Xiao et al., 2022).

In this study, we combine DAS and OBS data from the Oregon coast to attempt to pinpoint the locations of microseisms at different frequencies. By applying CCFs techniques, we track the propagation of seismic waves to infer the locations of their sources (Xiao et al., 2022). This integrated approach enables us to capture not only the direction of wave propagation but also the propagation process itself, allowing for a more accurate determination of microseism sources and a clearer understanding of their spatial distribution relative to the coastline. By leveraging the strengths of both DAS and OBS, we aim to investigate whether microseisms in the Oregon region primarily originate from deep oceanic areas or shallow coastal waters, while also providing insights that may inspire similar studies in other parts of the world.

Data and Methods

We utilized data from the 7D OBS network, deployed off the Oregon coast between 2011 and 2015 (Figure 1a). Additionally, we incorporated DAS data acquired from a seafloor fiber-optic cable located near Florence, Oregon. From August 6 to 1 December 2021, a FEBUS Optics A1-R interrogator unit was deployed to monitor the first 60 km segment of the Alaska-Oregon Network telecommunication cable (Figure 1b). This system recorded strain-rate data continuously at 20 m spatial intervals using a 40 m gauge length and a 100 Hz sampling frequency, resulting in data from 3,000 channels. A three-day interruption occurred between September 7 and 10, after which data collection resumed with a reduced configuration of 2,750 channels, corresponding to a 55 km segment of the cable. To ensure temporal consistency of the data set, we only used data from the initial 55 km segment throughout the analysis (red line in Figure 1b). According to the cable installation report, the initial 6.3 km of the cable-equivalent to the first 315 channels-runs through conduits along roadways, connecting the landing station to the coastline. The remaining 48.7 km extends offshore and is buried beneath 80-100 cm of sediment. Between 6.3 and 40 km from the interrogator unit, the cable lies on the continental shelf in waters shallower than approximately 150 m. Beyond the 40 km mark, the seafloor descends from 150 to 250 m, indicating the transition to the continental slope. Oriented at an average azimuth of approximately 254°from true north, the cable extends nearly perpendicular to the general trend of the shoreline (Viens et al., 2023;Xiao et al., 2024). For the purposes of our study and ease of interpretation, we used only the offshore portion of the DAS data and redefined 0 km as the location where the cable reaches the coastline.

We also compared our observations with predictions from the IFREMER model (Ardhuin et al., 2011(Ardhuin et al., , 2012)). This model uses WAVEWATCH III (WW3) to compute the power spectral density (PSD) of equivalent pressure generated by ocean surface gravity waves (Rascle & Ardhuin, 2013;Tolman, 2009), providing an independent benchmark for our observations. Figure 1c-1f present the PSD analysis of microseisms recorded by both DAS and OBS instruments. Figure 1c shows the PSD of DAS data at various locations along the cable. Figure 1d focuses on the PSD at the 35 km mark along the DAS cable, revealing two prominent microseisms spectral peaks: the first between 0.1 and 0.2 Hz, and the second between 0.2 and 1.5 Hz. Similarly, Figure 1e displays the PSD of OBS station J17D, which also exhibits two distinct peaks-one in the 0.1-0.2 Hz range and another in the 0.2-0.7 Hz range. At the deeper OBS station J18D (Figure 1f), the first peak at 0.1-0.2 Hz remains comparable to J17D, while the second peak (0.2-0.7 Hz) appears significantly weaker.

