Constraining the maximum depth of brittle deformation at slow- and ultraslow-spreading ridges using microseismicity

The depth of earthquakes along mid-ocean ridges is restricted by the relatively thin brittle lithosphere that overlies a hot, upwelling mantle. With decreasing spreading rate, earthquakes may occur deeper in the lithosphere, accommodating strain within a thicker brittle layer. New data from the ultraslow-spreading Mid-Cayman Spreading Center (MCSC) in the Caribbean Sea illustrate that earthquakes occur to 10 km depth below seafloor and, hence, occur deeper than along most other slow-spreading ridges. The MCSC spreads at 15 mm/yr full rate, while a similarly well-studied obliquely opening portion of the Southwest Indian Ridge (SWIR) spreads at an even slower rate of ∼ 8 mm/yr if the obliquity of spreading is considered. The SWIR has previously been proposed to have earthquakes occurring as deep as 32 km, but no shallower than 5 km. These characteristics have been attributed to the combined effect of stable deformation of serpentinized mantle and an extremely deep thermal boundary layer. In the context of our MCSC results, we reanalyze the SWIR data and find a maximum depth of seismicity of 17 km, consistent with compilations of spreading-rate dependence derived from slow- and ultraslow-spreading ridges. Together, the new MCSC data and SWIR reanalysis presented here support the hypothesis that depth-seismicity relationships at mid-ocean ridges are a function of their thermal-mechanical structure as reflected in their spreading rate.


INTRODUCTION
Seismicity is generally restricted to the mechanically strong lithosphere (Chen and Molnar, 1983), which, along mid-ocean ridges (MORs), may extend for several kilometers beneath the seafloor. At depth, increasing temperature eventually results in plastic deformation, defining a brittle-to-plastic transition and the maximum depth of seismic faulting (Searle and Escartin, 2004). The thermal structure of MOR lithosphere, and thus the thickness of the seismogenic zone, are expected to correlate with spreading rate (Morgan and Chen, 1993). Indeed, earthquakes 4.5 in magnitude or larger are absent along ridge segments spreading at fast and intermediate rates, but are observed along slow-and ultraslow-spreading ridges, where they show a deepening with decreasing spreading rate (Huang and Solomon, 1988).
To gain further insight into the mechanical behavior of ultraslow-spreading lithosphere, we conducted a microseismicity survey of the Mid-Cayman Spreading Center (MCSC), an ultraslowspreading center in the Caribbean Sea (Fig. 1). To test our MCSC findings, we also reanalyzed microseismicity data from the similarly ultraslow-spreading Southwest Indian Ridge (SWIR) between 13°E and 14°E, where unusually deep hypocenters have been reported Schmid and Schlindwein, 2016). Our results provide new insight into the depth dependence of ultraslow-spreading microseismicity and suggest that the unusually deep hypocenters observed at the SWIR result from a processing artifact. We place our results in context using a compilation of previously published microseismicity surveys from the global mid-ocean ridge system to assess worldwide trends.

