Ambient Noise Correlation Research Articles

Potential Pitfalls in the Analysis and Structural Interpretation of Seismic Data from the Mars InSight Mission.

The Seismic Experiment for Interior Structure (SEIS) of the InSight mission to Mars, has been providing direct information on Martian interior structure and dynamics of that planet since it landed. Compared to seismic recordings on Earth, ground motion measurements acquired by SEIS on Mars are made under dramatically different ambient noise conditions, but include idiosyncratic signals that arise from coupling between different InSight sensors and spacecraft components. This work is to synthesize what is known about these signal types, illustrate how they can manifest in waveforms and noise correlations, and present pitfalls in structural interpretations based on standard seismic analysis methods. We show that glitches, a type of prominent transient signal, can produce artifacts in ambient noise correlations. Sustained signals that vary in frequency, such as lander modes which are affected by variations in temperature and wind conditions over the course of the Martian Sol, can also contaminate ambient noise results. Therefore, both types of signals have the potential to bias interpretation in terms of subsurface layering. We illustrate that signal processing in the presence of identified nonseismic signals must be informed by an understanding of the underlying physical processes in order for high fidelity waveforms of ground motion to be extracted. While the origins of most idiosyncratic signals are well understood, the 2.4 Hz resonance remains debated and the literature does not contain an explanation of its fine spectral structure. Even though the selection of idiosyncratic signal types discussed in this paper may not be exhaustive, we provide guidance on best practices for enhancing the robustness of structural interpretations.

Diffuse field cross-correlations: Scattering theory and electromagnetic experiments.

The passive estimation of impulse responses from ambient noise correlations arouses increasing interest in seismology, acoustics, optics, and electromagnetism. Assuming the equipartition of the noise field, the cross-correlation function measured with noninvasive receiving probes converges towards the difference of the causal and anticausal Green's functions. Here, we consider the case when the receiving field probes are antennas which are well coupled to a complex medium-a scenario of practical relevance in electromagnetism. We propose a general approach based on the scattering matrix formalism to explore the convergence of the cross-correlation function. The analytically derived theoretical results for chaotic systems are confirmed in microwave measurements within a mode-stirred reverberation chamber. This study provides fundamental insight into the Green's function retrieval technique and paves the way for a new technique to characterize electromagnetic antennas.

The relationship between cross correlations and Green’s functions in ambient noise interferometry with Bayesian constraints

SUMMARYCross correlation of ambient seismic noise has begun to provide detailed images of the Earth that are not possible to attain using earthquake-based or active-source imaging techniques. However, the theoretical justification for the ambient noise correlation or seismic interferometry approach typically relies on an equipartition or diffuse field assumption that is not expected to be satisfied on Earth. Here we demonstrate that a Bayesian inference approach to calculating the expected cross correlation of seismic signals leads to an improved understanding of its relationship with the Green’s function. The Bayesian derivation is found to exactly replicate the equipartition result under a global energy constraint, and is able to accommodate any number of other constraints that typically cannot be utilized in other approaches. With these stronger constraints, the cross correlation is found to deviate more and more strongly from the ‘expected’ Green’s function, but nonetheless a specific prediction can be made regarding the expected cross correlation which can be used as a more reliable point of comparison with observations than the Green’s function. The approach therefore provides a path forward for how to use seismic cross correlation data under more and more realistic conditions and improvements in knowledge, and should eventually help improve the accuracy of resulting images of the Earth.

