Two video time-lapse cameras (VTLCs) were deployed by a remotely operated underwater vehicle (ROV) to observe the temporal and spatial variability of a natural hydrocarbon seep at 1180 m depth in the Green Canyon 600 lease block, Gulf of Mexico. The VTLCs were positioned approximately 60 and 90 cm away from the vent, each recording 15 s video bursts at 30 frames per second, illuminated by a 2000 lumen (lm) LED lamp. One camera functioned for 2 weeks; the second camera recorded 568 video bursts at 6 h intervals from 3 September 2017 to 2 February 2018 (153 days). Over the campaign period, seepage from three vents along a 10 cm cluster shifted toward a new fault line with up to nine intermittent individual vents shifting along 20 cm. We developed a semisupervised algorithm using Mathematica and ImageJ routines to resolve the rise velocity and size of individual bubbles. The algorithm was applied to the last 30 frames of each video burst. Bubble characteristics were also analyzed in the videos recorded by the ROV camera. Processing VTLC records yielded a bubble size distribution comparable (5% deviation) to the ROV camera, while the rise velocities were found to be 12% smaller than the ROV data. Hydrocarbon flux estimated from VTLC data was also compared favorably (2% difference) with synoptic physical collections of hydrocarbons into an ROV-held funnel. The long-term measurements indicate that bubble rise velocity was weakly correlated to the discharge rate as well as to the cross-flow velocity.
Natural hydrocarbon seeps are common on ocean margins throughout the world. Gas seeps with microbial and thermogenic methane sources are known in settings from continental margins to the deep ocean. Large-scale seepage has been identified along several continental boundaries, such as the U.S. Atlantic margin (Skarke et al. 2014), the West Spitsbergen margin (Mau et al. 2017), the Gulf of Mexico (MacDonald et al. 1994), and the U.S. Pacific margin (Baumberger et al. 2018). The global marine flux of methane from seeps to the atmosphere is estimated as 12 Tg yr−1 (range 5–20 Tg yr−1) (Rhee et al. 2009; Etiope 2012; Berchet et al. 2016). In deep ocean seeps, formation of gas hydrate at the seafloor interface is an important factor controlling the spatial and temporal characteristics of bubble venting from seeps (MacDonald et al. 2005). A subset of seeps, generally in deeper water, discharge petroleum. Kvenvolden and Cooper (2003) reported that the annual worldwide oil seepage rate—until better methods are applied and more oil seep rates are measured—is estimated to be between 0.2 and 2 Mt yr−1. With the “best estimate” of 0.6 Mt, 47% of the oil entering the oceans each year is from natural seepage.
In the Gulf of Mexico, persistent oil slicks imaged by satellite synthetic aperture radar (SAR) are evidence of an annual flux of oil to the ocean surface amounting to 2.5–9.4 × 104 m3 yr−1 (MacDonald et al. 2015)—equivalent to 0.02–0.09 Mt yr−1 assuming a nominal density of 895 kg m−3 for oil. The global and regional volume of hydrocarbons released remains largely unconstrained as hydrocarbon seeps are diverse, with a range of bubble/droplet size and rise velocity (Römer et al. 2012; Wang et al. 2016), emission rate (Judd et al. 2002; Leifer 2010), content (Clark et al. 2003, 2010), and gas-to-oil ratio, as well as a gas–fluid interface (Xu et al. 2001; Rehder et al. 2002) that constantly varies over time and space. The time scales for variation in these parameters can span from hours to months while the seep locations can change over spatial scales of order 1–10 m (Razaz et al. 2020).
Image-processing techniques can provide direct insight into important bubble characteristics such as bubble size, shape, and rise velocity. However, their application in deep-ocean settings are severely limited due to the reliance on remotely operated underwater vehicles (ROVs) to supply illumination, power, and communications. A stereoscope high-speed camera system called TAMU-CAM developed by Wang and Socolofsky (2015a) provides highly resolved details of bubble characteristics (i.e., methane hydrate formation, bubble size distributions, rise velocity, total gas flux, and void fraction) with high accuracy. Its stereoscopic configuration with a large depth of field (DOF) allows this system to adjust pixel-to-physical resolution so that the measurement errors caused by relative movement between an ROV and bubbles is minimized. A limitation, however, is that this ROV-based camera system only records relatively brief observations when the vehicle is in place. Data from this system, recorded at natural seeps at the Mississippi Canyon MC118 and Green Canyon GC600 lease blocks [lease block designations are given by the Bureau of Ocean Energy Management (BOEM)] in the northern Gulf of Mexico (Wang et al. 2016), have been used to verify the accuracy of results reported in the present work.
