Geophysical Research Letters

Transition from locked to creeping subduction in the Shumagin region, Alaska



[1] GPS velocities from the Alaska Peninsula are modeled to determine the extent of locking on the Alaska-Aleutian subduction interface. The observations, which span from the Semidi Islands to Sanak Island, encompass the 1938, Mw 8.3, rupture zone and the transition into the Shumagin gap. Model parameters are optimized using a simulated annealing method. Coupling variation along strike of the plate interface show a nearly fully locked (90%) subduction zone at the Semidi Islands, decreasing to about 30% locked at the Shumagin Islands, and freely slipping to the west of the Shumagins. Independent rupture of the Shumagin segment could produce repeated Mw 7.6 earthquakes, unless a significant fraction of the slip on the interface occurs as afterslip following large earthquakes. Southwest directed velocities at most of the sites may be attributed to clockwise rotation of a Bering block.

1. Introduction

[2] Subduction zone earthquakes are a major hazard for coastal populations both near and far from the epicenter, due to severe ground shaking and the potential to generate tsunamis. Pacheco et al. [1993] showed that seismic release at subduction zones varies worldwide. Identifying regions with low and high potential for sources of large subduction earthquakes is important for advancing our understanding of subduction zone seismicity, and local and regional hazards. Geodetic observations can be used to study the process of interseismic loading, and estimate the rate of slip deficit or moment deficit on the plate interface.

[3] The Shumagin segment of the Alaska-Aleutian subduction zone (Figure 1) has not experienced a large earthquake that has ruptured the majority of the plate interface since at least 1917, when an Ms 7.4 [Estabrook and Boyd, 1992] earthquake struck the region. That quake may have ruptured from the eastern edge of the Shumagins to somewhere west of the Shumagin Islands, but not as far as Sanak Island [Boyd et al., 1988]. Other moderate sized earthquakes have been attributed to rupturing a portion of the plate interface in the Shumagin segment, including an Ms 7.5 in 1948 and an Ms 7.1 in 1993 [Bufe et al., 1994]. No great earthquakes are known to have occurred in the Shumagin segment, and no large earthquakes are known from the western portion of the Shumagin segment.

Figure 1.

The Bering Block proposed by Mackey et al. [1997] is the bold line in the inset map, with a rectangle outlining the region shown in detail. The black vectors show the observed velocity at each site, with 95% confidence ellipses. The convergence rate of the Pacific plate and the North American plate, from REVEL [Sella et al., 2002], is shown at the approximate coordinate for which it was determined. The arc velocity that is applied as a correction to the modeled data is shown at the upper left. Left-lateral faults on Kodiak Island and the trench are shown as black lines.

[4] Fletcher et al. [2001] used a single plane model to examine the plate coupling (slip deficit) along the Semidi segment (Figure 1). They determined that the interface in this region is presently ∼80% coupled, that is, that the area of their 170-km-wide model plane on average slips at about 20% of the plate motion rate and thus has a slip deficit accumulating at a rate of 80% of the plate motion rate. Farther to the southwest along the subduction zone, near Sanak Island, the plates are freely slipping [Freymueller and Beavan, 1999]. The transition from the very wide, highly coupled zone in the north to the weakly coupled zone to the south occurs over a fairly small region surrounding the Shumagin Islands.

[5] Here we construct a model that explains the GPS observations along the Alaska Peninsula and offshore islands, and examine this zone of transition from a very wide locked zone on the plate interface, which has historically generated great earthquakes, to a freely-slipping segment that has no record of great earthquakes. We then interpret patterns of thrust-zone seismicity in the area in relation to the inferred locked and creeping segments along the subduction zone.

2. Data

[6] We use three component GPS velocities from 27 sites along the Alaska Peninsula, surveyed between 1991 and 2005 (Figure 1, Table S1 of the auxiliary material). Most of the data were collected during 1993–1996, and subsets of the data set were used earlier by Freymueller and Beavan [1999] and Fletcher et al. [2001]. However, new observations of several sites were made between 2003 and 2005, so we have new velocities for several sites and much greater precision for others. All phase and pseudorange data have been analyzed along with data from surrounding continuous GPS stations using the GIPSY-OASIS GOA4 software, and each daily solution is transformed into the ITRF2000 reference frame (IGSb00 realization). We used the JPL non-fiducial orbits for solutions 1995 and later, but estimated orbits from a global data set for earlier data. The ITRF velocities are then rotated into a North America-fixed reference frame using the REVEL model [Sella et al., 2002]. A sample time series is shown in Figure S1. A few sites have a long history of observations; these velocities have uncertainties of order 0.5 mm/yr, and as shown later, are fit by a model to a similar level of precision. All the surveys were conducted at roughly the same time in the summer; this mitigates the effects of seasonal variations.