Hybrid Simulation of Near-Fault Ground Motion for a Potential Mw 7 Earthquake in Lebanon

ABSTRACTLebanon is a densely populated country crossed by major faults. Historical seismicity shows the potential of earthquakes with magnitudes >7, but large earthquakes have never been instrumentally recorded in Lebanon. Here, we propose a method to simulate near-fault broadband ground motions for a potential Mw 7 earthquake on the Yammouneh fault (YF)—the largest branch of the Dead Sea Transform fault that bisects Lebanon from north to south. First, we performed the first 3D tomography study of Lebanon using ambient noise correlation, which showed that Lebanon could be approximated by a 1D velocity structure for low-frequency (LF) ground-motion simulation purposes. Second, we generated suites of kinematic rupture models on the YF, accounting for heterogeneity of the rupture process, and uncertainty of the rupture velocity and hypocenter location. The radiated seismic energy was next propagated in the inferred 1D velocity model to obtain suites of LF ground motions (<1 Hz) at four hypothetical near-fault seismic stations. These LF simulations included the main features of near-fault ground motions, such as the impulsive character of ground velocity due to the rupture directivity or fling-step effects (so-called pulse-like ground motions). Third, to obtain broadband ground motions (up to 10 Hz), we proposed a hybrid technique that combined the simulated LF ground motions with high-frequency (HF) stochastic simulations, which were empirically calibrated using a worldwide database of near-fault recordings. Contrary to other hybrid approaches, in which the LF and HF motions are generally computed independently, the characteristics of stochastic HF ground motions were conditioned on those of LF ground motions (namely on the characteristics of the velocity pulse, if it existed, or on the absence of a pulse). The simulated peak ground accelerations were in agreement with the ones reported in the Next Generation Attenuation-West2 (NGA-West2) database for similar magnitude and distances and with three NGA-West2 ground-motion prediction equations.

Using anisotropic and narrowband ambient noise for continuous measurements of relative clock drift between independent acoustic recorders.

Relative clock drift between instruments can be an issue for coherent processing of acoustic signals, which requires data to be time-synchronized between channels. This work shows how cross correlation of anisotropic narrowband ambient noise allows continuous estimation of the relative clock drift between independent acoustic recorders, under the assumption that the spatial distribution of the coherent noise sources is stationary. This method is applied to two pairs of commercial passive acoustic recorders deployed up to 14 m apart at 6 and 12 m depth, respectively, over a period of 10 days. Occasional calibration signals show that this method allows time-synchronizing the instruments to within ±1 ms. In addition to a large linear clock drift component on the order of tens of milliseconds per hour, the results reveal for these instruments non-linear excursions of up to 50 ms that cannot be measured by standard methods but are crucial for coherent processing. The noise field displays the highest coherence between 50 and 100 Hz, a bandwidth dominated by what are believed to be croaker fish, which are particularly vocal in the evenings. Both the passive and continuous nature of this method provide advantages over time-synchronization using active sources.

Subsurface Moisture Regulates Himalayan Groundwater Storage and Discharge

AbstractThrough the release of groundwater, most mountain rivers run year‐round despite their small catchments and sporadic precipitation. This makes mountain ranges important sources of reliable freshwater for downstream populations in many parts of the world. However, due to a lack of ground instrumentation, little is known about groundwater dynamics in mountainous landscapes. Recent research has shown that the amount of moisture trapped in the soil and weathered rocks in the vadose zone can significantly buffer groundwater recharge and runoff but the wider recognition of this effect on major mountain systems has not been yet established. In this study, we test whether the moisture reservoir has an impact on hydrological fluxes in a steep Himalayan catchment during three monsoon seasons. We measured an array of parameters including relative seismic velocity changes from ambient noise correlations. This noninvasive technique allows us to monitor groundwater dynamics in conjunction with classical hydrological measurements. We found that the moisture saturation in the vadose zone controls the onset of groundwater recharge and runoff and therefore determines the annual water availability supplied by monsoon precipitation. We model this dynamic using a surface layer that has a finite storage capacity that controls the connectivity of surface flux to groundwater. The extension of this concept, which is thought to apply widely in flat and undulating landscapes, to steep mountain topography with thin and discontinuous soils underlain by regolith and bedrock has important implications for mountain hydrology.

