Reconstructing paleoseismic deformation, 1: modern analogues from the 1960 and 2010 Chilean great earthquakes

The 1960, and 2010 Chilean great earthquakes provide modern analogues for the sedimentary signatures of the largest megathrust events and their accompanying tsunamis. This paper presents lithological and diatom assemblage data from five sites and provides key insights for the development of longer earthquake chronologies, essential for assessing the seismic hazards associated with a subduction zone. We find that the 1960 and 2010 tsunami deposits are fragmentary, variable and have no unique, diagnostic diatom assemblage. Where rapid postseismic sedimentation occurs, our diatom-based transfer function model gives estimates of coseismic deformation that agree with independent estimates of land-level change. Sedimentary hiatuses at two sites following the 2010 earthquake suggest that the magnitude of coseismic deformation may be underestimated in fossil records. Where sediment accumulation allows, criteria for distinguishing between seismic and non-seismic stratigraphies based on evidence for the largest plate boundary earthquakes are corroborated by the lesser magnitude earthquake of 2010. The key to reconstructing earthquake characteristics, such as rupture magnitude and differences between plate-boundary and upper plate sources, depends on applying explicit stratigraphic assessment criteria at multiple sites in order to identify the spatial pattern of deformation associated with each earthquake.


Introduction
The Chilean megathrust creates great earthquakes exceeding moment magnitude (M w ) 8, including the greatest magnitude ever recorded, the 1960 M w 9.5 rupture of the Valdivia segment, and the M w 8.8 Maule earthquake of 27th February 2010. These earthquakes are characterised by intense, long duration shaking, significant land surface deformation, generation of near-field tsunamis along the Chilean coast and may spawn destructive trans-Pacific tsunamis. Paleoseismic research at other subduction zones suggests historical and instrumental records may be too short to adequately assess the recurrence of the greatest magnitude seismic hazards, a factor contributing to inadequate anticipation of the 2004 Sumatra-Andaman and 2011 Tohoku megathrust earthquakes (Stein and Okal, 2011). Records kept by Spanish settlers and visiting Europeans indicate four megathrust earthquakes in the Valdivia segment over the last 500 years but current paleoseismic evidence records only some of these. Using evidence from tidal marshes, Cisternas et al. (2005) propose a 300-year recurrence interval between the largest ruptures, with the megathrust remaining partly loaded with accumulated plate motion through smaller intervening earthquakes. Differences between the historical and paleoseismic evidence may reflect variations in the size of the rupture zones of megathrust earthquakes; alternatively, interseismic land uplift may lead to low sediment accumulation or erosion of tidal marshes so the sediments record only a partial chronology of great earthquakes. Given the range of processes that may control the preservation of paleoseismic evidence (McCalpin and Carver, 2009), we require correlation of evidence from multiple sites in order to reconstruct the dimensions of land surface deformation and therefore estimate the extent of the segment rupture for each event (Nelson et al., 1996;Atwater and Hemphill-Haley, 1997;Atwater et al., 2005). Analysis of the sedimentary record of the 1960 and 2010 earthquakes provides potential modern analogues for building century to millennial scale paleoseismic records for different segments of the Chilean megathrust.
Following both the 1960 and 2010 earthquakes, measurements of the displacement of coastal landforms and biotic environments, such as shore platforms and intertidal encrusting molluscs, provided the first quantitative maps of coseismic land uplift and subsidence, essential for constraining models of slip distribution (Plafker and Savage, 1970;Moreno et al., 2009;Farías et al., 2010;Lorito et al., 2011;Vargas et al., 2011;Melnick et al., 2012). Such records can be fragmentary in both space and time and sedimentary records from tidal marshes can significantly enhance the paleoseismic record of both the coseismic deformation and changes through complete earthquake deformation cycles (Hamilton and Shennan, 2005). In this paper we aim to develop quantitative reconstructions of relative land surface deformation during the 1960 and 2010 earthquakes based on lithostratigraphy and diatom assemblages from five sites. We compare our results with other estimates of coseismic deformation and use our findings to test the applicability of criteria developed in studies of the Cascadia subduction zone to differentiate between sedimentary evidence of relative land-and sea-level changes of seismic and non-seismic origins (Nelson et al., 1996). The Cascadia studies, from Atwater's (1987) seminal paper onwards, made comparisons with the 1960 Chilean earthquake and the 1964 M w 9.2 earthquake in Alaska. Nelson et al. (1996) suggested their criteria would apply to magnitude 8 þ earthquakes and the 2010 M w 8.8 Maule earthquake provides the opportunity to directly test them. Finally, we assess the application and limitations of stratigraphic and microfossil approaches for developing Holocene records of multiple earthquake deformation cycles in Chile.

