Search USGS⌕ Search

SEARCH · Search USGS

Results for “Journal of Geophysical Research B: Solid Earth”

Search indexed USGS publications on groundwater, aquifers, geologic maps, mineral resources and earthquakes. Explore source records by subject and place.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 379 records · Page 21Linked to original sources
USGS2018

Quantifying uncertainty in cumulative surface slip along the Cucamonga Fault, a crustal thrust fault in southern California

Studies of historic earthquake ground surface ruptures show that displacements along strike are spatially variable. As a result, latest Quaternary slip rates developed from a spatially restricted set of cumulative displacement measurements may not accurately represent fault velocity. Here we examine the uncertainties associated with slip on the Cucamonga Fault, which is part of a network of faults that have generated damaging historical earthquakes in and around Los Angeles, California. Numerous scarps along its ~25‐km length are well expressed on alluvial fans. We make 310 measurements of vertical separation across the scarps using lidar data. We show that the dispersion of the vertical separations cannot be explained by our best estimates of analytical uncertainties alone. Additional epistemic uncertainties are required. We find that the magnitude of the required epistemic uncertainty is typically larger than analytical uncertainty by a factor of 3 and typically about 22% of the maximum vertical separation. These relationships appear to hold at several spatial scales. We examine three potential sources of epistemic uncertainty and find that none among surface age uncertainty, fault dip, and anthropogenic landscape alteration is likely sufficient to explain the overdispersion of the data, which suggests differences in cumulative strain along the strike of the fault. We calculate a range of dip‐slip rates between 0.4 and 2.6 mm/year. In light of our results, we suggest that future thrust‐fault slip‐rate studies adopt an epistemic uncertainty of 22% of the maximum value for vertical separation measurements, unless there are sufficient data to demonstrate otherwise.

California↗
USGS2019

Azimuthal seismic anisotropy of 70 Ma Pacific‐plate upper mantle

Plate formation and evolution processes are predicted to generate upper mantle seismic anisotropy and negative vertical velocity gradients in oceanic lithosphere. However, predictions for upper mantle seismic velocity structure do not fully agree with the results of seismic experiments. The strength of anisotropy observed in the upper mantle varies widely. Further, many refraction studies observe a fast direction of anisotropy rotated several degrees with respect to the paleospreading direction, suggesting that upper mantle anisotropy records processes other than 2D corner flow and plate‐driven shear near mid‐ocean ridges. We measure 6.0 ± 0.3% anisotropy at the Moho in 70 Ma lithosphere in the central Pacific with a fast direction parallel to paleospreading, consistent with mineral alignment by 2D mantle flow near a mid‐ocean ridge. We also find an increase in the strength of anisotropy with depth, with vertical velocity gradients estimated at 0.02 km/s/km in the fast direction and 0 km/s/km in the slow direction. The increase in anisotropy with depth can be explained by mechanisms for producing anisotropy other than intrinsic effects from mineral fabric, such as aligned cracks or other structures. This measurement of seismic anisotropy and gradients reflects the effects of both plate formation and evolution processes on seismic velocity structure in mature oceanic lithosphere, and can serve as a reference for future studies to investigate the processes involved in lithospheric formation and evolution.

Journal of Geophysical Research B: Solid Earth↗
USGS2018

Characteristic earthquake magnitude frequency distributions on faults calculated from consensus data in California

An estimate of the expected earthquake rate at all possible magnitudes is needed for seismic hazard forecasts. Regional earthquake magnitude frequency distributions obey a negative exponential law (Gutenberg‐Richter), but it's unclear if individual faults do. We add three new methods to calculate long‐term California earthquake rupture rates to the existing Uniform California Earthquake Rupture Forecast (UCERF3) efforts to assess method and parameter dependence on magnitude frequency results for individual faults. All solutions show strongly characteristic magnitude‐frequency distributions on the San Andreas and other faults, with higher rates of large earthquakes than would be expected from a Gutenberg‐Richter distribution. This is a necessary outcome that results from fitting high fault slip rates under the overall statewide earthquake rate budget. We find that input data choices can affect the nucleation magnitude‐frequency distribution shape for the San Andreas fault; solutions are closer to a Gutenberg‐Richter distribution if the maximum magnitude allowed for earthquakes that occur away from mapped faults (background events) is raised above the consensus threshold of M =7.6, if the moment rate for background events is reduced, or if the overall maximum magnitude is reduced from M =8.5. We also find that participation magnitude‐frequency distribution shapes can be strongly affected by slip‐rate discontinuities along faults that may be artifacts related to segment boundaries.

