Fiber optic distributed acoustic sensing and classification

Abstract

A fiber optic distributed sensing system and method. The fiber optic distributed sensing system includes: an optical fiber extending through an underwater gas field; an interrogator connected to the optical fiber so that light transmitted through the optical fiber is capable of being sensed by the interrogator; and a computer system configured to obtain sensor data captured by the interrogator and determine an amount or a characteristic of gas within the underwater gas field based on the sensor data. The method includes: capturing sensor data using a fiber optic distributed sensor extending through an acoustic field; performing a wavelet transform on the sensor data in order to obtain spatiotemporal data; and matching the spatiotemporal data to an event template of a predetermined set of event templates in order to identify a spatiotemporal event.

Description

FIBER OPTIC DISTRIBUTED ACOUSTIC SENSING AND CLASSIFICATIONGOVERNMENT FUNDING

[0001] This invention was made with government support under 2022716 awarded by the NationalScience Foundation. The government has certain rights in the invention.TECHNICAL FIELD

[0002] This disclosure relates to fiber optic sensing of spatiotemporal events and, more particularly, to fiber optic sensing of acoustic waves / vibrations and processing sensor data thereof.BACKGROUND

[0003] At most volcano observatories, continuous seismicity and ground deformation are monitored through ground-based sensors (e.g., [1]) and remote sensing approaches (e.g., [2]). However, besides the mechanical behavior, continuous volcanic gas release is another fundamental process to investigate (e.g., [3]). Volcanic gases are a main trigger of explosive eruptions. Still, the composition and flux of a fumarole, while highly relevant for interpreting volcanic unrest’s evolution (e.g., [4]), are generally only periodically determined from discrete direct sampling surveys (e.g., [5]) or through satellite images (e.g., [6]). Furthermore, most measurements are performed on known gas plumes, but gas can still escape diffusively through crustal faults [7], soil or water bodies [8, 9], even during quiescence [8], Consequently, diffuse volcanic gas emissions remain largely under-sampled at many volcanoes, and labor-intensive fieldwork is required to constrain them.

[0004] Of the dominant volcanic gases (CO2, SO2, H2O), CO2 seeping through the sub-surface is particularly challenging to detect due to its high concentration already present in the atmosphere [10, 11]. Therefore, global volcanic CO2 emissions are most likely underestimated [7, 12], The dynamics of the ideal subaerial volcanic tracer, SO2, remain hard to connect to source processes due to turbulent interactions in the atmosphere

[0013] , Besides, magmatic H2O is also challenging to detect as it is buffered by various processes in the subsurface and the atmosphere [8],

[0005] The presence of a large body of water nearby or within a volcano provides an underexploited opportunity to investigate volcanic degassing through non-conventional means. Contrary to subaerial settings, the acoustic detection of gas is straightforward in aquatic environments [14-16], This is because nucleating, oscillating, and collapsing gas bubbles are powerful acoustic sources which allow hydrophones to track gas fluxes at high temporal resolution before mixing and dilution in the atmosphere [17, 18], similar to hydrocarbon gas plume monitoring [19, 20], Yet, hydrophones provide single-point measurements, and acoustic waves quickly attenuate within a distance of a few meters from the point source [15, 21], limiting their monitoring capabilities. Therefore, degassingat non-measured locations is often unknown, and models rely on wide interpolation. This also limits the reliability of CO2 output estimates from volcanoes that remain today inaccurately known

[0022] .

[0006] There is no gas or other low-strength acoustic mechanism that is able to monitor large areas continuously in order to detect minor acoustic waves generated by small, but important events, such as gas bubble nucleation and detachment from a sediment bed on a seafloor or other underwater bed or floor, providing information regarding gas bubble size or estimating amount of gas released by a seafloor or other underwater floor or bed, for example.SUMMARY

[0007] In accordance with a first aspect of the invention, there is provided a fiber optic distributed sensing system. The fiber optic distributed sensing system includes: an optical fiber extending through an underwater gas field; an interrogator connected to the optical fiber so that light transmitted through the optical fiber is capable of being sensed by the interrogator; and a computer system configured to obtain sensor data captured by the interrogator and determine an amount or a characteristic of gas within the underwater gas field based on the sensor data.

[0008] The fiber optic distributed sensing system of the first aspect may further include any of the following features or any technically-feasible combination of two or more of the following features:- the computer system is configured to, as a part of determining the amount or the characteristic of gas being released, perform a wavelet transform on the sensor data to obtain spatiotemporal data;- the computer system is configured to, as a part of determining the amount or the characteristic of gas being released, compare the spatiotemporal data to template data;- the computer system is configured to, as a part of determining the amount or the characteristic of gas being released, use a convolutional neural network (CNN) to reduce dimensionality of the spatiotemporal data;- the computer system is configured to, as a part of determining the amount or the characteristic of gas being released, determine a gas bubble size based on a dominant frequency extracted from the sensor data;- the computer system is configured to, as a part of determining the amount or the characteristic of gas being released, use template matching in order to detect gas bubble events based on the sensor data;- the template matching includes performing a waveform similarity search based on the sensor data;- the computer system configured to determine a gas bubble type as the characteristic of gas within the underwater gas field;- the optical fiber includes a sensor portion that extends for a distributed sensor length and includes the plurality of linear lengths, and wherein the plurality of linear lengths extend for at least one half of the distributed sensor length;- the computer system configured to determine an amount of carbon dioxide gas as the amount of gas within the underwater gas field; and / or- comprising the fiber optic distributed sensing system is installed at an underwater volcanic site.

[0009] In accordance with another aspect of the invention, there is provided a method of detecting spatiotemporal events within an acoustic field using a fiber optic distributed sensing system. The method includes the steps of: capturing sensor data using a fiber optic distributed sensor extending through an acoustic field; performing a wavelet transform on the sensor data in order to obtain spatiotemporal data; and matching the spatiotemporal data to an event template of a predetermined set of event templates in order to identify a spatiotemporal event.

[0010] The method of detecting spatiotemporal events within an acoustic field using a fiber optic distributed sensing system may further include any of the following features or any technically- feasible combination of two or more of the following features- further comprising the step of: using a convolutional neural network (CNN) to reduce dimensionality of the spatiotemporal data;- the acoustic field is an underwater gas field, and wherein the fiber optic distributed sensor includes an optical fiber that extends through the underwater gas field;- the spatiotemporal event is a gas bubble event, wherein the gas bubble event indicates a time and a location of a gas bubble; and / or- the gas bubble event further indicates a gas bubble size or gas bubble type selected from a predetermined set of gas bubble types.

[0011] In accordance with a second aspect of the invention, there is provided a method of continuously monitoring a given site using a fiber optic distributed sensor. The method includes the steps of: installing a fiber optic distributed sensor at a target location, wherein the fiber optic distributed sensor includes an optical fiber and an interrogator; capturing sensor data using the fiber optic distributed sensor extending through an acoustic field in order to obtain spatiotemporal data; and determining a spatiotemporal event based on correlations between the captured sensor data and predefined template matching data.