[7] There are three obvious features in the data: (1) the velocities are largest closest to the trench and decrease with distance form the trench, the result of convergence of the Pacific and North America plates; (2) sites that are farther east generally have larger velocities, given the same distance from the trench; and (3) sites in the west move in a nearly trench-parallel direction and show little or no variation in velocity with distance from the trench. This results in a clear counter-clockwise rotation in the velocity field with the sites farthest from the trench moving in a more trench-parallel direction than sites closer to the trench. These features can be explained by a combination of a westward translation of all sites combined with along-strike variations in the extent of the locked subduction zone.

[8] We refer to the westward component to all of the data as the arc translation or arc velocity, because it is directed parallel to the arc, as previously noted by Mann and Freymueller [2003]. The arc velocity is 5.3 mm/yr directed toward 241°, based on the average velocity of the sites in the western part of the network. This motion may have to do with clockwise rotation of a Bering micro-plate [Mackey et al., 1997]. In this study we approximate block rotation with translation because the size of the network is small compared to the distance to the proposed Bering plate rotation pole. After removal of the arc velocity, all the velocities (except CHIR) are nearly parallel to the Pacific-North America relative plate motion direction, indicating that the counter-clockwise rotation results from the superposition of translation of the Alaska Peninsula relative to North America with strain resulting from the locked subduction zone. Site CHIR on Chirikof Island near the trench (Figure 2) does not show this same arc translation. We suggest that a network of active trench parallel, left lateral strike-slip faults on Kodiak Island [Sauber et al., 2006] extends to the southwest past Chirikof Island, and passes between Chirikof and the rest of the sites. This fault system may represent the southern limit of the Bering block, or of westward extrusion of southwest Alaska.

Figure 2.

Data (red) and model (black) velocity vectors are shown. All of the data have been corrected for arc translation, except CHIR. The blue vector shows the velocity at CHIR if an arc translation correction is applied at that site. The surface projections of the locked interfaces are shown as black rectangles with the amount of coupling shown on each plane as the percentages. The planes are numbered according to Table 1. For planes 1, 2, and 3 the locked area of the fault can be decreased to the area of the stipple pattern with the amount of coupling increased to 100%, 80% and 40% respectively. The gray region on interface 4 is the widest possible fully locked interface.

3. Methods

[9] We use a simple four plane model to represent the locked portion of the interface. This simple model geometry creates discontinuities in the plate interface, but these are located at sufficient depth, and the model is not sensitive enough to the dip angle, to represent more than a second order effect. Smoother model geometry would be needed, for example, to investigate stress changes near the interface. Although we do not incorporate a transition zone from locked to creeping downdip of the locked zone, we have found that for any model with a transition zone a model with an abrupt transition located at the midpoint of the transition zone produced almost identical surface displacements. The downdip end of the locked zone in our model should be interpreted as the midpoint of the transition from locked to fully creeping, and the downdip transition may be narrow or as wide as ∼50 km or perhaps more.

[10] We estimated the optimal model using the simulated annealing technique described by Cervelli et al. [2001]. The model space is loosely constrained around geometries based on previous geodetic and seismic studies [Von Huene et al., 1987; Abers, 1992; Freymueller and Beavan, 1999; Savage et al., 1999; Fletcher et al., 2001; Sauber et al., 2006]. For each plane we determine the length, width, dip and the slip deficit; the planes are constrained not to overlap or have gaps between them. Each fault plane is represented by a dislocation in a homogeneous elastic half-space [Okada, 1992]. Deformation is modeled by the superposition of steady slip on the interface at the plate convergence rate with slip deficit represented by backslip on the same interface [Savage, 1983]. The slip deficit is allowed to vary from 0 to 100 percent of the plate convergence rate, determined from the REVEL plate velocity model [Sella et al., 2002].

