Session 4
However, the peculiarities of the acquired data demand the customization of signal processing techniques. If on the one hand traditional algorithms in geosciences are tailored to handle either the high spatial or the temporal resolution, on the other hand, the combination of high spatio-temporal acquisition throughput has fostered the widespread adoption of recent breakthroughs in Big Data analysis and advanced data analytics engines.
The session aims to highlight the innovation in classical methods/procedures and the recent technological advances on fibre optic sensing data analysis in any field of geosciences: seismology, volcanology, glaciology, geodesy, geophysics, natural hazards, oceanography, urban environment, geothermal applications, laboratory studies, large-scale field tests, planetary exploration, gravitational wave detection, fundamental physics.
Contributions dealing with processing, analysis and modelling for fibre optic sensing users are equally solicited. The overarching objective is to gather ingenious approaches in the application of the state-of-the-art algorithms in the geophysical field as well as recent cutting-edge techniques, such as High Performance Computing and Artificial Intelligence processes, with particular emphasis on Machine Learning models.
Invited speaker: Martijn Van den Ende (Université Côte d'Azur, France)
Distributed Acoustic Sensing (DAS) enables the recording of seismic wavefields with an unprecedented spatial resolution, of the order of several metres, and with a sampling frequency that is uniform in both time and space. In spite of these unique characteristics, DAS data are commonly processed trace-by-trace, essentially treating the instrument as a set of conventional seismometers. Hence, one can imagine that such processing routines do not achieve their full potential by not considering the spatial dimension of DAS. To address this, we propose detection and classification methods leveraging the spatio-temporal coherence of the data, encoded in the form of covariance matrices. Through the use of Riemannian geometry, we propose a framework for change-point detection in a stream of such covariance matrices, detecting signals based on their coherence rather than their energy. We extend this framework to facilitate clustering and classification, opening the door to unsupervised data exploration. We apply these methods to borehole and offshore DAS data, and we conclude with an outlook on how this framework can be merged with Deep Learning methods.
How to cite: van den Ende, M., Wang, X., Borsoi, R., Rivet, D., Ferrari, A., and Richard, C.: Leveraging the spatio-temporal coherence of DAS data for detection and classification, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-16, https://doi.org/10.5194/egusphere-gc12-fibreoptic-16, 2024.
Extensive monitoring initiatives driven by the urgency to address climate change have led to the rise of long-term projects, particularly in offshore environments. These projects are evolving into complex, multi-decadal operations, necessitating comprehensive monitoring. Distributed Acoustic Sensing (DAS) arrays offer unique advantages in long-distance, high-density, real-time monitoring. However, the long-term archiving of DAS data presents significant challenges, due to the need for vast storage capacities (on the order of hundreds of terabytes per year). Innovative data compression techniques are essential to make continuous high-sample-rate DAS data storage feasible.
DAS data is composed of multiple channels carrying highly correlated and coherent signals. These characteristics allow us to exploit inter-channel compression techniques, which leverage the signal from consecutive channels for prediction-based compression. Inter-channel compression methods achieve a much higher compression ratio compared to compressing each channel separately and have been little studied. In this work, we present novel inter-channel compression algorithms and demonstrate state-of-the-art lossless compression. For this purpose, a lossless coding scheme was implemented inspired by successful video coding techniques, following a pipeline composed of intra-prediction, inter-prediction, transform, and entropy coding. The implementation is divided into an encoder and a decoder. The encoder uses a bitrate optimization search and can be tuned for either speed or high-compression modes, while the decoder is optimized for quick signal reconstruction. The designed algorithms and the provided implementation facilitate the deployment of long-term DAS recording and archiving.
How to cite: Seguí, A., Ugalde, A., and Ventosa, S.: Inter-channel lossless data compression for Distributed Acoustic Sensing, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-5, https://doi.org/10.5194/egusphere-gc12-fibreoptic-5, 2024.
Machine learning models require large amounts of training samples: in most cases, thousands of unique instances are required for each class of events.
Constructing such a dataset is an extremely expensive task in terms of the time it takes to identify and label meaningful events.