Other researchers (Torres et al. 2002; Sauter et al. 2006; Sahling et al. 2009; Leifer 2010; Römer et al. 2012) have reported in situ measurements of bubble size distributions and fluxes based on observations taken during short snapshots in time, i.e., seconds to minutes of video footage. Analysis of seepage observed using a standalone video time-lapse camera (VTLC) system, over longer periods was reported by Johansen et al. (2017), who quantified the release of hydrocarbons from a vent (27°22.204′N, 90°34.256′W) in the Mega Plume region of the GC600 seep zone over a 26 day observational period. None of these studies, however, addressed the variability of bubble size and rise velocity or the spatial changes in vent location over longer intervals. Significant changes in hydrocarbon bubble venting has been inferred to occur over intervals of months to years (MacDonald et al. 2005; Sahling and Ohling 2017). However, day-to-day variations in these processes have not been investigated because continuous observations did not exist.
The VTLC capability of autonomously recording videos over long periods comes at the cost of considerably reduced frame rate and illumination compared to laboratory (Zaruba et al. 2005) and ROV-based systems (Sauter et al. 2006; Wang and Socolofsky 2015a). The technical limitations imposed by this trade-off combined with presence of intermittent patches of oil on the image background, movement of aquatic fauna and marine snow, and large number of vents, made it impossible to process the records fully automatically. For the present work, we designed an experiment to obtain the first long-term (153 days) observation of seep characteristics based on VTLC technique (section 2) and developed a semisupervised algorithm to analyze the records (section 3).
2. Study site and experimental design
In collaboration with Oceaneering International Inc., we participated in three cruises from June 2017 to February 2018 in the Gulf of Mexico. Venting was observed in GC600 (Fig. 1a) during a field campaign in 2017 using acoustic multibeam mapping with an autonomous underwater vehicle (AUV) aboard Offshore Service Vessel (OSV) Ocean Project. The seep zone extends for approximately 1 km along the eastern flank of a large salt diapiric structure trending in a NW–SE direction (Fig. 1b). The seep zone is noteworthy for the prolific discharge of oil and gas from extensive gas hydrates that are exposed at the sediment–water interface and has been reported in numerous studies. The seabed includes large authigenic carbonate deposits precipitated due to microbial interaction with hydrocarbons (Roberts et al. 2010); the oil seeps generate persistent oil slicks on the surface that have been quantified by remote sensing studies (Garcia-Pineda et al. 2010; Daneshgar Asl et al. 2017); dynamics of discharge have previously been investigated based on seafloor mapping (Mitchell et al. 2018), in situ video monitoring by a VTLC (Johansen et al. 2017), and by ROV-based measurements (Wang et al. 2016).
In addition to water column and seafloor acoustic data collected with the AUV, we visited the potential seeps shown in Fig. 1b with the ROV Maxx aboard Multipurpose Service Vehicle (MSV) Ocean Intervention II. In the northern portion of the seep zone, oil and gas discharge is focused at a prominent vent known as Mega Plume at a depth of approximately 1180 m—highlighted in red along with other active seeps in black. To explore the time-space variability of a natural seep at the seafloor, two VTLCs were deployed using the ROV near the Mega Plume vent (Fig. 2a).
The VTLC system is composed of an anodized aluminum housing that holds an HD video camera, batteries, and the control circuitry, with an LED lamp mounted on a high-density polyethylene (HDPE) frame designed for deployment by a submersible. The high-definition 8 bit video camera (Replay XD 1080) has a spatial resolution of 1080 × 1920 pixels (px) and a 120° field of view with a time resolution limited to 30 frames per second (fps). Lighting is provided by a single 2000 lumn (lm) LED lamp (Lumos SeaLamp SLL-2000 Flood) and power is supplied with a 60 Ah lithium battery. The control unit can be programmed using the VTLC Terminal software with a computer to configure the camera system in time-lapse and/or triggered event modes. In time-lapse mode, the system is activated at specified times and records video bursts of specified length over a specified deployment period. In the trigger mode, the VTLC is manually activated by the submersible during deployment.
The cameras were positioned approximately 60 and 90 cm from the vent cluster. To prevent strong cross-flow from swaying the plume outside the camera frame and also due to technical constraints including limited illumination at heights above the bed, we directed the face of the cameras 30° down from the horizon to capture both the vents on the seafloor and the bubbles (Fig. 2a). Camera A was programmed to record 10 s of data every 3 h; however, it stopped functioning after two weeks; Camera B logged 15 s of data every 6 h starting 3 September 2017 and continued through 2 February 2018 (568 records in 153 days). Only the data recorded by Camera B are presented here. Concurrent to these measurements, an acoustic Doppler current profiler (ADCP) mounted on a bottom mount frame was positioned 68 m from the seep of interest to register background currents from the beginning of the VTLC deployment to 7 December 2017. Two principal vents were observed over an active venting area 30 cm across and were named as Clam Cluster and Fault Cluster (Figs. 2b and 2c, respectively).