Study area and methods
In August 2010, six months after the 27th February Maule earthquake, we completed field investigations of tidal marshes at five sites, which we number from north to south for ease of reference (Fig. 1). Sites 1 and 2 lie in the 2010 segment, sites 3 and 4 in the area where the 1960 and 2010 segments overlap, and site 5 is near the centre of the 1960 segment. The spatial pattern of deformation varies longitudinally within each rupture zone, with distance from the trench controlling the zones of coseismic uplift and subsidence (Fig. 1). We targeted tidal marshes as they may preserve evidence of land-level changes and tsunami inundation in their sediment stratigraphy (Atwater, 1987;Cisternas et al., 2005;Hamilton and Shennan, 2005). We employed transects of pits and short cores across marshes and adjacent coastal lowlands to assess the type, continuity and extent of 2010 tsunami deposition and the lateral extent and net accretion of sediments laid down since the earthquake. Any sub-surface sand layers were also documented and we use local testimony, comparison with other studies and caesium-137 ( 137 Cs) concentrations to suggest which relate to tsunami deposition following the 1960 earthquake.
We selected cores from each transect for further laboratory analyses, including diatom assemblages and grain size variations, to reconstruct relative sea-level changes and to assess the composition of tsunami deposits. Laboratory preparation of diatoms followed standard methods (Palmer and Abbott, 1986) with a minimum of 250 diatom valves counted per sample. Diatoms live at the sediment surface (epipelic or epipsammic), attached to vegetation (epiphytic) or in the water column (planktonic) and respond to variations in their environment, including salinity and frequency of tidal inundation (e.g. Vos and de Wolf, 1993). Diatoms may provide information on the depositional environment and source of sediments to a coastal area, for example they can be used to distinguish tsunami deposits from other sediments and help identify the source of tsunami lain sand (Dawson et al., 1996;Hemphill-Haley, 1996;Dawson, 2007;Horton et al., 2011).
Diatom assemblages can also provide estimates of coseismic land movement with decimetre precision (e.g. Shennan et al., 1996;Zong et al., 2003;Hamilton and Shennan, 2005). We estimate coseismic deformation for each site by comparing the diatom assemblages from sediments deposited before and after the earthquake with the modern distribution of diatoms at two tidal marshes in northern Isla de Chiloé, site 5, and data collected by Nelson et al. (2009). We use modern diatom samples collected in 2010 from site 5 only as it was not affected by elevation change or tsunami inundation in 2010. To account for variations in tidal ranges between sites, we convert the elevation of each modern sample to a standardized water level index (SWLI), whereby a SWLI value of 100 represents mean sea level and 200 represents mean higher high water (Hamilton and Shennan, 2005). We define: where: SWLI n is the standardised water level index for sample n h n is the elevation of sample n h MSL is Mean Sea Level at the site h MHHW is Mean Higher High Water at the site We follow the transfer function approach outlined by Hamilton and Shennan (2005), first using detrended canonical correspondence analysis (DCCA) to determine the requirement for a unimodal method, then developing a transfer function model that can produce reconstructions of marsh surface elevations, with error terms, from fossil diatom sequences (Software: C2 version 1.7.2, Juggins, 2011). Estimates of coseismic deformation compare preand post-earthquake marsh surface elevations, accounting for the thickness of any tsunami deposit. We define the uncertainty for each estimate of coseismic deformation: where: CD error is the 1s error of the coseismic deformation estimate E PRE error and E POST error are the sample specific standard errors for samples preceding and following the deformation event respectively.

