Non-volcanic tremor associated with the March 2010 Gisborne slow slip event at the Hikurangi subduction margin, New Zealand



[1] Non-volcanic tremor accompanies slow slip events at most circum-Pacific subduction zones. Previously, the Hikurangi subduction margin, New Zealand, and the Boso Peninsula of Japan have been noted as exceptions where slow slip, which occurs at shallow depth compared to other subduction zones, does not generate seismic tremor. The lack of tremor observations raises questions whether tremor and slow slip are manifestations of the same process and whether different temperature and pressure conditions influence the physical mechanism of tremor production. Analysis of seismic data during the March 2010 Gisborne slow slip event, New Zealand, reveals that this slow slip event at the northern Hikurangi margin is accompanied by tremor with distinctive characteristics compared to local microearthquakes. Both the Gisborne area of New Zealand and the Boso Peninsula of Japan are underlain by thick accumulations of sedimentary rocks that likely inhibit the observation of low amplitude tremor due to high attenuation. Discovery of tremor in New Zealand suggests that tremor may always accompany slow slip and that its apparent absence at the Boso Peninsula, the Hikurangi margin prior to this study, and any future regions discovered to experience slow slip without concurrent tremor, may be the effect of highly-attenuating sediments.

1. Introduction

[2] Like many circum-Pacific subduction zones, slow slip events have been repeatedly observed at the Hikurangi margin, east of North Island, New Zealand (see Wallace and Beavan [2010] for review). Most subduction related slow slip events (SSEs) around the world are accompanied by non-volcanic tremor (NVT), as observed in southwest Japan [e.g., Obara and Hirose, 2006], Cascadia [e.g., Dragert et al., 2004], Alaska [Peterson and Christensen, 2009], Mexico [Payero et al., 2008], and Costa Rica [Walter et al., 2011]. While a strong correlation between slow slip and tremor has allowed the moment of small SSEs in Cascadia to be estimated based on tremor productivity [Aguiar et al., 2009], no globally applicable relationship between seismic tremor and slow slip has been established. In the northern Hikurangi subduction zone, similar to the Boso peninsula, Japan, slow slip has been shown to trigger seismicity without accompanying tremor [Ozawa et al., 2007; Delahaye et al., 2009]. Slow slip at these two subduction zones occurs at much shallower depth compared to most regions where tremor accompanies slow slip. This has lead to conjecture that temperature/pressure conditions, and thus rheological behavior, dictate the seismological response to slow slip with triggered microseismicity favored at lower P/T conditions [Delahaye et al., 2009]. We report the first observation of NVT at the Hikurangi margin associated with the March 2010 Gisborne, NZ shallow slow slip event, demonstrating that slow slip regardless of its depth can be accompanied by tremor.

[3] At the eastern North Island of New Zealand, the Pacific plate subducts westward beneath the Australian plate at the Hikurangi margin at a rate of between 2–6 cm/yr [Wallace et al., 2004]. In Gisborne and the surrounding area (Figure 1) SSEs were first documented in 2002 [Douglas et al., 2005] and have repeated with a two year interval with approximate Mw ∼ 6.6 and slip durations of two weeks [Wallace and Beavan, 2010]. SSEs in this region occur at shallow depth (<10–15 km) on the plate interface at the downdip edge of the locked zone [Wallace and Beavan, 2010] where fluid pressure related to subducting seamounts is believed to be high [Bell et al., 2010].

Figure 1.

Map of the study area with GeoNet broadband stations shown with red triangles and short-period stations with magenta and yellow inverted triangles. Data during our study period was only partially available from stations with yellow markers. Station MWZ, used for the spectral analysis, is labeled. The slip distribution of the March 2010 slow slip event is contoured [from Wallance and Beavan, 2010]. Light and dark blue circles are approximate tremor locations prior to and during the slow slip event, respectively. The large green diamond is the epicenter of the earthquake shown in Figure S2. Earthquakes within the shaded region were used to produce Figure 3. The dashed line indicates the approximate location of the transition between high Vp/Vs (trench-ward) and lower Vp/Vs (arc-ward) imaged by seismic tomography [Reyners et al., 1999]. Approximate convergence rate and direction are also indicated in the figure [Wallace et al., 2004].

[4] Since the study documenting an increase in microseismicity associated with the 2004 Gisborne slow slip event using only a sparse broadband network [Delahaye et al., 2009], the New Zealand seismic network has been densified with additional short period stations [Petersen et al., 2011]. We believe that tremor likely accompanied the 2004 Gisborne SSE, but was not identified by Delahaye et al. [2009] due to the low station density at that time. Recently, NVT was reported in the central North Island, triggered by the Chilean earthquake (Mw = 8.8) of February 2010 [Fry et al., 2010] demonstrating the capacity of this region to generate tremor. In this study, we show that NVT occurred during the 2010 Gisborne slow slip event. Tremor detection required inclusion of data from the new short period stations; tremor was undetectable if only the data from the broadband stations were used.

