The physical state of the shallow plate-boundary fault governs the updip extent of seismic rupture during powerful subduction zone earthquakes and thus on a first order impacts on the tsunamigenic hazard of such events. During the 2004 Mw 9.2 Aceh-Andaman Earthquake seismic rupture extended unusually far seaward below the accretionary prism causing the disastrous Indian Ocean Tsunami. Here we show that the formation of a strong bulk sediment section and a high fluid-pressured predécollement, that likely enabled the 2004 rupture to reach the shallow plate-boundary, result from thermally controlled diagenetic processes in the upper oceanic basement and overlying sediments. Thickening of the sediment section to >2 km ∼160 km seaward of the subduction zone increases temperatures at the sediment basement interface and triggers mineral transformation and dehydration (e.g., smectite-illite) prior to subduction. The liberated fluids migrate into a layer that likely host high porosity and permeability and that is unique to the 2004 rupture area where they generate a distinct overpressured predécollement. Clay mineral transformation further supports processes of semilithification, induration of sediments, and coupled with compaction dewatering all amplified by the thick sediment section together strengthens the bulk sediments. Farther south, where the 2005 Sumatra Earthquake did not include similar shallow rupture, sediment thickness on the oceanic plate is significantly smaller. Therefore, similar diagenetic processes occur later and deeper in the subduction zone. Hence, we propose that shallow seismic rupture during the 2004 earthquake is primarily controlled by the thickness and composition of oceanic plate sediments.
 The amount, composition, and physical properties of sediments entering a subduction zone affect the physical state of the plate-boundary fault and thus influence the rupture pattern of megathrust earthquakes [e.g., Ruff and Kanamori, 1980; Underwood, 2009]. In 2004 and 2005, two of the largest ever instrumentally recorded earthquakes ruptured large parts of the Sunda Subduction Zone with significant differences in their respective slip patterns. In 2004, the Mw 9.2 Aceh-Andaman Earthquake ruptured ∼1300 km of the plate-interface involving seismic slip close to the trench that triggered the disastrous Indian Ocean Tsunami [Ishii et al., 2005]. Three months later the plate-boundary segment immediately to the Southeast ruptured during the 2005 Sumatra Earthquake (Mw. 8.7). During this event, seismic slip did not extend as far seaward [Ishii et al., 2007] and the following tsunami was significantly smaller. Coseismic rupture within deeper waters has potential to generate greater tsunami amplitudes, as recently also highlighted by the 2011 Tohoku-Oki Earthquake and tsunami. This underlines the need to understand the geologic factors that support or prevent shallow seismic rupture.
 The boundary between the 2004 Aceh-Andaman and the 2005 Sumatra earthquakes appears stable over multiple earthquake cycles [Meltzner et al., 2012] and may be explained by the presence of topography on the oceanic plate [Franke et al., 2008] or thickened oceanic crust and associated changes in physical properties [Tang et al., 2013]. However, the origin of the segment boundary does not explain the differences in the updip extent of seismic rupture. A common explanation for the updip limit of the seismogenic zone is that it is controlled by temperature induced variations in sediment mineralogy at the base of the sediment section, especially the dehydration of stable-sliding clays, and resulting changes in bulk sediment strength [e.g., Oleskevich et al., 1999]. In the southern 2004 rupture area, Gulick et al.  seismically imaged thick (>4 km) coherent sediment piles that have been accreted to the margin and interpret them as unusually competent and strong with the latter being the result of intense dewatering. Close to the base of the coherent sediment piles Dean et al.  showed a distinct high amplitude negative polarity (HANP) predécollement seismic reflector along a 100 km stretch. This anomalous reflector indicates the presence of a porous and probably fluid-rich layer, in an overall dense and strong sediment section, that appears to act as the décollement upon subduction. They precluded the existence of a similar predécollement in the 2005 rupture area. However, the genetic origin as well as the spatial extent of the HANP reflector seaward of and along the subduction zone remains unclear.
 We investigate ∼50 multichannel reflection seismic (MCS) lines extending from the northern tip of Sumatra (∼5°N) to south of South Pagai Island (∼4°S) covering the southern ∼400 km of the 2004 rupture area, the entire 2005 rupture area, as well as a ∼650 km long stretch that has been unruptured by a large (MW >8.5) earthquake since 1833 [Meltzner et al., 2012] (Figure 1). Some of the MCS lines extend far (∼300 km) into the Wharton Basin imaging the oceanic plate up to ∼6 Ma before it enters the subduction zone [Royer and Sandwell, 1989]. We analyze the nature and spatial distribution of the HANP predécollement reflector and its correlation with sediment thickness on the oceanic plate and propose a conceptual model that includes the thermal regime and dehydration processes within the oceanic plate sediments to explain the physical properties of both this reflector and the surrounding input sediments.