[0012] The method of continuously monitoring a given site using a fiber optic distributed sensor may further include any of the following features or any technically-feasible combination of two or more of the following features- further comprising using template matching in which the spatiotemporal data pertaining to the spatiotemporal event is matched to an event template of a predetermined set of event templates in order to detect the spatiotemporal event;- the target location is underwater at a seafloor or other underwater bed, and wherein the optical fiber is located at the seafloor or other underwater bed; and / or- the acoustic field is an underwater gas field, and wherein the fiber optic distributed sensor includes an optical fiber that extends through the underwater gas field.BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Illustrative embodiments will hereinafter be described in conjunction with the appended drawings, wherein:

[0014] FIG. 1 is a block diagram depicting a fiber optic distributed sensing system with an optical fiber extending through an underwater acoustic field and an underwater gas field, according to one embodiment;

[0015] FIG. 2 is a schematic top-down plan view of the fiber optic distributed sensing system installed in a body of water, according to one embodiment;

[0016] FIGS. 3 A and 3B illustrate an experimental embodiment of the fiberoptic distributed system carried out at Laacher See in Germany, according to one embodiment;

[0017] FIGS. 4A-E depict graphs of exemplary waveforms manually extracted;

[0018] FIG. 5 depicts graphs of multi-channel waveform data, power spectral density (PSD) over frequency, and individual frequency spectra for each channel, illustrating how bubble event signals vary in waveform and frequency content across channels, attenuating and losing high-frequency components with distance from the seep, according to one embodiment;

[0019] FIGS. 6A-C depict graphs for matched waveforms amongst channels and an average correlation coefficient, according to one embodiment;

[0020] FIGS. 7A-H depict graphs of showing centroid matrices and multiple events for each cluster, according to one embodiment;

[0021] FIG. 8 shows a plan view depth graph and spacing of an optical fiber installed within a body of water, according to one embodiment;

[0022] FIG. 9 depicts the channels recording the earliest arrival of a tap event, which were identified and linked to their corresponding tap-test locations, according to one embodiment;

[0023] FIG. 10 depicts a flowchart illustrating an embodiment of a method 200 of detecting spatiotemporal events within an acoustic field using a fiber optic distributed sensing system, according to one embodiment;

[0024] FIG. 11 shows test results with the left panel showing the elbow test of the K-means algorithm with original large dimension spectrograms as the input, and the right panel shows theelbow test of the K-means algorithm with reduced dimensionality matrices by CNN as the input, according to one embodiment;

[0025] FIG. 12 depicts waveforms of a large bubble along channels 140-149, with the smooth traces being waveforms bandpass fdtered between 2-10 Hz for each raw waveform thereabove, according to one embodiment; and

[0026] FIG. 13 depicts template matching results of Zone 1 by small batches of templates, with detection numbers being represented as a function of the number of templates used, according to one embodiment.DETAILED DESCRIPTION

[0027] The system and method described herein enables determining a spatiotemporal event based on sensor data captured from a fiber optic distributed sensor extending through a gas field (and an acoustic field) in order to obtain spatiotemporal data. Further, at least in embodiments, the system and method enables classification and / or quantization of spatiotemporal events, such as, for example, classification indicating a bubble size and / or a particular nucleation process or detail and quantization indicating a gas volume (e.g., gas volume released by the seafloor in a particular zone or region). According to embodiments, the system provided herein is characterized as a fiber optic distributed sensing system, and includes an optical fiber extending through an underwater acoustic field, an interrogator connected to the optical fiber so that light transmitted through the optical fiber is capable of being sensed by the interrogator, and a computer system configured to obtain sensor data captured by the interrogator and to determine an amount of gas being released within the underwater acoustic field based on the sensor data.

[0028] According to embodiments, there is provided a method of detecting spatiotemporal events within an acoustic field using a fiber optic distributed sensing system and, at least in some embodiments, this method may be used for detecting an amount of gas being released within the underwater acoustic field based on fiber optic distributed sensor data. This method includes: capturing sensor data using a fiber optic distributed sensor extending through an acoustic field; performing a wavelet transform on the sensor data in order to obtain spatiotemporal data; and matching the spatiotemporal data to an event template of a predetermined set of event templates in order to identify a spatiotemporal event. The previously-discussed method, or another method, of detecting spatiotemporal events within an acoustic field using a fiber optic distributed sensing system may include other steps, including, for example, installing the optical fiber and the interrogator as a fiber optic distributed sensor at a target location, such as crater lakes formed in volcanic craters or calderas.

[0029] The system and method provided herein enable continuous underwater gas monitoring for a large area, such as, for example, monitoring underwater volcanic degassing. The presence of alarge body of water nearby or within a volcano provides an underexploited opportunity to investigate volcanic degassing, and embodiments discussed herein are directed toward such underwater degassing monitoring; however, it will be appreciated by those skilled in the art that the system and method discussed herein are applicable to various applications using fiber optic sensing of acoustic wave s / vibrations .

[0030] Discussion of one or more embodiments of the system and method are discussed below, with reference to FIGS. 1-10 and the Appendix being submitted herewith, which is hereby incorporated in its entirety as it is part of the specification. Any reference to a FIG. having a number beginning with the character “S” may be found in the Appendix.

[0031] With reference to FIGS. 1 and 2, there is shown an embodiment of a fiber optic distributed sensing system 10 having an optical fiber 12 extending through an underwater acoustic field A and an underwater gas field G, an interrogator 14 connected to the optical fiber 12 so that light transmitted through the optical fiber 12 is capable of being sensed by the interrogator 14, and a computer system 16 configured to obtain sensor data captured by the interrogator 14 and determine an amount of gas being released within the underwater gas field G based on the sensor data.

[0032] In particular, the illustrated embodiment shown in FIG. 1 depicts the optical fiber 12 positioned underwater at a seafloor (or underwater floor) F of a body of water W. The body of water W may be a lake, sea, ocean, or other body of water. In one embodiment, the body of water W is next to a volcano and gas rising from the seafloor F is monitored, which may provide information regarding volcanic activity, for example.

[0033] The optical fiber 12 is comprised of a transmissive core surrounded by a cladding layer, and may further include a sheathing so as to provide protection. The optical fiber 12 receives light from the interrogator 14 and this light travels through the transmissive core of the optical fiber 12 from an interrogator end 20 to a free end 22 of the optical fiber 12. The optical fiber 12 used may be selected according to particulars suited for the environment or implementation that the fiber optic distributed sensing system 10 is to be used. It will be appreciated that the optical fiber 12 is depicted in FIG. 1 for illustrative purposes and those skilled in the art will appreciate that the optical fiber 12 may be a very long cable (z.e., 100 meters or more) and routed according to various configurations, very often dependent on the particular application for which the fiber optic distributed sensing system 10 is being used.

[0034] According to one embodiment, the optical fiber 12 is a 3.2 millimeter (mm) diameter double-coated tactical armored cable with a central metal tube containing one single optical fiber strand. This optical fiber 12 in one particular embodiment has a net weight of 10.5 kilograms (kg) per kilometer (km), meaning its density is slightly higher than freshwater, suitably causing the optical fiber 12 to naturally sink to the seafloor F. According to embodiments, the optical fiber 12 is fixed to anchor points 26. For example, in one embodiment where the fiber optic distributedsensing system 10 is used to monitor underwater degassing near a volcano, professional divers install the optical fiber 12, situating the optical fiber 12 within the underwater gas field F at a selected position, which in this example corresponds to a subsurface underwater floor position whereat the optical fiber 12 rests under a few centimeters of alluvial sediments, as this configuration improves ground coupling. At the free end 22 of the optical fiber 12, an optical damper 24 may be installed so as to prevent or at least mitigate undesirable laser / light pulse reflections.

[0035] The optical fiber 12, being optically coupled to the interrogator 14, enables the interrogator14 to use its light sensor 18 to capture light, particularly backscattered light, and obtain information about the light, which is then used to determine characteristics of the underwater gas field G. Together, the optical fiber 12 and the interrogator 14 (including its light sensor 18) constitute a fiber optic distributed sensor 19. The light sensor 18 of the interrogator 14 refers to a light sensing device that is able to capture information about light, such as intensity of light. As used herein, the term “light” (including when in its various forms) refers to visible light, near-infrared (NIR) light, infrared (IR) light, and other suitable electromagnetic waves employed for achieving a suitable Rayleigh backscatter profile, as the particular waves employed vary depending on the application’s specific requirements, the type of fiber used for the optical fiber 12, and desired performance characteristics.

