Plate-Wide Deformation Before the Sumatra-Andaman Earthquake

SAEEAPb– 08/05.11

Plate-wide deformation before the Sumatra-Andaman Earthquake

Stuart Crampin1,2 and Yuan Gao3

1British Geological Survey, EdinburghEH9 3LA, UK. E-mail:

2also at School of GeoSciences, University of Edinburgh, EdinburghEH9 3JW, UK.

3Institute of Earthquake Science, CEA, 100036 Beijing, China. E-mail:

ABSTRACT

Rock is weak to shear-stress and the energy released by the 26th December, 2004,M≈9 Sumatra-Andaman Earthquake, the largest earthquake for four decades, must have accumulated over enormous volumes of crust and mantle, certainly plate-wide, possibly world-wide. Here we report evidence for plate-wide stress accumulation. Changes in seismic shear-wave splitting monitor stress-induced changes inthe geometry of the microcrack distributions in almost all rocks in the Earth’s crust. Such changes observed in Iceland show stress-accumulation beginning at least four years before the Sumatra-Andaman Earthquake. These changes were recognised as monitoring stress-accumulation before an impending large earthquake and 10 ‘stress-forecasts’ were emailed to Iceland Meteorological Office for some 27 months forecasting an impending large earthquake. The remarkable sensitivity of critical-systems of microcrack geometry to miniscule changes of stress had not been recognised at that time and the stress-accumulation was expected to lead to aM≥ 7 earthquake somewhere in Iceland. Only now is it recognised that the changes in shear-wave splitting were monitoring stress-accumulation which would eventually lead to the Sumatra-Andaman Earthquake at a distance of some 10,500km on the opposite side of the Eurasian Plate. This extreme sensitivity confirms the critical nature of fluid-saturated stress-aligned microcracks in the Earth’s crust.

Key words: Crack-coalescence, critical-systems, fluid-saturated stress-aligned microcracks, plate-wide deformation, shear-wave splitting, stress-accumulation, stress-relaxation.

1. INTRODUCTION

Stress-aligned shear-wave splitting (SWS) above swarms of small earthquakes is widely observed in the Earth’s crust (CRAMPIN and PEACOCK 2008). Caused by propagation through the distributions of fluid-saturated stress-aligned (typically-vertical)microcracks pervading almost all in situ rocks, changes in SWS time-delays monitor stress-induced changes to microcrack geometry. We have found that, whenever there is appropriate source-to-seismometer recording-geometry, stress-accumulation eventually leading to a large or larger earthquake has always been observed by analysing characteristic changes in SWS. Such characteristic changes are often observed at substantial distances from the impending source zone (CRAMPINPEACOCK 2005, 2008). Increases in tectonic stress are typically the result of subduction, magma generation, and other interactions at plate margins. Initially such increasing stress fields carry no information about where the stress will eventually be released, thus the observed stress-accumulations are a widespread volumetric effect andnot precursory to any particular earthquake. Additionally, whenever there are sufficient shear-wave source data to show the behaviour immediately before an impending earthquake, the stress-accumulation is observed to abruptly stop and start to decrease (GAO & CRAMPIN 2004). Thisdecrease is interpreted as stress relaxation as microcracks begin to coalesce onto the eventual fault-break. Consequently,stress-relaxation decreases are precursory to specific earthquakes, and measurements of stress-accumulation increases and crack-coalescence decreases allow the fault breaks of impending earthquakes to be stress-forecast (CRAMPIN et al.1999; CRAMPINPEACOCK 2005, 2008; CRAMPINet al.2008).

We refer to such estimates of earthquake occurrence as stress-forecasts rather than earthquake forecasts orearthquake predictions to emphasise the different formalisms. Note that stress-forecasting is an extraordinary simplifying assumption. Stress-accumulation occurs in aheterogeneous rock mass and varies irregularly. If stress accumulates over a small rock volume, the accumulation is rapid, but the resulting earthquake is small. If stress accumulates over a large rock volume, the accumulation is slower, but the resulting earthquake is larger. This link with earthquake magnitude is confirmed as logarithms of the durations of both increases and decreases are each proportional (self-similar) to the earthquake magnitude (see Figure 2, below) (CRAMPIN et al.2008; CRAMPIN & PEACOCK 2008). It is our experience that stress-accumulation increases have always been followed by large or large earthquakes, often at substantial distances from impending source zone.

Here we present evidence that stress-accumulation before the26th December, 2004 M≈9 Sumatra-Andaman Earthquake (SAE) was monitored bychanges in SWS in Iceland starting at least four years earlier. Iceland ison the opposite side of the Eurasian Plate from Sumatra at a distance of ~10.500kmwhere this extreme sensitivityis the result of the critical nature of microcracks in the Earth’s crust.