Additionally, the large amount of data produced by DAS interrogator units makes storage prohibitively expensive: the mediums need to be both fast (storing one second of data should take less than one second), and of high capacity (spanning long periods of time, such that rarely-occurring events of interest can be captured). To alleviate these constraints, a typical approach is to only store "interesting" events. But how to know which events to store without having a classifier in the first place?
To solve this chicken-and-egg problem, we propose a framework for constructing datasets used to train classifiers of events detectable through the acoustic fingerprinting of DAS measurement data.
Our framework progressively builds up a dataset starting from one or more hard-coded anomaly detection rules (even as simple as energy thresholding) to solve the initial problem of managing limited storage space. This creates an intermediate dataset of unlabeled events.
The intermediate dataset is then evaluated by means of acoustic fingerprinting, which assigns a feature vector to each event. To facilitate further user input, the feature vectors are projected onto a two-dimensional space. Supervised clustering is performed on the projected representations, where users can select the granularity of the clustering process, and consequently the number of resulting intermediate classes.
By assigning labels to entire clusters instead of individual samples, the time required to annotate the dataset is significantly reduced.
We evaluate the outcomes of this framework on a dataset of more than 50000 events initially detected on an inner-city measurement line, and discuss the possibilities of developing more sophisticated models that can be bootstrapped from this approach.
How to cite: Dumitru, V., Strasser, L., and Lienhart, W.: A Framework for Bootstrapping Event Classification Datasets, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-18, https://doi.org/10.5194/egusphere-gc12-fibreoptic-18, 2024.
Due to its dense spatial sampling of the wavefield, Distributed Acoustic Sensing (DAS) holds appeal for various applications in seismology. However, it produces large and complex datasets that are practically unfeasible to analyze manually. Additionally, we often lack prior knowledge of expected signals and their source distribution. This leads to the necessity of new processing tools, particularly for early data exploration. To address this challenge, we leverage signal coherence to detect and characterize seismic sources recorded by DAS. We segment the dataset into short time windows and compute the coherence matrix for each window at all frequencies, followed by averaging across different frequency bands of interest. Subsequently, we employ non-negative matrix factorization to isolate sources and retrieve their time-dependent coefficients. The resulting features display locally well-defined regions of elevated coherence. Different frequency bands can be analyzed simultaneously, which helps discriminate between signals originating from similar locations. Additionally, there is no need to make any assumptions about the data prior to applying the processing workflow. We apply this methodology to an urban DAS dataset, demonstrating its capability to detect coherent signals in complex settings. Notably, one feature reveals stable and repetitive sources of surface waves suitable for time-lapse monitoring of the subsurface. This makes our approach particularly interesting for applications of seismic interferometry, where understanding source distribution is crucial. We further demonstrate the value of coherence-based methods by applying them to a volcano dataset. Eigendecomposition of coherence matrices enables the detection and characterization of various volcanic signals such as explosions, earthquakes, and tremor pulses.
How to cite: Grimm, J. and Poli, P.: Coherence-based methods for early data exploration of DAS data, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-61, https://doi.org/10.5194/egusphere-gc12-fibreoptic-61, 2024.
Detecting earthquakes in continuous seismic data is essential for various applications. Such applications include real-time natural or anthropogenic hazard monitoring and generating earthquake catalogues for body-wave tomography. Although numerous detection algorithms have been developed specifically for fibreoptic data, they are typically only applicable for certain fibre geometries, and generally struggle with P and S wave phase association. Furthermore, integrating fibreoptic and conventional seismometer data into current algorithms remains challenging. Here, we present an automated, faster-than-real-time earthquake back-migration method, adapted to incorporate arbitrary 3D fibre geometries and include all available seismic observations from any instrumentation. Crucially, the strength of this method lies in stacking energy based on physics-derived time-shifts, locating the event during the detection process. Unlike machine learning methods, it does not require a training dataset so is therefore readily applicable to new scenarios. We demonstrate the performance of the method to detect near-surface seismicity at an alpine glacier and crustal seismicity from the ongoing Sundhnúkur eruption near Grindavik, Iceland. We also show how subsequent automated phase-arrival refinement can provide sufficient quality picks for travel-time tomography. As we present these results, we highlight how fibreoptic sensitivity limitations can be quantified and accounted for during earthquake detection, how different data sources can be combined, and we identify areas where advances are yet to be made. Generic earthquake detection algorithms such as that presented here are essential for harnessing the dense spatial sampling that fibreoptic sensing provides, while at the same time accounting for fundamental fibreoptic sensitivity limitations.