[0036] The light sensor 18 is used to capture information about light, and is or includes one or more photodetectors or photodiodes. Generally, the optical fiber 12 is optically coupled to the light sensor 18, which is located at a linear length L and a cord length C from a given point P along the optical fiber 12. In order to determine a location at which the backscattered light BL relates to (was excited from), the interrogator 14 uses time of flight, which is the time difference between transmission of a light pulse and receipt of backscattered light resulting from transmission of the light pulse through the optical fiber 12. At least in embodiments, sensor data captured by the light sensor 18 is grouped based on segments of a fixed length, such as 1 meter, and each such group may be referred to as a channel, such as where consecutive channels represent consecutive segments along the optical fiber 12. The final output at each channel is continuous phase, strain, strain-rate, or velocity based on the manufacturer’s design.

[0037] With reference to FIG. 2, and with continued reference to FIG. 1, the interrogator 14 is a distributor acoustic signal (DAS) interrogator, at least according to embodiments, and its light sensor 18 is used to capture sensor data derived from light transmitted through the optical fiber 12, as discussed above. In one embodiment, the interrogator 14 is a Silixa™ iDAS interrogator and, in an embodiment discussed below regarding an experimental proof-of-concept system, this interrogator was used to obtain data saved in native measurement units (proportional to strain rate) at 5,000 and 8,000 Hz for a duration of 18 and 22 hours (with 40 minutes pause time), respectively.

[0038] The computer system 16 includes at least one processor and memory having computer instructions that are executable by the at least one processor. The computer instructions, whencarried out by the at least one processor, cause the computer system 16 to perform its functionality as discussed herein. In embodiments, the computer system 16 is configured to perform a method of detecting spatiotemporal events within an acoustic field using a fiber optic distributed sensing system, or at least steps included as a part thereof. For example, in one embodiment, the computer system 16, which is given access to sensor data captured by the light sensor 18 of the interrogator 14, determines a spatiotemporal event by performing a wavelet transform on the sensor data captured by the interrogator 14, which is used to obtain spatiotemporal data. The spatiotemporal data may then be clustered to obtain spatiotemporal data that is of a reduced dimension and that may be referred to as clustered spatiotemporal data.

[0039] The computer system 16 may be comprised of one or more computers (each having a processor and memory) that may be co-located and / or remotely located. For example, with reference to FIG. 2, the computer system 16 includes a first computer subsystem 16a and a second computer subsystem 16b, which is shown as being located away from the first computer subsystem 16a. The first computer subsystem 16a and the second computer subsystem 16b communicate with one another via an interconnected data network 30, which may be a global or regional electronic data communications infrastructure, such as the Internet, comprising a plurality of interconnected devices, generally a vast array of such devices, operating in order to enable exchange and transmission of data across geographically-dispersed (or remote / non-co-located) locations. In embodiments, the computer system 16 includes an interferometer that concerts backscattered light measurements (raw sensor data from the interrogator 14) into time series or strain-rate time series data, and the interferometer may constitute or be a part of the first computer subsystem 16a.

[0040] According to at least some embodiments, the computer system 16 employs template matching, where spatiotemporal data pertaining to the spatiotemporal event, such as un-clustered or clustered spatiotemporal data, is matched to an event template of a predetermined set of event templates in order to detect and identify the spatiotemporal event. Thus, template matching results in the detection and classification of a spatiotemporal event when the spatiotemporal data is matched to one of a plurality of event templates. For example, the spatiotemporal event is matched to a predetermined event template that is associated with a gas bubble event that indicates a gas bubble size or gas bubble type selected from a predetermined set of gas bubble types. Further, the sensor data (including its derivations thereof such as the spatiotemporal data) indicates a time and a location of the event or, here, gas bubble, which provides useful information when considered with physical properties of the event (e.g., size) that are determined through template matching, for example.

[0041] In addition to attributes such as gas bubble size, other gas bubble physical properties may be determined, such as a gas bubble nucleation / sediment detachment type, which may more generally be referred to as a gas bubble type. One or more, and in many instances a plurality of, predetermined gas bubble types are determined based on experimental observations and empiricaldata, for example. Provided below is a discussion of an experimental exemplary implementation of a fiber optic distributed sensing system, including an example of a process for determining a set of predetermined gas bubble types.

[0042] Experiment at Lacicher See. FIGS. 3A and 3B illustrate an experimental embodiment that was carried out at Laacher See in Germany. FIG. 3A illustrates a map of the Laacher See volcano, including a fiber optic distributed sensing system 10'. FIG. 3B shows a zoomed-in portion of Laacher See at which the fiber optic distributed sensing system 10' is installed. A 500 meter (m) optical fiber 12' was installed by professional divers under a few centimeters of alluvial sediments. The subsections or zones of the cable (having optical fiber 12') are identified in FIG. 3B, with: a first subsection (or length or zone) labeled as Zone 1 and extending between land and the first dashed line crossing the optical fiber path (or a “first optical fiber reference position”); a second zone labeled as Zone 2 and extending between a second optical fiber reference position and a third optical fiber reference position; and a third zone labeled as Zone 3 and extending between the third optical fiber reference position and a fourth optical fiber reference position. These three zones correspond to three zones with enhanced gas emissions (same as FIGS. 6A-6C). The hydrophone location inside the fiber optic loop is also shown in FIG. 3B. Each dot (white circle with black outline, labeled as CO2 gas seep) represents a CO2 gas seep identified by Goepel et al.

[0034] , and it is noted that the bathymetric map is from Goepel et al.

[0034] .

[0043] A 500-m fiber-optic cable (the optical fiber 12') was immersed in the southeast region of the lake (Laacher See, as depicted in FIGS. 3A and 3B) where CO2 gas seeps were previously identified [34, 36; FIGS. 3A and 3B], The cable is a 3.2 mm diameter double-coated tactical armored cable with a central metal tube containing one single optical fiber strand. This cable has a net weight of 10.5 kg / km, meaning its density is slightly higher than fresh water, and it naturally sinks to the lake bottom. Professional divers set the cable at maximum depths of ~25 meters under a few centimeters of alluvial sediments, similar to what is shown in FIG. 1, to guarantee a better ground coupling. An optical damper was installed at the end of the cable to decrease unwanted laser reflections. The georeferencing of the cable (having the optical fiber) is described below in the Methods section.

[0044] The fiber was then interrogated with a Silixa™ iDAS interrogator installed in a near-shore house, and data was saved in native measurement units (proportional to strain rate) at 5000 and 8000 Hz for a duration of 18 and 22 hours (40 minutes pause time), respectively. The data was collected at a I-m spatial sampling with a 10-m gauge length for the optical fiber 12' with a total of 640 recording channels, where the channels 0-125 are inside the DAS interrogator 14' and function as the reference section, channels 126-140 are inland, and channels 141-190 are near the lake edge with uncertain coupling conditions and extensive optical noise, leaving 480 channels situated underwater and acting as hydroacoustic sensors.