Determination of Near Surface Shear‐Wave Velocities in the Central Los Angeles Basin With Dense Arrays

AbstractIn this study, we investigate the shallow shear wave velocity structure of the Los Angeles Basin in southern California, using ambient noise correlations between 5 dense arrays and 21 broadband stations from the Southern California Seismic Network (SCSN). We observe clear fundamental mode and first overtone Rayleigh waves in the frequency band 0.25–2.0 Hz, and obtain group velocity maps through tomography. We further derive a 3D shear wave velocity model, covering a large portion of the central LA Basin for the depths shallower than 3 km. We found that the small scale shallow velocity structure heterogeneities are better resolved compared with the SCEC Community velocity models. Our model captures the presence of the Newport‐Inglewood fault by a NW–SE trending high velocity belt. Our model provides more accurate constraints on local ground motion predictions with detailed mapping of structural heterogeneities.

Distinct lithospheres in the Bay of Bengal inferred from ambient noise and earthquake tomography

We present an improved 3-D shear wave velocity image of the uppermost mantle beneath the Bay of Bengal (BoB), the Bengal basin and the adjoining Indian shield to their west. Shear wave velocities were inverted from fundamental mode Rayleigh wave group velocities calculated along 21,600 crisscrossing paths from cross correlation of ambient noise as well earthquake seismograms. This study shows the hitherto unsuspected existence of distinct lithospheres in the eastern and western Bay of Bengal, on either side of 86° E longitude, but understandable in terms of their different origins and ages. The western Bay of Bengal, with greater than 120 km thick layered lithosphere has a shear wave velocity of 4.7 km/s beyond the depth of 90 km. This velocity structure is in lateral continuation with the high velocity in the adjacent cratonic India. The lithosphere thickening can be explained to be the result of conductive cooling of an oceanic plate with a temperature of 1300 ± 50 °C at its base. The remarkable similarity between the velocity structure of the western BoB and the adjoining Indian craton, which could be fortuitous, however, suggests the untested possibility that the fracture plane mediating the initial break-up of India from Antarctica might have been inclined towards the former creating asymmetrically spreading oceanic crust over a cratonic upper mantle. The eastern Bay of Bengal (BoB), in contrast, has thinner lithosphere (60–75 km) with minimum velocity of ~4.2 km/s which is anomalously low for an old ocean. This significant thinning could have been caused by a number of factors such as reheating of the original lithosphere arising from a thermal boundary layer instability or westward flow of mantle due to slab rollback of subducting Indian lithosphere in the Andaman arc.

Near-Field Body-Wave Extraction From Ambient Seafloor Noise in the Nankai Subduction Zone

Ambient noise correlation is capable of retrieving waves propagating between two receivers. Although waves retrieved using this technique are primarily surface waves, the retrieval of body waves, including direct, refracted, and reflected waves, has also been reported from land-based observations. The difficulty of body wave extraction may be caused by large amplitudes and little attenuation of surface waves excited by microseisms, indicating that body wave extraction using seafloor records is very challenging because microseisms are generated in ocean areas and large amplitudes of surface waves are presumably observed at the seafloor. In this study, we used a unique dataset acquired by dense arrays deployed in the Nankai subduction zone, including a permanent cabled-network of 49 stations, a borehole sensor, and 150 temporary stations, to attempt to extract near-field body waves from ambient seafloor noise observed by multivariate sensors of broadband and short-period seismometers, differential pressure gauges (DPGs), and hydrophones. Our results show thatPwaves are extracted only in the DPG-record correlations at a frequency of 0.2–0.5 Hz, which can be seen up to a separation distance of two stations of 17 km with an apparent velocity of 3.2 km/s. At 1–3 Hz,Pwaves are observed only in the vertical-record correlations up to a separation distance of 11 km with an apparent velocity of 2.0 km/s. These velocity differences reflect the vertical velocity gradient of the accretionary prism, because thePwaves at low frequencies propagate at relatively long distances and therefore the turning depth is greater. Moreover, the long-period and short-periodPwaves are observed at the slope and flat regions on the accretionary prism, respectively. To investigate the retrieved wavefield characteristics, we conducted a two-dimensional numerical simulation for wave propagations, where we located single sources at the sea surface above the flat and slope bathymetry regions. Based on our observations and simulations, we suggest that the retrieval of near-field body waves from ambient seafloor noises depends on the relative amplitudes ofPand other surface waves in the ambient noise wavefield, and those are controlled by the subseafloor velocity structure, seafloor topography, and water depth.

