Seismic rate variations prior to the 2010 Maule, Chile MW 8.8 giant megathrust earthquake

The MW 8.8 Maule earthquake is the largest well-recorded megathrust earthquake reported in South America. It is known to have had very few foreshocks due to its locking degree, and a strong aftershock activity. We analyze seismic activity in the area of the 27 February 2010, MW 8.8 Maule earthquake at different time scales from 2000 to 2019. We differentiate the seismicity located inside the coseismic rupture zone of the main shock from that located in the areas surrounding the rupture zone. Using an original spatial and temporal method of seismic comparison, we find that after a period of seismic activity, the rupture zone at the plate interface experienced a long-term seismic quiescence before the main shock. Furthermore, a few days before the main shock, a set of seismic bursts of foreshocks located within the highest coseismic displacement area is observed. We show that after the main shock, the seismic rate decelerates during a period of 3 years, until reaching its initial interseismic value. We conclude that this megathrust earthquake is the consequence of various preparation stages increasing the locking degree at the plate interface and following an irregular pattern of seismic activity at large and short time scales.

www.nature.com/scientificreports/ shock), slow slip events (SSE) [25][26][27][28][29] within the seismogenic zone, or episodic tremors and slip (ETS) 30,31 below the seismogenic zone may occur, indicating that unusual semi-aseismic slip can act as precursors of large ruptures. However, the irregular nature 3,31,32 of the sequences of seismic activity followed or synchronized with aseismic slow slip at depth prohibits us from using them as an unambiguous proxy to predict brutal and large ruptures. At larger time scales before the main seismic event, some author found that seismic rate changes can be observed in the whole coseismic area or in the vicinity of the rupture zone. Huang 33 indicates for example that for some major earthquakes of the Asia and Asia-Pacific regions, a seismic quiescence, corresponding to a substantial seismic rate decreasing, occurs a few years before the mainshock. In the case of moderate earthquakes (5 < M W < 6), Gentili et al. 34 also found in most of their studied cases seismic quiescence processes, alternating with regain of seismic activity from months to a few years before the main event. As for the foreshock activity in a moderate to short time before the main event, the rapid quiescence of usually seismogenic zones could play the role of potential precursors of major ruptures. Various studies indicate the occurrence of seismic foreshocks (classic earthquakes, SSE, ETS), or possible shutdown of seismic activity for decades 35 before large earthquakes, but still no observations of these combined seismic rate changes in the short time before a subduction Giant Mega-Earthquake (GME in the following) exist.
In the work presented here, we analyze seismic activity in the area of the 27 February 2010, M W 8.8 Maule GME at different time scales from 2000 to 2019 in order to more precisely characterize the potential changes in the seismic rate preceding this earthquake. We differentiate the seismicity located inside the coseismic rupture zone of the main shock from that located in the areas surrounding the rupture zone, using equivalent spatial areas and similar time windows of observations. Our original combined method of spatial-temporal analysis of the seismicity within and around the coseismic rupture zone allows us to identify potential seismic precursors of the Maule earthquake, beginning years before the mainshock as well as a few days before it.