[0045] In addition, two hydrophones were installed at the center of the loop (FIGS. 3A and 3B).The microphones (SNAP from Loggerhead company) recorded the ambient acoustic noise at 44. 1 kHz. The gain level was set to the minimum based on the noise level recorded during previous experiments

[0036] ,

[0046] Visual inspection of the records revealed the presence of countless impulsive events continuously occurring throughout the recording period (FIG. 4A). Both DAS and hydrophone data show that these acoustic signals hold the characteristic signatures of bubbles rising to the surface [18, 37], and these signals characteristic of bubbles are referred to as “bubble events.” FIG. 4A is for a 15 -meter record section over a 15 -second period recorded with DAS. The amplitudes of the signals in the dashed-line boxes of FIG. 4A are multiplied by a factor of 10 and 5, respectively. FIG. 4B shows typical monochromatic bubble acoustic signals recorded with DAS. FIG. 4C shows typical monochromatic bubble acoustic signals recorded with a hydrophone. FIGS. 4D and 4E show typical multimodal bubble acoustic signals recorded with DAS.

[0047] Over the wealth of waveforms observed, three main classes of bubble events stand out(FIGS. 4B, D, and E). FIGS. 4B and C (DAS vs. hydrophone) show a typical oscillating lightly damped bubble -acoustic signal. It reaches its maximum amplitude in the first milliseconds and then quickly decays as a classical damped oscillator

[0038] , These bubble-acoustic signals reveal one dominant frequency peak, generally between ~ 100 and ~2000 Hz. The largest and clearest events have a duration of up to ~100 ms, but the vast majority have a duration of ~30 ms. Besides, other bubble events often start with a high-frequency impulsive onset (~kHz) followed by lower frequency oscillations (~ 100-1000 Hz), and have a duration ranging between 50 ms up to 200 ms (FIG. 4D). Other less frequent bubble events start with a low-frequency onset under ~50 Hz, followed by a high-frequency wave packet between ~100 and ~2000 Hz (FIG. 4E). These bubble events tend to have a longer duration over ~ 100 ms.

[0048] These text-book bubble events

[0039] reflect three distinct bubble nucleation processes [15,18, 37], The monochromatic bubble event characterizes a bubble detaching from a rigid surface, such as a we 11 -formed sediment bed orifice, but without extended interactions with this surface. The bimodal bubble events indicate an interaction with the sediment bed

[0037] , In FIG. 4D, the high- frequency onset is generated by a void space opening and subsequent sediment motion. Then, the lower-frequency signal is caused by free oscillations of the newly formed bubble, right after it detaches from the sediments. In FIG. 4E, the very low-frequency oscillation may reflect a turbulent interaction of the fluid with conduits in the sediments before the bubble escapes in the water column

[0018] .

[0049] The bubble waveforms hold both the signature of their interactions with the sediment bed during or after nucleation

[0018] , as well as an indication of their sizes

[0038] , The shallow sediments at Laacher See are weakly cohesive. Therefore, the particles at the surface can easily be displacedalong with the gas motion over short time scales. Thus, even at a constant gas throughput, an orifice can rapidly change its shape, giving rise to a myriad of processes and, consequently, bubble waveform onsets over time. Once the bubble is formed and oscillates in the water column, its dominant frequency is proportional to its size

[0039] , Therefore, following the Minnaert

[0038] equation (Eq. 1 in Methods), it is possible to estimate the volumes of gas associated with a given bubble event.

[0050] Overall, bubble events close to each other exhibit similar waveforms and spectra. Yet, they can significantly change along the channels and rapidly attenuate. A given bubble event depicts a slightly different waveform based on where it is recorded on the sediment bed (FIG. 5). Indeed, both increasing and decreasing frequencies along the distance from the bubble seep are observed, but the onset high-frequency signal is often lost moving away from the seeps (e.g., FIG. 5). This likely relates to bubble attenuation, different coupling conditions, and / or local site effects. Overall, the implementation was only able to detect the same bubble signals over a 30-m section. In addition, a smearing effect is caused by a gauge length of 10 meters. Yet, unlike conventional hydrophones, DAS provides unprecedented spatial resolution and indications of the interactions with the lake bed, thereby delivering a more complete picture of gas-sediment interactions and degassing processes.

[0051] Spatiotemporal analysis of bubble events. Based on an initial population of 30 manually extracted waveforms (e.g., FIGS. 4A-E), a waveform similarity search is applied, followed by a channel-to-channel template matching over the entire length of the immersed fiber array (see Method section). As the first 140 channels are reference or inland section, this subsection recorded few informative signals. The last 50 channels were near the lake edge. After manual inspections, many abrupt and suspicious signals with incoherent spectral contents were found. Therefore, considering the locations of these channels and the smearing effect, the detection analysis was performed only on channels 130-580. The bubble event search enabled the identification of three different main zones with active seepage along the cable (Zone 1, Zone 2, and Zone 3 in FIGS. 3A, 3B, and 6A). The total number of bubble events detected after association exceeds 470,000 from three zones aggregated over a recording period of 40 hours (FIG. 6B). On average, the rate at which bubble events were detected is about one bubble per second, which is in accordance with some of the videographic data collected near the shore. Then, the temporal trend of gas emission volume was estimated for further calculations using Eq. 1. FIG. 6C shows the evolution of gas seepage volume from three zones. Knowing that three zones have comparable amounts of bubble events (FIG. 6B), the significantly higher estimation of gas volume in Zone 1 indicates a greater proportion of low- frequency events in this area.

[0052] Understanding the time evolution within the bubble groups can help better constrain gas seepage patterns. A clustering algorithm or technique (referred to collectively as “clustering technique”) was implemented for high-SNR bubble events based on the temporal-spectral characteristics of bubbles, and this clustering technique includes using a convolutional neuralnetwork (CNN) for dimensionality reduction of spectrograms and a K-means cluster estimator, which is discussed more below. Yet, because of the impracticability of mix-frequency clustering, bubble events sampled at 8000 Hz were focused on, and such analysis was applied to channel 144 from Zone 1, which has the most high-SNR detections (4,784 events). By both manually scrutinizing the bubble waveforms and carrying elbow tests on the K-means clustering algorithm (FIG. 11), four clusters were defined for the dataset. And, by only including the first 1,000 bubbles, the algorithm reached an accuracy of 95.5%. FIG. 11 shows test results with the left panel showing the elbow test of the K-means algorithm with original large dimension spectrograms as the input, and the right panel shows the elbow test of the K-means algorithm with reduced dimensionality matrices by CNN as the input. A relevant elbow point corresponds to three or four clusters in the present embodiment.

[0053] FIGS. 7A-H show centroid matrices and multiple events for each cluster. While three commonly-recognized bubbles are shown in FIGS. 4A-E, such events are archetypal examples showing features comparable with other studies. However, the four groups from the clustering algorithm do not have the exact same features and sometimes show hybrid bubble characteristics (low and high frequencies) from different groups, likely due to more complex sediment / water interaction with bubbles. Nevertheless, these four clusters exhibit distinctly observable characteristics both in the temporal and spectral domains. Cluster 1 features more low-frequency (SS 600 Hz) energy. Events in Cluster 2 have temporally stable middle-frequency (~ 600-1600 Hz) energy. The impulsive arrival and higher frequency (^ 1600 Hz) component characterize Cluster 3. Cluster 4 has a strong middle-frequency component following the arrival of high-frequency energy. The analysis was then expanded to the complete set of events. FIG. 71 shows the temporal distribution of the events. The interruption of Cluster 1 and Cluster 2 was observed at around 3 hours after start time. The occurrence rates of Cluster 3 and Cluster 4 also dramatically decreased after 10 hours. Low-frequency Cluster 1 signals have a much larger bubble radius and contribute a significant amount of gas seepage volume. The stoppage of Cluster 1 signal after 3 hours corresponds well to the Zone 1 gas volume pattern (dashed line in FIG. 6C). When including all events, retraining the model, and fine-tuning the parameters, the best accuracy rate reaches 77%. Compared with the model using the first 1,000 events, this performance degradation is attributed to the uneven distribution of bubble groups across time and intra-cluster waveform temporal variations. Permanent changes in physical processes such as pressure, temperature, and soil conditions may cause such intra-cluster instability.