To accurately locate the sources of microseisms, we performed CCFs between adjacent seismic stations or DAS channels with fixed distance to track the propagation of seismic waves. CCFs has long been used by researchers to infer the direction of incoming seismic energy, with the intersection of source directions from multiple stations pairs used to estimate the source location (Chen et al., 2016;Stehly et al., 2006;Tian & Ritzwoller, 2015;Xiao, Xue, Pan, & Gao, 2018, Xiao, Xue, Yang et al., 2018). However, this triangulation method is often unreliable for locating microseism sources, as these microseisms sources are typically widely distributed and can originate from any point along a given direction. Instead, our approach takes a different perspective: we use CCFs between adjacent stations or DAS channels with fixed spacing to trace the direction of wave propagation, as indicated by the acausal and causal parts of the signals. By analyzing a continuous series of station pairs, we are able to track both the origin and the propagation path of the seismic waves (Xiao et al., 2022). Because our seismic stations and DAS array are situated on potential microseism sources, we can use the observed propagation direction of the microseisms to infer source locations. For example, when a station pair or DAS channel pair is located directly above a microseism source, we expect to observe seismic energy propagating in both directions-oceanward and landward. In contrast, if energy is observed propagating in only one direction, the location is unlikely to coincide with the source of the microseisms. More specifically, we performed CCFs of DAS data at intervals of 7 km along the array (Figure 2a). This spacing was chosen based on the consideration of the longest wavelengths associated with the microseisms under investigation. For the OBS data, CCFs were performed between the nearest pairs of stations due to the long interstation distances. We computed CCFs using 30-min time windows and stacked the results to enhance signal clarity. For the DAS data, we stacked CCFs from August-December 2021. For the OBS data, we calculated and stacked one full year of recordings. We used the signal-to-noise ratio (SNR) of CCF to evaluate the amplitude of seismic noise. The SNR was calculated separately for the acausal and causal parts by dividing the maximum amplitude of the CCF's signal by the amplitude of the noise. Figure 2a shows an example of a stacked CCF in the 0.35-1.5 Hz frequency band. The top-left panel displays the CCF between DAS channels at 6 and 13 km along the array (i.e., channels at 6 km and 6 + 7 km). The orange (acausal) part represents seismic waves propagating landward, while the blue (causal) part represents waves propagating oceanward. This result indicates that, at this location, seismic waves propagate both landward and oceanward, suggesting the presence of a potential local microseism source. However, in such cases, it is important to first rule out contributions from distant sources in order to confidently identify a local microseism source. In the bottom-left panel, the CCFs between channels at 40 and 47 km is shown. In this case, the seismic energy predominantly propagates oceanward (causal), suggesting that the microseism source is likely located closer to the shallow-water region, rather than at this location or in more distant deep-ocean areas. Otherwise, we would expect to observe landward-propagating seismic waves. Overall, this method allows us to approximate the locations of microseism noise sources.

Results

Figure 2b shows the stacked CCFs of DAS data in the 0.35-1.5 Hz frequency band from August to December 2021. The top-right panel of Figure 2a presents the SNR of CCFs at different locations along the array. Near the coastline, oceanward SNR is very low, indicating that microseisms from the Atlantic Ocean on the opposite side of the continent cannot effectively propagate to this region. As distance from the coast increases, oceanward SNR gradually rises, suggesting a progressive buildup of microseism energy. Conversely, the landward SNR initially increases and then decreases to background levels farther offshore, implying that the deep ocean does not generate significant seismic noise at this frequency. Within the 3-35 km range, seismic energy propagates in both directions, while between 0 and 3 km, only landward signals are seen, and beyond ∼35 km (depth ∼150 m), only oceanward propagation is dominant (Figure 2b). These patterns suggest that the source region of microseisms at 0.35-1.5 Hz lies between 3 and 35 km from the coast. Given the 7 km spacing between correlated DAS channel pairs, the actual source region likely extends from approximately 3-42 km. This result shows that the majority of the microseism energy is generated within this range, although a small amount of energy may originate from beyond it (Figure 2b). Figure S1 in Supporting Information S1 shows the temporal variation of the microseism source. It can be seen that the source remains within the 3-42 km range, though its location shifts over time. Previous studies have shown that microseisms in this frequency band are primarily generated by local winddriven ocean waves, and their generation location is closely related to wind direction (McCreery et al., 1993;Tanimoto & Anderson, 2023;Xiao et al., 2022). In addition, Figure S2 in Supporting Information S1 presents results across different narrow frequency bands, demonstrating that microseism source locations vary with frequency-higher-frequency sources are located closer to the coastline, while lower-frequency sources can originate farther offshore.

For SPDF microseisms (0.2-0.35 Hz; Figures 3a and 3b), oceanward-propagating energy progressively increases with distance from the coastline. In contrast, landward-propagating energy initially rises but then decreases, though it remains above background noise levels. Notably, in shallow water regions, low-frequency DASrecorded seismic signals are more vulnerable to interference from ocean wave motion (Figure 1b), which may contribute to the reduced CCF amplitudes observed in both directions (Figure 3a). This occurs because pressure fluctuations from ocean waves can reach the seafloor when the water depth is less than half the wavelength. Moreover, the results suggest that the SPDF microseism source extends continuously from the coastline to beyond the far end of the DAS array. Overall, the relative amplitudes of oceanward-and landward-propagating waves vary across different locations, which may reflect varying contributions from overlapping coastal and deep-ocean SPDF microseism sources.

Similarly, for LPDF microseisms in the frequency band 0.1-0.2 Hz (Figures 3c and 3d), both oceanward-and landward-propagating energy first increase and then decrease along the array. Notably, the peak amplitude of landward-propagating signals is approximately 1.5-2 times greater than that of oceanward-propagating signals.