California↗
USGS2019

Earthquake stress drop and Arias Intensity

We determine earthquake stress drops directly from the Arias intensity database of NGA-West2. Arias intensity (Arias, 1970) is an engineering measure proportional to the integral of the absolute value of acceleration squared, over the significant duration of the signal. As such, it is closely related to root-mean-square acceleration, and can readily be connected to earthquake stress drop (Hanks and McGuire, 1981). Arias intensity records out to 100 km yield stable stress drops for moderate-to-large magnitude earthquakes, M6.5+; for smaller events ~M4.5 – 6.5, only closer-in records yield stable results. For the 116 events considered, stress drops are about 35% larger for Class 1 mainshocks than for traditional on-fault Class 2 aftershocks, and smaller for those aftershocks close to the main fault plane. Aftershock stress drops show large variability, however, implying that on average they re-rupture weakened patches, but can also rupture intact rock or high-stress asperities. We observe an increase of stress drop with earthquake depth similar to that of other studies but do not find any significant faulting mechanism dependence. The variability of the Arias intensity-based stress drop is lower than that of eGf-based stress drops from Baltay et al. (2010, 2011), and nearly on par with variability seen in ground-motion prediction equations. The Arias intensity stress drop is a novel and promising method to estimate stress drop without the need for path and site corrections, and yields further insight into the connection between source physics and ground-motion.

Journal of Geophysical Research B: Solid Earth↗
USGS2018

Earthquakes induced by hydraulic fracturing are pervasive in Oklahoma

Wastewater disposal is generally accepted to be the primary cause of the increased seismicity rate in Oklahoma within the past decade, but no statewide analysis has investigated the contribution of hydraulic fracturing (HF) to the observed seismicity or the seismic hazard. Utilizing an enhanced seismicity catalog generated with multi‐station template matching from 2010‐2016 and all available hydraulic fracturing information, we identified 274 HF wells that are spatiotemporally correlated with bursts of seismicity. The majority of HF induced seismicity cases occurred in the SCOOP/STACK plays, but we also identified prominent cases in the Arkoma Basin as well as some more complex potential cases along the edge of the Anadarko Platform. For HF treatments where we have access to injection parameters, modeling suggests poroelastic stresses are likely responsible for seismicity, but we cannot rule out direct pore pressure effects as a contributing factor. In all of the 16 regions we identified, ≥75% of the seismicity correlated with reported HF wells. In some regions, >95% of seismicity correlated with HF wells and >50% of the HF wells correlated with seismicity. Overall, we found ~700 HF induced earthquakes with M ≥ 2.0, including 12 events with M 3.0‐3.5. These findings suggest state regulations implemented in 2018 that require operators in the SCOOP/STACK plays to take action if a M > 2 earthquake occurs could have a significant impact on future operations.

Oklahoma↗
USGS2019

Ground-motion residuals, path effects, and crustal properties: A pilot study in southern California

To improve models of ground motion estimation and probabilistic seismic hazard analyses, the engineering seismology field is moving toward developing fully nonergodic ground motion models, models specific for individual source‐to‐site paths. Previous work on this topic has examined systematic variations in ground‐motion along particular paths (from either recorded or simulated earthquake data) and has not included physical properties of the path. We present here a framework to include physical path properties, by seeking correlations between ground motion amplitudes along specific paths and crustal properties, specifically seismic velocity and anelastic attenuation, along that path. Using a large data set of small‐magnitude earthquakes recorded in Southern California, we find a correlation between the gradient of seismic S wave velocity and the path term residual, after accounting for an average geometric spreading and anelastic attenuation, indicating that heterogeneity in crustal velocity primarily controls the path‐specific attenuation. Even in aseismic regions, details of path‐specific ground motion prediction equations can be developed from crustal structure and property data.