[0054] Bubble characteristics at Laacher See. The gas bubbles rise in the water column and are therefore dominated by a non-condensable gas rather than water vapor. The compositions of the bubbles during the field experiment at these locations is unknown, but existing studies havedominantly found magmatic CO2 in bubbles [34, 41], Due to their similarity with bubbles studied elsewhere using hydrophones [e.g., 18, 37, 42], these signals are ascribed to CO2 bubbles.

[0055] Contrary to the hybrid frequencies bubble signals recorded by hydrophones in YellowstoneLake

[0018] , the onset of the low-frequency (100-1000 Hz) signals at Laacher See only starts when the high-frequency amplitude vanishes (FIG. 4C). High-frequency onset and overall signals at Laacher See also last longer than the signals at Y ellowstone (an order of magnitude higher compared to Yellowstone). Finally, the absence of very low frequencies associated with the high-frequency onsets (2-40 Hz) is noted, except for the largest bubbles (FIG. 12). FIG. 12 depicts waveforms of a large bubble along channels 140-149, with the smooth traces being waveforms bandpass fdtered between 2-10 Hz for each raw waveform thereabove. Laboratory experiments

[0037] instead show similar waveforms to Laacher See with high-frequency signals followed by low-frequency oscillations only starting when the bubbles are detached from the sediments. Taken altogether, these observations suggest that bubbles are generally not being nucleated beneath the fiber optic cable at Laacher See and seep through the watery sediments as already-formed CO2 gas bubbles rather than dissolved CO2. Bubble nucleation events typically start with an impulsive, high-frequency onset generated by the opening of void space in the liquid, and the subsequent, lower-frequency signal is generated by free oscillations of the newly formed bubble [18, 43], Yet, an exception corresponding to the largest amplitude bubble signals having very low-frequency oscillations associated with the high-frequency onsets is noted, as observed in Yellowstone (FIG. 12). Such a signal possibly bears the characteristics of bubble nucleation within the sediments.

[0056] The potential of DAS for hydroacoustic monitoring on active volcanoes. Optical fiber seismic hydrophones have recently been successfully designed to detect earthquakes (below 100 Hz)

[0044] , The study described herein indicates that fiber-optic cables interrogated by the DAS technology are sensitive to gas bubbles that are tiny acoustic events in the water column also sensed by single hydrophones. DAS technology, however, provides unprecedented spatial and spectral resolution. Compared to conventional hydrophones, fiber-optic cables therefore not only allow us to characterize degassing but also study its spatial evolution. Consequently, DAS opens new perspectives to study attenuation in three dimensions and the possible influence of the water column on wave propagation. DAS also provided detailed information on ground vibrations during the nucleation and seepage processes. Being coupled with the ground, critical improvement in the understanding of gas-sediment interactions, such as gas migration pathways or nucleation, is anticipated.

[0057] In addition, some fiber-optic cables can be operated in various environments, at large depths with high pressures, high temperatures, and even in acidic environments. They would also open new ways to monitor underwater volcanoes as the cables can be left permanently without much maintenance required. In addition, the interrogation can be carried out inland. At underwatercalderas or craters, such as Santorini (Greece) or Hunga Tonga-Hunga (Tonga), the interrogation would provide complementary monitoring observations, such as gas volume over time, but also improved estimates of volcanic CO2 emissions

[0022] . Yet, the coupling appears critical and should be carefully mitigated. Overall, DAS technology holds great promise to monitor underwater volcano degassing at high resolution and 70% of volcanoes are found underwater. But this technology also opens the door to monitoring underwater CO2 sequestration operations or CH4 leaks associated with permafrost or hydrates destabilization.

[0058] Materials and methods. Cable geolocalization. Geophysical measurements with DAS require knowing the geographic trajectory of the fiber and the distribution of the virtual sensors (each centered at their respective gauge). If the fiber trajectory is known and accessible, the task of georeferencing may be accomplished by simply taping the fiber (tap-test) and associating the stimulated channel with the respective geolocation of the tap. In this experiment, the fiber is submerged in up to 25 meters of water (FIG. 8) and, therefore, is not directly accessible.

[0059] The fiber array was initially designed as an equilateral triangle and, to do so, two buoys and the near-shore house were used to mark the comers of a triangle with a perimeter almost equal to the fiber’s length (e.g. , 500 m). While rolling out the fiber from the boat traveling from one buoy to another, maintaining a straight course was not always possible due to obstacles near the shore and the drift induced by strong winds. The fiber was also laid around each buoy in a wider radius to prevent sharp bends and possible damage.

[0060] After divers coupled the fiber on the lake floor, tap tests were performed by hitting the boat’s hull at six locations (at the buoys and halfway between them (FIG. 8)) . Then, the channels recording the earliest arrival of the tap event (FIG. 9) were picked up and associated with the respective taptest location. Locations of the channels in between the buoys were preliminarily obtained by equidistantly distributing them along the straight connecting the respective tap locations.

[0061] The uncertainty of the obtained sensor locations can be attributed to several sources. First, the GPS locations were retrieved with an ordinary smartphone yielding an accuracy down to several meters only. The taps were not timed, so their travel time to the lake floor could not be directly assessed. Thus, it was assumed that the picked channels are the nearest to the tap location, but they were not directly associated with it. Due to the bathymetry, the nearest sensors may not lie vertically below the boat.

[0062] Second, a misfit between the optical and the mapped fiber length was observed. Between two picked channels, the optical and mapped fiber lengths are given by the number of channels between the spatial sampling and the arclength of the lake floor across the line between the corresponding tap locations, respectively. The optical fiber length between picked channels resulted in longer than the mapped trajectory between tap locations on all fiber segments. This process is equivalent to mapping the channel locations at an inter-channel distance that is smaller than thechosen spatial sampling period (FIG. 8). The fiber trajectory was manipulated manually, accounting for the previously-described detours during the laying out, thus effectively increasing the mapped fiber length until it matched the optical length (that is, mapped inter-channel distance equals the chosen spatial sampling period; FIG. 8). However, for fiber segments containing curves, there is a multitude of possible trajectories satisfying this condition, thus increasing the uncertainty of channel locations on such.

[0063] Several measurements could improve and facilitate future deployments in a lake. Projecting the geo-spatial trace of the divers and fixing the fiber on the lake floor would probably yield the most accurate trajectory. If the fiber lies at depths unreachable by divers, spatially dense tap tests should be accurately timed and located with triggered GPS clocks and differential GPS locations, respectively. The travel times of the induced acoustic waves could then be accurately picked, and their inversion could yield reasonably good virtual sensor locations. Systematic variation of the retrieved fiber trajectory could then yield an ultimate assessment of the trajectory’s uncertainty. Each alternative trajectory must meet the condition that its mapped arclength and the optical distance remain numerically equal (between tap-test locations and the corresponding pair of picked channels) .