How to cite: Hudson, T., Klaasen, S., Fontaine, O., Zunino, A., van Meulebrouck, S., Walter, F., Jonsdottir, K., and Fichtner, A.: Towards a generic fibreoptic earthquake detection and location algorithm for arbitrary fibre geometries and hybrid fibre-seismometer networks, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-26, https://doi.org/10.5194/egusphere-gc12-fibreoptic-26, 2024.
During 2018, a study was conducted to understand the response of new seismic instrumentation to the complex seismo-acoustic wavefield of Mt. Etna. The study consisted on deploying a multi-instrumental network at Pizzi Deneri (PDN) observatory, near the main craters of Mt. Etna. The multi-instrumental network comprised infrasound sensors, broad-band seismometers (BB) and a fiber optic cable buried within the local loosed scoria surface. The cable was connected to a Distributed Dynamic Strain (DDSS) interrogator. Part of the collected data reveals, what is believed to be, a case of a non-linear ground response from an air-to-ground coupling from an acoustic wave. An infrasound sensor registered the arrival of a signal from a volcanic explosion with a main frequency of ~2 Hz. Immediately, a BB and fibre optic virtual sensor (DDSS channel), co-located with the infrasound sensor, registered a signal linked to the acoustic arrival. However, the dominant frequencies captured by the BB and DDSS range between 15 and 20 Hz. To further study this phenomenon, a second experiment was conducted in 2019 in the same place (PDN) and using the same type of instrumentation, but in a different spatial arrangement. In a total of three months, we obtained more than 65000 examples of acoustic signals linked to volcanic explosions. Embedded in the examples, there are cases of non-linear ground response. However, the dataset also contains cases with no ground response triggered by acoustic signals. To understand which acoustic inputs could trigger the ground response, we performed a classification of the acoustic signals based on waveform similarity. In addition, to understand the resulted ground response, we extended the waveform similarity classification to the DDSS records to achieve spatial-temporal characterization of the phenomenon in study. The outcomes of this method allows us to understand the spatial effect of acoustic signals on the ground, monitor temporal variations, and discriminate between reliable data and DDSS signal artifacts such as saturation.
How to cite: Diaz-Meza, S., Jousset, P., Currenti, G., Costes, L., and Krawczyk, C.: Characterizing non-linear ground response using signal classification from acoustic signals and Distributed Dynamic Strain Sensing (DDSS) at Mt. Etna volcano, Sicily., Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-76, https://doi.org/10.5194/egusphere-gc12-fibreoptic-76, 2024.
An active experiment was held in Fürstenfeldbruck, Germany, to investigate and compare the performance of rotational sensors. The active sources were vibroseis truck sweeps and explosions. In this study we focused on monitoring the movement of a vibroseis truck using six-degree-of-freedom (6-DoF) measurements, which combine rotational sensors and seismometers. The investigation spanned from 20 November 2019, at 11:00 UTC, to 21 November 2019, at 14:00 UTC. Throughout this period, 480 sweep signals were emitted, each lasting 15 seconds and covering frequencies from 7 to 120 Hz. Sweeps were emitted at 160 different locations.
During the second day of measurements, SV-type of waves dominated at frequencies up to 60 Hz while SH waves were dominant between 60 Hz to 120 Hz. At lower frequencies, we estimate the direction based on a method that uses SV-type waves. The accuracy of estimates declined with increasing distance between the truck and sensors. To gain insights into this phenomenon, we scrutinized the wavefield itself. Upon separating the wavefield using a fingerprinting algorithm, we observed that only a small portion of the wavefield exhibited a strong dominance of SV waves. This is surprising as the source generated only SV waves. Thus, we were able to track the movement of the truck only after separating the wavefield and determine the portion corresponding solely on SV-type of waves.
How to cite: Izgi, G., Eibl, E. P. S., Krüger, F., and Bernauer, F.: How Does Wavefield Separation Affect Direction Estimates Using a Rotational Sensor?, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-66, https://doi.org/10.5194/egusphere-gc12-fibreoptic-66, 2024.