2. BRIEF SUMMARY OF SHEAR-WAVE SPLITTING ANALYSIS

Only shear-waves recorded within the effective shear-wave window above swarms of small earthquakes can be successfully recorded at the free-surface without severe waveform distortion due to S-to-P conversions (BOOTH & CRAMPIN 1985). [The shear-wave window is defined by incidence angles less than sin-1(VS/VP) at a horizontal free-surface, which is ~35º for a Poisson’s ratio of 0.25.] Stress-induced changes to microcrack geometry modify crack aspect-ratios, and SWS time-delays in Band-1 directionswithin the shear-wave window are sensitive to aspect-ratios (CRAMPIN 1999). Band-1 directions are the solid angle subtending 15º-to-45º to the vertical crack plane. Time-delays in Band-2 directions, ±(0º-to-15º) to the crack plane, are sensitive to crack density but crack density is not directly related to increases of stress (CRAMPIN 1999). The effects of the shear-wave window and Band directions are specified more fully in Appendix A.

Swarms of small earthquakes in the shear-wave windows of seismic recording networks can be used as ‘stress-monitoring-stations’ to measure changes in SWS caused bystress-inducedchanges to the geometry of the microcracks pervading the rock mass (CRAMPIN PEACOCK 2005, 2008). Using seismic swarms as stress-monitoring-stations, characteristic stress-accumulation changes in time-delays between split shear waves have been observed retrospectively before 14 earthquakes worldwide (CRAMPIN & PEACOCK 2008). On one occasion when the seismic network in Iceland was monitored routinely, the time, magnitude, and fault-break of aM= 5 earthquake in SW Iceland were successfully ‘stress-forecast’ in a comparatively narrow time/magnitude/location window (CRAMPINet al.1999, 2008). Stress-forecasts are usually posed in Smaller-Earlier to Larger-Later (SELL) windows (as in Table 1).

At various times it has been suggested that variations in SWS time-delays,on which stress-forecasting is based, could be caused by variations in: rainfall and glacier runoff; source earthquake migration; systematic variations in source earthquake magnitude or frequency; and by magmatic flux variations, and other local disturbances. Initially such possible variations were considered by VOLTI & CRAMPIN (2003a, 2003b). Many of the effects are eliminated or weakened by only using source events deeper than 5km-depth. However, the principle demonstration that such effects are negligible is that characteristic stress-accumulation increases and crack-coalescent decreases have now been observed with consistent parameters before 15 earthquakes world-wide (including the SAE in this paper) in a wide variety of different environmental, geological, and tectonic regimes. There are no exceptions. Had these local variations been significant, the characteristic stress-accumulation increases and crack-coalescent decreases would have been seriously disturbed. The migration issue is discussed inCRAMPIN &GAO (2005).

TABLE 1 HERE

3. MEASURING SHEAR-WAVE SPLITTING IN ICELAND

Iceland is on an offset of the Mid-Atlantic Ridge where several transform faults uniquely run onshore. Onshore transform faults provide the persistent swarm seismicity recorded by the comprehensive SIL seismic network (STEFÁNSSONet al.1993) necessary for measuring temporal variations in SWS in stress-monitoring-stations. Such persistent onshore seismicity is rare, and in this optimum environment, many properties of SWS were first identified in data from Iceland(VOLTI CRAMPIN 2003a, 2003b; CRAMPIN & PEACOCK 2005, 2008).

The data in this paper are records of shear waves from earthquakes within the shear-wave window beneath the specified seismic stations in Figure 1 which shows the measured SWS time-delays. The data are from all earthquakes below 5km-depth, to avoid near-surface and other anomalies, and ~10% are omitted when seismograms are too complicated to yield good SWS measurements. The complications are caused by: irregular surface and sub-surface topography distorting the shear-wave window (Appendix A); P-wave coda overrunning shear-wave arrivals so that waveforms are distorted; and where earthquakes arenot well separated in time so that again waveforms are distorted.

FIGURE 1 HERE

We use earthquake locations from the Iceland Meteorological Office (IMO) Seismic Bulletin. It is difficult to provide statistical errors for SWS measurements that are meaningful in geophysics. The vertical lines in the plots of normalised time-delays in Figure 1 are the observed time-delays (in ms) divided by the range of standard errors in hypocentral distance (in km) from the IMO locations (VOLTI & CRAMPIN 2003a, 2003b). The nearly negligible error bars in the majority of the time-delay plots in Figure 1 suggest that most source events are sufficiently well located to provide meaningful measures of time-delays.