2. Analysis and Results

[5] In order to identify and locate NVT associated with the 2010 Gisborne SSE, we applied a modified version of the automated envelope cross-correlation algorithm of Wech and Creager [2008]. We analyzed the east component of motion from broadband and short-period seismic stations in the greater Raukumara Peninsula region (shown in Figure 1) during March and April of 2010, available through New Zealand GeoNet. We band-pass filtered data in two frequency ranges between 2 to 5 Hz and 8 to 20 Hz, created envelope functions, low-pass filtered at 0.1 Hz, and decimated to 1 Hz. We used a 2-minute time window, shifting with 50% overlap, to cross-correlate envelope functions and located tremor when cross-correlation coefficients exceeded a value of 0.6 on more than 10 station pairs. We considered any signal with significant energy above 8 Hz to originate from local earthquakes. As such, we did not locate tremor if correlation coefficients computed using the higher frequency envelopes (8–20 Hz) exceeded 0.6 on more than five station pairs, We also discarded time windows where envelopes from more than 5 stations correlated well with a reference station, located far outside of the slow slip region, to prevent signals from regional and teleseismic earthquakes from being misidentified as tremor. Using a 1-D local velocity model for the Gisborne area [Reyners et al., 1999], we located tremor by inverting for the position in a 3-D grid that produced the best cross-correlation among stations, following Wech and Creager [2008]. We only kept locations with epicentral bootstrapping error estimates less than 0.1 degrees; the tremor location method provides poor depth constraints. In order to further eliminate isolated earthquakes or randomly scattered erroneous detections, we applied a clustering requirement that each event has at least one additional event in 24 hours and 5 additional events in 72 hours within .1 degrees.

[6] A low level of tremor activity that increases during the 2010 Gisborne slow slip event (Julian Days 78–93) peaks at the beginning and at the end of the SSE (Figure 2). Tremor locations are at the downdip edge of the northeastern section of the slow slip patch determined by Wallace and Beavan [2010] (Figure 1). From our analysis, the total duration of tremor during the 16 day SSE was estimated to be less than 2 hours; however, we suspect the actual duration of tremor is longer since several of the time windows eliminated due to earthquake occurrence likely also contain tremor. Also, given the sparse network coverage (average ∼30 km station spacing in 2010 [Petersen et al., 2011]), low amplitude tremor events are likely recorded at too few stations to be detected. In addition to the total duration of tremor during the SSE being low compared to slow slip and tremor episodes elsewhere (e.g., total ∼200 hours of tremor for an SSE with Mw = 6.6, at the rate of ∼90 hours/week in Cascadia [Aguiar et al., 2009; Szeliga et al., 2004]) individual tremor signals from the 2010 Gisborne slow slip event are short in duration, with each individual burst of tremor lasting between 20–60 seconds (Figure S1 of the auxiliary material). Comparison of tremor characteristics with closely located microearthquakes show that while both earthquake and tremor signal have equivalent amplitude between 2–3 Hz, tremor is reduced to the noise level at the higher 7–8 Hz band (Figure S2 of the auxiliary material). While P and S wave arrivals are always visible for earthquake signals, some tremor episodes also display distinct phases at specific stations with S wave moveout (Figures S2 and S3 of the auxiliary material), similar to low frequency earthquakes observed in other subduction zones [Brown et al., 2009]. This suggests that tremor in the Gisborne region, like that in many other subduction zones, consists of swarms of low frequency earthquakes (LFE). Distinct S wave arrival times could be picked for a few LFEs, improving their hypocentral determinations and revealing locations consistent with occurrence on or near the plate interface (Figure S3 of the auxiliary material).

Figure 2.

Histogram showing the number of detected tremor windows in each Julian day of 2010. Light blue corresponds to dates prior to the geodetically determined SSE, and blue corresponds to the dates during the SSE. Tremor locations using the same color coding are shown in Figure 1. The number of detections increases greatly during the SSE. No tremor was detected after the SSE.

[7] The stations that we used in this study are located on thick Neogene marine sedimentary rocks found to have high attenuation [Eberhart-Phillips and Chadwick, 2002]. In order to confirm that our tremor detections are not local microearthquakes having lost their higher frequency content to attenuation, we compared stacked power spectral density for 51 15-second tremor windows (from 9 tremor bursts) to nearby microearthquakes and background noise at station MWZ. Figure 1 shows the general location of the 34 microearthquakes (1.7 < ML < 2.7) from the GeoNet catalog and the tremor episodes (blue dots) used in the spectral analysis (shaded box). Both earthquake and tremor sources have depths that range between 20–40 km. Tremor signals are strongest between 2–5 Hz and fall to noise levels at higher frequencies (Figure 3). The power spectral densities of nearby microearthquakes, stacked in two different magnitude ranges, retain high amplitude between 2–8 Hz, and show a higher corner frequency and lower amplitude for the smaller events, as expected. If the tremor signals were very small earthquakes with higher frequencies lost through attenuation, the stacked power spectral density of nearly co-located small earthquakes would exhibit a similar pattern compared to that of tremor. However, local microearthquakes have significant energy above 10 Hz, whereas tremor signals have comparable amplitude to the microearthquakes below 2 Hz but fall off much more rapidly above 3 Hz, plunging down to the noise level by 8 Hz.