The analysis of signals acquired through Distributed Acoustic Sensing (DAS) technology offers an innovative method for seismic monitoring. However, owing to the high noise levels, the analysis of DAS data presents significant challenges in taking full advantage of dense temporal and spatial sampling. This is particularly true in accurately picking phase arrival times on DAS data.
Currently, some techniques have been proposed to address the picking problem on DAS data both from classical methods and through use of machine learning approaches, including the notable model named PhaseNet-DAS. Despite this, challenges persist, especially in real-time seismic monitoring applications and in the presence of high-frequency and high-density data.
In this context, we propose a novel model that leverages visual features and is based on the fundamental principles of Transformers, a class of Artificial Intelligence models, widely recognized for their ability to model complex relationships in sequential data. Our proposed model shows its effectiveness in learning seismic wave characteristics from DAS data, enabling an efficient phase picking.
To demonstrate the effectiveness of our approach, we present preliminary results on the use of our model to DAS data acquired in the seismically active area of Campi Flegrei caldera. The experimental results show the benefits of our method in exploiting DAS technology for enhancing seismic monitoring.
How to cite: Corsaro, M., Cannavò, F., Currenti, G., Palazzo, S., Allegra, M., Jousset, P., Prestifilippo, M., and Spampinato, C.: Seismic Phases Picking with Artificial Intelligence: A Novel Approach for Distributed Acoustic Sensing Data Analysis, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-82, https://doi.org/10.5194/egusphere-gc12-fibreoptic-82, 2024.
Distributed Acoustic Sensing (DAS) can provide unprecedented spatial resolution and sensitivity to a wide frequency band. The instrument response of the interrogated optical fibre cables, however, is largely unknown and difficult to separate from source, path, and directivity effects on seismic records. This prevents us from using DAS in many staple seismological techniques that require either absolute amplitude values or a complete understanding of the full waveform (e.g., earthquake magnitude estimation, waveform tomography).
Here we present a full-waveform simulation scheme developed to model the DAS instrument response using a particle-based Elastic Lattice Model (ELM-DAS). The scheme allows us to simulate a virtual cable embedded in the medium and made of a string of connected particles. By measuring the strain along these particles, we are able to replicate the axial strain natively measured by DAS as well as the effects of irregular cable geometries. Analysing synthetic DAS data allows us to focus on the main factors that are believed to determine the instrument response: cable-ground coupling and local site effects. The particle-based numerical scheme allows us to easily simulate complex properties of the material around the cable (e.g., unconsolidated sediments, nonlinear materials) as well as different degrees of cable-ground coupling.
By simulating DAS cables in 2D, we observe that at the meter scale, realistic DAS materials, cable-ground coupling, and the presence of unconsolidated trench materials around it dramatically affect wave propagation, each change affecting the synthetic DAS record, with differences exceeding at times the magnitude of the recorded signal. By expanding the scheme to 3D, we can accurately include the effects of realistic, complex–and at times sub-wavelength–cable geometries and how they influence DAS records. Our observations show that cable coupling and local site effects have to be considered both when designing a DAS deployment and analysing its data when either true or along-cable relative amplitudes are considered.
How to cite: Celli, N. L., Bean, C. J., and O'Brien, G.: Understanding DAS records and their response with full-waveform modelling, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-2, https://doi.org/10.5194/egusphere-gc12-fibreoptic-2, 2024.
Traditional rain detection methods are often constrained by their limited spatial coverage and lack of adaptability in diverse environmental conditions. This gap highlights the need for more advanced, adaptable solutions for rain detection. Recent advances in distributed acoustic sensing (DAS) have shown potential for overcoming these limitations. DAS transforms existing optical fiber cables into distributed sensornetworks, which detects external perturbations, such as vibrations, and acoustic pulses, along the fiber by utilizing the coherent Rayleighbackscattering. Each fiber segment with a length of spatial resolution can be considered a point sensor that enables continuous, real-time measurements across the entire route. Prior studies have demonstrated the potential of DAS in detecting and classifying rain intensity under controlled laboratory conditions. Building upon this research, further work expanded the application of DAS to real-world environments, employing pre-trained supervised Convolutional Neural Networks (CNNs) for field trials in rain detection and classification. However,existing research primarily focus on environments with consistent, labeled datasets and often struggle to adapt to variable environmental conditions. Using DAS, we propose a CNN with unsupervised domain adaptation technique that utilizes already-laid optical fiber networks for rain intensity classification. Specifically, firstly we collected the field data from a live telecommunication network using DAS underdifferent field environment, which was recorded over eight different days spanning five months. After filtering and feature extraction, a Deep Reconstruction Classification Network (DRCN) is implemented to concurrently minimize both the classification loss and reconstruction lossduring the learning process, aiming to capture a domain adaptive feature representa- tion applicable to new environmental conditions. Experimental results show that our approach effectively identifies rain status and adapts well to rain intensity classification across newdomains, addressing the gap left by machine learning methods in the context of unlabeled field data. This adaptability is crucial fordeveloping more accurate and reliable rain detection systems that capable of functioning effectively across various domains.