[0064] Bubble detection. A template-matching (TM) procedure was designed to detect bubbles in an automated way along the fiber. The TM technique includes using previously detected events (the templates) to scan the continuous data by performing cross-correlations to find new events (the detections) [45, 46], Because the dataset has data with different sampling rates, the detection was conducted separately for 5000 Hz and 8000 Hz data with the same procedure. First, a preliminary Waveform Similarity Search (WSS) was run over all channels to create a template database, by utilizing 60 manually-selected high-SNR bubble waveforms and computing cross-correlation with the continuous time series. Second, a Median Absolute Deviation (MAD) threshold of 9 was used to identify similar waveforms

[0045] , FIG. 6A summarizes the WSS results, highlighting three 40- meter zones of higher seepage activity along the cable (at channels 130-170, 220-260, and 260- 300). In total, WSS matched 112.32 million waveforms from the 640 channels combined. Yet, in the processing workflow, a given bubble event can be matched by several templates over multiple channels. Hence, most events are counted multiple times. Consequently, the large number of matched waveforms needed to be further constrained. Thus, the WSS detection results were examined and only events with high SNR and spatial consistency (e.g., observed over a minimum of 20 channels) were kept as templates. In total, 5210, 893, and 1573 templates for three zones, respectively, were kept. The same MAD = 9 threshold for TM was used. Considering the relatively large number of templates for the first section, TM was run on Zone 1 by small batches of templates, demonstrating the relative completeness of the detection (FIG. 13). FIG. 13 depicts template matching results of Zone 1 by small batches of templates, with detection numbers being represented as a function of the number of templates used. A general trend of diminishingly increasing and atendency of convergency is observed. TM results are shown in FIG. 6B. From the 40-hour experiment period, 148,341, 128,710, and 197,397 events are found for the three zones by TM, respectively.

[0065] Gas throughput estimation. Using the dominant frequency, bubble sizes may be estimated assuming they are spherical. The diameters of these bubbles are on the order of 0.5 to 1.6 mm using Minnaert’s

[0038] equation:Equation (1) where P is the hydrostatic pressure (1082 kPa), y is the ratio of specific heats of the gas in the bubble (1.3 for CO2), p is the density of the liquid (1000 kg / m3), and f is the dominant frequency. Estimated sizes are compatible with bubble sizes estimated using independent active and passive hydroacoustic instruments at nearby lake locations (e.g. , 20 meters away), where bubbles rising towards the surface were able to be visually tracked. The volume for each bubble ranges between 0.002 and 0.007 m .

[0066] Bubble clustering. A bubble can be characterized based on its dominant frequencies and arrival time move-outs of different frequency groups (FIGS. 4A-E). Therefore, high-SNR events were selected and wavelet transform of those events was computed to retrieve the temporal-spectral information. Such analysis was applied on the single channel with the most detections (z.e., channel 144; where video of a bubble rising towards the surface was obtained), where 4,784 bubble events pass the SNR threshold. The bubbles were shifted and trimmed so that each trace is set to be 0.2 seconds and all bubbles are well aligned. Wavelet transform converts the time-series of bubbles into 128 x 1600 2D matrices (128 spectral estimates, 1600 temporal estimates).

[0067] Raw 2D matrices were fed from wavelet transform to the K-means clustering algorithm. By doing an elbow test, it was determined to cluster the bubbles into four groups (FIG. 11). Then, 400 events were manually labeled for clustering accuracy estimation. This K-means cluster separator serves as a baseline model and yields an accuracy of 62.5%. Mousavi et al.

[0047] designed an unsupervised learning algorithm for clustering teleseismic events and local earthquakes, and it includes a set of convolutional layers and a K-means clustering layer, integrating the loss from both components. To improve the clustering performance, a similar CNN structure to reduce the dimensionality of the raw inputs and concentrate the temporal-spectral information to 4x50 size matrices was adopted. Four groups for subsequent K-means analysis were selected. After training the model for multiple iterations for the matrix reconstruction loss to converge, bottleneck layers were used to feed the K-means clustering algorithm. Such CNN+K-means model was initially used with the first 1,000 bubble events. By doing so, 200 events included in the 1,000-event training dataset yielded an accuracy of 95.5% while the other 200 events serving as a test group returned a lower accuracy of 75%. These significant accuracy improvements to the K-means-only modeldemonstrates the efficacy of the convolutional network for dimension reduction. In addition, such results also suggest a preference for the complete inclusion of bubble events into the model training set, providing access to large-memo computational resources. Therefore, the same procedure was repeated with the inclusion of all bubbles. Knowing the stochastic nature of the convolutional network, the model was repeatedly trained on a GPU card 100 times and the performance evaluated with labeled data. The best-performance model is then adopted and shown in FIGS. 7A-I.

[0068] With reference to FIG. 10, there is shown a flowchart illustrating an embodiment of a method 200 of detecting spatiotemporal events within an acoustic field using a fiber optic distributed sensing system. The method 200 may be carried out by the fiber optic distributed sensing system 10.

[0069] The method 200 begins with step 210, wherein sensor data is captured using a fiber optic distributed sensor extending through an acoustic field in order to obtain spatiotemporal data. When using the interrogator 14, for example, the fiber optic cables are used to detect acoustic signals based on phase and amplitude changes in backscattered light resulting from local strains in the cable, typically caused by nearby acoustic disturbances, such as gas bubble activity, as discussed above. When an acoustic event, such as a gas bubble being formed and released / detached from a sediment bed or seafloor, occurs near the optical fiber, it induces a strain change detectable in the backscattered light. The sensor may be stored in memory of the interrogator 14 or other memory of the system 10, such as the memory 30 of the computer system 16. The method 200 continues to step 220.

[0070] In step 220, a wavelet transform is performed on sensor data captured from a fiber optic distributed sensor extending through an acoustic field in order to obtain spatiotemporal data. The acoustic field refers to any region having acoustic waves passing therethrough that is selected for monitoring. As mentioned above, when an acoustic event, such as a gas bubble being formed and released / detached from a sediment bed or seafloor, occurs near the optical fiber, it induces a strain change detectable in the backscattered light. Applying the wavelet transform to the raw DAS data aids in breaking down the non-stationary signal into various frequency components for enhanced resolution. This process not only assists in noise reduction and signal enhancement but also facilitates the extraction of specific signal features, making event classification or anomaly detection more precise. According to embodiments, a continuous wavelet transform (CWT) or a discrete wavelet transform (DWT) may be used to provide information indicating which frequencies are present and how strong or dominant each frequency is at a given moment. The method 200 continues to step 230.

[0071] In step 230, the spatiotemporal data is matched to an event template of a predetermined set of event templates in order to identify a spatiotemporal event. As discussed above, template matching enables detecting and classification of a spatiotemporal event by matching processedsensor data to a predetermined event template, such as one that has been empirically-derived and / or otherwise predetermined as being indicative of a characteristic of a physical phenomenon, such as gas bubble size, nucleation, and / or detachment from a sediment bed. In addition to attributes such as gas bubble size, other gas bubble physical properties or other characteristics may be determined, such as a gas bubble nucleation / sediment detachment type. The method 200 continues to step 240.

[0072] In step 240, information about the spatiotemporal event is stored in memory and / or provided to a user. This information, which may be referred to as spatiotemporal event information, may include indications of a characteristic of a physical phenomenon, which may be described as a discrete characterization (e.g., bubble type selected from a predefined definite number of bubble types) and / or a continuous characterization (or quantization), such as an amount (e.g., volume) of gas. The spatiotemporal event information may be stored in memory of the interrogator 14, the computer system 16, and / or sent to a third party for use . In embodiments, a multitude (i. e., 1000 or more) of spatiotemporal events are detected and recorded, and the information for these events is used together to determine an aggregate characteristic of a region that the fiber optic distributed sensor is monitoring. This aggregate spatiotemporal data may be stored in memory as well. The method 200 then ends.