2. Data and Method
 The investigated MCS lines were recorded during R/V SONNE cruises SO186-2, SO189-1, and SO198 (Figure 1). Processing of the seismic lines collected during SO186-2 and SO189-1 was conducted at the BGR and seismic lines collected during SO198 were processed at the University of Southampton. Processing varies slightly between the individual datasets but generally includes deconvolution, gain recovery, and prestack time or depth migration. Sediment thickness on the oceanic plate (Figure 1) was calculated with a constant average velocity of 2.75 km/s over the entire sediment section (based on seismic reflection derived velocities in Dean et al. ). The choice of a constant velocity results in an overestimation of the sediment thickness in areas with thin sediments and an underestimation in areas with thick sediment sections. However, when compared to the “true” sediment thickness along two prestack depth migrated MCS lines (blue lines in Figure 1), we found total errors in the range of ±5%. This uncertainty range is acceptable in the context and resolution of our investigations.
 We also developed a temperature model for the oceanic plate sediment section along MCS line BGR06-102 (Figure 2) based on the GDH1 model for the variation of heat flow with lithospheric age [Stein and Stein, 1992; see also Hippchen and Hyndman, 2008; Klingelhoefer et al., 2010]. For the thermal model, we used a plate age of 60 Ma [Singh et al., 2011] and an average thermal conductivity of 2.2 W/m2C. The latter is based on measurements at ODP Site 718C (Leg 116), located west of the Ninety-East-Ridge, where an average thermal conductivity of 1.9 W/m2C has been measured between 350 and 600 mbsf [Cochran et al., 1989]. As we are looking at a sediment thickness of about 3 km, we expect a slightly higher average thermal conductivity.
3.1. Spatial Distribution of the HANP Reflector and Its Correlation With Sediment Thickness and Temperature
Dean et al.  were able to correlate the HANP reflector, which has amplitudes higher than the top of the oceanic basement and higher than any other reflector within the sediment section, throughout their profiles due to its consistent clear reflection character. The seismic data do not resolve whether the reflector represents a discrete layer significantly thinner than the seismic wavelength (∼100 m) or an interface between two units with contrasting seismic velocities and densities. At other subduction zones where drilling has taken place comparable predécollements have been identified as discrete thin layers (e.g., Barbados) [Moore et al., 1998], and we thus favor the hypothesis of the HANP reflector representing an individual layer. In some profiles shown by Dean et al. , the horizon onlaps onto oceanic basement. Singh et al.  further showed that the HANP reflector is almost continuously present west of the North Sumatra subduction zone, and only interrupted where it onlaps onto basement highs, along a seismic line located between 2 and 4°N (Figure 1). Both authors concluded that the reflector has a lithological origin and Dean et al.  suggested that the nonexistence of the reflector in the 2005 rupture area might be controlled by differences in the input section on either side of the 2004–2005 earthquake segment boundary.
 Extending the investigations of Dean et al.  and Singh et al. , we are able to identify and correlate the HANP reflector in the extensive set of MCS lines used in this study. The HANP reflector appears in all MCS lines located west of the southern 2004 rupture area (red parts of MCS lines shown in Figure 1). The reflector is not present in any of the lines south of Simeulue over 1000 km along the margin. Off northern Sumatra, where MCS data extends far into the Wharton Basin (Figure 1), we are further able to analyze the extent of the HANP reflector seaward from the subduction zone. Along MCS line BGR06–102 (Figure 2), the reflector can be traced to CMP 17100 (∼100 km seaward of the deformation front). Here the reflector abruptly changes to normal polarity and lower amplitudes across a fault (inset A in Figure 2). Between CMPs 10500 and 14000, the reflector is interrupted by a basement high. Farther westward the reflector remains clearly visible but retains normal polarity with moderate amplitudes. Interestingly, the reflector changes back to negative polarity but retains moderate amplitudes to both sides of a seamount that is located around CMP 23700. The polarity change again occurs across distinct fault zones about 10–20 km to both sides from where the reflector onlaps onto the seamount (inset B in Figure 2).