Transfer function model development
The modern training set comprises 96 samples collected in 2010 and 32 collected by Nelson et al. (2009) Birks, 1995) and we use weighted averaging partial least squares regression (WA-PLS) with bootstrapping crossvalidation (ter Braak and Juggins, 1993;Birks, 1995). We chose the three-component model over one and two-component models, as it has the highest r 2 value, a more linear distribution of observed against predicted values and a RMSEP improvement of at least 5% with the addition of each extra component (Fig. 2). Increased precision can be obtained by reducing the range of sampled elevations; however we currently have no independent measure to select a model using a narrower elevation range, as applied in Alaska where the data set is more than twice the size (Hamilton and Shennan, 2005).
Given the distance between sites and their different environmental conditions, we may not expect the modern training set to fully reflect the range of diatom assemblages and environments that exist in fossil samples, even though the number of samples exceeds the minimum required to give sample-specific error terms in WA-PLS. We use an analogue measure, modern analogue technique (MAT) to quantify the similarity between each fossil sample and the modern training set using a squared chord distance dissimilarity method (Birks, 1995). We use the 5th percentile of the dissimilarity values for the modern samples as the threshold between a 'good' and 'close' modern analogues for each fossil sample and the 20th percentile as the cut-off for a 'poor' modern analogue.

Site 1: Río Mataquito
The site lies within the 2010 segment and north of the 1960 segment ( Fig. 1). The Río Mataquito is deflected northwards by an 8 km long supratidal sand spit at its confluence with the Pacific. The spit was largely submerged following coseismic deformation in 2010 or eroded by the ensuing tsunami which crested 11 m above tide level ( channel, and thins in a landward direction (Fig. 4). The lower contact is abrupt, with occasional above ground parts of terrestrial plants preserved immediately beneath and within the sand, pointing in the direction of flow as they were flattened by tsunami inundation. The deposit is normally graded and is both coarser grained and less organic than the underlying tidal marsh sediments.
Diatom assemblages from the tsunami deposit are characterised by epipelic and epipsammic taxa, with a significant minority, w30%, of planktonic forms (Fig. 5). Almost 65% of the tsunami assemblage is characterised by diatom species identified by the transfer function model as reflecting environments above mean higher high water. A further 25% of the assemblage consists of species not encountered in the modern tidal marshes, with the remaining 10% corresponding to species indicative of environments below mean higher high water.
Despite an increase in accommodation space resulting from the reported coseismic subsidence, (Vargas et al., 2011;Vigny et al., 2011), we do not record any post-tsunami sediment accumulation in the six month interval between the earthquake and field sampling and, therefore, no estimate of coseismic deformation is possible using the diatom-based transfer function.

Site 2: Río Andalién
The site lies within the 2010 segment and immediately north of the 1960 segment ( Fig. 1). Tidal marshes occupy the broad northfacing embayment of the Bahía de Concepción between Talcahuano and Penco. Estimates of coseismic movement in 2010 indicate uplift to the west (Fritz et al., 2011) and subsidence to the south (Vigny et al., 2011). Watermarks in Talcahuano and Penco indicate 2010 tsunami flow depths of between 4 and 7 m (Fritz et al., 2011) and a maximum inundation distance across the low lying tidal marshes of 2.6 km (Morton et al., 2011). We investigated two transects from the eastern edge of the embayment, alongside the Río Andalién (Fig. 2).

2010 Earthquake and tsunami
The 2010 tsunami deposit is a largely continuous sand sheet with occasional rounded mud rip-up clasts. The normally graded deposit is generally less than 0.1 m thick and both transects exhibit landward thinning (Fig. 4). The lower boundary is abrupt, however vegetation was generally found to have remained in growth position, rather than having been flattened beneath the sand layer. Diatoms that live attached to sediment account for more than 75% of the 2010 tsunami assemblage (Fig. 5). The summary assemblages (Fig. 6A) show that the tsunami deposit has a greater proportion of species indicative of lower elevations than the underlying tidal marsh sediment, reflecting net sediment transport from the lower intertidal and perhaps subtidal zone during the tsunami.
Six months after the earthquake, postseismic sediment accumulation had reached a maximum of 20 mm. Pre-and post-tsunami diatom assemblages indicate subsidence of 0.75 AE 0.43 m (Fig. 6A), but we note the poor modern analogues and discuss this further in Section 5.3.