4. Results

[11] Table 1 shows the parameter values of the optimal model. This model has locking depths ranging from ∼20–30 km, which is consistent with previous studies of this area [Fletcher et al., 2001; Savage and Lisowski, 1986] and subduction zones in general [Oleskevich et al., 1999]. The model is not very sensitive to variations in dip, but seismic observations constrain the dip angles to a relatively narrow range. The 10°–13° dip along the western segment agrees with seismicity in the Shumagin area which indicates a dip of 10–15° [Savage and Lisowski, 1986; Abers, 1992; Zheng et al., 1996]. In the Semidi Island region, the results determined here for dip, locking depth and amount of coupling agree well with the model reported by Fletcher et al. [2001]. This is expected, because in this region we have very little new data beyond what they used. At the Shumagin Islands the slip deficit drops to 30% of the plate rate, and then to nearly zero west of the Shumagins.

Table 1. Model Parametersa
PlaneLongitudeLatitudeLength, kmWidth, kmDip, degStrike, °EDepth, kmCoupling, %
  • a

    The longitude and latitude describe the eastern-most corner of each plane. Depth is the vertical distance to the top and bottom of each plane. The planes are numbered 1–4 from east to west. Parameter uncertainties and tradeoffs are discussed in the text.


[12] We estimated uncertainties through a series of parameter searches centered on the optimal model. We varied parameters independently or together to study tradeoffs, but in all cases only one plane is modified at a time, while the other planes are fixed to the best model values (Table 1). We estimated the range of acceptable parameters from the increase in misfit over the optimal model (95% confidence limits). Except for the fault lengths, which cause all faults in the model to shift, varying the parameters for one fault has only a minor effect on the parameters of the others.

[13] The dip is not very well constrained by the observations (Figure S2), but variations in dip have only a modest affect other parameters. For example if the dip on plane #1 is changed from 5° to 12°, there is a ∼20% decrease in the amount of locking on the interface. A comparable change in dip on planes #2 and #3 results in only ∼10% change in coupling. Plane #4 has essentially no coupling, so all other parameters for this plane can be varied with no change to misfit, but an increase in coupling significantly increases misfit.

[14] Width and slip deficit show a negative correlation (Figure S3). Plane #1 can be varied by up to ±40 km (95% confidence range) in width with a slip deficit of almost 100% locked for the narrower plane to less than 80% locked for the wider plane. The tradeoff is more dramatic for plane #2 where the locking amount decreases from 90% to 50% over the range of widths, 130 km to 210 km. Plane #3 has a larger width uncertainty, ±70 km, than the other planes due to sparse data in this region. Over this range of widths the slip deficit only varies from 30% to 20%. Considering all uncertainties, we estimate the uncertainty in the plate coupling fraction to be ∼20%.

[15] Resolution of the state of the plate interface close to the trench is very poor, due to a lack of observations there. The sensitivity of the updip limit of the locked zone was assessed using a grid search in which the depth to the updip edge of the locked zone was increased and the slip deficit was re-estimated. Figure 2 shows the narrowest possible downdip locked region, at the 95% confidence level, with a stippled pattern. If planes 1–3 are reduced to their narrowest, the coupling on each plane increases to 100%, 80%, and 40% respectively. This shows that the data are not sensitive to the shallow portion of the plate interface. The top of the locked zone in this case is at 12 km, 19 km, and 21 km depth from east to west respectively.

[16] The western most plane has nearly zero slip deficit, so instead of determining the sensitivity to a shallow creeping zone we determine the data sensitivity to a shallow locked region. The gray-shaded box in Figure 2 shows the largest fully locked interface that can be allowed within the 95% confidence limits of the data. The downdip edge of this zone is at 11 km. Oleskevich et al. [1999] used thermal modeling to show that in the Cook Inlet region of Alaska the plate interface probably is not locked shallower than 12 km, and if the same is valid for the Sanak region there is probably no locked zone at all.