California↗
USGS2019

Four major Holocene earthquakes on the Reelfoot fault recorded by sackungen in the New Madrid seismic zone, USA

Three sequences of well-documented, major ~M7+ earthquakes (1811-1812 CE, ~1450 CE, and ~900 CE) in the New Madrid seismic zone, USA, contribute significantly to seismic hazard in the region. However, it is unknown whether this <550 yr recurrence interval has been constant throughout the Holocene given limited geomorphic evidence of prior earthquakes. We extend the record of paleoearthquakes along the Reelfoot fault via investigation of ridge-top gravitational failure features, interpreted as sackungen. The sackungen occur in bluffs along the eastern margin of the Mississippi River floodplain and are concentrated near (<15 km) the southwest-dipping Reelfoot reverse fault. A paleoseismic trench excavated across sackungen at the Paw Paw site exposed four packages of colluvial sediment that postdate 30-11 ka Peoria loess. We interpret the colluvial packages to have been deposited following episodic failure of the sackungen as a result of strong ground motions from the following sequence of earthquakes: event 4, 1640 ± 1730 BCE; event 3, 340 ± 670 CE; event 2, 1430 ± 380 CE; and event 1, 1810 ± 50 CE (2-sigma). Event timing corresponds to previously documented earthquakes and represents the longest archive of paleoearthquakes on the Reelfoot fault. If the trenched sackungen record all major Reelfoot fault earthquakes, our observations in combination with prior investigations indicate a period of quiescence from at least 11 – 4.7 ka, followed by four major seismic events culminating in the 1811-1812 CE sequence. This clustered earthquake recurrence helps place bounds on seismic-hazard and geodynamic models in the New Madrid seismic zone.

Arkansas, Missouri↗
USGS2019

Depth determination of the 2010 El Mayor‐Cucapah earthquake sequence (M ≥ 4.0)

The 2010 M W 7.2 El Mayor‐Cucapah earthquake ruptured a zone of ~120 km in length in northern Baja California. The geographic distribution of this earthquake sequence was well constrained by waveform relocation. The depth distribution, however, was poorly determined as it is near the edge of, or outside, the Southern California Seismic Network. Here we use two complementary methods to constrain the focal depths of moderate‐sized events ( M ≥ 4.0) in this sequence. We first determine the absolute earthquake depth by modeling the regional depth phases at high frequencies (~1 Hz). We mainly focus on Pn and its depth phases pPn and sPn , which arrive early at regional distance and are less contaminated by crustal multiples. To facilitate depth phase identification and to improve signal‐to‐noise ratio, we take advantage of the dense Southern California Seismic Network and use array analysis to align and stack Pn waveforms. For events without clear depth phases, we further determine their relative depths with respect to those with known depths using differential travel times of the Pn , direct P , and direct S phases recorded for event pairs. Focal depths of 93 out of 122 M ≥ 4.0 events are tightly constrained with absolute uncertainty of about 1 km. Aftershocks are clustered in the depth range of 3–10 km, suggesting a relatively shallow seismogenic zone, consistent with high surface heat flow in this region. Most aftershocks are located outside or near the lower terminus of coseismic high‐slip patches of the main shock, which may be governed by residual strains, local stress concentration, or postseismic slip.

Baja California↗
USGS2019

Peak ground displacement saturates exactly when expected: Implications for earthquake early warning

The scaling of rupture properties with magnitude is of critical importance to earthquake early warning (EEW) systems that rely on source characterization using limited snapshots of waveform data. ShakeAlert, a prototype EEW system that is being developed for the western United States, provides real-time estimates of earthquake magnitude based on P-wave peak ground displacements measured at stations triggered by the event. The algorithms used in ShakeAlert assume that the displacement measurements at each station are statistically independent and that there exists a linear and time-independent relation between log peak ground displacement and earthquake magnitude. Here we challenge this basic assumption using a comprehensive database of more than 130,000 vertical component waveforms from M4.5-M9 earthquakes occurring near Japan from 1997 through 2017 and recorded by the K-NET and KiK-net strong-motion networks. By analyzing the time-evolution of P-wave peak ground displacements for these earthquakes, we show that there is a break, or saturation, in the magnitude-displacement scaling that depends on the length of the measurement time window. We demonstrate that the magnitude at which this saturation occurs is well-explained by a simple and non-deterministic model of earthquake rupture growth. We then use the predictions of this saturation model to develop a Bayesian framework for estimating posterior uncertainties in real-time magnitude estimates which incorporates the expected time-dependence of the peak displacement measurements.

