Deep Transient Slow Slip Detected by Survey GPS in the Region of Atacama, Chile

Abstract

We detected a long‐term transient deformation signal between 2014 and 2016 in the Atacama region (Chile) using survey Global Positioning System (GPS) observations. Over an ∼150 km along‐strike region, survey GPS measurements in 2014 and 2016 deviate significantly from the interseismic trend estimated using previous observations. This deviation from steady state deformation is spatially coherent and reveals a horizontal westward diverging motion of several centimeters, along with a significant uplift. It is confirmed by continuous measurements of recently installed GPS stations. We discard instrumental, hydrological, oceanic, or atmospheric loading effects and show that the transient is likely due to deep slow slip in the transition zone of the subduction interface (∼40‐ to 60‐km depth). In addition, daily observations recorded by a continuous GPS station operating between 2002 and 2015 highlight similar transient signals in 2005 and 2009, suggesting a recurrent pattern.

Key Points

• We detected a transient signal that occurred between 2014 and 2016 in Central North Chile using sGPS data

• We rejected any nontectonic source such as instrumental or hydrological processes

• These observations confirmed by cGPS are explained by slow slip events in the transition zone of the megathrust

1. Introduction

Thanks to the development of Global Positioning System (GPS) networks along plate boundaries in the last 25 years, slow slip events (SSEs) have been detected within the brittle‐ductile transition region of various subduction zone interfaces: Japan, Cascadia, Mexico, Costa Rica, New Zealand, Alaska, among others (see Beroza & Ide, 2011; Peng & Gomberg, 2010; Schwartz & Rokosky, 2007, for reviews). These SSEs are characterized by slow aseimic fault creep and are often associated with seismicity on the form of nonvolcanic tremor activity (e.g., Miller et al., 2002; Obara, 2002; Payero et al., 2008; Rogers & Dragert, 2003) and low‐frequency earthquakes (e.g., Frank et al., 2013; Hirose & Obara, 2005; Shelly et al., 2007). Although SSEs have been extensively studied over the last decade, physical mechanisms controlling slow slip remain quite elusive (Ikari et al., 2013; Leeman et al., 2016; Romanet et al., 2018; Segall et al., 2010). The influence of SSEs on the seismic cycle and their impact in terms of seismic hazard are also not yet fully understood. Slow slip seems to be closely related to earthquakes as several observations suggest that SSEs may have happened just before or after large seismic events (Kato et al., 2012; Radiguet et al., 2016; Ruiz et al., 2014; Wallace et al., 2017; Zigone et al., 2012). On the other hand, some SSEs occur quasiperiodically independently of large earthquakes (Rogers & Dragert, 2003; Wallace et al., 2016). Although the Chilean megathrust is one of the most active, with three megathrust earthquakes of M _w≥8 over the last decade (Illapel 2015, Iquique 2014, Maule 2010, Ruiz et al., 2016, 2014, Vigny et al., 2011), it is one of the only subduction zone where recurrent deep SSEs have not been observed yet. Until recently, the poor coverage of continuous GPS over this 4,000‐km‐long plate boundary prevented our ability to determine if SSEs occur in the region. In recent years, two aseismic events were observed on the shallow part of the interface (<40 km), both associated with low to moderate seismic activity, and were followed by a significant earthquake: a few weeks to a month‐long slip event preceding the 2014 M _w=8.1 Iquique earthquake (Ruiz et al., 2014; Socquet et al., 2017) and a few days‐long slip event associated with the 2017 M _w=6.9 Valparaiso earthquake (Ruiz et al., 2017).

Despite the poor continuous GPS instrumentation, decadal‐scale interseismic deformation has been intensely monitored thanks to the deployment and yearly measurements of several dense survey GPS (sGPS) networks overlying the plate interface, in the regions of South Central Chile (Moreno et al., 2010; Ruegg et al., 2009), Coquimbo (Khazaradze & Klotz, 2003; Klotz et al., 2001; Métois et al., 2012; Vigny et al., 2009), Atacama (Klein et al., 2018; Métois et al., 2014), and North Chile (Kendrick et al., 2003; Klotz et al., 1999; Métois et al., 2013). The coupling pattern constrained by the resulting interseismic velocity field highlights a heterogeneous segmentation, alternating between coupled segments where megathrust earthquakes are likely to occur and Low Coupling Zones (LCZ), which might act as seismic barriers (Métois et al., 2016).

