Earthquakes track subduction fluids from slab source to mantle wedge sink

Fluids trigger earthquakes as they flow upward along the subducted slab and escape into the overlying rocks.


This PDF file includes:
Supplementary Text Mantle wedge seismicity in other subduction zones Episodicity of mantle wedge earthquakes and velocity of fluid migration Fig. S1. Along-trench profiles of mantle wedge seismicity and P-velocity structure, plotted as seen from the trench. Fig. S2. Temporal evolution of seismicity in the Tripoli cluster. Fig. S3. P-velocity to S-velocity (V p /V s ) ratio structure beneath western Greece.  Table S1. Seismograph networks from western Greece used in the waveform processing. Table S2. Focal mechanism solutions of deep earthquakes in the Western Hellenic subduction zone. Table S3. Locations of mantle wedge seismicity displayed in cross sections in Fig. 4. External Data file S1. Deep earthquake hypocenters in Greece. External Data file S2. Deep earthquake focal mechanisms in Greece. External Data file S3. Earthquake arrival time picks. External Data file S4. Model of the subduction plate interface. External Data file S5. Thermal structure model of the subduction zone. References (59-81)

Supplementary Text
Mantle wedge seismicity in other subduction zones Figure 4 shows the worldwide locations where clusters of mantle wedge seismicity have been observed. Cross sections from these locations reveal how the Tripoli cluster (A) compares to selected clusters offshore Sanriku (Japan, Fig. 4 B), eastern Crete (Greece, C), below Raukumara (New Zealand, D), and below Martinique (lesser Antilles, E). All occurrences appear in cold subduction zones with generally dry mantle wedge corners (see map in Fig. 4). Here, we briefly describe each region.
Cross section A, through the Tripoli cluster, shows the deep earthquakes depicted in Fig. 3 of the main text. In addition, it includes shallow earthquake hypocenters published by NOA (2011 -2017, (33)) that fulfill the following criteria: location errors smaller than 3 km, minimum number of 12 phase picks or 8 observing stations, and azimuthal gap of less than 180º.
Cross section B, eastern Crete, comprises earthquakes from two sources: i) the NOA catalog, and ii) the catalog from regional studies in refs. (23,59). We select NOA hypocenters with the same criteria as in cross section A, except that we plot hypocenters from all depths and loosen the gap criterion to 225º for earthquakes observed with more than 18 picks or 12 stations, to include offshore earthquakes south of Crete. The second catalog is based on multiple dense temporary seismic experiments on-and offshore Crete. These were deployed to study microseismic activity of the subduction zone where permanent monitoring is hampered by sparse land station coverage. With these temporary networks, Sodoudi et al. (23) imaged the plate interface and Moho discontinuities in the region with PS-receiver function analysis. The location of the plate interface and hypocenters in Fig. 4 B indicate two regions of high seismicity within subducting crust: a shallow segment updip from the ~35 km interface depth, and a deep segment downdip from the ~50 km interface depth. The two segments are separated by a ~60 km wide region of reduced seismicity. At the upper tip of the deep segment, earthquake activity appears to stretch into the mantle wedge and connect to seismicity clusters in the lower part of the overriding crust. The high-low-high seismicity pattern does not affect seismicity within the subducting mantle (> 10 km below the interface), which extends continuously below the crustal gap.
Cross section C, offshore Sanriku, shows hypocenters relocated by double-difference inversion in ref. (20). Seismic discontinuities in the region are constrained by P-S conversions from deep local earthquakes (20). The upper and lower seismic plane of the WBZ dipping westward are clearly separated. A dense cluster of seismicity clearly exists on the interface at 45 km depth, above which a large region of seismic activity stretches into the trenchward part of the mantle wedge. A ~20 km wide gap in crustal seismicity is present ~20 km updip from the interface cluster, but some earthquakes occur right updip from the interface cluster. We note that this region falls right above the shallow limit of the lower seismicity plane, where some hypocenters appear to from a link between the lower and upper planes. This may reflect the existence of a vertical fluid pathway from the lower to the upper seismic zone.
Cross section D, at Raukumara, shows mantle wedge seismicity analyzed in ref. (21). Hypocenters were obtained by relocation in a 3-D velocity model (60), with plate interface and Moho discontinuities constrained by seismicity, local earthquake arrival time tomography, and a seismic reflection image. The cross section clearly shows an intense cluster of seismicity extending ~15 km upward from a ~40 km thick WBZ dipping to the west. WBZ seismicity stretches from the interface well into the subducting mantle. Updip of the mantle wedge cluster, there are hardly any earthquakes in a ~30 km wide section of subducting crust. This seismicity gap only exists in the subducting crust (as in Crete), with earthquake activity appearing uninterrupted in the subducting mantle below. In the context of our proposed fluid migration model, this dichotomy suggests two clearly separated migration pathways in the subducting crust and mantle.
Cross section E, across Martinique, shows seismicity that was first analyzed in ref.
(61) and relocated in a 3-D velocity model in ref. (22). The plate interface and Moho discontinuities are from ref. (22) and were constrained by seismicity, local earthquake arrival time tomography, and seismic reflection imaging (22,61). A dense cluster of seismicity appears on the interface at ~45 km depth and is overlain by a region of distributed mantle wedge seismicity. It is not immediately clear why the mantle wedge seismicity is more diffuse than in other regions. But it appears that interface seismicity extends over a larger depth segment here. This could indicate that venting of fluids occurs over a broader region below Martinique. Updip from the main cluster of interface seismicity (~45 km depth), the subducting crust is nearly aseismic. Figure S1 shows the occurrence in time of earthquakes in the Tripoli cluster, with a distinction between earthquakes in the slab, on the interface and in the mantle wedge. These earthquakes, unlike their counterparts in Japan (10), do not appear to show any timedependent behavior. To investigate whether this discrepancy is due to a difference in subduction rate, we turn to recent studies that have researched the mechanisms by which fluids produced by dehydration reactions flow through the slab and into the mantle wedge. The mechanisms are debated, but a strong candidate is compaction-driven flow in the form of porosity waves (9, 27, 62).