Journal of Geophysical Research B: Solid Earth↗
USGS2019

Vertical coseismic offsets from differential high-resolution stereogrammetric DSMs: The 2013 Baluchistan, Pakistan earthquake

The recent proliferation of high-resolution (< 3-m spatial resolution) digital topography datasets opens a spectrum of geodetic applications in differential topography, including the quantification of coseismic vertical displacement fields. Most investigations of coseismic vertical displacements to date rely, in part, on pre- or post-event lidar surveys that are intractable or non-existent in many locales. Stereogrammetric digital surface models (DSMs) derived from high-resolution satellite optical imagery provide a new avenue for the retrieval of spatially-dense vertical coseismic displacements on a global scale. In this study, we generated 2-m resolution pre- and post-seismic DSMs from satellite optical imagery spanning the 2013 Mw7.7 Baluchistan strike-slip earthquake that occurred on the Hoshab fault in southern Pakistan. We applied the Iterative Closest Point algorithm to the DSMs to quantify the coseismic vertical displacement field at a spatial resolution of 10-30 m and to generate 3D coseismic strain tensors. We found that across-fault vertical offsets alternated between uplift and subsidence and varied between ~1-3 m in a non-systematic manner along the Hoshab fault. We show that the pre-existing topography and near-fault geomorphology are variably consistent and inconsistent with the displacement kinematics of the 2013 earthquake, and we argue that these relationships highlight varied slip sense history along the Hoshab fault. Notably, topography along the southern extents of the Hoshab fault requires different surface displacement kinematics than occurred in the 2013 earthquake, suggesting that the Hoshab fault accommodates varying senses of slip (bimodal slip) through time.

Baluchistan Province↗
USGS2019

Delayed dynamic triggering of disposal-induced earthquakes observed by a dense array in Northern Oklahoma

Recent increases in earthquake occurrence rates in Oklahoma have been linked to the injection of large volumes of saltwater, a byproduct of oil and gas extraction. Here we present a detailed study of remote earthquake triggering in an area of active injection‐induced seismicity in northern Oklahoma using data from the LArge‐n Seismic Survey in Oklahoma (LASSO) temporary array and nearby permanent broadband seismic stations. We estimate changes in earthquake rates and calculate the Coulomb failure stress changes on potential receiver faults due to passing teleseismic surface waves. A statistically significant increase in seismicity is observed ∼8 hr after the 16 April 2016 M w 7.8 Ecuador earthquake. The Coulomb stress changes associated with the Ecuador earthquake are on the order of ∼1 kPa. Physical mechanisms consistent with the observed dynamic stress threshold include failure driven by activation of aseismic slip or hydrological response of the fault system.

Oklahoma↗
USGS2019

Offshore landslide hazard curves from mapped landslide size distributions

We present a method to calculate landslide hazard curves along offshore margins based on size distributions of submarine landslides. The method analyzes ten different continental margins, that were mapped by high-resolution multibeam sonar with landslide scar areas measured by a consistent GIS procedure. Statistical tests of several different probability distribution models indicate that the lognormal model is most appropriate for these siliciclastic environments, consistent with an earlier study of the U.S. Atlantic margin [Chaytor et al., 2009]. Parameter estimation is performed using the maximum likelihood technique and confidence intervals are determined using likelihood profiles. Pairwise comparison of size distributions for the ten margins indicates that the U.S. Atlantic and Queen Charlotte margins are different than most other margins. These margins represent end members, with the U.S. Atlantic margin having the highest mean scar area and the Queen Charlotte margin, the lowest. We demonstrate that empirical, offshore landslide hazard curves can be developed from the landslide size distributions, if the duration of mapped landslide activity is known. This study indicates that the shape parameter of the size distribution is similar among all ten margins and thus the shape of the hazard curves is also similar. Significant differences in hazard curves among the margins are therefore related to differences in mean sizes and, potentially, differences in the duration of landslide activity.