The region of Atacama (29–25°S) possesses somewhat different characteristics compared to the rest of the subduction zone. Seismic activity on the shallower part of the interface (<30 km) is quite low (Métois et al., 2016) with the exception of the occurrence of regular swarms offshore Caldera (27°S in 1973, 1979, 2006; Holtkamp et al., 2011; Figure 1a). A clear gap in seismicity is also visible at depths greater than 40 km compared to adjacent regions (Figure 1a). Historical seismicity reports two events of M _w≥8.5 (1819, 1922), both characterized by complex rupture sequences (Willis, 1929). In this region, a sGPS network was installed starting in 2010 and measured yearly to estimate the steady state velocity field and derive the coupling ratio on the megathrust interface (Figure 1a and Figure S1 in the supporting information). The two highly coupled Atacama and Chañaral segments are separated by the Baranquilla LCZ (Klein et al., 2018; Métois et al., 2014), where the aforementioned swarm events occurred (Métois et al., 2016; Figures 1a and 1b; 28°S). However, in a region ∼150 km long centered on Copiapó and Caldera (Figure 1), GPS measurements made in 2015 and 2016 deviate from the steady state interseismic trend (Figure 2a). Estimating the static offset from available survey data reveals a spatially coherent pattern of displacements with centimeter‐level amplitude. This transient deformation is also consistent with the observations made by the few continuous stations in the area (Figures 1b and 2b), which discard any instrumental errors related to the sGPS processing.

Figure 1.

Seismotectonic context. (a) Coupling model and seismicity in the study region. Circles represent earthquake locations from the United States Geological Survey (USGS) catalog (1973–2017) that are colored as function of depth. Blue stars depict locations of seismic swarms in 1973, 1979, 2006, (Holtkamp et al., 2011) and 2015. The red color map shows the coupling model from Klein et al. (2018) obtained using interseismic velocities (see Figure S1). White lines represent main bathymetric features of the Nazca plate (Álvarez et al., 2015). Our preferred SSE slip model is represented with blue contour lines (+5 cm). Inset: Location of the study zone in South America. (b) Static horizontal (red) and vertical (blue) offsets estimated from sGPS time series. Seismicity from the CSN catalog (2014.5–2016) is represented as function of depth (same color scale as in a). SSE = slow slip event; sGPS = survey GPS; CSN = Centro Sismologico Nacional de Chile; GPS = Global Positioning System.

Figure 2.

Survey and continuous GPS time series. (a) sGPS annual time series and (b) cGPS weekly time series (downsampled from University of Nevada, Reno (UNR) daily time series). Time series are ordered by latitude of sites and are represented in ITRF08 (Altamimi et al., 2012). Duration of the 2010 and 2014 transients is depicted by gray areas. (c) Location of GPS sites and seismic stations. Detrended time series are presented in Figure S2 of the supporting information. GPS = Global Positioning System.

In this study, we investigate the possible origins of this transient signal that can be associated with tectonic or nontectonic sources. In particular, we assess potential effects due to hydrological, oceanic, or atmospheric sources and we investigate if this signal corresponds to episodic aseismic slip.