Extracting Dispersion Curves From Ambient Noise Correlations Using Deep Learning

We present a machine-learning approach to classifying the phases of surface wave dispersion curves. Standard FTAN analysis of surfaces observed on an array of receivers is converted to an image, of which, each pixel is classified as fundamental mode, first overtone, or noise. We use a convolutional neural network (U-net) architecture with a supervised learning objective and incorporate transfer learning. The training is initially performed with synthetic data to learn coarse structure, followed by fine-tuning of the network using approximately 10% of the real data based on human classification. The results show that the machine classification is nearly identical to the human picked phases. Expanding the method to process multiple images at once did not improve the performance. The developed technique will faciliate automated processing of large dispersion curve datasets.

Velocity structure and radial anisotropy of the lithosphere in southern Madagascar from surface wave dispersion

SUMMARYWe investigate the upper mantle seismic structure beneath southern Madagascar and infer the imprint of geodynamic events since Madagascar’s break-up from Africa and India and earlier rifting episodes. Rayleigh and Love wave phase velocities along a profile across southern Madagascar were determined by application of the two-station method to teleseismic earthquake data. For shorter periods (<20 s), these data were supplemented by previously published dispersion curves determined from ambient noise correlation. First, tomographic models of the phase velocities were determined. In a second step, 1-D models of SV and SH wave velocities were inverted based on the dispersion curves extracted from the tomographic models. As the lithospheric mantle is represented by high velocities we identify the lithosphere–asthenosphere boundary by the strongest negative velocity gradient. Finally, the radial anisotropy (RA) is derived from the difference between the SV and SH velocity models. An additional constraint on the lithospheric thickness is provided by the presence of a negative conversion seen in S receiver functions, which results in comparable estimates under most of Madagascar. We infer a lithospheric thickness of 110−150 km beneath southern Madagascar, significantly thinner than beneath the mobile belts in East Africa (150−200 km), where the crust is of comparable age and which were located close to Madagascar in Gondwanaland. The lithospheric thickness is correlated with the geological domains. The thinnest lithosphere (∼110 km) is found beneath the Morondava basin. The pre-breakup Karoo failed rifting, the rifting and breakup of Gondwanaland have likely thinned the lithosphere there. The thickness of the lithosphere in the Proterozoic terranes (Androyen and Anosyen domains) ranges from 125 to 140 km, which is still ∼30 km thinner than in the Mozambique belt in Tanzania. The lithosphere is the thickest beneath Ikalamavony domain (Proterozoic) and the west part of the Antananarivo domain (Archean) with a thickness of ∼150 km. Below the eastern part of Archean domain the lithosphere thickness reduces to ∼130 km. The lithosphere below the entire profile is characterized by positive RA. The strongest RA is observed in the uppermost mantle beneath the Morondava basin (maximum value of ∼9 per cent), which is understandable from the strong stretching that the basin was exposed to during the Karoo and subsequent rifting episode. Anisotropy is still significantly positive below the Proterozoic (maximum value of ∼5 per cent) and Archean (maximum value of ∼6 per cent) domains, which may result from lithospheric extension during the Mesozoic and/or thereafter. In the asthenosphere, a positive RA is observed beneath the eastern part Morondava sedimentary basin and the Proterozoic domain, indicating a horizontal asthenospheric flow pattern. Negative RA is found beneath the Archean in the east, suggesting a small-scale asthenospheric upwelling, consistent with previous studies. Alternatively, the relatively high shear wave velocity in the asthenosphere in this region indicate that the negative RA could be associated to the Réunion mantle plume, at least beneath the volcanic formation, along the eastern coast.