Figure 3.

Stacked power spectral density for earthquakes and tremor events shown in Figure 1 recorded at station MWZ. Instrument response was removed and power spectral density of velocity waveforms was computed using the multi-taper method. 17 earthquakes were stacked in each magnitude range. Local microearthquakes show significant amplitudes at higher frequencies. In contrast, tremor signals start to fall off around 2 Hz, reaching the noise level at ∼7 Hz.

3. Discussion

[8] The downdip edge of the tremor locations coincides with a transition in Vp/Vs from high toward the trench to low arcward [Reyners et al., 1999], suggesting that tremor activity in this area may be related to high fluid pressure. A seismic reflection study offshore of Gisborne [Barker et al., 2009] suggested that the Gisborne slow-slip patch is related to the subduction of seamounts and underthrust sediments that elevated fluid pressure [Bell et al., 2010]. Some SSEs and tremor phenomena in other subduction zones have been shown to be spatially correlated with areas with high fluid pressure [e.g., Shelly et al., 2007; Audet et al., 2010] and the 2010 Gisborne slow slip and tremor event seems to fit in this category.

[9] Although we located very little tremor offshore, where the amount of slip in the 2010 SSE was the greatest (Figure 1), we cannot rule out its more extensive existence. The application of our tremor location procedure without the earthquake elimination steps identifies numerous earthquakes in the offshore region, suggesting that a location bias for onshore tremor events does not exist. However, thick Neogene marine sediments with very low seismic quality factor, Q, overlie the plate interface above the slow-slip patch [Eberhart-Phillips and Chadwick, 2002] and may reduce offshore tremor amplitude to undetectable levels (see auxiliary material).

[10] Besides the Hikurangi margin, shallow SSEs (<20 km) have been observed in Boso Peninsula, Japan [Ozawa et al., 2007] and Nicoya Peninsula, Costa Rica [Outerbridge et al., 2010]. No tremor activity has been documented so far at the Boso Peninsula; however, similar to the Hikurangi margin, this region lies on top of a low velocity accretionary prism [Nishida et al., 2008] that likely attenuates tremor signals. Strong tremor generated at shallow depth (<20 km off-shore) has accompanied slow slip beneath the Nicoya Peninsula, Costa Rica [Walter et al., 2011]. The Pacific coast of Costa Rica is an erosive convergent margin [Ranero and von Huene, 2000], lacking the accumulation of a thick sedimentary sequence above the shallow plate interface. We hypothesize that tremor always accompanies slow slip and that the absence of observable tremor at the Boso Peninsula may be the effect of highly-attenuating sediments above slow slip events along the shallow plate interface, rather than particular P/T conditions.

[11] Similar to the 2004 Gisborne SSE [Delahaye et al., 2009], analysis of GeoNet catalogued earthquakes shows an increase in seismicity near the Mahia peninsula (Figure 1) at the onset of the 2010 Gisborne SSE (see Figures S4 and S5 of the auxiliary material). Delahaye et al. [2009] attributed the increase in seismicity near the Mahia Peninsula to stress triggering of the 2004 slow slip patch; this appears to repeat following the 2010 slow slip event. In Boso, Japan, seismic swarms occurred in the vicinity of the 2007 slow slip area, indicating that the stress change due to slow slip caused the triggered seismicity [Ozawa et al., 2007]. With our discovery of tremor associated with slow slip at the Hikurangi margin, we infer that earthquakes are triggered in the adjacent seismogenic zone when slow slip induces positive change in the stress field, regardless of generation of seismic tremor in the slow-slipping areas.

4. Conclusion

[12] Our study reveals the presence of non-volcanic tremor associated with the March 2010 Gisborne slow slip event at the Hikurangi subduction margin, in a region previously reported to lack tremor. Despite the greater amount of geodetic slip offshore, our tremor locations are limited to the downdip area of the slow-slip patch near shore and on land, close to station locations. Both the Boso Peninsula of Japan and the Gisborne area of New Zealand, where no tremor signals associated with SSEs were reported prior to this paper, lie on thick slow velocity, high attenuating sedimentary deposits as imaged in tomographic studies [Nishida et al., 2008; Eberhart-Phillips and Chadwick, 2002]. With our discovery of tremor in New Zealand, we hypothesize that tremor always accompanies slow slip and that the absence of observable tremor at the Boso Peninsula and the Hikurangi margin, prior to this paper, is the effect of highly-attenuating sediments.


[13] We acknowledge the New Zealand GeoNet project and its sponsors EQC, GNS Science and LINZ, for providing data/images used in this study. We thank Jake Walter for first alerting us to the likely existence of tremor at the Hikurangi margin, Aaron Wech for providing his tremor location algorithms and Laura Wallace for the slip distribution data. This research was supported by NSF award OCE-0841061 to SYS.