Episodicity of mantle wedge earthquakes and velocity of fluid migration
Even though the two-phase flow models currently used are strictly only valid for the ductile domain of (hot) mantle wedge and lower crust, recent theoretical studies suggest that reaction-induced compaction changes may lead to analogous porosity waves in the brittleelastic domain (63, 64). These findings are important, because if we can assume that fluid is transported by porosity waves, then we can evaluate how a difference in subduction rate may affect the velocity at which fluid migrates through the system.
Here, we assume that compaction-driven flow can occur in the subducting crust and estimate how the fluid velocity varies as a function of subduction rate. First, we can establish that an increase in subduction rate linearly affects the fluid production rate, as slab rocks pass the dehydration window in a shorter time span. Second, since the formation of porosity is caused by the production of fluids (65), we can establish that porosity is also linearly related to subduction rate. Third, based on the knowledge that, at low porosities, permeability is approximately related to the cube of porosity (eq. 17 in (65)), we can establish that permeability is related to the cube of the subduction rate. Lastly, as the fluid velocity is equal to the ratio of permeability to porosity, we find that fluid velocity is proportional to the square of the subduction rate. We can now compare fluid migration rates between NE Japan with western Greece. In NE Japan, where the subduction velocity is 82.7 mm/yr (8), mantle wedge seismicity has been shown to occur episodically with a 1-year cycle. The subduction velocity is about 2.4 times higher than in western Greece (35 mm/yr (13)), where we have not detected any episodicity. A 2.4-fold slower subduction velocity in western Greece would result in a fluid migration velocity that is considerably smaller, i.e., by a factor of 2.4 2 6, than in NE Japan. With an expected cycle of 6 years, this episodicity would be very difficult to detect robustly in Greece yet, where our high-resolution data are limited to a time span of 11 years. and Kremidi cluster along one main subvertical feature. In contrast, below Tripoli, they form four subclusters that branch out from the interface patch in different directions (A). These observations suggest that fluids generally escape the slab through highly localized interface vents, but once in the mantle wedge they can sometimes follow multiple pathways upwards.       profile ran parallel to the scattered wave image shown in Fig. 3 A, about 20 km farther southeast. A vertically stretched low-resistivity anomaly coincides precisely with the cluster of interface seismicity, suggesting the presence of highly conductive material, e.g. saline slab fluid at that location. Vertically, the anomaly is likely stretched because depth tends to be poorly constrained in such magnetotelluric images. Table S1. Seismograph networks from western Greece used in the waveform processing.
Here we indicate the time periods that we used for the processing of waveform data from each network. We only used data that were publicly available through data services of the Federation of Digital Seismograph Networks at the time of the study.   External Data file S1. Deep earthquake hypocenters in Greece.
Link: https://doi.pangaea.de/10.1594/PANGAEA.894348 List of relocated hypocenters of 2172 deep earthquakes of western Greece. The columns in the list contain information on the earthquake's time, latitude, longitude, depth, local magnitude, number of observations (number of P plus S-picks), the horizontal distance along the dip direction in km (x), the horizontal distance along the strike direction in km (y), error in x-direction (km), error in y-direction (km), error in depth (km), relative error in xdirection (km), relative error in y-direction (km), relative error in depth, logical indicator whether the hypocenter stems from double-difference relocation, distance from the slab top in km (positive is above the slab top).
External Data file S2. Deep earthquake focal mechanisms in Greece. Here we corrected the model slightly such that the clusters of interface earthquakes fall precisely on the interface model. The interplate nature of these earthquakes is supported by their focal mechanisms (thrust solutions) and by their alignment parallel to the plate interface within their respective clusters.

Link
External Data file S5. Thermal structure model of the subduction zone.
Link: https://doi.pangaea.de/10.1594/PANGAEA.894350 Temperature model for the present-day Western Hellenic Subduction Zone, as displayed in Fig. 3A. The model is provided in netCDF format and contains a grid of distance along the dip direction (x, in km), depth (km) and temperature-values (ºC) that describe the twodimensional temperature field. The creation of this model is described in the "Thermal and phase stability Modeling" section in the methods.