 The distinct amplitude and polarity changes of the HANP reflector along MCS line BGR06–102 indicate that the physical properties causing its peculiar reflection pattern are not entirely a function of the primary properties of the horizon but likely the product of postburial digenetic alterations and fluid distribution. If high-fluid pressure causes the negative polarity, which seems to be the most plausible explanation [Dean et al., 2010], then there must be a mechanism that releases fluids in the deeper sediment section or upper oceanic basement and traps them in this layer. Unraveling this mechanism is essential to understanding the processes that generate a strong bulk sediment section and the location of the décollement close to the base of the strong sediments.
 Sediment thickness on the oceanic plate along the deformation front decreases from maximum values of 5 km at ∼4°N to less than 1 km at ∼4°S (Figure 1). Off northern Sumatra, the sediment thickness decreases gradually seaward toward the Ninety-East-Ridge. The thermal model that we calculated for the oceanic sediments suggests a temperature increase of 30°C per kilometer depth. When compared with the spatial occurrence of the HANP reflector, a first-order correlation is identified of the HANP reflector being restricted to areas where the sediment thickness exceeds 2–3 km and temperatures at the base of the sediments exceed 60–90°C over some tens of kilometers distance.
3.2. Thermal Structure and Fluid Generation Within the Input Section
 Our results combined with previous observations of Dean et al.  and Singh et al.  suggest that a number of circumstances have to be fulfilled to generate the probably focused high fluid content in the horizon represented by the HANP reflector. First, the horizon itself must have a high porosity and permeability. Although the presence of these physical properties is difficult to prove in the absence of drill core data, at other margins such as Barbados or Nankai, where drilling has taken place, comparable predécollement and décollement reflectors host reduced density, high-porosity and high pore-fluid content. In Barbados, a negative polarity décollement results from reduced density and high porosity caused by high-fluid content and underconsolidation in a weak radiolarian-rich claystone [e.g., Moore et al., 1998; Bangs et al., 1999]. In the Nankai Trough (Muroto transect), the décollement is negative polarity beneath the outer prism [Bangs et al., 2004] and drilling and seismic data show high porosity, reducing velocity and density at and below the décollement [Mikada et al., 2005]. Because the negative polarities in the northern Sumatra oceanic sediment section along MCS line BGR06–102 (Figure 2) are first observed within about 20 km distance from where the horizon onlaps onto oceanic basement, we suggest that fluids generated in the shallow oceanic basement are contributing to feeding the horizon. Therefore, second, the reflector has to be in contact with oceanic basement as a pathway for fluids to migrate laterally from the oceanic basement into the horizon, and third, it needs an overburden sediment thickness in the range of 2–3 km or more. In the following, we present a conceptual model for the formation of the HANP reflector (Figure 3) and summarize the different processes that affect prism strength, fault development, and slip behavior in the southern 2004 rupture area (Figure 4).
 Clay dehydration processes, especially the transition from stable-sliding smectite to illite at temperatures between 60 and 150°C [Kastner et al., 1991] and/or pressures above 1.3 MPa [Hüpers and Kopf, 2012], are thought to strongly impact on the updip limit of seismic rupture [e.g., Oleskevich et al., 1999]. Applying our thermal model, temperature-controlled dehydration in the oceanic plate sediments and/or shallow oceanic basalts and volcanigenics should start once a minimum sediment thickness of 2 km (∼1.5 s two-way traveltime [TWT]) is reached. For the 2004 rupture area, this requirement means that the transition from smectite to illite could initiate 160 km seaward of the subduction zone in the central Wharton Basin. In contrast, off the 2005 rupture area and further south, sediment thickness in the trench rarely exceeds 3 km. Where it does exceed 3 km, a significant part of the sediment section has built up very rapidly close to the subduction zone.
 There is ample evidence for smectite in the shallow oceanic basement and in the overburden sediments in the study area. Samples from the distal Bengal and Nicobar Fan showed high clay concentrations with relative abundances of smectite of up to 80% that were measured on Quaternary to Eocene sediments recovered from DSDP Sites 211, 212, and 213 [Venkatarathnam, 1974]. Along the Ninety-East-Ridge DSDP Leg 22 recovered Paleocene and Cretaceous sediments with notably high (partially >90%) smectite concentrations [Matti et al., 1974]. Similar prefan pelagic sediments are expected at the base of the sediment section in the study area overlying the 60–70 Ma old oceanic basement. Smectite in the oceanic basement is formed by the reaction of oceanic basalt with seawater during hydrothermal circulation [Seyfried and Bischoff, 1979]. Where drilling has reached the oceanic basement in the wider study area (e.g., DSDP Sites 211, 212; ODP Site 758) the recovered basalts show evidence for intense weathering and alteration and high smectite concentrations [Hekinian, 1974; Peirce et al., 1989].