1960 Earthquake and tsunami
A second sand layer occurs approximately 0.2 m below the present ground surface (Fig. 4). Comparison of the colour, grain size and nature of the lower contact with the 2010 tsunami deposit suggests that this lower sand layer was also deposited by a tsunami. Numerous significant tsunamis have struck the coast of central Chile, including major events in 1960, 1835, 1751, 1730, 1657, 1575and 1570(Lomnitz, 2004Cisternas et al., 2005). Tide gauge data from Talcahuano indicate that the 1960 tsunami reached heights of 3 m within Bahía de Concepción (Sievers et al., 1963). Comparable tide gauge measurements of the 2010 tsunami were approximately 0.65 m lower, suggesting that the 1960 tsunami was also of sufficient size to erode, transport and deposit intertidal sediments within the bay. We find elevated 137 Cs concentrations immediately below the sand layer; this indicates deposition no earlier than the beginning of atmospheric nuclear testing in 1952, suggesting that the sand layer was deposited in 1960.   Denys, 1991;Stoermer, 1980;Van Dam et al., 1994;de Wolf, 1993, 1988). Epipelic and epipsammic species are grouped as "attached to sediment". We do not report any tsunami deposit from site 3, Tubul.
Diatom assemblages from this deposit are highly mixed, with epiphytic species and those that live attached to sediment contributing the majority of the assemblage (Fig. 5). Planktonic species are rare, accounting for less than 5% of the assemblage. Species indicative of elevation classes above and below mean higher high water occur approximately equally (Fig. 7A), giving no clear indication of predominant sediment source.
Pre-and post-tsunami diatom assemblages are generally similar, with the largest changes in species that are not characterised by our modern assemblage data (Fig. 7A). Quantitative reconstructions indicate uplift of 0.11 AE 0.45 m, again noting the poor modern analogue classifications.

Site 3: Tubul
Tubul lies in the area where the 2010 and 1960 segments overlap (Fig. 1). The Tubul and Raqui rivers drain a substantial sheltered tidal and freshwater marsh on the northern edge of the Arauco Peninsula (Figs. 1 and 3). Estimates of coseismic movement in 2010 indicate uplift of between 1 and 2 m (e.g. Farías et al., 2010;Melnick et al., 2012). Our coring transects ranged from intertidal mud and sand flat to freshwater marsh above the influence of tides.

2010 Earthquake and tsunami
Despite tsunami flow depths estimated at over 5 m (Fritz et al., 2011), we did not observe a surficial or sub-surface tsunami deposit at any location on our transects. This may reflect local effects of coseismic uplift reducing the potential tsunami inundation distance inland and the position of our sampling area with respect to the open coast.
Although we noted no change in sediment lithology in the field, laboratory analysis of the uppermost 5 mm of the recovered sediment profile shows a significant change in diatom assemblage (Fig. 6B). Our reconstruction indicates coseismic uplift of 0.64 AE 0.67 m, with close modern analogues for the pre-earthquake samples.

Site 4: Río Tirua
The Río Tirua meanders through a low-lying coastal plain, characterised by tidal and freshwater marsh environments. The site is close to the southern limit of surface deformation in 2010 and within the 1960 segment (Fig. 1). Intertidal mussels indicate coseismic uplift of between 0.5 and 1 m in 2010 (Melnick et al., 2012). Our transect lies w1 km from the open coast (Fig. 3).

2010 Earthquake and tsunami
While tsunami runup reached 20 m on the exposed rocky shoreline to the southwest, heights closer to the river mouth were approximately half as large (Fritz et al., 2011;Vargas et al., 2011;Bahlburg and Spiske, 2012). The 2010 tsunami deposited a grey sand layer which we traced along the incised banks of the Río Tirua. The deposit exceeds 0.1 m in thickness and thins upstream and, more rapidly, away from the river channel (Fig. 4). The lower contact is abrupt, with frequent flattened stems below and encased within the base of the deposit.
The tsunami diatom assemblage is predominately composed of species that live attached to sediment, with some epiphytic forms (Fig. 5). Taxa favouring the lower elevations of the modern transects contribute 25e45% of the assemblage, with species not found in the modern marshes providing a further 30%.
There was no identifiable postseismic sedimentation six months after the 2010 earthquake. Tsunami-lain sand was still visible at the surface in January 2012, almost two years after the earthquake. No estimate of coseismic deformation is possible using the diatombased transfer function. Assemblage summary based on modern species coefficients derived from the WA-PLS transfer function model. We use the distance to the closest modern analogue from the modern analogue technique in C2 (Juggins, 2011) to assess the similarity between modern and fossil assemblages.