MID-CAYMAN SPREADING CENTER MICROSEISMICITY
The MCSC accommodates 15 mm/yr of east-west extension between the North American plate and the Caribbean plate (Hayman et al., 2011), with a deep axial valley of up to 5.5 km in depth and corresponding cold mantle potential temperature (Klein and Langmuir, 1987). In April 2015, we deployed 21 short-period and four broadband ocean-bottom seismographs (OBSs) in the axial region of the Cayman Trough (Fig. 1). These OBSs recorded continuously for up to 16 days. Most OBSs were located near the center of the spreading segment around the Mount Dent oceanic core complex at 18°25′N (Hayman et al., 2011), covering the median valley for 60 km along its length. Instrument spacing was 5-7 km, with an additional three OBSs deployed at the segment ends to assess seismicity along the entire 100-km-long MCSC.
We detected 292 earthquakes, an average of ∼20 events per day, with moment magnitudes of 0.4 < M w < 3.2. These events were located using P-wave and S-wave phase arrival times picked from waveform data (see the GSA Data Repository 1 ). Focal parameters were calculated using a probabilistic, nonlinear earthquake location algorithm (Lomax et al., 2000), and travel times were calculated using a one-dimensional velocity-depth profile derived from a seismic line shot through the seismic network . In this model, P-wave velocity (Vp) increases from ∼3-4 km/s near the seafloor to 7-8 km/s at 3-5 km depth. The velocity-depth structure as well as geological investigations suggest that volcanic crust occupies most of the axial valley seafloor (Haughton et al., 2019), although exhumed mantle occurs locally (Hayman et al., 2011). S-wave velocity (Vs) was derived using the ratio between P-and S-wave velocity (Vp/Vs), approximated from a cross plot between S-P time and origin time ( Fig. DR1 in the Data Repository). This yielded a Vp/Vs ratio of ∼1.8 for the uppermost crust (<4 km) and ∼1.75 for the mantle. To reduce the impact of non-modeled three-dimensional wave propagation effects, station corrections were calculated and iteratively updated for earthquakes observed by six or more stations that had a station gap of <180°. Station corrections reduced the initial root mean square (RMS) misfit from ∼0.12 s to ∼0.07 s, and the Gaussian distribution of residuals shows that the resulting station terms are appropriate (Fig. DR2).
Microseismicity clusters at the Mount Dent oceanic core complex ( Fig. 1), highlighting the detachment, and along a linear feature interpreted as an axial-volcanic ridge (AVR; Harding et al., 2017). In the south, the AVR, interpreted as a horst by Haughton et al. (2019), traverses the median valley, running obliquely to the overall trend of the MCSC and approaching Mount Dent. To the north, the AVR runs roughly parallel to the trend of the MCSC. Both the Von Damm hydrothermal vent field at Mount Dent and the Beebe vent field to its northeast within the median valley (Connelly et al., 2012) show little seismicity (Fig. 1). However, with an OBS spacing of 5-7 km, this station spacing is likely too large to record small earthquakes caused by thermal cracking (e.g., Sohn et al., 1998).
Well-located earthquakes occur between 4 and 8 km depth, reaching a maximum of 10 km below seafloor (Figs. 1C and 1D; Fig. DR3). The depth of microseismicity also appears to increase toward the segment ends. Most focal mechanisms (Table DR1), derived from P-wave onset first motions, suggest that normal faulting predominates, as might be expected in an extensional tectonic regime. There is evidence for transfer (strike-slip) faulting and compressional earthquakes as well. However, most importantly, the depth range of earthquakes observed at the MCSC is consistent with that expected from the global seismicity temperature-depth relationship. Schlindwein and Schmid (2016) suggested that, in contrast to the MCSC, the SWIR seismicity pattern is significantly different from that expected from the global temperature-depth relationship. The SWIR forms the boundary between the African and Antarctic plates, diverging at an average velocity of 14-16 mm/yr. The area between 13°E and 14°E (Fig. DR4) is part of the so-called oblique super-segment, where the spreading direction is oblique to the trend of the ridge axis (e.g., Dick et al., 2003). In December 2012, ten OBSs were deployed for 1 yr . Schlindwein and Schmid (2016) analyzed the recorded microseismicity to reveal an aseismic zone in the upper 5-10 km of the lithosphere and unusually deep earthquakes below that, extending to depths of 32 km (Figs. 2B and 2C). This result contradicts the implications of existing thermal-mechanical models of the lithosphere. Here, we test this result against that of the MCSC by reanalyzing the first six months of the deployment when coverage was best.

SWIR OBLIQUE SUPER-SEGMENT MICROSEISMICITY
In our analysis, we adopted the same methodology as applied at the MCSC above. The iterative update of station terms reveals unusually large delays of the S-waves observed at most OBSs, reaching a maximum of 2 s. Any location procedure not accounting for station terms results in biased Wadati diagrams and travel-time residuals (Figs. DR5 and DR6). We believe that the observed large S-wave delays are caused by unconsolidated sediment covering the seafloor; Schmid and Schlindwein (2016) reported deposits of >150 m within the median valley. A 200-400-m-thick layer of unconsolidated, low S-wave velocity, marine sediment of Vs ∼200 m/s would cause a delay of 1-2 s. Relocating the events in the catalog and accounting for both Pand S-wave station corrections result in an overall RMS misfit of 0.09 s, more than 4× smaller than the RMS of 0.4 s reported for the original location procedure . Hypocenters are now located at depths extending from ∼1 km to ∼17 km below seafloor (Figs. 2A and 2B). In the relocated catalog, neither an aseismic zone in the uppermost lithosphere nor very deep-seated seismicity is revealed.