 The HANP reflector, only continuously present in sediment depth >3 km, correlates to a window where our estimated temperatures may drive smectite dehydration in both the upper oceanic basement and the sediments. The fact that the reflector changes to normal polarity and moderate amplitudes ∼10–20 km away from where it onlaps onto an oceanic basement high (CMP 14000) or seamount (CMP 23700) (Figure 2) further shows that sediment dehydration alone is likely insufficient to release enough fluids to generate the negative polarities and that a basement source probably has to be involved. As the observed polarity changes occur across distinct faults, it seems that many of the faults imaged along MCS line BGR06–102 (Figure 2) represent (at least currently) impermeable structures.
 Presubduction clay dehydration does not only seem to control the formation and location of the predécollement but potentially also impacts on the physical state of the accretionary prism. To cause locking and seismic rupture of the shallow plate-interface, the basal accretionary prism has to be competent and strong. Gulick et al.  interpreted the strong prism in the southern 2004 rupture area to result from the accretion of thick indurated sediment piles that have undergone significant compaction dewatering and lithification. The process of clay dehydration demonstrated here, i.e. the transformation of sediment derived smectite clays to stiff illite, in the lower sediment section further supports processes of semilithification and induration of sediments. In the outer accretionary prism, further compaction and fracturing may drive fluids from the décollement strengthening the fault zone, with the negative polarity of the seismic reflector diminishing landward [see Dean et al., 2010]. These processes together represent effective mechanisms to create such strong sediment packages (Figure 4). Similarly, the Makran subduction zone, with an input sediment section up to 7.5 km thick, has recently been demonstrated to have the potential for very shallow seismic rupture and large magnitudes earthquakes due to strengthening of basal sediments at high temperatures [Smith et al., 2013].
 The documented amplitude and polarity variations of the predécollement reflector off northern Sumatra characterize the situation long before (up to 160 km, 3.2 Ma), the sediments reach the subduction zone. On the journey toward the subduction zone the increase in sediment thickness will intensify dehydration processes (Figure 3). Therefore, when reaching the subduction zone the sediments will have obtained particular characteristics, i.e., (I) strong and coherent bulk sediment packages in which most of the sediment-derived smectite clays have been transformed to stiff illite and (II) a high fluid-pressured horizon at the base of the strong sediments that acts as the preferred décollement (Figure 4). This way coseismic rupture will likely continue to reach the shallow part of the Aceh-Andaman subduction zone that ruptured in 2004 over the next millions of years.
 An extensive seismic reflection dataset has highlighted processes that control the existence of a predécollement horizon and a strong coherent bulk sediment section in the rupture area of the 2004 Mw 9.2 Aceh-Andaman Earthquake as well as the absence of similar characteristic structures further south in the area that ruptured during the 2005 Mw 8.7 Sumatra Earthquake. Our results suggest that the development of the predécollement within the sediment input section as well as of the surrounding strong bulk sediment section is the result of thermally induced diagenetic reactions in the shallow oceanic basement and overlying sediment section prior to subduction. Sediment thicknesses on the oceanic plate that exceed 2–3 km over tens of kilometers induce temperatures that are high enough (>60°C) to mobilize fluids in the basal sediments and the basement through dehydration of clay minerals. In combination with preferential physical properties, i.e., a high porous and permeable layer that is also locally in contact with the oceanic basement, this results in the formation of an overpressured predécollement. Furthermore, early clay digenesis supports semilithification and induration of the sediments and together with compaction dewatering causes the development of a strong coherent sediment sections. In the subduction zone, this shifts the updip limit of seismic rupture far seaward toward the trench. In contrast, farther south where the 2005 Sumatra Earthquake did not involve comparable shallow seismic rupture, sediment thickness on the oceanic plate is significantly smaller and mineral transformation and dehydration processes likely occur further landward and deeper in the subduction zone. On a global perspective, our results can help to evaluate the likelihood of seismic rupture on the shallow plate-boundary fault in subduction zones with a comparable geologic regime (sediment thickness, lithology, and thermal structure).
 This research was supported by a Marie Curie Intra European Fellowship within the 7th European Community Framework Programme. R/V SONNE cruises SO186-2 and SO189-1 were carried out with grants 03G0186A (SeaCause) and 03G0189A (SUMATRA) of the Federal Ministry of Education and Research (BMBF), Germany. R/V SONNE cruise 198 was funded by the UK Natural Environment Research Council (NE/D004381/1). We thank BPPT, Jakarta for their logistical assistance. We thank reviewers Sean Gulick and Roland von Huene for their careful reviews which helped us to improve the manuscript.