Methods
Geometry of the studied zone. To analyze the seismic activity in the near area of the M W 8.8 Maule earthquake, we differentiated two zones located within the 33°S-40°S/69°W-76°W region, corresponding to the center-coast of Chile (Fig. 1). The first zone (IN-zone, blue in the figure) was defined as the limit of the coseismic slip of the Maule Earthquake, obtained by Moreno et al. 36 using joint inversion of GPS, InSAR and land-level changes 36,37 . This model was selected amongst the existing coseismic models for its robustness, using all the relevant GPS data available at this time. In order to study the specificity of the seismic activity of the Maule coseismic zone before and after the main shock, we defined an OUT-zone encompassing the seismicity outside the coseismic area, but inside the surrounding region (Fig. 1a). The OUT-zone had been chosen to have an equivalent area to the IN-zone for consistency of the comparisons. The geometry of the closed polygon delimitating the OUT-zone includes more than 95% of the seismicity in the region during the studied period. Earthquake catalogue. We used the recent updated catalogue made by the International Seismological Center (ISC, http://www.isc.ac.uk, last reviewed on February 2017) to determine the location and magnitude of the events during the period 2000-2018 of the study. In order to maximize the coherence of the data during this period, we prioritized, when available, M W and M S magnitudes and location estimations of the ISC, GFZ, NEIC, GUC, and SJA respectively (see http://www.isc.ac.uk/iscbu lleti n/agenc ies/ for more information about the contributing agencies). In order to diminish the impact of the non-thrust seismicity on our observations (such as www.nature.com/scientificreports/ intra-accretionary wedge seismicity or intraplate seismicity), we only considered seismicity located at depths larger than 20 km. Figure 2 shows the cumulative and non-cumulative Frequency-Magnitude Density (FMD) of our seismic dataset for the three regions defined in Fig. 1b. We evaluated the magnitude of completeness M C of our datasets by using two specific catalog-based techniques, and selected for each dataset the highest value in order to comfort the selected range of magnitude used in this study: (1) the non-parametric maximum curvature method (MAXC), consisting in computing the maximum value of the first derivative of the FMD curve, and (2) the parametric magnitude of completeness estimation by b-value stability (MBS) defined by Cao and Gao 38 and improved by Woessner et al. 39 . The MAXC technique is known to underestimate M C in bulk data, whereas the MBS overestimates M C , so that computing both methods can provide a reliable range of M C estimations 40 . The b-value is then computed by using the maximum likelihood method parametrized with the estimated M C : where M corresponds to the mean magnitude of the sample for magnitudes higher than M C , and M to the binning width of the catalog 41 . The values of M C obtained for each period and dataset with both MAXC and MBS methods and the corresponding b-values are shown in the supplementary material (Supplementary Table T1).
Statistical analysis of the earthquakes sequence. In order to obtain a quantitative value reflecting if the seismicity observed within the IN-zone a few days before the 2010 main shock is connected or independent of the following major event, we compute a simple statistical test based on the random Poisson distribution. Basically, we calculate for our dataset the parameter λ of the Poisson distribution for different periods of time before the main event, and deduced the probability that the observed seismicity before the main shock occurs. If the probability is high, then the observed foreshock sequence is not specific to the following main shock, as such sequence could frequently and randomly occur. If it is low, then this type of sequence is rare or unique and could be related to the subsequent mainshock. The Poisson probability can be written as: www.nature.com/scientificreports/ where N is the number of events, t = S R × t with S R the seismic rate for the chosen period of the dataset and Δt the period in which we evaluate the probability that N events occur. The probability that a sequence of at least k events occur within a period Δt for a selected dataset can then be written as: Figure 3 shows the results of the Poisson probability P(t), function of the considered periods t, for k = 6 events occurring in Δt = 5 days for both studied regions. These results are discussed in the later part (see results and Fig. 5).

Results
Seismic quiescence. The 27 February 2010 Maule M W 8.8 earthquake is known to have had very few earthquakes of M W > 5 in the large rupture zone during a long period of time, starting at least in the 1990's 4,5 . In a preliminary study Madariaga et al. 2 identified only ten events of M W > 5 in the 10-year period before the main shock. During this last part of the interseismic period before the Maule earthquake, the seismic activity (M > 4) within our defined OUT-zone shows a regular trend of about 25 events per year (Fig. 4), followed by a regular trend of about 32 events per year starting a few months after the main-shock, involving a large reactivation zone at depth within the slab. As a result of the Maule GME, the seismic station network substantially improved after 2010 in the studied region, which could explain the increase of ~ 30% of the seismic rate of the OUT-zone. However, as confirmed in Fig. 2, the cumulative FMD of both OUT and entire region (red and black, left panel Fig. 2) shows that the local seismic network was already sufficient to catch the seismicity with M > 4 in this particular zone, confirming a physical process behind the observed rate increase. The IN-zone, on the other hand, shows a different behavior (blue curves, Fig. 4c,d): a seismic rate of around 20 events per year from the years 2000 to 2007, followed by a quiet time of 2-3 years before the main shock, where the seismic rate strongly decreased down to about 7 events per year. Only one event with M W > 5 was recorded during this period, whereas this zone experienced ten M W > 5 in the first 5 years of the twenty-first century, as explained before. Two years after the main shock and its numerous aftershocks, the rupture zone of the Maule earthquake came back to its normal seismic rate value of 20-22 events per year, highlighting the unusual aspect of the seismic quiescence observed between 2007 and 2010. The appearance of a seismic gap during the period from 2007 to 2010 within the IN-zone, not observed in the OUT-zone, seems to indicate that the plate-interface locally locked, entailing an increase of the stress acting on the major patches of "silent" locked asperities, located within the IN-zone. Indeed, deeper earthquakes are still observed during this period (east part of the OUT-zone, Figs. 1b and 4a), indicating that processes of slab plunge continues, while the upper part is resisting the traction. Furthermore, various geodetic studies confirm that before the main shock, the locking degree was indeed very high at the coseismic interface 36,42,43 . The quiescence of the IN-zone being not "absolute", slip at the interface probably still occurs, releasing a small part of the accumulated energy.
Foreshock activity. Five to six weeks before the main shock, two events with M > 4 where recorded within the rupture zone on 14 and 21 January, as already observed by Bouchon et al. 3 , followed a week later by a M W = 5 event very close to the previously defined IN-zone, on 29 January 2010 (Fig. 5a). Both M > 4 events were located very close to the epicenter of the 2010 earthquake in front of the town of Cobquecura. Slip in this patch was close to 10 m during the main shock (Fig. 5b). After January 29 (day of the last M W = 5 foreshock), a decrease of about a month in the activity within the IN-zone was observed, until the initiation of a seismic reactivation   (Fig. 5b). These events can be considered as foreshocks of the Maule main event, as the probability of such seismic rate changes in this zone is extremely low (p < 0.1% for the IN-zone, and p < 0.01% if we only focus on the major coseismic displacement patch). After the main shock, this zone remained the most active, as observed in Fig. 4b (and Supplementary Fig. S1).  www.nature.com/scientificreports/ Postseismic sequence. Immediately after the main shock, the seismic rate within the IN-zone drastically increased, concentrating most of the aftershocks sequence 7,8 , and confirming the location of the major seismic coalescent patches at the interface. Two to three years after the main shock, the seismic rate within the IN-zone returned to its initial interseismic value, whereas the OUT-zone presented a slight increase probably connected with the post-seismic reactivation of preexisting seismic faults ( Fig. 4 and Supplementary Fig. S1). Interestingly, during the period 2013-2018, while the OUT-zone presents an increase of its seismic rate for M > 4, a decrease of its seismic rate for M < 4 can be observed (Fig. 1b, three upper panels versus three lower panels, and yellow dots Fig. 4a,b). As the seismic network improved during this period compared to the 2000-2009 time window, this decrease of the small earthquake rate in this region is unlikely to be connected to a lack of data. Deeper large earthquakes (z > 50 km) are indeed known to have anomalously low aftershock productivity, specifically in the Andean subduction region 44,45 due to the temperature and stress regime a depth, leading to a deficit of lower earthquakes magnitudes, compared to medium-to-high magnitudes.