The VTLC records indicate that the seepage was very dynamic on daily to monthly time scales. Initially—from September to late November 2017—gas bubbles, oil droplets, and/or a mixture of the two were released from the edge of an exposed gas hydrate deposit (identified as the Clam Cluster in Fig. 2b). The dense cluster included one to three vents shifting intermittently over a region approximately 10 cm long. From early December 2017, emission from the Clam Cluster started to decrease; concurrently, a mixture of gas and oil started to seep along a new crack or fault line, hereafter known as the Fault Cluster (Fig. 2c) on the seafloor. Within approximately 20 cm of the Fault Cluster captured by the Camera B, up to nine individual vents could be identified. Over the measurement period, we observed significant variations of release frequency among the vents and possibly gas/oil contents from the color appearance of the hydrocarbon bubbles/droplets. Note that we use “bubble” as a general term to represent gas bubbles and/or oil droplets in our observation.
On the recovery cruise aboard MSV Connor Bordelon, a 10 min video of the seep at the source was also recorded using the Ocean ProHD TV camera (1080i/59.4) on the ROV Millennium. To surmount the parallax error associated with a single-camera system, a screen with size-scale markings was placed behind the seep. Also, a custom-made funnel (Fig. 3) was used to measure the emission rate over the time (310 s) it took to fill the volume compartment marked with eight visible sections, each indicating 250 cm3. Using the ROV arm, the funnel was kept on top of the seep at approximately 0.15 m above bed (mab) such that no hydrocarbons could escape the funnel.
3. Image processing and analysis
The first step in obtaining quantitative information on the bubble properties was to identify individual bubbles in each video frame. To partially automate this process, we used a combination of Mathematica and ImageJ routines. Before processing the images, Mathematica was used to crop the images and remove areas of the frame that did not contain any seepage in order to reduce interference from marine snow or fauna moving around the vents and to reduce the computational background size (cf. Fig. 4a with Fig. 2c). Because both Clam and Fault Clusters were sometimes simultaneously active, the selected frame size was kept constant at 1650 × 700 px, which spanned across both clusters (Fig. 4a). The cropped images were processed with ImageJ to remove the background using “maximum intensity Z-projection”(Rueden et al. 2017). Processing too many frames at a time using this technique produces blurred images with significant interference from moving fauna and marine snow. By trial and error, it was found that processing 30 frames at a time results in bubbles with sharp edges in each frame. After converting 8 bit images to grayscale from 0 to 1, their contrast was increased by 3% (Fig. 4b). Then, as it follows, the image sequences are processed further by Mathematica to determine the bubble size and rise velocity.
To facilitate the bubble tracking in relatively low-quality images and taking into account that not all the bubbles were ellipsoidal (e.g., see Fig. 4), we digitally replaced the bubbles with nodes at the centroid of each bubble, then attributed the area-equivalent circular diameter De to each node. To smooth out the small inhomogeneities due to uneven illumination, ImageJ’s effect of subtracting the background medium, and the need to sharpen the edges of the bubbles, the images were passed through a curvature flow filter (Huisken 1984, 1990; White 2002) using Mathematica built-in functions (Fig. 4c). The filter applies the partial differential equation ∂f/∂t = κ|∇f| with the contour curvature of κ = ∇ ⋅ (∇f/|∇f|) to every image channel f. The flow curvature time t parameterizes the evolution of the curvature flow and thereby the spatial range of the filter. Here, we selected t = 5. To remove the noise caused by shadows and motion blur, the grayscale images were converted into binary scale using the strict threshold of 0.1. This threshold effectively sets pixels with values between 0.0 and 0.1 to 1, and all other pixels to 0 (Fig. 4d).