Ambient noise correlation analysis of S-net records: extracting surface wave signals below instrument noise levels

SUMMARY We applied ambient noise cross-correlation analysis to the cabled ocean bottom seismic network offshore northeast Japan (Seafloor observation network for earthquakes and tsunamis along the Japan Trench: S-net) to extract surface waves propagating in the ocean area of the forearc region. We found two types of peculiar pulses in the cross-correlation functions (CCFs) of ambient seismic noise records: periodic pulses mainly every minute and sharp pulses around the lag time zero. These pulses strongly contaminate the surface wave signals in the CCFs at frequencies below ∼0.1 Hz. The periodic pulses originate from periodic instrument noises, while the zero-lag pulses originate from random instrument noises which are coherent within station pairs. By developing solutions to remove the periodic and zero-lag pulses based on the characteristics of the pulses, we succeeded in extracting Rayleigh and Love wave signals from the S-net records at 0.03–0.3 Hz, while the surface wave signals at 0.03–0.1 Hz were not visible without the application of these solutions. These solutions widen the frequency range of analysis, and may be applicable to other seismic networks, particularly to recent dense but non-broad-band networks. We identified the fundamental and first higher modes of Rayleigh waves and the fundamental mode of the Love wave. The extracted surface wave signals can constrain the shear wave velocity structure from the sediment to seismogenic zone around the megathrust plate boundary in the forearc region.

Effect of centimetric freezing of the near subsurface on Rayleigh and Love wave velocity in ambient seismic noise correlations

SUMMARY About a decade ago, noise-based monitoring became a key tool in seismology. One of the tools is passive image interferometry (PII), which uses noise correlation functions (NCF) to retrieve seismic velocity variations. Most studies apply PII to vertical components recording oceanic low-frequent ambient noise ( < 1 Hz). In this work, PII is applied to high-frequent urban ambient noise ( > 1 Hz) on three three-component sensors. With environmental sensors inside the subsurface and in the air, we are able to connect observed velocity variations with environmental parameters. Temperatures below 0 °C correlate well with strong shear wave velocity increases. The temperature sensors inside the ground suggest that a frozen layer of less than 5 cm thickness causes apparent velocity increases above 2 % , depending on the channel pair. The observations indicate that the different velocity variation retrieved from the different channel pairs are due to different surface wave responses inherent in the channel pairs. With dispersion curve modelling in a 1-D medium we can verify that surfaces waves of several tens of metres wavelength experience a velocity increase of several percent due to a centimetres thick frozen layer. Moreover, the model verifies that Love waves show larger velocity increases than Rayleigh waves. The findings of this study provide new insights for monitoring with PII. A few days with temperature below 0 °C can already mask other potential targets (e.g. faults or storage sites). Here, we suggest to use vertical components, which is less sensitive to the frozen layer at the surface. If the target is the seasonal freezing, like in permafrost studies, we suggest to use three-component sensors in order to retrieve the Love wave response. This opens the possibility to study other small-scale processes at the shallow subsurface with surface wave responses.

Ambient-Noise Tomography of the Baiyun Gold Deposit in Liaoning, China

Abstract Ambient-noise tomography (ANT) has become an effective method for determining the fine velocity structure of the shallow crust. However, studies on metal mines using this method are rarely reported. To investigate the tectonic background and prospecting of the deep mine in the Baiyun gold deposit (BYGD) of eastern Liaoning Province, China, we use ANT to determine a 3D S-wave velocity structure model of the BYGD. A total of 21 broadband seismic stations were installed in an area of 15×14 km, centered at the BYGD. Continuous observations for approximately three months were made. After single-station preprocessing, cross correlation of ambient noise, and phase-weighted stacking, the empirical Green’s function for the Rayleigh waves between stations was recovered. Next, group-velocity dispersion with 0.8–3 s periods was measured. A direct inversion method of surface-wave dispersion based on raytracing was then adopted to determine a 3D S-wave velocity structure of the BYGD from the ground surface to a depth of 1.8 km. The results show that the distribution of S-wave velocities in the study area well reflected the geological characteristics of the surface. The velocities were significantly low within the “ore field” and the regional ore-controlling Jianshanzi fault. Combining this with the fact that a large number of magmatic veins were visible inside both structures, it was deduced that both structures had experienced large-scale magmatic intrusion activities, thus confirming that BYGD was a magmatic hydrothermal deposit. The significantly low S-wave velocities beneath the gold deposit extended to a depth of 1.8 km. This might imply the occurrence of blind ore bodies at that depth. The fine velocity structure of the BYGD reconstructed by this study provided a direction for subsequent prospecting of deep regions and demonstrated that ANT has good potential in metal mine exploration.