How to cite: Ding, Y., Shi, S., Tian, Y., Jiang, Z., Ozharar, S., Wang, T., and Moore, J.: Fiber Optic Sensing in Rain Detection Using Unsupervised Domain Adaptation, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-87, https://doi.org/10.5194/egusphere-gc12-fibreoptic-87, 2024.
The vulnerability of urban assets, including soils, buildings, and infrastructure, is influenced by human activities, environmental factors, and societal vulnerabilities. Leveraging Distributed Acoustic Sensing (DAS) technology deployed on telecom fiber networks, we have developed an information model that facilitates data extraction, exchange, and networking to aid decision-making regarding civil infrastructure assets. In collaboration with the French company APRR-AREA, we focus on a 25 km stretch of telecom optic fiber along the A40 motorway concession — known as the "autoroute des Titans" — in eastern France. Ensuring the control of vehicle weight is crucial for the safety of highway transportation systems as it helps assess structural fatigue and traffic impact on roadways. Our initial goal is to predict heavy vehicle weight, speed, and lane using DAS records. For this prediction task, we designed various convolutional neural network (CNN) architectures, which take a 2-dimensional DAS panel as input. The ground truth labels are provided by a dynamic axle weighing system. The data is collected over 17 346 vehicles with weights ranging from 1,660 to 73,123 kilograms, over a 68-hour DAS acquisition period. We first train the networks over purely synthetic datasets. The forward problem, used to produce the synthetic panels, involves a Flamant-Boussinesq approximation for predicting the quasi-static road deformation caused by passing vehicles with different speeds and weights in the two lanes of circulation. We model DAS strain rate signal accounting for signal variability due to vehicles deviations in their lanes and a spatially variable path of the optic fiber. Our synthetic experiments yield promising results in predicting vehicle weight, speed and lane, allowing to disentangle the influence of vehicle weight and distance on fiber data. Moving forward, we plan to further refine our model by training, after image segmentation, the CNN on the A40 DAS dataset. We anticipate that this study will underscore the potential of DAS technology in complementing dedicated instrumentation for enforcing load limits in areas with heavy traffic.
How to cite: Santos, T., Rodet, J., Amin Panah, M., Tauzin, B., Bodin, T., and Pittet, R.: A Deep Learning Neural Network for Controlling Vehicle Weight, Speed, and Lane from Telecom Dark Fiber Distributed Acoustic Sensing, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-40, https://doi.org/10.5194/egusphere-gc12-fibreoptic-40, 2024.
Managing large datasets in distributed acoustic sensing (DAS) presents significant challenges due to their terabyte-scale or larger size. Effectively addressing these challenges necessitates proactive measures to mitigate the impact of bad data before processing. Moreover, the dynamic nature of data quality in recorded channels requires continuous monitoring for efficient data cleaning.
This study proposes a novel approach building upon prior research to automatically detect and eliminate clusters of flawed channels. Leveraging a metric that identifies channels exhibiting dissimilarity from their neighboring data points, we then employ machine-learning-based anomaly detection techniques to establish a threshold for data exclusion. Our method incorporates user input to accommodate the variability of threshold selection across different use cases, aiming to enhance the likelihood of excluding data considered inadequate for subsequent processing.