5. Discussion and Conclusions

[17] The model shows a transition from the wide locked zone of the 1938 rupture to the creeping zone south of Sanak (Figure 3). The edge between the wide locked zone and the weakly coupled Shumagin segment in the model agrees with the edge of the 1938 rupture zone estimated from the aftershock distribution. Between the 1938 rupture zone and the Sanak segment, the plate interface becomes more dominated by creep. Velocities of sites along the Alaska Peninsula between longitudes 201°E and 202°E show no variation, so the transition from dominantly locked to dominantly creeping (plane 2 to plane 3) must be abrupt, occurring over an along-strike distance that is very small compared to the width of the locked zone. Over a distance of ∼50 km SW of the Shumagin islands, the remaining locked regions on the interface disappear.

Figure 3.

The partially locked interfaces from the model are shown along with 100 years of seismicity. The dashed lines show the minimum locked region as in Figure 2. Events with M ≥ 5.0 and depth < 50 km from the AEIC seismic catalog are plotted as circles. Black stars show significant earthquakes in the region; 1917 Ms 7.4 [Estabrook and Boyd, 1992], 1938 Mw 8.3, 1948 Ms 7.5, and 1993 Ms 6.9 [Abers et al., 1995]. The extent of the 1938 rupture is shown as a black outline.

[18] Past estimates of the extent of the locked region on the plate interface in the Shumagin Islands by Savage and Lisowski [1986] and Larson and Lisowski [1994] found a smaller locked zone width than we find here, but they also assumed the slip deficit had to equal the full plate convergence rate. Zheng et al. [1996] found that 20% coupling worked well to describe the data. Their assumed locked zone extended farther down dip and thus closer to the geodetic network, which has the effect of requiring less coupling to produce the observed strains. Increasing the downdip length of the locked zone and decreasing the amount of slip deficit results in a constant moment deficit, so all of these have similar moment deficit rates. Since both the Zheng et al. [1996] and Savage and Lisowski [1986] models are two dimensional we can compare the yearly energy stored on the length equivalent to the Shumagin segment (plane 3) in the model determined here. Although our model extends the locked interface all the way to the trench, this is a modeling convenience and we have shown that this is not required by the data. Based on the distribution of microseismicity Zheng et al. [1996] used ∼16 km for the updip end of the locked zone. Since this is shallower than the sensitivity limit of our data (21 km in this region) and we do not expect that the locked interface extends to the trench, we use 16 km as the upper limit to the locked zone. With a rigidity of 3E10 Pa this provides a moment deficit rate of 6.4E18 Nm/yr for our model compared to 5.2E18 Nm/yr for the Zheng et al. [1996] model and 9E18 Nm/yr for the Savage and Lisowski [1986] model. In terms of magnitude this amount of energy is equivalent to an Mw 6.47, 6.41 and 6.57 each year, respectively.

[19] Boyd et al. [1988] estimated the average recurrence interval for major earthquakes in the Shumagin region to be about 65 years, and Nishenko and Jacob [1990] estimated 64 years. If the Shumagin plane determined in this model is ruptured every 65 years, this would amount to an average slip of 1.4 m, and would produce roughly an Mw 7.6 quake, which is in agreement with the size and rupture area reported by Estabrook and Boyd [1992] for the 1917 earthquake. The relatively low average slip deficit, and the occurrence of two M ∼ 7 earthquakes in the region since 1917, makes it possible that instead of rupturing in one large earthquake, more frequent moderate earthquakes may be likely. The magnitudes of the 1948 and 1993 earthquakes (Ms 7.5 and Ms 7.1) are roughly consistent with rupture of this entire zone every ∼40 years, making up the full moment deficit. In a region characterized by creep, it is also possible that a significant fraction of the slip on the interface may occur as afterslip following large earthquakes, as has been observed for parts of the plate interface in northern Japan [Heki and Tamura, 1997; Heki et al., 1997; Igarashi et al., 2003].

[20] The seismicity during the last century and the geodetic observations of the past decade show a similar transition in character from east to west. Figure 3 shows Alaska Earthquake Information Center (AEIC) catalog (M > 5) events from the last century. In the east the 1938 rupture zone has relatively few earthquakes but does have several large sized shocks and one great earthquake. In the west there are many small earthquakes but no large events. The Shumagin transition zone has a high density of smaller quakes and a few large tremors. This contrast is similar to that of creeping and non-creeping segments of the San Andreas fault in California [Wiemer and Wyss, 1997].