Since energy scales with the square of amplitude, this corresponds to a 2.25-4 times greater energy for landwardpropagating waves. This implies that deep-ocean sources contribute significantly more to the observed LPDF microseism energy than coastal sources. However, due to the limited length of the DAS array, the extent to which nearshore energy propagates toward the deep ocean remains uncertain.

To gain a more comprehensive understanding of the DF microseism sources, we applied the same CCFs strategy to the OBS data. We selected an OBS array oriented approximately perpendicular to the coastline and performed

Geophysical Research Letters

10.1029/2025GL118297

pairwise CCFs between stations (Figure 1a). From the OBS data, the 1-year stacked results in the 0.1-0.2 Hz frequency band show clearly defined CCFs (Figure 4a). From the first station pair near the coastline (FN02C and FN04C), the CCF clearly shows the absence of oceanward-propagating seismic waves and the presence of only landward-propagating signals. This indicates that LPDF microseisms generated on the far side of the continent, such as in the Atlantic Ocean, do not propagate to this region. Additionally, the results suggest that no significant LPDF microseism sources exist in the shallower regions east of FN02C. In the result from the FN04C-FN06C station pair, oceanward-propagating microseism signals begin to appear, indicating that LPDF microseisms start to be generated at this depth (104 m, 88 km from the coastline). Additionally, the landward-propagating signal becomes significantly stronger, with the SNR increasing from 15.3 (FN02C-FN04C) to 28.4 (FN04C-FN06C), under similar stacking durations (Figure 4a). It is important to note that differences in stacking time may arise due to variations in when different station pairs record coherent seismic signals. Thus, in this analysis, we focus primarily on the relative energy between oceanward-and landward-propagating signals, which remains robust even if stacking times vary. As water depth increases, oceanward-propagating microseism signals initially strengthen and then gradually weaken. In the station pair J52A-J53A, although the landward-propagating microseism signals exhibit relatively high SNR, the oceanward-propagating signals are nearly absent. This suggests that beyond J52A (2640 m, 197 km from the coastline), LPDF microseisms generated near the coast cannot effectively reach this region. To compare the strength of coastal and deep-ocean LPDF microseism sources, we examine the SNR amplitudes of oceanward-and landward-propagating signals. It is clearly observed that the landward SNR is consistently much higher than the oceanward SNR-often more than twice as large (Figure 4a). Figure S3a in Supporting Information S1 also shows similar results for a different station line. Considering that seismic energy scales with the square of the amplitude, this implies that the energy of landwardpropagating signals is more than four times greater than that of oceanward-propagating signals, indicating that deep-ocean LPDF microseisms sources generate significantly more energy than nearshore sources.

For the SPDF microseisms (0.2-0.35 Hz), a similar pattern is observed. In the station pair J51A-J52A, and in deeper pairs beyond, oceanward-propagating microseism energy disappears (Figure 4b). This indicates that beyond J51A (2626 m water depth, 166 km from the coastline), coastal microseisms in this band are no longer generated. Compared to LPDF microseisms, the nearshore source region for SPDF microseisms is therefore narrower. As shown in Figure 4b and Figure S3b in Supporting Information S1, the SNRs of oceanward-and landward-propagating signals vary across different locations, with neither direction consistently dominant. This suggests that, SPDF microseisms observed on land are influenced by both coastal and deep-ocean sources.

Discussion and Conclusions

By integrating DAS and OBS data, our study reveals the dominant sources of microseisms across different frequency bands, providing new insights into a long-standing debate. Through the use of CCFs between DAS channels and OBS station pairs, we are able to trace both landward-and oceanward-propagating microseisms across a range of frequencies and water depths. Our results show that microseism sources are strongly frequencydependent: high-frequency microseisms (0.35-1.5 Hz) are predominantly generated in the coastal region; SPDF microseisms (0.2-0.35 Hz) are influenced by both coastal and deep-ocean sources to a similar degree; and LPDF microseisms (0.1-0.2 Hz) originate mainly in the deep ocean.