1960 Earthquake and tsunami
We observed a second sand layer, analogous to that deposited by the 2010 tsunami, at a depth varying between 0.1 and 0.4 m alongside the Río Tirua. The contact with underlying marsh sediments is abrupt. Elevated 137 Cs concentrations in the buried marsh surface and the testimony of local residents who experienced the event suggest that this layer relates to the tsunami associated with the 1960 earthquake. Tsunami runup on Isla Assemblage summary based on modern species coefficients derived from the WA-PLS transfer function model. We use the distance to the closest modern analogue from the modern analogue technique in C2 (Juggins, 2011) to assess the similarity between modern and fossil assemblages.
Mocha, 30 km offshore from the mouth of the Río Tirua, exceeded 15 m (Sievers et al., 1963), but the wave height as it approached the mainland in this sector is unknown. By way of comparison, runup during 2010 exceeded 20 m on Isla Mocha (Fritz et al., 2011); however the difference may result from the different approach directions of the two tsunamis. The 1960 tsunami deposit is generally thinner and more fragmented than the 2010 deposit (Fig. 4), as we also noted for site 2, Río Andalién.
The 1960 sand layer is characterised by a higher proportion of epiphytic and planktonic species than the 2010 deposit, although epipelic and epipsammic species still contribute more than half of the total assemblage (Fig. 5). Species indicative of elevations above mean higher high water are dominant in most of the tsunami samples; however a peak in the abundance of one species results in almost 50% of one sample consisting of diatoms of unknown elevation preference (Fig. 7B).
Pre-and post-tsunami diatom assemblages are generally similar; however there are no close modern analogues for any of the samples. Our marsh surface reconstructions are indistinguishable from zero, 0.03 AE 0.72 m (Fig. 7B).

Site 5: Chucalen
On the north west of Isla de Chiloé, tidal marshes line the western fringe of Bahía Quetalmahue, sheltered from the Pacific Ocean by the Lacui Peninsula (Fig. 1). This region is close to the centre of the 1960 segment and w400 km south of the 2010 segment. There was no tsunami recorded here in 2010.

1960 Earthquake and tsunami
At Chucalen we traced a grey sandy deposit through a 100 m long transect at depths of between 0.1 and 0.35 m below the present marsh surface (Fig. 4). The normally graded deposit decreases in thickness with increasing elevation and distance from the marsh front. The contact with the underlying marsh sediments is abrupt. Through comparison with preliminary investigations by Bartsch-Winkler and Schmoll (1993), 137 Cs concentrations and statements from local residents, we correlate this deposit with the 1960 tsunami. Witnesses suggest that a series of three waves resulted in runup exceeding 15 m on exposed headlands on the northern edge of the Lacui Peninsula, with 5 m waves striking Ancud, decreasing to 1.5 m in Bahía Quetalmahue (Sievers et al., 1963). Bartsch-Winkler and Schmoll (1993), however, suggested waves of several times this magnitude may have entered the Quetalmahue estuary across the isthmus that joins the Lacui Peninsula, close to our sampling area.
Diatoms that live attached to sediment account for almost three quarters of the tsunami deposit assemblage at Chucalen, with diatoms of unknown life form making up the second largest component (Fig. 5). When classified by modern distribution, the tsunami deposit exhibits a greater proportion of species indicative of lower elevations than those from the underlying tidal marsh sediment (Fig. 7C).
Diatom assemblages in sediments immediately above and below the tsunami deposit show a change from species characteristic of the highest elevations of modern tidal marshes to taxa more tolerant of regular tidal inundation (Fig. 7C). Paleomarsh surface elevation reconstructions indicate land subsidence of 1.12 AE 0.53 m.