Converting the image into binary scale makes it easier to distinguish the bubbles from the background when using a combination of morphological component analyses (Starck et al. 2005) and connected-component analyses (Devijver and Ronse 1984). The morphological component analysis is a new method which allows us to delineate features contained in an image when these features present different morphological aspects. The morphological analyses were not sensitive to the threshold for distinguishing bubbles from the background as the analyses were performed in binary space. The connected-component analysis method identifies connected pixels to detect regions (in this case bubbles) in primarily binary digital images. Due to the reflection of the light by the bubbles, this technique leads to separation of only the darker part of the bubble. Together with connected-component analysis, we used a boundary representation (BRep) model (Stroud 2006) to calculate the geometrical properties of bubbles to account for the missing parts of the bubble image. A BRep of an object is a geometric and topological description of its boundary which allows integral properties of the object to be easily and efficiently computed. Figure 5a shows the resulting shape of the bubbles detected for five frames.
Some overlapped bubbles were inevitably identified as an individual bubble; however, it is not possible to identify directly these anomalies in the existing image frames. Instead, a shape parameter, s = 4πA/P2, where A and P are, respectively, bubble area and perimeter, provided a first estimate of the number of overlapped bubbles. Given that s = 1 represents a perfectly spherical bubble, we assume bubbles with s ≤ 0.33 as overlapped. As such less than 8% of the bubbles found in the data were considered overlapped or incorrectly identified. The primary causes were inherent shortcomings involved with the application of single-camera systems, low image quality due to limited illumination, and interference of oil patches in the background.
Once the bubbles were identified, different strategies were used to track the individual bubbles, depending on the venting clusters. In the Fault Cluster, the clutter caused by the large number of individual vents—particularly in presence of lateral currents—hinders the bubble tracking process. Thus, it was necessary to analyze the bubbles emanating from individual orifice clusters separately. The position of bubbles is referred to a set of rectangular Cartesian axes x, y, and z whose origin is an arbitrary point on the seafloor as shown in Fig. 4c, with the x and z measured in the camera frame plane, and y along the fault. The bubble streams were separated by applying a Bayesian–Gaussian mixture model (Figueiredo and Jain 2002). For this purpose, the centroid of bubbles identified in all image sequences (30 images) for each stream were projected onto the x axis. However, to minimize the overlap between the projections of each two neighboring streams, it was necessary to remove the cross-flow effect which tilts the bubble stream from the vertical. The bubble streams were rotated simultaneously around an arbitrary point with an angle varying between 0° and 15° with 1° stride to find the optimal rotation angle at which all the orientation of bubble streams was closest to the vertical. The rotation angle leading to the highest density of centroid projections onto the x axis over an arbitrary distance was selected as the optimal rotation angle. The upper limit of rotation is determined arbitrarily. Figure 5b displays the centroids of all the bubbles detected in 30 frames in the Fault Cluster with the horizontal current removed (i.e., an angle rotation of 12° was applied to bubble streams shown in Fig. 5a) and color-coded based on the identified subclusters. Once this separation is achieved, the rotation angle representing the horizontal cross-flow is added back in to preserve the natural state of the bubble streams before tracking (e.g., Fig. 5c). It was not possible to separate individual bubble streams in the Clam Cluster as the vents were too close and thus considered as one merged stream from more than one vent.
After separating the bubble streams, the next step was tracking the bubbles. In stagnant water, bubbles typically accelerate immediately after release from an orifice within a very short distance (Di Marco et al. 2003; Celata et al. 2007) and reach a terminal velocity; however, in turbulent natural flows, the rise velocity of a bubble is a combination of its inherit terminal velocity and the local velocity of water. To minimize the impact of local velocity fluctuation, we tracked bubbles from the source using “3 frame: minimum acceleration (3MA)” (Ouellette et al. 2006) technique and reduced tracking conflicts by adding some additional conditions. In the original 3MA technique, the position of the bubble in frame f−1 is used along with the position in frame f to estimate a velocity and therefore a position for the bubble in frame f+1. The nearest bubble to the predicted position is specified as the position of bubble in the f+1 frame. In the absence of features such as bubble shape, an additional condition to restrict the direction of bubble movement between the frames was necessary. For this purpose, we used the general direction of the bubble stream determined by fitting a second- or third-order polynomial to the centroid of all the bubbles detected through the 30 frames, sample of which is shown in Fig. 5b. A pie-shaped area with a width of ±θ and a radial distance of Rmax= 45 px (roughly allowing a maximum vertical velocity of 35 cm s−1), was scanned for identifying the potential matches for each bubble. The angle θ is measured relative to the general bubble stream direction and was selected empirically to be 15°. We devised a simple scoring system, Ssr and Sbp, based on the size ratio of the bubble area to that of a potential match and the angle of the bubble path with respect to the general direction of the plume, respectively. The closer the size ratio is to 1, the higher the Ssr score. When the angle between the potential bubble path and the tangent of the general bubble stream direction at the bubble height is smallest, the score Sbp is highest. The highest total score with the least acceleration possible is finally chosen as the best match (Fig. 5c).