Regarding the spatial extent of coastal microseism sources, we find that as frequency decreases, the noise sources shift progressively toward deeper water (Figure S2 in Supporting Information S1). For example, in the 0.35-1.5 Hz band, microseism energy is largely confined to within 3-42 km of the coastline (Figure 5a). However, for DF microseisms (0.1-0.35 Hz), the generation zone associated with reflected waves may extend farther offshore. For instance, at 0.1-0.2 Hz (LPDF), oceanward-propagating microseisms are observed up to approximately 197 km offshore, in water depths of around 2640 m, based on OBS data. For the 0.2-0.35 Hz (SPDF) frequency band, oceanward-propagating signals are observed up to about 166 km from the coastline (Figure 5a). However, the coastal microseisms sources regions at these low frequencies are harder to delineate, since landward-propagating energy nearshore includes contributions from the deep ocean. Therefore, the true offshore extent of coastal microseism generation can only be bounded by the presence of oceanwardpropagating energy, which tends to disappear at large distances from the shore, setting an upper bound on the coastal source region. The actual generation zone is likely narrower than this observed extent. This also indicates that microseisms generated near the coastline do not effectively propagate into deep-ocean regions. Since the energy of landward-propagating microseisms includes contributions from deep-ocean sources, we cannot determine the propagation distance of the energy originating specifically from the coastline toward land.

We compared the PSD of ocean wave-induced equivalent pressure predicted by the IFREMER source model in the 0.1-0.2 Hz frequency band for 2012 with our observations (Figure 5b). The comparison shows that the main energy generated in the deep North Pacific, with some contribution from coastal reflections. However, due to the relatively coarse spatial resolution of the model (0.5°), exact correspondence with our observations is not expected. The excitation also depends on site-specific effects. Previous studies have shown that the sediment thickness at the seafloor and the water depth can significantly influence the excitation coefficients (Gualtieri et al., 2014(Gualtieri et al., , 2015)). Our observations are consistent with the calculated excitation efficiency in shallow regions with a sedimentary layer (Figure S6 in Supporting Information S1 in Xiao et al. (2022)). In this scenario, the excitation efficiency is higher at frequencies typically above 0.35 Hz, gradually decreasing with increasing water depth. However, for the 0.1-0.2 Hz frequency band, the excitation coefficients remain relatively low, which also agrees with our observations.

To evaluate the influence of sedimentary layers on our method, we also compared their effect on the amplitudes of the landward and oceanward CCFs in the 0.1-0.2 Hz frequency band. To obtain a more accurate estimate of group velocity, we used a station spacing of 17 km in this analysis. In Figure S4c in Supporting Information S1, the black dashed line shows the variation in group velocity. The results indicate that in regions (25-30 km) with lower group velocity, or equivalently thicker sediments, the amplitudes of seismic waves increase in both directions. However, since this effect influences both directions simultaneously, it does not compromise the reliability of our source localization method.

In addition, we tested different station spacings. Overall, smaller spacings provide higher spatial resolution for source localization, although the spacing is constrained by the seismic wavelength. For high-frequency microseisms (0.6-1 Hz), a station spacing of about 3 km allows us to resolve a finer structure of the noise sources (Figure S5 in Supporting Information S1). However, for LPDF microseisms (0.1-0.2 Hz), overly small spacings cause the causal and acausal parts of the CCF to overlap, making it difficult to distinguish the direction of wave propagation. At the same time, considering energy attenuation, excessively large station spacings in the high-frequency band (0.5-1.5 Hz) result in poor CCF quality, limiting their usefulness for reliable analysis (Figure S6 in Supporting Information S1).

It is worth noting that spurious phases appear at approximately 22.5 and 30 km in Figure 2b. Previous studies have suggested that such phases may be associated with persistent noise sources (Retailleau et al., 2017). To investigate this phenomenon, we performed autocorrelations on all channels in the 1-4 Hz frequency band. Two centers of spurious phases can be observed at around 26 and 33.5 km (Figure S7 in Supporting Information S1). The results suggest that these spurious phases may be caused by subsurface structures, such as reflections from faults (Yang Geophysical Research Letters 10.1029/2025GL118297 et al., 2022). If they were indeed two distinct noise sources, such spurious phases would not appear in the autocorrelations. This also explains the positional differences of approximately 3.5 km between the autocorrelations and the CCFs, which correspond to half of the 7 km spacing between CCF channels. Additionally, reflections from the coastline also contribute noticeably to these observations. Our results underscore the complexity and frequency sensitivity of microseism generation processes, and demonstrate the power of combining DAS and OBS data to disentangle overlapping source contributions across the continental margin. To better understand these mechanisms, future work should incorporate more comprehensive, higher resolution numerical modeling efforts. In addition, this method is not only suitable for locating microseism sources but also effective for identifying other seismic signals that lack clear onsets, such as volcanic tremor and slow slip events. These signals typically have long durations and lack distinct impulsive arrivals, making them difficult to locate using traditional phase-picking techniques. By analyzing source-region CCFs between adjacent stations, we can track the direction of wave propagation and infer source regions, providing a powerful approach for studying such non-impulsive seismic phenomena.