Journal of Geophysical Research B: Solid Earth↗
USGS2019

Locating shallow seismic sources with waves scattered by surface topography: Validation of the method at the Nevada Test Site

Accurate and robust source location is fundamental in seismology. Previously, we proposed a new full waveform location method using waves scattered by topography near the source, and we demonstrated its potential in obtaining accurate source location using synthetic data. In the work presented here, we validate this method with observed seismic data from the 1993 Non-Proliferation Experiment, a chemical explosion with a precisely known location in a region with moderate topography at the Nevada Test Site. We select the first arrivals (Pn/Pg) and their immediate codas to determine the source location and compare our solution with the known location. We use a collocated-grid finite-difference method to calculate the strain Green’s tensor in a grid-search volume containing the source, and obtain three-component synthetic waveforms at 12 broadband seismic stations at regional distances using source-receiver reciprocity. We assess the solution using a least-squares misfit between the observed and synthetic waveforms. When scattered coda waves are incorporated, the best solution is within a few hundreds of meters of the exact source location, and the estimated uncertainty of the solution is reduced compared to the waveform based solution using only the P waves. The solution is robust to the choice of the frequency content and to the addition of random velocity heterogeneity. We conclude that the full waveform source location method is effective for real seismic data, though more validation tests like this one are needed to further understand its efficacy for source-station geometry, roughness of topography, signal quality and other factors.

Nevada↗
USGS2019

Interseismic quiescence and triggered slip of active normal faults of Kīlauea Volcano’s south flank during 2001-2018

The mobile south flank of Kīlauea Volcano hosts two normal fault systems, the Koa'e fault system (KFS) and the Hilina fault system (HFS). In historical time, at least three M>6.5 earthquakes have occurred on the basal detachment of the Kīlauea Volcano's south flank, with the most recent being the 4 May 2018 M6.9 earthquake. Here we analyze kinematic Global Positioning System data collected from 2001 to 2017 and interferometric synthetic aperture radar data before, during, and after the 2018 M6.9 earthquake to determine the crustal motion across the HFS and KFS faults. Our results indicate that the HFS faults did not significantly slip during the interseismic period from 2007 to 2011. Despite its substantial magnitude, interferometric synthetic aperture radar (InSAR) data show that the 2018 M6.9 earthquake triggered subcentimeter level slip along sections of the previously mapped HFS branches. Up to 20 cm of offset occurred on what appears to be a newly formed (or previously unknown) fault near the eastern end of the HFS. During the 3 months following the M6.9 earthquake, up to ~30 cm of slip occurred along the KFS, which helps accommodate rapid large‐scale subsidence of Kīlauea's summit region as large volumes of summit reservoir magma fed the lower East Rift Zone eruption. The HFS appears to activate only in concert with large earthquakes on the basal detachment. The KFS, on the other hand, moves both seismically during small local earthquakes and aseismically in response to nearby earthquakes and caldera subsidence.

Hawaii↗
USGS2019

Slow-growing and extended-duration seismicity swarms: Reactivating joints or foliations in the Cahuilla Valley Pluton, Central Peninsular Ranges, Southern California

Three prolific earthquake swarms and numerous smaller ones have occurred since 1980 in the Mesozoic igneous plutonic rocks of the Perris block of the Peninsular Ranges, Southern California. The major swarms occurred in 1980–1981, 1983–1984, and 2016–2018, with the latest swarm still ongoing. These swarms have no clear mainshock, with the largest events of M L 3.6, M L 3.7, and M w 4.4. Each successive swarm had larger cumulative seismic moment release with about 314 and 411 events of M ≥ 1.5, while the third swarm has produced about 451 events of M ≥ 1.5 (as of September 2018). The concurrent strike‐slip faulting occurred on north to northwest striking planes but with no orthogonal northeast trending seismicity alignments. These shallow swarms are probably driven by intrablock Pacific‐North America plate boundary stress loading of the two bounding major late Quaternary strike‐slip faults, the Elsinore and San Jacinto faults. The state of stress within the Cahuilla Valley pluton has a ~40° angle between the maximum principal stress and the average trend of the swarms, suggesting that migrating pore fluid pressures aid in the formation and growth of zones of weakness. These swarms, which last more than 600 days each, exhibit clear bilateral spatial migration for distances of up to ~7–8 km and reach their full length in about 20 months. The slow spatial‐temporal development of the swarms corresponds to a fluid diffusivity of 0.006 to 0.01 m 2 /s, consistent with very low permeability rocks as expected for this block. There is no geodetic or other evidence for a slow slip event driving the swarms.