Discussion
The observation of pre-seismic quiescence within the 2010 Maule earthquake rupture zone during 2 years, followed by a seismic reactivation a few weeks and days before the main shock, confirm that this megathrust earthquake was preceded by preparation phases reflecting various states of deformation accommodation, and leading to the major rupture it experienced on 27 February 2010 (Fig. 6). The seismic quiescence, illustrated by the stage 2 in Fig. 6, could be interpreted as a period of SSE or ETS 25,31 , but no geodetic data support the existence of such patterns in this zone. Rather, several studies have shown that the coseismic 27F zone was strongly locked before the main shock 6,36,43 , experiencing a very low strain rate compared to the surrounding zones 46 . During the period of low seismic rate, the largest asperities on the plate interface then resist to slip, increasing the locking degree of the entire interface. Such pattern of stress and strain rate changes along the plate interface years before a main shock is in good accordance with some previous studies in other contexts of subduction zones 47,48 . A short time before the rupture, bursts of seismicity occur within the main locked zones corresponding to the largest ready-to-slip friction patches, and reflecting the high degree of accumulated stress in these places. These foreshocks, that occur in localized areas that are usually quiet, are unlikely to trigger the strong instability resulting in a large rupture, because of the very small Coulomb stress changes involved (only a few mbars of stress change in the case of M < 4), but seem to indicate the state of maturity of the contact. After the main rupture, the rapid deceleration of the seismic rate, reaching its regular rhythm in less than 3 year inside the coseismic area, reflects that the plate interface rapidly re-locked 49,50 in its quasi-entire surface.
The irregularity of the time scale with which these different preparation phases occur during the seismic cycle has been observed for past large earthquakes 31,51,52 , as well as foreshock occurrence 3,53,54 . However, the sequence of seismicity-quiescence-foreshocks-mainshock-aftershocks-seismicity scheme, illustrated by substantial seismic 0 1 4 -J a n 2 1 -J a n 2 9 -J a n 0 www.nature.com/scientificreports/ rate changes (Fig. 6c), similar to the "A-type Mogi Doughnut" seismic pattern observed for medium-to-large earthquakes 51,54 , has never been documented at the scale of a subduction GME. In the case of the M W 8.8 Maule earthquake, these first observations of changes in the seismic rate at different time and space scales, put in  The locking degree increases at the interface, and the stress strongly increases on the main asperities. Stage 3 corresponds to the foreshock phase occurring a few days before the main shock. This activity occurred close to the areas where the maximum co-seismic slip were observed, indicating the maturity of the main asperities to fail soon, and Stage 4 images the main shock occurring on 27 February 2010. On Stage 5, numerous aftershocks occur during the early relaxation phase within the main shock rupture zone, before the main asperities grip again, increasing the local locking degree. Stage 6. At this stage, the seismic rate reaches its steady-state, going back to the first stage. (b) Conceptual scheme of the stress change on the plate interface during the seismic cycle of the M W 8.8 Maule earthquake. (c) Conceptual scheme of the seismic rate change.