This approach is applied to two distinct sets of DAS data collected from the Canary Islands. The first dataset spans two months in 2020, while the second dataset covers a period of six months between 2022 and 2023. While both datasets share substantial location overlap, one main difference is an updated interrogator for the second experiment. We demonstrate the capacity of our method to identify the evolution of channel quality over the acquisition of both datasets, thereby illustrating its efficacy in adapting to dynamic environmental and equipment-related factors.
How to cite: Rodríguez, T., Cubas Armas, M., Latorre, H., Ventosa, S., and Ugalde, A.: Enhancing Data Analysis in Distributed Acoustic Sensing through Implementation of a Channel Quality Index, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-11, https://doi.org/10.5194/egusphere-gc12-fibreoptic-11, 2024.
Distributed Acoustic Sensing (DAS), renowned for its high spatio-temporal resolution, offers a way to detect and identify a large variety of event sources hard to measure with conventional seismometers, as anthropogenic or environmental event sources. In view of the large amount of data generated by DAS networks, detecting and cataloging such events appears to be a challenge. To this end, machine learning is an opportunity and can help to automate this task. In this work, we implement a processing chain able to operate quasi-real-time classification of anthropogenic and natural seismic event sources.
Our approach works on continuous data streams to avoid dependency of a third-party detection algorithm. The preprocessing step consists in computing hand-crafted features that encapsulate the observed seismic signal characteristics into quantities suitable for source classification. These features include, amongst other, temporal and spatial standard deviation, kurtosis, skewness, temporal power spectrum density of the seismic signals, and cross-correlation and dynamic time warping of multiple seismic traces. The processing step performs classification tasks using the XGBoost machine learning algorithm. XGBoost quantifies the contribution of each feature and the certainty of the produced classification, which helps to speed up the processing chain using only discriminative features, and to reduce the false alarm rate. The post-processing step, Markov Random Field, accommodates spatial and temporal information redundancy.
We tested our proposed processing chain on two scales: locally, with tests in a controlled environment, and regionally, for real-field event detection. Both have been recorded with a FEBUS A1-R DAS. The first dataset was obtained on the FEBUS Optics test bench for simulated seismic anthropogenic sources over a 600m long fiber optic. The catalog includes six anthropogenic seismic sources denoted as footsteps, impacts, backhoe, compactor, and leaks. The second dataset contains cataloged earthquakes and quarry blasts events that were collected in the Pyrenees along a 91 km-long fiber optic cable between August 30 and September 20, 2022 with the support of TotalEnergies. The conducted tests show that features related to signal temporal content are enough to perform classification on the test bench and reach a F1-score of 84% for streamed data, and of 88% after application of the post-processing algorithm. The processing chain also shows its interest for real-field data analysis, as 12 of the 13 earthquakes of magnitude above 0.4 were correctly detected despite the natural and anthropogenic noises.
The promising outcomes achieved in both datasets indicate that the method is likely applicable to newly obtained data, but further data is needed to enhance the robustness of the algorithm. Working on these two datasets highlights the difficulties to work from events measured in controlled conditions to events acquired in the field. In particular, building a catalog from a continuous dataset is time-consuming and necessitates tools to identify events of interest. In future works, we aim at exploring the potential of self-supervised learning to help fasten the exploration of future newly acquired DAS datasets.
How to cite: Huynh, C., Hibert, C., Jestin, C., Malet, J.-P., and Lanticq, V.: Leveraging Hand-Crafted Features and Continuous DAS Data Stream for Automated Event Classification, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-28, https://doi.org/10.5194/egusphere-gc12-fibreoptic-28, 2024.
The technology of Distributed Optical Fibre Sensing (DOFS) has attracted a lot of attention for monitoring and studying geophysical processes, including earthquakes. This is partly because using DOFS earthquakes can be studied in greater detail, as Earth ground motions are captured on a synchronised network of hundreds of thousands of sensing locations along the fibre optic cable. Using DOFS, seismologists and other experts can monitor earthquakes with a sensing range spanning hundreds of kilometres and over harsh environments, e.g., across oceans.
These experimental studies are typically carried out for long time-periods, and therefore, are known to generate large amounts of data. At the same time, seismic signals captured by DOFS are typically of lower signal-to-noise ratio in comparison to the conventional seismic sensors. Hence, devising effective signal processing and machine learning technologies that can be used to process DOFS signals has become critical for fibre optic seismology studies. Of particular importance are regions whereby naturally occurring earthquakes may be rare, but there is nevertheless the need to deploy capable earthquake detection and classification systems using DOFS.