California↗
USGS2019

Heat flow in the Western Arctic Ocean (Amerasian Basin)

From 1963 to 1973 the U.S. Geological Survey (USGS) measured heat flow at 356 sites in the Amerasian Basin (Western Arctic Ocean) from a drifting ice island (T-3). The resulting measurements, which are unevenly distributed on Alpha-Mendeleev Ridge (AMR) and in Canada and Nautilus basins, greatly expand available heat flow data for the Arctic Ocean. Average T-3 heat flow is ~54.7 ± 11.3 mW m-2, and Nautilus Basin, including Mendeleev Plain, is the only well-surveyed area (~13% of data) with significantly higher average heat flow (63.8 mW m-2). Heat flow and bathymetry are not correlated at a large scale, and turbiditic surficial sediments (Canada and Nautilus basins) have higher heat flow than the sediments that blanket the AMR. Thermal gradients are mostly near-linear, implying that conductive heat transport dominates and that near-seafloor sediments are in thermal equilibrium with overlying bottom waters. Combining the heat flow data with modern seismic imagery suggests that some of the observed heat flow variability may be explained by local changes in sediment thickness or lithology or the presence of basement faults that channel circulating seawater. A thermal model that incorporates thermal conductivity variations along a profile from Canada Basin (thick sediment on mostly oceanic crust) to Alpha Ridge (thin sediment over thick magmatic units associated with the High Arctic Large Igneous Province) predicts heat flow lower than that observed on Alpha Ridge. This, along with other observations, implies that circulating fluids modulate conductive heat flow and contribute to high variability in the T-3 dataset. .

Journal of Geophysical Research B: Solid Earth↗
USGS2019

Crustal magmatism and anisotropy beneath the Arabian Shield - A cautionary tale

Volcanism in Saudi Arabia includes a historic eruption close to the holy city of Al Madinah. As part of a volcanic hazard assessment of this area, magnetotelluric (MT) data were collected to investigate the structural setting, the distribution of melt within the crust, and the mantle source of volcanism. Interpretation of a new 3‐D resistivity model includes a shallow graben beneath thin lava fields (Harrats), a melt‐free upper crust, and decompression melting in the asthenosphere below thin lithosphere. Within the lower crust the model images elongate conductivity anomalies, one of which was attributed in a previous MT study to melt. The regional MT data, combined with perspective from geology and geophysical modeling, suggest the lower crust is anisotropic with no interconnected melt zones. These divergent interpretations have distinct hazard implications and highlight the importance of large survey aperture and anisotropic modeling to MT studies of volcanic regions. Lower‐crustal anisotropy extends beyond the Harrat, with the most conductive direction oriented N10°E and a factor of 3–5, determined from 2‐D anisotropic inversion, between the most and least conductive directions. The enhanced conductivity is likely due to interconnected grain boundary graphite, while the anisotropy direction reflects either frozen‐in fabric from Neoproterozoic stabilization of the Arabian Shield or modern ductile deformation driven by channelized asthenospheric flow coupled through a thin rigid mantle lid. Asthenospheric melt is interpreted to transect the crust primarily through diking, with limited melt storage and short residence times in the crust.

Arabian Shield↗
USGS2019

Introduction to special issue on gas hydrate in porous media: Linking laboratory and field‐scale phenomena

The proliferation of drilling expeditions focused on characterizing natural gas hydrate as a potential energy resource has spawned widespread interest in gas hydrate reservoir properties and associated porous media phenomena. Between 2017 and 2019, a Special Section of this journal compiled contributed papers elucidating interactions between gas hydrate and sediment based on laboratory, numerical modeling, and field studies. Motivated mostly by field observations in the northern Gulf of Mexico and offshore Japan, several papers focus on the mechanisms for gas hydrate formation and accumulation, particularly with vapor phase gas, not dissolved gas, as the precursor to hydrate. These studies rely on numerical modeling or laboratory experiments using sediment packs or benchtop micromodels. A second focus of the Special Section is the role of fines in inhibiting production of gas from methane hydrate, controlling the distribution of hydrate at a pore scale, and influencing the bulk behavior of seafloor sediments. Other papers fill knowledge gaps related to the physical properties of hydrate-bearing sediments and advance new approaches in coupled thermal-mechanical modeling of these sediments during hydrate dissociation. Finally, one study addresses the long-standing question about the fate of methane hydrate at the molecular level when CO2 is injected into natural reservoirs under hydrate-forming conditions.

Journal of Geophysical Research B: Solid Earth↗