a. Converting px to mm
Particle tracking velocimetry (PTV) analysis described in Wang and Socolofsky (2015a) was used to determine the bubble size distribution and vertical velocity from 350 s of the 10 min video recorded by the ROV Millennium. Motion blurs were present in the rest of ROV videos and hence this part of the data were excluded from the analysis. Figure 6 demonstrates the lognormal probability density function (PDF) fitted to the distribution of the area-equivalent circular diameter, Deυ weighted by the bubble volume for the ROV-based data. The weighted PDF in pixel space is also calculated for the last four VTLC records (spanning 24 h with the last record taken 2 h before the ROV-based record) taking into account the variable distance of bubbles emanated from vents along the Fault Cluster (Fig. 2c). Figure 7 shows a snapshot photo taken with the ROV Millennium on 3 February 2018, in which the known dimensions of the camera frame are used to measure the distances of the Clam and Fault Clusters to the VTLC-B. This additional information is used to define a dynamic pixel resolution for all the Fault Cluster vents shifting position intermittently along the y axis from the constant resolution at the x–z origin (pm0). A linear transformation is applied to all the distances and dimensions for the Fault Cluster according to
where is a linear function that varies between 1.0 for x = 0 and 1.5 (=97/65) for an x value corresponding to the last point on the y axis, and is the average distance of all the bubbles released from the pth vent. Selecting a conversion rate of 4.4 px = 1.0 mm (pm0 = 1/4.4) yields the maximum similarity between the two weighted PDFs as illustrated in Fig. 6. The mean volume-weighted bubble diameter for the VTLC data is then calculated to be Deυ = 7.74 ± 1.41 mm and compares favorably with the corresponding ROV-based value of Deυ = 7.37 ± 1.31 mm by about a 5% difference. The standard deviations are derived from fitting the lognormal distribution to the size of the bubbles, as a measure of scale such as standard deviation is not robust and highly susceptible to outliers (Razaz and Kawanisi 2011). In case of ROV-based PTV, the misalignment of the scaled screen with respect to vertical, slightly tilted viewing angle, and the unknown distance offset between the scale screen and the actual bubble stream are some of the adverse factors that might have contributed to the uncertainty in the results.
b. Sensitivity analysis
The cost of computation for delineating the bubbles from the background increases substantially with the number of frames being processed because the algorithms developed are semiautomatic. In this section we determine the rise velocity and size of the bubbles by processing 30 and 300 frames (from a total 450 frames in each 15 s video) of three randomly selected video records using the algorithms described above. These videos are selected from when the Fault Cluster was active and due to the complexity of the geometry and number of active vents, more sophisticated analyses were required. Figure 8 qualitatively compares the differences between the dynamic properties as well as size distribution of the detected bubbles deduced from different number of frames processed; Table 1 gives a summary of quantified differences. The conformity between the bubble characteristics detected from records with significantly different lengths suggests that it is reasonable to process only 30 frames (1 s) of the records in return of less than 1% and 10% average uncertainty, respectively, for the bubble size and rise velocity. Although with 30 frames processed the temporal length was not always enough to capture the largest bubbles. Note that the overall uncertainties are likely to be smaller than that given in Table 1 as the averaging is performed over only three records. We have visually examined all video data, and did not observe significant temporal variation in each burst (15 s). Considering the computation costs and relatively small difference between key parameters determined from significantly different number of frames processed, we conclude that processing 30 frames per burst should provide similar results.
4. VTLC results
Results from our long-term observations reveal short-term changes in venting characteristics that would have not been detected during snapshot observations from an ROV. Figure 9 presents two separate 4 week intervals of the VTLC image processing results for the Clam and Fault Clusters: 3 September–2 October 2017 and 5 January–2 February 2018, respectively. Where possible, the results are compared against the ROV-based PTV analysis and physical collection rates into a calibrated funnel deployed by the ROV during the recovery at the end of the experiment.