Tsunami deposition
The tidal marshes investigated here demonstrate a variable and fragmentary record of tsunamis associated with two great earthquakes in 1960 and 2010. We found no 2010 tsunami deposit at one of the four marshes adjacent to the rupture zone. Where present, it is composed of more than 85% sand and abruptly overlies finer grained, more organic tidal marsh sediments, frequently preserving flattened but still rooted terrestrial plants at their contact. Flattened vegetation assists in determining tsunami flow direction (e.g. Morton et al., 2011). The extent and continuity of the deposit is highly variable. At Río Mataquito, site 1, the sand layer reaches a maximum of 0.40 m in thickness, however accumulations of 0.05e0.15 m are more common both at this site, and at the other sites investigated. Comparable 2010 tsunami deposit thicknesses are reported by Horton et al. (2011) and Morton et al. (2011). At our sites the 2010 deposit fines in a landward direction, reflecting decreasing sediment transport as the wave train moved inland. We attribute further along-transect variability in the thickness of the deposit to variable vegetation cover and pre-tsunami surface topography (Morton et al., 2007).
The 1960 tsunami deposit is preserved as a continuous stratigraphic layer only at Chucalen, site 5, in the central part of the rupture segment. Towards the northern limit of the 1960 segment it is absent from site 3, Tubul and fragmented at site 4, Río Tirua. It is present, though fragmented at site 2, Río Andalién, just north of the segment boundary, and we did not encounter any buried sand layers at site 1. Where present, the deposit is similar to the 2010 tsunami layer: sand-rich and abruptly overlying organic tidal marsh sediment. The thin and fragmentary nature of the deposit, despite the significant size of the 1960 tsunami, may indicate postseismic erosion prior to burial and encasement into the sediment record.
Diatom assemblages from tsunami sand layers vary both between different sites for the same tsunami and between different tsunamis at the same site. There is no unique tsunami diatom assemblage. Rather, the assemblage reflects the local sediment source, with mixed assemblages of different salinity preferences, different life forms and different habitats. Diatom assemblages are likely to be mixed as tsunamis inundate inland areas and erode, transport and redeposit marine, intertidal and non-marine sediments. Similar mixed assemblages including freshwater, brackish and marine species are observed in tsunami deposits at Pichilemu, central Chile, reported by Horton et al. (2011), and are consistent with observations in modern and paleotsunami deposits elsewhere (Hemphill-Haley, 1996;Atwater and Hemphill-Haley, 1997;Tuttle et al., 2004;Dawson, 2007;Sawai et al., 2008).

Estimating coseismic land-level change
Transfer function models provide estimates of coseismic land surface deformation for two of the five sites in 2010 and three sites in 1960. The lack of post-earthquake sedimentation or the absence of a 1960 tsunami deposit to guide our sampling approach precludes the quantification of deformation at the remaining sites. Despite our reservations based on the lack of good modern analogues, our reconstructions compare favourably with published estimates of coseismic land-level change (Fig. 8). This confirms the potential of using diatom-based transfer function models to quantify coseismic movement from previous great earthquakes in this region. The apparent offset for the 2010 data (Fig. 8) may be the effect of the lack of good modern analogues, but it may also reflect the different measures of coseismic subsidence. The GPS and benchmark relevelling data relate to the vertical movement of rock surfaces, whereas marsh sediments may undergo additional local scale subsidence due to ground shaking and dewatering leading to sediment consolidation. This was observed at numerous locations in Alaska during the 1964 M w 9.2 earthquake (Plafker, 1969) and the M w 8.1 and 8.2 earthquakes in 1899 (Plafker and Thatcher, 2008).
The estimates by Plafker and Savage (1970) for coseismic motions in 1960 may also incorporate some local sediment consolidation, for example their estimate from close to our site 5 based on comparison of the lower growth limit of pre-and post-earthquake vegetation.