When Clocks Are Not Working: OBS Time Correction

AbstractOcean-bottom seismographs (OBSs) are used to obtain seismic recordings offshore and are an increasingly important tool for investigating the globe. However, because OBS data cannot be time stamped using Global Positioning System (GPS) during deployment, correction for drift of the internal clock is required. This time drift is typically derived by synchronizing the clock before and after deployment. Linear correction is then applied using the timing deviation between GPS and the instrument’s internal clock at recovery, that is, the skew measurement. If synchronization measurements are missing, ambient noise cross-correlation functions (CCFs) are commonly used for time correction. When investigating recordings from a small-scale OBS network located on the Mohn’s mid-ocean ridge, we observed a remaining drift on the skew-corrected data. After recalculating the drift of the raw data using CCFs, we found that the skew-based time correction was incorrect. This was also verified with the observation of teleseismic P-wave arrivals. We describe a method to obtain properly time-corrected data and discuss the OBS timing issues in detail. The results shown were obtained using a software package that we developed for this specific purpose and made available as open-source software. Although we cannot explain the technical reason for the failure of skew correction, this study shows that skew corrections should not be trusted alone, and OBS timing should always be verified by either ambient noise correlations or P-wave arrival times.

Improving cross-correlations of ambient noise using an rms-ratio selection stacking method

SUMMARY Stacking of ambient noise correlations is a crucial step to extract empirical Green's functions (EGFs) between station pairs. The traditional method is to linearly stack all short-duration cross-correlation functions (CCFs) over a long period of time to obtain final stacks. It requires at least several months of ambient noise data to obtain reliable phase velocities at periods of several to tens of seconds from CCFs. In this study, we develop a new stacking method named root-mean-square ratio selection stacking (RMSR_SS) to reduce the time duration required for the recovery of EGFs from ambient noise. In our RMSR_SS method, rather than stacking all short-duration CCFs, we first judge if each of the short-duration CCF constructively contributes to the recovery of EGFs or not. Then, we only stack those CCFs which constructively contribute to the convergence of EGFs. By applying our method to synthetic noise data, we demonstrate how our method works in enhancing the signal-to-noise ratio of CCFs by rejecting noise sources which do not positively contribute to the recovery of EGFs. Then, we apply our method to real noise data recorded in western USA. We show that reliable and accurate phase velocities can be measured from 15-d long ambient noise data using our RMSR_SS method. By applying our method to ambient noise tomography (ANT), we can reduce the deployment duration of seismic stations from several months or years to a few tens of days, significantly improving the efficiency of ANT in imaging crust and upper-mantle structures.

Effects of islands and downslope seafloors on underwater noise in the northern South China Sea during a typhoon

The correlation of ambient noise with wind speed, and the depth dependence of ambient noise are both investigated, where the ocean noise data were recorded by a vertical line array in the northern South China Sea. It is shown that the correlation coefficients increase with increasing hydrophone depth during typhoon periods when the frequency ⩾ 250 Hz, which opposes the generally accepted knowledge that the correlation coefficients of noise level and wind speed decrease with increasing depth during non-typhoon periods. Particularly at frequencies of 250 Hz, 315 Hz and 400 Hz, the correlation coefficients increase by more than 0.05 at depths ranging from 155 m to 875 m. At the three frequencies, the average noise levels also increase with increasing depth during typhoon periods. It is suggested that these differences are attributed to the wind-generated noise in shallow waters and the effect of “downslope enhancement” to sound propagation. During typhoon periods, the surf breaking and surf beat upon the shores and reefs are strengthened, and the source levels are increased. The wind-generated noise in shallow waters interacts with the downslope sea floor, with the noise-depth distribution changed by a “downslope enhancement” effect promoting noise propagation.