Due to the complexity and temporal and spatial variations of SWS waveforms (CRAMPIN et al.2004),wholly automatic measuring techniques cannot successfully identify and measureSWS (CRAMPIN & GAO 2006). We measureorientations and time-delays of SWS on the Iceland seismograms with thesemi-automatic Shear-Wave Analysis System (SWAS) of HAO et al. (2008). SWAS automatically selects and plots polarisation diagrams (hodograms) of three-component SWS waveforms in screen images, which allow visual assessments and manual adjustments of picks for optimising time-delay arrivals. Specifically designed for Iceland data, SWAS gives rapid reliable measures of polarisations and time-delays for about 80% of all earthquakes below 5km-depth in the shear-wave windows of stress-monitoring-stations (HAO et al.2008).

During routine analysis of IMOdata from Iceland, we noticed what we thought was long-term stress-accumulation before an impending large earthquake at seven seismic stations above swarms of small earthquakes. Figure 1shows variations in measured time-delays for six and a half years from January 2000 at the seven seismic stations in Iceland where there were sufficient small earthquakes within the shear-wave window to monitor temporal variations in SWStime-delays. The lower half diagrams are arrivals in Band-1 directions (sensitive to aspect-ratio and hence to changes of stress, CRAMPIN 1999). Band-1 directions are specified in Appendix A. The upper half diagrams are arrivals in Band-2 directions specified in Appendix A that are sensitive to crack density, but crack density does not appear to have a direct relationship with changes of stress, and Band-2 variations are not used (CRAMPIN 1999). The least-squares lines in Band-1 show increases and decreases before the 2004 Sumatra-Andaman Earthquake (marked by a bar in the time axis). Diagrams to the left (BJA, KRI, SAU) of Figure 1 are stations from southern Iceland above seismicity associated with the SW Iceland Transform Fault. Diagrams to the right (BRE, FLA, GRI, HED) are stations over 300km to the north above seismicity associate with the Húsavík-Flatey Transform Fault.

The relative north-south locations of the seismic stationswere used in the initial interpretation of emails in Table 1,when we thought the stress-accumulation referred to an impending large earthquake in Iceland, and are recorded for completeness in Table 1. This interpretation was inappropriatewhen it was realised in retrospect that the stress-accumulation referred to the Sumatra-Andaman Earthquake. The network of seismic stations in Iceland subtends less than 2º at Sumatraand is essentially a single direction. Consequently, the variations in Figure 1 are believed to be caused by local variations in the shear-wave source seismicity in Iceland, local geology and tectonics, and interactions with irregular surface topography.

The diagrams in Figure 1 show characteristic increases and decreases in time-delays in a time window spanning the Sumatra-Andaman Earthquake that are typical of temporal variations in SWS time-delays observed before impending earthquakes elsewhere (CRAMPIN et al.1999, 2008; VOLTI & CRAMPIN 2003a, 2003b; CRAMPIN & PEACOCK 2008), but with much longer time scales. The large scatterabout the mean (sometimes referred to as a‘±80%’ scatter) is a real geophysical effect observed in all measurements of SWStime-delaysthat is understood and can be modelled but the scatter cannot be eliminated from the measurements(discussed in Appendix B) (CRAMPIN et al.2002, 2004).

4. INTERPRETING STRESS-FORECASTS EMAILED TO ICELAND

The long term increases in time-delays in Figure 1 were first recognised in September, 2002, at Stations BJA, SAU, and BRE, and the first of 10 stress-forecasts was emailed to the IMO warning of an impending large earthquake. Table 1 summarises these emails.

TABLE 1 HERE

Note that the least-squares fits to increases and decreases in time-delays in Figure 1 were cited retrospectively to the impending earthquake. Increased understanding of the phenomena was ongoing during this period, and some of the earlier stress-forecasts in Table 1 are not wholly consistent with the up-dated least-squares lines in Figure 1.

The initial stress-forecast, SF1 (13/09.2002), Table 1, was based principally on stress-accumulation increases at Stations BJA, SAU, and BRE. Table 1 distinguishes reports from stations in SW Iceland by bold-face font from those in northern Iceland in regular-face font. These geographical locations guided the interpretation of stress-forecasts summarised in Table 1. The cause of the large scatter in time-delays was not initially recognised(see discussion in Appendix B),nor were the implications of increasing time-delays abruptly decreasing in stress-relaxation as microcracks coalesce onto the fault-plane (GAO & CRAMPIN 2004). In the analyses, crack-coalescent decreases were first identified in SF4 (07/03.2003) at northern Stations BRE and FLA, which suggested that the impending earthquake would be in northern Iceland, SF5 (04/04.2003). Later, in SF9 (29/09.2004), crack-coalescence decreases were identified on all seven available stations, and SF10 (22/12.2004) suggested that the impending M = 7+ earthquake could be anywhere in Iceland. The situation did not change in the final stress-forecast SF11 (18/02.2005), nor subsequently to the time of writing in May, 2011. The anticipated M ≥ 7 earthquake in Iceland has not occurred, and by now would be far too late to be associated with the stress-forecasts in Table 1. Consequently, the variations in Figure 1 are now associated with the only candidate earthquake, the Sumatra-Andaman Earthquake of December, 2004 The stress-accumulation increases and crack-coalescence decreases in Figure 1 are characteristic of the changes in SWS observed before 15 earthquakes worldwide(CRAMPIN PEACOCK 2008), but with significantly larger time durations appropriate to the larger magnitude earthquake.