DISCUSSION
Microseismicity along the MCSC is clearly related to the structure of the median valley, with the majority of earthquakes either wrapping around the Mount Dent core complex, in similar pattern to that observed at other oceanic core complexes (e.g., deMartin et al., 2007;Parnell-Turner et al., 2017), or clustering along the AVR. Some local strike-slip earthquakes appear to reflect oblique kinematics near the associated transform fault segment boundaries. At the MCSC, the region between the seafloor and ∼2 km depth is effectively aseismic, similar to results from many studies from the Mid-Atlantic Ridge (MAR) (e.g., Wolfe et al., 1995;Grevemeyer et al., 2013). The deepest micro-earthquakes occur at ∼8-10 km below seafloor. This depth is somewhat deeper than observed along most segments of the MAR, but shallower than found along the ultraslowspreading Gakkel Ridge (Arctic Ocean) (Korger and Schlindwein, 2014), suggesting that microearthquakes extend to greater depth as spreading rate decreases (Fig. 3). Such spreading rate-depth relationships for mid-ocean ridge earthquakes are predicted by models that relate the thickness of the brittle lithosphere to its temperature (Morgan and Chen, 1993). Schlindwein and Schmid's (2016) study of SWIR microseismicity concluded that an extensive aseismic region exists between the seafloor and 5-10 km depth below seafloor, and that such a shallow aseismic regime could result from serpentinization of exhumed mantle, which favors stable sliding on weak faults. However, this interpretation contradicts observations from other settings where serpentinization is pervasive, such as areas affected by subduction-related bend faulting (e.g., Grevemeyer et al., 2018a) or portions of the MAR (e.g., deMartin et al., 2007) where oceanic core complexes exhume deep-seated lithologies to the seafloor.
Microseismicity data from bend-fault settings suggests that serpentinized faults within the mantle are seismically very active and rupture frequently but under low-stress conditions (Lefeldt et al., 2009). At the Rainbow Massif on the MAR, with its high-temperature hydrothermal discharge through ultramafic rocks, micro-earthquakes occur immediately below the vent field and extend to 8-10 km below seafloor (Horning et al., 2018). Similar features were observed at the Logatchev Massif, also on the MAR (Grevemeyer et al., 2013), suggesting that serpentinized domains support elevated levels of seismicity instead of promoting an aseismic regime.
Our new analysis of the SWIR data set indicates that the previously proposed thick aseismic region in the upper lithosphere was simply a consequence of a model embedded in  the hypocenter location procedure that did not include a thickness of few hundred meters of unconsolidated sediment. Once this sediment is included, a focal depth range of 1-17 km results, and the overall pattern of relocated events is very similar to that observed along other ultraslow-spreading ridges (Fig. 3). However, we note that 17 km maximum depth of microearthquakes at the SWIR is still deeper than observed along the MCSC (Fig. 1). We suggest that the greater maximum depth of seismicity simply reflects the lower effective spreading rate. The spreading rate of 15 mm/yr at the MCSC is only slightly faster than 11.2 mm/yr at the SWIR, but, if its obliquity is considered, the SWIR effective spreading rate is much lower at ∼8 mm/yr (Dick et al., 2003). This may suggest that a decreased effective spreading rate with high associated obliquity mechanically affects increasingly deeper levels of the lithosphere to, or even below, the brittle-ductile transition. Furthermore, microseismicity may occur within a region defined by the seafloor and the maximum depth of faulting, supporting the hypothesis that temperature is the primary factor controlling the rheology of the oceanic lithosphere (Searle and Escartín, 2004) rather than lithology and hydration . However, an alternative interpretation is that differences in the maximum depth of seismicity are, instead, related to temporal variations in spreading process rather than spreading rate itself. In this case, phases of spreading dominated by magmatic activity, related to injection of melt into the shallow lithosphere, would lead to a high heat flux and, hence, higher lithospheric temperature. In contrast, magmatically starved phases would be dominated by tectonic extension with faulting extending to a greater depth, thus promoting mining of heat by circulating fluids and resulting in a cooler lithosphere.
At the MCSC, seafloor spreading accompanied by magmatic activity clearly occurred during accretion of the relatively thick (∼5 km) crust under the Mount Dent oceanic core complex  and distributed basaltic crust within the median valley Haughton et al., 2019). However, both within the axial rift as well as off axis, large domains are characterized by serpentinized mantle exposed at the seafloor (e.g., Hayman et al., 2011), suggesting that the style of spreading has repeatedly switched, over time and in position along axis, between magma-rich and magma-poor oceanic crustal accretion (Grevemeyer et al., 2018b). On this basis, it is reasonable to hypothesize that seismicity extending to greater depth may characterize phases of little magma and heat flux into the lithosphere and, hence, a deep-seated brittle-to-plastic transition. In contrast, higher magmatic and heat flux may increase lithospheric temperature, causing a shallower brittle-to-plastic transition zone and, in turn, shallower seismicity. Yet, even at its most robust, magmatic accretion at the MCSC has been below global average rates Van Avendonk et al., 2017). Consequently, the detachment faulting observed on the eastern flank of the Mount Dent oceanic core complex could result in anomalously deep seismicity, such as that observed at some oceanic core complexes along the MAR (deMartin et al., 2007;Parnell-Turner et al., 2017).
In conclusion, spreading rate, phases of magma-poor seafloor spreading, and deeply rooted detachment faulting may all affect and control the maximum depth of faulting along mid-ocean ridges, such that it is not necessary to invoke pervasive serpentinization and hydration to create a regime of shallow aseismic creep and deep seismicity along ultraslow-spreading centers.