To demonstrate the temporal variability in hydrocarbon seeps in a collective fashion, Fig. 9 presents the rise velocities averaged over each frame (1/30 s), (thin gray curve), regardless of the bubble sizes, as well as values averaged over each video burst (1 s), (solid black dots). The overbar denotes averaging over time, and F = 30 is the total number of frames processed in each video record. The scatter in frame-averaged quantities may arise from a combination of variability in bubble characteristics (size, shape, and composition), properties of gas–liquid systems (density, viscosity, surface tension, density difference between gas and liquid, oil coating, and formation of methane hydrate), and hydrographic conditions (temperature, pressure, and cross-flow velocity). A comprehensive review of the effect of these parameters on bubble rise velocity in laboratory conditions is given by Kulkarni and Joshi (2005). Nevertheless, the vertical rise velocities from the VTLC records fit well within the broad range of the ROV-based PTV analyses for the Fault Cluster; however, the average value of 22.1 cm s−1 over the period shown is 12% smaller than the average 25.1 cm s−1 obtained from our ROV data (Fig. 9a). Some contributing factors that can lead to an underestimation of wb from the VTLC records may include the parallax error, the rough estimate of dynamic resolution, and neglecting the effect of cross-flow (in directions other than the x–z plane) on the distance between chains of bubbles and the camera. We could detect significantly more bubbles in the Fault Cluster compared to the Clam Cluster because of larger number of active vents and slightly better background conditions—no patches of oil and bright microbial mat—as demonstrated in Fig. 9b. The effects are evident in considerably less fluctuation in frame-averaged rise velocities determined for the Fault Cluster compared to the Clam Cluster.
The mean and standard deviation of volume-weighted bubble diameter Deυ for each sequence is derived by fitting a lognormal distribution as explained in section 3a and are demonstrated in Fig. 9c. The temporal variability of the mean Deυ indicates that while the distribution of bubble size was more or less constant on daily basis, there was notable variability on longer (weekly) time scales. Note that the standard deviation is not sensitive to the sample size and therefore an increase in the number of detected bubbles is not reflected in the temporal variability of standard deviations shown in Fig. 9c.
The volumetric hydrocarbon flow rate Q is one of the key parameters we sought to measure. It should be noted here that we estimated the total hydrocarbon flux which includes oil, gas and a mixture of the two. From the total volume of bubbles in F frames of each image sequence and given the maximum time tmax it takes for a bubble to travel the distance ΔzF from the vent to the upper boundary of the image area with an average rise velocity of , Q is calculated as a summation,
where P is the number of individual bubble streams, B is the number of bubbles identified in each frame, tF is the length of each interval in seconds, ΦF = 30 fps is the camera frame rate, and ς is the maximum number of times same bubble volume is calculated in F frames. The variation in the volumetric hydrocarbon flow rate, shown in Fig. 9d, compares well with low frequency variations in the total number of bubbles identified (Fig. 9b), and in the bubble sizes (Fig. 9c).
There are several sources of uncertainty in calculating the flow rate including the assumption of spherical bubbles, and the uncertainties involved in detecting the area of bubbles and their rise velocity. To approximate the errors involved in the Q estimates from Eq. (2), we take the total derivative and simplify the result, which yields
Comparing the ROV-based PTV results with VTLC suggests that δDeυ/Deυ ≈ 5% and . Therefore, the uncertainty involved in discharge estimates is approximately 13%. Processing the video of our customized funnel filling with hydrocarbons as recorded by the ROV, leads to a discharge estimate of QFunnel = 6.4 cm3 s−1, indicating a 2% difference with the last QVTLC separated by 2.5 h (Fig. 9d). Note that it was not possible to repeat the funnel sampling more than once since the compartment was filled with a solid mixture of methane hydrate and oil.
Presented in Fig. 10 is variability of the rise velocity, ⟨wb⟩, corresponding to the bubble size averaged over 0.5 mm bins, ⟨Deυ⟩, for the Clam and Fault Clusters. Also shown are the ROV-based PTV results averaged over 0.5 mm bins. The VTLC results closely follow the empirical prediction of the rise velocity of “dirty” natural gas bubbles in stagnant water (Clift et al. 2005) calculated by Wang et al. (2016). The only exception is the data correspond to the short period before the emission from the Clam Cluster ceased completely; visual inspection of the videos suggests that they represent oil droplets slowly released from chimney-like tubes. This group of observations show similar rise velocities as those of hydrate, oily bubbles in Wang et al. (2016) measurements from a vent (Confetti) located at 27°22.1954′N, 90°34.2624′W—about 10 m away from the vent we performed our measurements. In general, our estimates of rise velocity as a function of size align well with findings of Maini and Bishnoi (1981), Rehder et al. (2002), Greinert et al. (2006), and Römer et al. (2012), who reported velocities less than 30 cm s−1 for bubbles smaller than 26 mm in diameter (released within the hydrate stability zone).
a. Upwelling effect
Upwelling fluid can increase the vertical rise velocity of bubbles and thus decrease the transit time across the water column. Upwelling due to the entrainment of ambient water into the bubbly flow through turbulent shear, for example, can occur either at the edge of the bubble stream or at the edge of the individual bubble wakes. For air bubbles released continuously from a single orifice forming a chain-like structure in the water, Marks (1973) measured the rise velocity of bubbles with sizes ranging from 0.6 to 10 mm in diameter to approximate the effect of upwelling on bubble rise velocity. He suggested an empirical relation in the general form of wb ∝ Qβ, with β = 0.141, for flow rates exceeding 0.33 cm3 s−1. Note that the relationship is not dimensionally consistent.