Limitations and improvement of quantitative reconstructions of relative land-and sea-level change
Although the current modern training set includes samples from four transects in two locations and from modern environments ranging from unvegetated tidal flat below mean sea level to above the highest limits of tidal inundation, many of the fossil samples do not have a 'good' or 'close' modern analogue. While this is, in part, a result of our selected percentile thresholds that are stricter than those used in some other studies, the apparent dissimilarity between modern and fossil diatom assemblages remains a cause for concern. We highlight this lack of modern analogues as a limitation of the current study and advocate the need for larger training sets, preferably from a wide range of sites. The modern samples are from locations well to the south of sites 1e4 and there may be a spatial control on diatom assemblages that we are currently unable to assess. Ongoing postseismic deformation and the lack of significant sedimentation preclude the collection and use of samples from sites within the 2010 rupture zone and further investigations should focus on the northern half of the 1960 rupture zone.
Accurate estimates of the magnitude of coseismic deformation depend on the recommencement of sediment accumulation before significant postseismic deformation has occurred. The four marshes in the 2010 rupture zone showed variable responses six months after the earthquake. At site 1, no postseismic sedimentation followed coseismic subsidence and tsunami deposition. Up to 20 mm of sedimentation followed coseismic subsidence and tsunami deposition at site 2. At site 3, up to 5 mm of sedimentation followed coseismic uplift and no tsunami sedimentation and by early 2012, vegetated marsh was developing on previously unvegetated tidal flat. At site 4, no postseismic sedimentation had occurred almost two years after coseismic uplift and tsunami sedimentation in 2010. These postseismic accumulation rates are several orders of magnitude less than after the 1964 Alaskan earthquake (Atwater et al., 2001) and closer to estimates from the channels of the Cruces river following the 1960 Chilean earthquake (Reinhardt et al., 2010). Postseismic vertical movements in the six months following the Maule 2010 earthquake were small, estimated from GPS data at <15 mm at Concepción and <50 mm elsewhere along the rupture zone (Baez et al., 2010). Consequently, postseismic movements are unlikely to significantly affect estimates of coseismic land motions based on sediment biostratigraphy at sites 2 and 3. At sites with a sedimentary hiatus, coseismic deformation estimates will include both coseismic and some postseismic movements, resulting in potential underestimation of the coseismic movement. In fossil sequences it can be very difficult to identify the duration of any hiatus. Radiocarbon dated samples either side of the stratigraphic boundary will give maximum and minimum ages and may identify a large hiatus (e.g. Carver and Plafker, 2008), while reconstructions for the same episode from multiple locations on the same marsh also help (e.g. Shennan and Hamilton, 2006).