In this study, we demonstrate the concept of detecting earthquakes, using hundreds of spatiotemporal images, as obtained using a commercial DOFS system. Approximately two months of data were collected using the 55 km subsea fibre optic cable, which is located off the coast of Muroto in Japan. Each of the 9800 locations along the fibre optic cable was sampled at a rate of 500 Hz. Although earthquake events have been historically shown to occur frequently in this region, we demonstrate a monitoring capability that does not require a priori “labels” in constructing an earthquake detection model. The abundance of non-earthquake data generating sources in the region, e.g., fishing vessels, marine life, and ambient noise, allowed us to compute representations that are specific to those sources. Consequently, the amount of deviation from these representations can be calculated on new spatiotemporal images and be used to flag potential seismic activity. A Variational Autoencoder was used to obtain the representations of non-earthquake sources. Our earthquake detector correctly identifies more than 90 % of available earthquakes in our dataset using metrics such as the reconstruction error.
How to cite: Matthaiou, I., Masoudi, A., and Brambilla, G.: Variational autoencoders in earthquake detection and processing for submarine distributed optical fibre sensing, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-32, https://doi.org/10.5194/egusphere-gc12-fibreoptic-32, 2024.
The EU-INFRATECH funded SUBMERSE project will establish continuous monitoring of several oceanic telecom cables for landing sites in Portugal, Greece, and Svalbard. We develop tools for the automated analysis of these upcoming data sets as well as other submarine DAS data. We try to leverage DeepLab v3, a Deep Neural Network (DNN) architecture for semantic segmentation, to train a machine learning model for earthquake detection and P and S wave picking using submarine Distributed Acoustic Sensing (DAS) data. The input data being two-dimensional necessitated the adoption of the DeepLab v3 model, known for its superior performance in image segmentation. The complexity of submarine DAS data, characterized by diverse ocean noise environment and levels, as well as varying parameters such as cable length, shape, channel spacing, deployment environment, and location, led us to employ a larger model for effective earthquake detection.
Given the scarcity of submarine DAS seismic records, we adopted a strategy to pre-train our model using terrestrial DAS seismic records before fine-tuning it with submarine DAS records including Madeira Island, Svalbard Island, Chile and Greece coast. This approach aimed to leverage the abundant and diverse seismic event data available from land-based DAS system to establish a robust base model. Subsequently, the model was fine-tuned using the scarcer, yet critically important, submarine DAS data to adapt its earthquake detection capabilities to the unique characteristics and challenges presented by the submarine environment. This two-step training process allowed us to efficiently exploit the available data resources, ensuring that our model benefited from a broad learning base while achieving specialized performance for submarine earthquake detection.
Our results demonstrate the model's robust ability to identify seismic events and label P and S waves accurately. For three-component seismometer data it is generally assumed, that distinction of P and S waves relies primarily on the polarization of the arrivals. Our model's capacity to recognize P and S waves in single-component DAS data is therefore intriguing. We conducted tests under various scenarios to understand how the model discriminates between P and S waves: inverting the sequence of P and S waves did not affect identification performance, suggesting that order does not play a role. Even when the input consisted solely of P or S waves, the model could still identify them, indicating the identification is not based on their simultaneous or nonsimultaneous appearance. Amplifying the P wave's amplitude by five or ten times—surpassing the S wave's amplitude—still allowed for discrimination of P and S waves, albeit with diminished accuracy, highlighting that amplitude information is significant. However, our model also learned additional aspects we have yet to understand, suggesting it captures more complex patterns
How to cite: Xiao, H. and Tilmann, F.: DeepDAS: An Earthquake Phase Identification Tool Using Submarine Distributed Acoustic Sensing Data, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-51, https://doi.org/10.5194/egusphere-gc12-fibreoptic-51, 2024.
Among the strengths of Distributed Acoustic Sensing (DAS) applications, the sensing of existing fibre optic cable has certainly aided its extensive adoption in all fields, without excluding the geophysical domain. As a matter of fact, the high-quality of data recording, at high spatial-temporal resolution, has enabled the detection of a variety of seismic-volcanic events, especially in poorly or not at all instrumented environments.