Retrieval of amplitude and attenuation from ambient seismic noise: synthetic data and practical considerations

SUMMARY Ambient noise correlation has been used extensively to retrieve traveltimes of surface waves. However, studies of retrieving amplitude information and attenuation from ambient noise are limited. In this study, we develop methods and strategies to extract Rayleigh wave amplitude and attenuation from ambient noise correlation, based on theoretical derivation, numerical simulation, and practical considerations of real seismic data. The synthetic data included a numerical simulation of a highly anisotropic noise source and Earth-like temporally varying strength. Results from synthetic data validate that amplitudes and attenuations can indeed be extracted from noise correlations for a linear array. A temporal flattening procedure is effective in speeding up convergence while preserving relative amplitudes. The traditional one-bit normalization and other types of temporal normalization that are applied to each individual station separately are problematic in recovering attenuation and should be avoided. In this study, we propose an ‘asynchronous’ temporal flattening procedure for real data that does not require all stations to have data at the same time. Furthermore, we present the detailed procedure for amplitude retrieval from ambient noise. Tests on real data suggest attenuations extracted from our noise-based methods are comparable with those from earthquakes. Our study shows an exciting promise of retrieving amplitude and attenuation information from ambient noise correlations and suggests practical considerations for applications to real data.

Global‐Scale Full‐Waveform Ambient Noise Inversion

AbstractWe present the first application of full‐waveform ambient noise inversion to observed correlation functions that jointly constrains 3‐D Earth structure and heterogeneous noise sources. For this, we model and interpret ambient noise correlations as recordings of correlation wavefields, which completely eliminates the limiting assumptions of Green's function retrieval, such as equipartitioning and homogeneous random noise sources. Our method accounts for seismic wave propagation physics in 3‐D heterogeneous and attenuating media and also for the heterogeneous and nonstationary nature of the ambient noise field. Designed as a proof of concept, the study considers long periods from 100 to 300 s, thus focusing on the Earth's hum. Treating correlations as self‐consistent observables allows us to make separate measurements on the causal and acausal branches of correlation functions, without any need to choose one of them or form the average. We validate our approach by assessing the quality of the obtained models and by comparing them to previous studies. This work is a step toward the establishment of full‐waveform ambient noise inversion as a tomographic technique with the goal to go beyond ambient noise tomography based on Green's function retrieval.

Seismic monitoring in the Gugla rock glacier (Switzerland): ambient noise correlation, microseismicity and modelling

SUMMARY A network of seismometers has been installed on the Gugla rock glacier since October 2015 to estimate seismic velocity changes and detect microseismicity. These two processes are related to mechanical and structural variations occurring within the rock glacier. Seismic monitoring thus allows a better understanding of the dynamics of rock glaciers throughout the year. We observed seasonal variations in seismic wave velocity and microseismic activity over the 3 yr of the study. In the first part of our analysis, we used ambient noise correlations to compute daily changes of surface wave velocity. In winter, seismic wave velocities were higher, probably due to refreezing of the permafrost active layer and cooling of the uppermost permafrost layers, leading to increased overall rigidity of the medium. This assumption was verified using a seismic model of wave propagation that estimates the depth of P- and S-wave velocity changes from 0 down to 10 m. During melting periods, both a sudden velocity decrease and a decorrelation of the seismic responses were observed. These effects can probably be explained by the increased water content of the active layer. In the second part of our study, we focused on detecting microseismic signals generated in and around the rock glacier. This seismic activity (microquakes and rockfalls) also exhibits seasonal variations, with a maximum in spring and summer, which correlates principally with an exacerbated post-winter erosional phase of the front and a faster rock glacier displacement rate. In addition, we observed short bursts of microseismicity, both during snowfall and during rapid melting periods, probably due to pore pressure increase.