[0073] In some embodiments, in the method 200, for example, redundant spatiotemporal event information represented by the spatiotemporal data is removed. In some implementations, a single bubble may cause a signal to be received at multiple channels. For example, in one embodiment, an optical fiber having a gauge length of 10m and a channel spacing of Im, and with a sampling rate of 8000 Hz, may be used, and signals from bubbles may be generated at adjacent channels, such as when smaller channel spacing is used.

[0074] In some embodiments, a clustering algorithm is used to cluster a plurality of pre-classified spatiotemporal events into a plurality of clusters, where each cluster is associated with a different attribute or property, such as a physical attribute or property, for example, bubble volume range or other amount. A neural network technique may be used to cluster the bubbles, as discussed above. In embodiments, each of the plurality of clusters the classifications is associated with a dominant frequency, which may be used to determine a gas bubble size, as discussed above. When the system is in use, gas bubbles may be detected and their volume determined / estimated based on classifying the spatiotemporal event.

[0075] It is to be understood that the foregoing description is of one or more embodiments of the invention. The invention is not limited to the particular embodiment(s) disclosed herein, but rather is defined solely by the claims below. Furthermore, the statements contained in the foregoing description relate to the disclosed embodiment(s) and are not to be construed as limitations on the scope of the invention or on the definition of terms used in the claims, except where a term or phraseis expressly defined above. Various other embodiments and various changes and modifications to the disclosed embodiment(s) will become apparent to those skilled in the art.

[0076] As used in this specification and claims, the terms “e.g.,” “for example,” “for instance,”“such as,” and “like,” and the verbs “comprising,” “having,” “including,” and their other verb forms, when used in conjunction with a listing of one or more components or other items, are each to be construed as open-ended, meaning that the listing is not to be considered as excluding other, additional components or items. Other terms are to be construed using their broadest reasonable meaning unless they are used in a context that requires a different interpretation. In addition, the term “and / or” is to be construed as an inclusive OR. Therefore, for example, the phrase “A, B, and / or C” is to be interpreted as covering all of the following: “A”; “B”; “C”; “A and B”; “A and C”; “B and C”; and “A, B, and C.”REFERENCES[1] Robin S Matoza and Diana C Roman. One hundred years of advances in volcano seismology and acoustics. Bulletin of Volcanology, 84(9): 86, 2022.[2] Kevin Reath, Matthew Pritchard, M Poland, F Delgado, S Cam, D Coppola, B Andrews, SK Ebmeier, E Rumpf, S Henderson, et al. Thermal, deformation, and degassing remote sensing time series (ce 2000-2017) at the 47 most active volcanoes in latin america: implications for volcanic systems. Journal of Geophysical Research: Solid Earth, 124(1): 195-218, 2019.[3] Alessandro Aiuppa, Marcello Bitetto, Dario Delle Donne, Francesco Paolo La Monica, Giancarlo Tamburello, Diego Coppola, Massimo Della Schiava, Lorenzo Innocenti, Giorgio Lacanna, Marco Laiolo, et al. Volcanic co2 tracks the incubation period of basaltic paroxysms. Science advances, 7(38): eabh0191, 2021.[4] Giovanni Chiodini, Antonio Paonita, Alessandro Aiuppa, Antonio Costa, Stefano Caliro, Prospero De Martino, Valerio Acocella, and Jean Vandemeulebrouck. Magmas near the critical degassing pressure drive volcanic unrest towards a critical state. Nature communications, 7(1): 1-9, 2016.[5] Christoph Kern, Alessandro Aiuppa, and J Maarten de Moor. A golden era for volcanic gas geochemistry? Bulletin of Volcanology, 84(5):43, 2022.[6] SA Cam, Lieven Clarisse, and Alfred J Prata. Multi-decadal satellite measurements of global volcanic degassing. Journal of Volcanology and Geothermal Research, 311:99-134, 2016.[7] Michael R Burton, Georgina M Sawyer, and Domenico Granieri. Deep carbon emissions from volcanoes. Reviews in Mineralogy and Geochemistry, 75(1): 323-354, 2013.[8] Tarsilo Girona, Fidel Costa, and Gerald Schubert. Degassing during quiescence as a trigger of magma ascent and volcanic eruptions. Scientific Reports, 5(1): 1-7, 2015.[9] Tobias P Fischer, Santiago Arellano, Simon Cam, Alessandro Aiuppa, Bo Galle, Patrick Allard, Taryn Lopez, Hiroshi Shinohara, Peter Kelly, Cynthia Werner, et al. The emissions of co2and other volatiles from the world’s subaerial volcanoes. Scientific reports, 9(1): 1-11, 2019.

[0010] Florian M Schwandner, Michael R Gunson, Charles E Miller, Simon A Cam, Annmarie Eldering, Thomas Krings, Kristal R Verhulst, David S Schimel, Hai M Nguyen, David Crisp, et al. Spacebome detection of localized carbon dioxide sources. Science, 358(6360), 2017.

[0011] Alessandro Aiuppa, Tobias P Fischer, Terry Plank, and Philipson Bani. CO2flux emissions from the Earth’s most actively degassing volcanoes, 2005-2015. Scientific Reports, 9(1): 1-17, 2019.

[0012] Nemesio M Perez, Pedro A Hernandez, German Padilla, Dacil Nolasco, Jose Barrancos, Gladys Melian, Eleazar Padron, Samara Dionis, David Calvo, Fatima Rodriguez, et al. Global co2 emission from volcanic lakes. Geology, 39 (3):235-238, 2011.

[0013] Julia Woitischek, Nicola Mingotti, Marie Edmonds, and Andrew W Woods. On the use of plume models to estimate the flux in volcanic gas plumes. Nature Communications, 12(1): 1-8, 2021.

[0014] Corentin Caudron, Agnes Mazot, and Alain Bernard. Carbon dioxide dynamics in Kelud volcanic lake. Journal of Geophysical Research: Solid Earth, 117(B5), 2012.

[0015] Jianghui Li, Ben Roche, Jonathan M Bull, Paul R White, John W Davis, Michele Deponte, Emiliano Gordini, and Diego Cotterle. Passive acoustic monitoring of a natural co2 seep siteimplications for carbon capture and storage. International Journal of Greenhouse Gas Control, 93: 102899, 2020.

[0016] Manfredi Longo, Gianluca Lazzaro, Cinzia Giuseppina Camso, Andrea Corbo, S Scire Scappuzzo, Francesco Italiano, Alessandro Gattuso, and Davide Romano. Hydro-acoustic signals from the panarea shallow hydrothermal field: New inferences of a direct link with stromboli. Geological Society, London, Special Publications , 519(1):SP519-2020, 2022.

[0017] J Vandemeulebrouck, J-C Sabroux, M Halbwachs, N Poussielgue, J Grangeon, J Tabbagh, et al. Hydroacoustic noise precursors of the 1990 eruption of Kelut Volcano, Indonesia. Journal of Volcanology and Geothermal Research, 97 (l-4):443-456, 2000.

[0018] Corentin Caudron, Jean Vandemeulebrouck, and Robert A Sohn. Turbulence-induced bubble nucleation in hydrothermal fluids beneath Yellowstone Lake. Communications Earth & Environment, 3(1): 1-6, 2022.

[0019] Jerry Blackford, Henrik Stahl, Jonathan M Bull, Benoit JP Berges, Melis Cevatoglu, Anna Lichtschlag, Douglas Connelly, Rachael H James, Jun Kita, Dave Long, et al. Detection and impacts of leakage from sub-seafloor deep geological carbon dioxide storage. Nature climate change, 4(11): 1011—1016, 2014.