Although our field measurements did not observe a perfect chain-like structure, the continuously released bubbles still form streams of leading-trailing trajectories very close to the source. Hence, we applied the empirical power-law relationship between wb and Q, and found wb to increase proportionally with the flow rate but with β = 0.09 and 0.10 for the Fault and Clam Clusters, respectively (Fig. 11). These exponents are smaller than the laboratory result in Marks (1973) for chain-like bubble structures, likely due to the presence of background cross-flows.
The majority of entrainment upwelling studies have, however, focused on the strong plumes that are commonly observed in lake aerations with a much shallower water depth and larger gas discharge (McGinnis et al. 2004) compared to our study here. The exponent β is reported to be 0.25 and 0.33 in Matsunashi and Miyanaga (1990) and Lemckert and Imberger (1993), respectively, for depths less than 60 m and plume discharges of <0.1 m3 s−1. Leifer (2010) reported a much steeper curve, β = 0.66, from measurements in the Cole Oil Point seep field at shallow depths (<85 m) and for flow rates ranging from 0.06 to 68 cm3 s−1. Note that instead of the rise velocity, Leifer used the upwelling velocity (the difference between rise velocity and terminal velocity). The difference in these literatures reported β values from our measurement data is largely due to significant difference in source discharge, as the natural seeps presented here have much smaller flux than those in lake aeration plumes. Wang et al. (2019) and Wang and Socolofsky (2019) are among the few who address the fundamentally different behavior of weak bubble plume or bubbly flow that cannot be named “plume.” For the strong seeps, bubble plumes have strong upwelling and behave like coherent buoyant plumes in which the shear entrainment is dominant. For the weak seeps, bubbly flow may have strong upwelling only within a short height of rise.
b. Effect of cross-flow on rise velocity
Bubble rise velocity as a function of the horizontal cross-flow Uc and bubble diameter Deυ over the period when ADCP data were available, is illustrated in Fig. 12a. Assuming that the thickness of the boundary layer was larger than the profiling range of the ADCP (4.6 to 25 mab), we applied the law of the wall logarithmic velocity profile (Hussain and Reynolds 1975; Nezu and Rodi 1986) to the entire velocity profile and estimated the current speed at 0.5 mab to give cross-flow speeds closer to the seafloor. The dashed line denotes the average terminal velocity approximated for the range of bubble size shown following Jamialahmadi et al. (1994) and Liu et al. (2016). There is a large uncertainty involved in estimates of terminal velocity in our work as it was not possible to determine the bubble composition, the formation of methane hydrate, and oil coating on the surface of bubbles released from a natural seep at approximately 1180 m depth for 153 days. Therefore, some of values are smaller than the rough estimate of average terminal velocity.
The laboratory measurements of Wang and Socolofsky (2015b) of bubble rise velocity at different cross-flow speeds suggest that when bubbles are released in chain-like structures the rise velocity approaches the terminal velocity rapidly as the cross-flow intensifies. In our in situ measurements no statistically robust relationship can be found between the bubble rise velocity and the cross-flow of 0–12 cm s−1 (Fig. 12a). Note that our plot here is a collective measure (i.e., burst-average rise velocity) of whether the cross-flow would impact the rise of the seep bubbles. One more interesting feature in this figure is that, in contrast to expectations, smaller bubbles (Deυ < 6.5 mm) are found only when the cross-flow was very small (less than approximately 5 cm s−1).
Cross-flow affects the rise velocity through degradation of bubble–bubble interactions by drifting the wake of leading bubbles away from the trailing ones (Marks 1973; Katz and Meneveau 1996). Using the turbulent wake theory (Schlichting and Gersten 2016), Wang and Socolofsky (2015b) developed a simple model based on the findings of Marks (1973) to predict the mean rise velocity considering the cumulative effect of multiple wakes on in-chain bubble dynamics, effect of velocity disturbance on drag coefficient, and the displacement caused by the cross-flow as
where the plus superscript denotes a normalized value (see the appendix for definitions of Q+ and ). As shown in Fig. 12b, our data appear to cluster around the relation with an R2 = 0.37. The negative values for arise from the uncertainties involved in estimating the terminal velocity of bubbles, such as bubble composition, oil content, and contamination by surfactants. Nevertheless, our results based on Eq. (4) suggests that the bubble rise velocity is somewhat correlated to the cross-flow in a nondimensional format, despite that the direct correlation between rise velocity and cross-flow is low (Fig. 12a). The agreement between our in situ measurements and the prediction model for in-chain bubbles [Eq. (4)] indicate the flow regimes in both cases were similar. In this flow regime, the upwelling effect is highly localized due to bubble–bubble interactions and the cross-flow can efficiently destruct the bubble–bubble interactions.