A test of the criteria to differentiate between sedimentary evidence of seismic and non-seismic relative land-and sea-level changes
Based on observations of sedimentary responses to the largest plate boundary earthquakes and similar stratigraphies found in tectonically stable locations, Nelson et al. (1996) propose a series of criteria for differentiating between evidence for seismic and non-seismic relative land-and sea-level changes. Atwater and Hemphill-Haley (1997) apply the same criteria alongside geophysical approaches and structural geology to discuss the differences between plate-boundary and upper-plate sources for the earthquakes, the sizes of the earthquakes, the dimensions of plateboundary ruptures and the trade-off between size and frequency. Both of these papers point to differences between M w 7.5 or 8.0 earthquakes and great earthquakes, M w 8þ. The 2010 earthquake provides an opportunity to test the criteria on a smaller plateboundary rupture than the 1960 Chilean M w 9.5 and 1964 Alaskan M w 9.2 earthquakes.
The key criteria outlined by Nelson et al. (1996) are: lateral extent of peat-mud couplets with sharp upper contacts, suddenness of subsidence, amount of subsidence, synchroneity of subsidence with other sites and, for some locations, presence of tsunami sediments. Although these criteria were developed for areas undergoing coseismic subsidence, they are equally applicable to identifying coseismic uplift ). Section 5.1, above, demonstrates the variable pattern of tsunami deposition, but confirms that, where present, it is a valuable line of evidence and may occur in either uplifted or subsided locations. Our work on the 2010 earthquake deposits suggests that, in south central Chile, sedimentary hiatuses and low rates of sedimentation may temporarily postpone the formation of the characteristic subsidence stratigraphy (Section 5.2), but the burial and preservation of the 1960 deposits shows that any hiatus is brief in the context of multiple earthquake cycles. Our diatom-based reconstructions of relative land/sea-level change are promising, but require a greater range of modern samples in order to provide better modern analogues and greater confidence in the elevation estimates and associated error terms. Nelson et al. (1996) and Atwater and Hemphill-Haley (1997) draw attention to a lower limit of resolution for identifying coseismic deformation; approximately 0.5 m. Our estimates, Section 5.2, currently do not suggest any finer  (4) and Chucalen (5). Diagonal line is the 1:1 line, not a best-fit regression line. Due to hiatuses or the absence of tsunami deposits, we cannot estimate deformation at sites 1 and 3 for 1960 and at sites 1, 4 and 5 for 2010. For 1960 we compare our estimates with Plafker and Savage's (1970) sampling locations 1, 9 and 45 and for 2010 we compare with the continuous GPS station at Concepción (Vigny et al., 2011) and benchmark relevelling at Tubul (Melnick et al., 2012). resolution. Comparison of the marsh sediment sequences for the 2010 and 1960 earthquakes suggests that they produce similar stratigraphic records and that the key to reconstructing earthquake characteristics, such as the rupture magnitude and differences between plate-boundary and upper-plate sources, depends on applying the stratigraphic criteria at multiple sites in order to identify the spatial pattern of deformation associated with each earthquake.
5.5. Implications for reconstructing Holocene megathrust earthquake rupture zones Precise estimates of coseismic deformation are required from multiple sites to constrain models of the location, dimensions and slip distribution of megathrust earthquakes (Atwater and Hemphill-Haley, 1997;Atwater et al., 2005). To date, elastic deformation models of the 1960 and 2010 earthquakes have been constrained by GPS vectors and measurements of displaced coastal landforms and biotic environments (e.g. Plafker and Savage, 1970;Moreno et al., 2009;Farías et al., 2010;Lorito et al., 2011). Further development of the diatom-based transfer function approach detailed here provides an additional viable method for validating models of these earthquakes, with the added benefit of applicability to older ruptures. Databases of coseismic deformation index points, each characterised by a vertical deformation estimate with an associated error term and spatial and chronological attributes, may be used to assist estimation of the magnitude of historical and prehistoric earthquakes. Such databases have already been successfully employed to constrain the magnitude of past great earthquakes in Cascadia (Atwater and Hemphill-Haley, 1997;Leonard et al., 2004Leonard et al., , 2010Hawkes et al., 2011). The development of this approach along the Chilean subduction zone may allow further investigation of the variability in rupture mode in the 1960 segment (Cisternas et al., 2005) and confirm or refute the permanence of the Arauco Peninsula as a segment boundary over multiple seismic cycles.
Differentiation between the closely temporally spaced rupture of two adjacent segments and a single, multi-segment rupture may prove crucial to interpreting evidence for the largest Holocene megathrust earthquakes (Atwater and Hemphill-Haley, 1997;Atwater et al., 2005;Shennan, 2009;Shennan et al., 2009). While radiocarbon dating alone may be insufficient to distinguish between single and multi-segment ruptures, the work presented here establishes the stratigraphic separation of two closely timed earthquakes in locations close to a seismic segment boundary. We suggest that detailed marsh surface elevation reconstructions from boundary locations, combined with precise dating approaches, should form an integral part of establishing the long-term history of the seismic hazards associated with the Chilean subduction zone.

Conclusions
The 1960 and 2010 Chilean great earthquakes provide critical modern analogues for sedimentary processes during seismic cycles at plate boundaries. The major conclusions of this work are: 1) Deposits from the 1960 and 2010 tsunamis are fragmentary, variable and have no unique, diagnostic diatom assemblage. 2) Our transfer function method provides estimates of coseismic land surface deformation for two sites in 2010 and three sites in 1960. Reconstructions agree with independent estimates, confirming the potential for our approach to be used to quantify coseismic deformation for previous great earthquakes in south central Chile.
3) Sedimentary hiatuses at two sites following the 2010 earthquake indicate that the magnitude of coseismic deformation may be underestimated in fossil records. 4) A lack of close modern analogues for fossil diatom assemblages remains a limitation of the current study and we advocate the need for larger training sets, preferably from a wide range of sites. 5) Where sediment accumulation allows, criteria for distinguishing between seismic and non-seismic stratigraphies developed from evidence for the largest plate boundary earthquakes (Nelson et al., 1996;Atwater and Hemphill-Haley, 1997) are corroborated by the lesser magnitude earthquake of 2010.