[0020] M Veloso, Jens Greinert, Jurgen Mienert, and Marc De Batist. A new methodology for quantifying bubble flow rates in deep water using splitbeam echosounders: Examples from the a retie offshore nw-s valbard. Limnology and Oceanography: methods, 13(6):267-287, 2015.

[0021] Richard Manasseh, Guillaume Riboux, and Frederic Risso. Sound generation on bubble coalescence following detachment. International Journal of Multiphase Flow, 34(10):938-949, 2008.

[0022] Tobias P Fischer and Alessandro Aiuppa. Agu centennial grand challenge: volcanoes and deep carbon global co2 emissions from subaerial volcanism — recent progress and future challenges. Geochemistry, Geophysics, Geosystems, 21(3):e2019GC008690, 2020.

[0034] Andreas Goepel, Martin Lonschinski, Lothar Viereck, Georg Biichel, and Nina Kukowski. Volcano-tectonic structures and co2-degassing patterns in the laacher see basin, germany. International Journal of Earth Sciences, 104(5): 1483-1495, 2015.

[0036] Corentin Caudron, Marc De Batist, Guillaume Jouve, Guillaume Matte, Thomas Hermans, Adrian Flores-Orozco, Wim Versteeg, Zakaria Ghazoui, Philippe Roux, Jean Vandemeulebrouck, et al. Messages in the bubbles. EOS, 101, 2020.

[0037] A Vazquez, R Manasseh, and R Chicharro. Can acoustic emissions be used to size bubbles seeping from a sediment bed? Chemical Engineering Science, 131 : 187- 196, 2015.

[0038] Marcel Minnaert. Xvi. On musical air-bubbles and the sounds of running water. The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 16(104):235-248, 1933.

[0039] Timothy Leighton. The acoustic bubble. Academic press, 2012.

[0041] Karin Brauer, Horst Kampf, Samuel Niedermann, and Gerhard Strauch. Indications for the existence of different magmatic reservoirs beneath the eifel area (germany): A multi -isotope (c, n, he, ne, ar) approach. Chemical Geology, 356: 193-208, 2013.

[0042] Ben Roche, Paul R White, Jonathan M Bull, Timothy G Leighton, Jianghui Li, Colin Christie, and Joseph Fone. Methods of acoustic gas flux inversion — investigation into the initial amplitude of bubble excitation. The Journal of the Acoustical Society of America, 152(2):799-806, 2022.

[0043] Robert H Mellen. Ultrasonic spectrum of cavitation noise in water. The Journal of the acoustical Society of America, 26(3):356-360, 1954.

[0044] M Janneh, FA Bruno, S Guardato, GP Donnarumma, G lannaccone, G Gruca, S Werzinger, A Gunda, N Rijnveld, A Cutolo, et al. Field demonstration of an optical fiber hydrophone for seismic monitoring at campi-flegrei caldera. Optics & Laser Technology, 158: 108920, 2023.

[0045] David R Shelly, Gregory C Beroza, and Satoshi Ide. Non-volcanic tremor and low-frequency earthquake swarms. Nature, 446(7133):305-307, 2007.

[0046] Zhigang Peng and Joan Gomberg. An integrated perspective of the continuum between earthquakes and slow-slip phenomena. Nature geoscience, 3 (9): 599- 607, 2010.

[0047] S Mostafa Mousavi, Weiqiang Zhu, Yixiao Sheng, and Gregory C Beroza. Cred: A deep residual network of convolutional and recurrent units for earth-quake signal detection. Scientific Reports, 9(1): 10267, 2019.

Claims

CLAIMS1. A fiber optic distributed sensing system, comprising: an optical fiber extending through an underwater gas field; an interrogator connected to the optical fiber so that light transmitted through the optical fiber is capable of being sensed by the interrogator; and a computer system configured to obtain sensor data captured by the interrogator and determine an amount or a characteristic of gas within the underwater gas field based on the sensor data.

2. The fiber optic distributed sensing system of claim 1, wherein the computer system is configured to, as a part of determining the amount or the characteristic of gas being released, perform a wavelet transform on the sensor data to obtain spatiotemporal data.

3. The fiber optic distributed sensing system of claim 1, wherein the computer system is configured to, as a part of determining the amount or the characteristic of gas being released, compare the spatiotemporal data to template data.

4. The fiber optic distributed sensing system of claim 3, wherein the computer system is configured to, as a part of determining the amount or the characteristic of gas being released, use a convolutional neural network (CNN) to reduce dimensionality of the spatiotemporal data.

5. The fiber optic distributed sensing system of claim 1, wherein the computer system is configured to, as a part of determining the amount or the characteristic of gas being released, determine a gas bubble size based on a dominant frequency extracted from the sensor data.

6. The fiber optic distributed sensing system of claim 5, wherein the computer system is configured to, as a part of determining the amount or the characteristic of gas being released, use template matching in order to detect gas bubble events based on the sensor data.

7. The fiber optic distributed sensing system of claim 5, wherein the template matching includes performing a waveform similarity search based on the sensor data.

8. The fiber optic distributed sensing system of claim 1, wherein the computer system configured to determine a gas bubble type as the characteristic of gas within the underwater gas field.

9. The fiber optic distributed sensing system of claim 1, wherein the optical fiber includes a sensor portion that extends for a distributed sensor length and includes the plurality of linear lengths, and wherein the plurality of linear lengths extend for at least one half of the distributed sensor length.

10. The fiber optic distributed sensing system of claim 1, wherein the computer system configured to determine an amount of carbon dioxide gas as the amount of gas within the underwater gas field.

11. The fiber optic distributed sensing system of claim 1, comprising the fiber optic distributed sensing system is installed at an underwater volcanic site.

12. A method of detecting spatiotemporal events within an acoustic field using a fiber optic distributed sensing system, comprising the steps of: capturing sensor data using a fiber optic distributed sensor extending through an acoustic field; performing a wavelet transform on the sensor data in order to obtain spatiotemporal data; and matching the spatiotemporal data to an event template of a predetermined set of event templates in order to identify a spatiotemporal event.

13. The method of claim 12, further comprising the step of: using a convolutional neural network (CNN) to reduce dimensionality of the spatiotemporal data.

14. The method of claim 13, wherein the acoustic field is an underwater gas field, and wherein the fiber optic distributed sensor includes an optical fiber that extends through the underwater gas field.

15. The method of claim 12, wherein the spatiotemporal event is a gas bubble event, wherein the gas bubble event indicates a time and a location of a gas bubble.

16. The method of claim 15, wherein the gas bubble event further indicates a gas bubble size or gas bubble type selected from a predetermined set of gas bubble types.

17. A method of continuously monitoring a given site using a fiber optic distributed sensor, comprising the steps of:installing a fiber optic distributed sensor at a target location, wherein the fiber optic distributed sensor includes an optical fiber and an interrogator; capturing sensor data using the fiber optic distributed sensor extending through an acoustic field in order to obtain spatiotemporal data; and determining a spatiotemporal event based on correlations between the captured sensor data and predefined template matching data.

18. The method of claim 17, further comprising using template matching in which the spatiotemporal data pertaining to the spatiotemporal event is matched to an event template of a predetermined set of event templates in order to detect the spatiotemporal event.

19. The method of claim 17, wherein the target location is underwater at a seafloor or other underwater bed, and wherein the optical fiber is located at the seafloor or other underwater bed.

20. The method of claim 17, wherein the acoustic field is an underwater gas field, and wherein the fiber optic distributed sensor includes an optical fiber that extends through the underwater gas field.

Images