6. Summary and conclusions
This work outlines our efforts to estimate the spatial and temporal variation in emission of hydrocarbons from an offshore deep-sea marine seep over a long period of time. To record the release of hydrocarbon bubbles from the Mega Plume seep from 3 September 2017 to 2 February 2018, two newly designed VTLC systems were deployed in the Gulf of Mexico at GC600, at approximately 1180 m depth. To support the long-term measurements, Camera B recorded 15 s videos at 30 fps every 6 h for the entire campaign period, with illumination provided by a 2000 lm LED lamp over a period of 153 days. Here, we developed a sequence of algorithms for effectively processing the videos collected at qualities and quantities much lower than typical laboratory conditions, to deduce the dynamic properties of individual bubbles. Utilizing these algorithms, 30 consecutive frames (1 s) of all the records (568 records) were processed. Through dives with the ROV Millennium a 10 min video was collected and processed to provide independent reference data for the bubble sizes and velocities. We also used a customized funnel to measure the instantaneous discharge of hydrocarbons corresponding to a value of 6.4 cm3 s−1.
The VTLC records reveal finescale changes in the physical and spatial constraints on bubble venting that represent inherent variability for seeps in the deep ocean. The VTLC bubble rise velocities for the Clam Cluster were slightly faster than those for the Fault Cluster with an average of speed of and 21.0 ± 2.3 cm s−1, respectively; the mean rise velocities are attributed to the entire campaign. Over the last 4 weeks, the average rise velocity for the Fault Cluster is found to be 12% lower than those obtained from the ROV-based PTV analyses; however, the relative difference for the average size of the bubbles determined from the two techniques was less than 5%. The flux rate deduced from the VTLC records was in excellent agreement with our funnel measurements with approximately 2% difference. In general, our measurements of rise velocity were consistent with previous studies. In particular, our results are in excellent agreement with the rise velocities measured by Wang et al. (2016) at the Confetti vent, located about 10 m away from where our current measurements were taken, and was not active during our campaign.
Small values of β = O(0.1) in wb ∝ Qβ suggest that upwelling does not have a profound effect in enhancing the rise velocity of bubbles so close to the source in the range of in situ observed fluxes (up to 13.5 cm3 s−1) and cross-flow conditions (0–12 cm s−1). Further, our results indicate that the bubble rise velocity is moderately correlated to the cross-flow velocity at 0.5 mab, which signifies the elimination of bubble–bubble interactions, similar to the organized in-chain bubbly flow.
This work was funded by the Gulf of Mexico Research Initiative as RFP V and VI individual investigator grants. All video imagery has been submitted to GRIIDC with a unique dataset identifier number [VTLC data: Daniela Di Iorio, Andreas Thurnherr, Mahdi Razaz, James Kelly (2020) Video imaging of a naturally occurring hydrocarbon seep in the GC600 block of the Gulf of Mexico from 3 September 2017 to 2 February 2018. Distributed by Gulf of Mexico Research Initiative Information and Data Cooperative (GRIIDC), Harte Research Institute, Texas A&M University–Corpus Christi. doi:10.7266/n7-3466-rn36; ADCP data: Daniela Di Iorio, Andreas Thurnherr, Mahdi Razaz, James Kelly (2019) Hydrographic and ADCP data obtained from Oceaneering vessel Ocean Intervention II in the Gulf of Mexico from 18 June 2017 to 2 February 2018. Distributed by Gulf of Mexico Research Initiative Information and Data Cooperative (GRIIDC), Harte Research Institute, Texas A&M University–Corpus Christi. doi:10.7266/n7-qyn1-sx24]. We are grateful to the staff and ship crew of Oceaneering Inc., who made it possible to collect the data for this analysis. We are also thankful to the ECOGIG2 GOMRI funded consortia who loaned us their VTLCs for use in our project and AquaPix LLC, the manufacturer of the VTLC hardware. Thanks goes to James Kelly and Dr. Andreas Thurnherr, who helped with the deployment of all instruments.