In densely populated areas, the seismic exploration through the deployment of traditional seismic arrays would require high maintenance at prohibitive costs. In contrast, the interrogation of commercial fiber optic infrastructure through DAS technology results in minimal intrusion into urban life in a cost-effective but equally efficient manner.
However, data collection in urban contexts has to deal with unavoidable human interferences that frequently corrupt the seismic signal with anthropogenic noise. Indeed, ground vibrations induced by transportation, industrial and construction activities significantly reduce the signal-to-noise ratio by masking target events.
With the purpose of cleaning the DAS signal, anthropogenic noise removal has been approached with a deep learning technique. Both the neural network architecture and the ad-hoc training procedure have been developed with the aim of maintaining the meaningful features of the original recordings while removing man-induced noise.
Promising validation results on real low-frequency seismic events, detected during the 2021 Vulcano Island unrest, highlight the potential of the suggested method as a pre-processing step to help with the subsequent DAS signal analysis.
How to cite: Allegra, M., Cannavò, F., Corsaro, M., Currenti, G., Jousset, P., Palazzo, S., Prestifilippo, M., and Spampinato, C.: A Deep Learning Approach for Denoising DAS data in urban volcanic areas, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-81, https://doi.org/10.5194/egusphere-gc12-fibreoptic-81, 2024.
Utilizing Seafloor Fiber Optic Cables with Distributed Acoustic Sensing (DAS) provides a cost-effective alternative to traditional cabled ocean bottom seismic networks by offering seismic data with high-density and extensive coverage near seismic sources. This innovative method holds great potential for advancing offshore monitoring systems aimed at hazard mitigation and enhancing our understanding of earthquake mechanics. In this study, we present a methodological approach for back-projection imaging of earthquake ruptures using DAS data from the ocean floor off the coast of Chile. Our approach leverages the unique characteristics of DAS data to significantly improve the resolution and precision in estimating the source parameters of local earthquakes.
Our methodology encompasses several steps to optimize back-projection performance. We employ spatial integration to convert DAS strains into displacements, mitigating the adverse effects of wave scattering on waveform coherence. We enhance travel time accuracy through shallow-sediment time corrections and utilize array processing on overlapping cable segments (sub-arrays) to determine apparent slowness. The collective information from all sub-arrays is then used to localize earthquakes employing a 1D local velocity model.
Through extensive synthetic testing with the 120-km-long cable configuration off Chile's coast, we identified a "high-precision, high-resolution source region," less affected by velocity structure uncertainties. This region spans approximately 80 km laterally from the cable and reaches depths of up to 15 km, likely due to optimal signal focusing from various angles, extendable with increased cable length. Our method applied to data from around 50 local earthquakes with magnitudes ranging from 1.5 to 3 consistently yields sharp back-projection images with high spatial accuracy, within 1 to 4 km, for earthquakes within this defined region, comparable to seismic catalog location uncertainties.
Our approach's real potential lies in its capacity to image the rupture process of larger earthquakes. Applying our method to synthetic waveforms of a magnitude 7 earthquake constructed from multiple empirical Green's functions, we demonstrate that strong coda waves do not hinder the precise detection and localization of subsequent sub-sources, provided travel time calibration is applied. The rupture speeds and locations of sub-sources are accurately recovered, even for concurrent multiple sources. Ongoing enhancements to travel time calibration aim to further increase location accuracy and resolution, including waveform alignment with static calibration, 3D velocity model travel time tables, and slowness bias measurements and calibrations for each source-subarray pair. Together, these improvements will boost the resolution and accuracy of our method, alongside more sophisticated back-propagation methods for individual arrays. Our work shows promise for earthquake and tsunami early warning development, contingent upon effectively addressing the amplitude saturation issue of DAS data.
How to cite: Xie, Y., Ampuero, J.-P., van den Ende1, M., Trabattoni, A., Baillet, M., and Rivet, D.: Advancing Earthquake Rupture Imaging through Ocean Bottom Distributed Acoustic Sensing Data, Galileo conference: Fibre Optic Sensing in Geosciences, Catania, Italy, 16–20 Jun 2024, GC12-FibreOptic-59, https://doi.org/10.5194/egusphere-gc12-fibreoptic-59, 2024.