5. PLOTS OF DURATIONS OF STRESS INCREASES AND STRESS DECREASES

The association between changes in shear-wave-splitting time-delays and earthquakes is confirmed by the linearity (self-similarity) of logarithms of the durations of both increases and decreases with earthquake magnitudes, analogous to the linearity of the Gutenberg-Richter relationship (GUTENBERG & RICHTER 1954). Figure 2 shows plots against magnitude of logarithms of the durations of stress-accumulation increases (Figure 2a) and crack-coalescence decreases (Figure 2b) of all available data, where the solid data points are identified in CRAMPIN & PEACOCK (2008). The data for the open circles for the Sumatra-Andaman Earthquake in Figure 2 are taken from Figure 1. The initial points of these Sumatra-Andaman increases in Figure 1 are hidden in noisy data, and are likely to have begun possibly tens of years before the start of the time interval shown in Figure 1. Thus the arrow head to the right of the open circles in Figure 2a indicates that the data points should be located at (unspecified) greater durations. Extrapolation of the least-squares line reaches M ≈ 9 at ~10000days (~27 years) which may be associated with the world-wide return period of M ≈ 9 earthquakes.

FIGURE 2 HERE

The least-square line through the solid data points in Figure 2a is thought to represent the rate of increasing tectonic stress generated at plate boundaries. The solid-circle outliers at M= 6 (1986, North Palm Springs Earthquake, California) and M= 7.7 (1999, Chi-Chi Earthquake, Taiwan), respectively, refer to earthquakes at different plate boundary faults, where the rates of stress-accumulation are likely to be different from those elsewhere (CRAMPIN & PEACOCK 2008).

In Figure 2b, the open circles at M≈ 9 for the Sumatra-Andaman Earthquake from Figure 1 are believed to be correctly located. The stress-relaxation least-squares line in Figure 2b is thought to be related to the rate at which networks of cracks coalesce onto a principal fault break. This suggests that the linear relationship for the smaller magnitude data ends at about M= 6 and curves upwards for larger magnitude events. Such effects might be expected. The linear data below M= ~6 are expected to refer to data from coalescing crack networks in the Earth’s crust, whereas data for larger earthquakes include crack-coalescence in crack networks which are at least partially located in the upper-mantle. Thus the upward trending curve in Figure 2b may indicate that crack distributions in the mantle coalesce more rapidly than those in crust. Since the fluid in the cracks in the mantle is likely to be films of high-temperature liquid melt (CRAMPIN 2003)rather than the lower-temperature water or water-based salt solutions common in the crust, faster crack coalescence in the mantle may be expected.

Estimates based on duration/magnitude plots, both for stress-accumulation and crack-coalescence as in Figure 2, were also included in the emailed stress-forecasts from SF9 (29/09.2004) onwards listed in Table 1. Current studies of the Sumatra-Andaman Earthquake reported elsewhere generally focus on co-seismic and post-seismic behaviour are discussed in the Conclusions. To our knowledge, this is the first report of plate-wide pre-earthquake stress-induced deformation.

6. TECTONIC IMPLICATIONS

FIGURE 3 HERE

The map in Figure 3 shows approximate plate boundaries and the great-circle path from SAE to Iceland. Plate-wide observations of stress-accumulation at seismic stations in Iceland at ~10,500km from SAE confirm the extreme sensitivity expected of pervasive distributions of microcracks in crustal rocks (CRAMPIN & PEACOCK 2005; CRAMPIN 2006). Note that stress is a volumetric effect not confined to straight-line transmission. The huge stress-accumulation before the SAE was clearly plate-wide as indicated by this paper, and potentially worldwide, as is suggested by the approximate coincidence of start of stress-accumulation (~27 years) with the ~40 yearssince the last M ~9 earthquake. This was the 1964, M ~9.2, Alaskan Earthquake - although the relationship is complicated by the 2011, M ~9, Tohoku Earthquake in offshore Japan.