2. GPS Observations

2.1. Survey GPS Observations

The survey network in Atacama was first installed and measured in 2010, right after the M _w=8.8 Maule earthquake (Métois et al., 2014). Annual surveys have been conducted ever since, jointly with the northward extension of the network in Taltal (Klein et al., 2018). The Central Andes GPS Project (Brooks et al., 2003) and South America Geodynamic Activities (Klotz et al., 2001) GPS marks in the area were remeasured at the same time to gather a data set of almost 70 survey points, measured at least 3 times and up to 7 times (see Table S1 and Klein et al., 2018, for more details about the network and measurements). The sGPS data set is augmented by observations from recently installed cGPS stations operated by the Centro Sismologico Nacional de Chile (CSN; Báez et al., 2018), together with a selection of well‐distributed permanent stations across the South American continent (International GPS Service; Beutler et al., 1999), (Red Argentina de Monitoreo Satelital Continuo; Piñón et al., 2018), and Rede Brasileira de Monitoramento Contínuo networks. This data set is processed using a differential approach (GAMIT/GLOBK software), following methods outlined in Herring et al. (2010a, 2010b) and with the same approach used by previous studies in Central North Chile (Klein et al., 2018; Métois et al., 2016; Vigny et al., 2009). Position time series are estimated by combining daily solutions using global sites to define the reference frame in GLOBK because of postseismic deformation following the Maule earthquake which affects many reference stations on the South American continent (Klein et al., 2016).

After 7 years of yearly sGPS observations in a very localized area of the Atacama region, the 2016 position deviates significantly from the interseismic trend, both in horizontal and vertical (Figures 2a and S2). While the effect on the horizontal components is significantly smaller, we also observe a similar deviation in the vertical trend in 2015.

We estimate the deviation between the 2016 position predicted by the 2010–2014 trend and the 2016 measured position. The resulting static offsets show a very coherent pattern over an ∼150‐km region, on both horizontal and vertical components. Horizontal displacement vectors are diverging: northwestward north of Caldera and southwestward south of Caldera (Figure 1b). Vertical offsets are showing a centimeter‐level uplift with amplitudes clearly above the noise level, while the interseismic regime is characterized by subsidence (Figure 2a, Up). The northern boundary of this transient signal is marked by a lack of measurable deviation from the 2010–2014 interseismic trend at the latitude of Chañaral (Figure 1b), while the southern one is more difficult to define, due to the coseismic and postseismic deformations associated with the 2015 Illapel earthquake.

Because of such a spatial coherency of the signal, we exclude both instrumental errors during the data acquisition (equipment issues or artificial signal due to the change of equipment; see Table S1) or random errors in the processing parameterization. Furthermore, we rule out any issue in the definition of the reference frame, which would result in a systematic error over the whole area processed and not only in this region.

2.2. Continuous GPS Observations

Until 2015, the Atacama region had poor coverage of cGPS instruments with only one station installed in 2002 in Copiapó (COPO), which was decommissioned in 2015. A few months before COPO was decommissioned, two new nearby stations (BN03 and UDAT) were installed by the CSN. Two other stations were installed at the same period, TAM3 located about 25 km southeast of COPO and LLCH 120 km southwest of COPO (Figure 2c).

To confirm that the signal extracted from survey mode data does not artificially result from our differential processing, we investigate the time series of CSN cGPS stations provided by the Nevada Geodetic Laboratory, University of Nevada, Reno (UNR), (Figure 2b) and processed following a different processing strategy (PPP, for more details see http://geodesy.unr.edu/gps/ngl.acn.txt).

All the stations in the region of Copiapó show a clear transient signal starting in mid‐2014 that is visible on vertical and horizontal components (Figures 3a and S3). The cGPS station LLCH located south of the region also shows such a signal but with a smaller amplitude, in agreement with the amplitude decrease observed by nearby campaign measurements. Similar to sGPS observations at the latitude of Chañaral (26.4°S), no clear transient displacement is detected on the time series of the station PAZU located 140 km north of Copiapó (Figure 2b). Most of stations shown in Figure 2b have been installed shortly before (PAZU, TAM3, and LLCH) or during (BN03 and UDAT) the transient signal, while COPO was decommissioned before the conclusion of the event. To avoid any inconsistent amplitudes due to such temporal incoherences, we do not estimate any static offset from the cGPS observations. Although sGPS offsets are probably underestimated because they were measured before the end of the transient, they span a common time period and can therefore be used together to constrain a slip model representing that time period. Figure S4 shows that sGPS and cGPS solutions are consistent, which confirms the existence of a transient signal. In the time series of COPO which started in 2002, we observe a similar transient signal in 2009 with a shorter duration but comparable amplitude (Figure 3). We also observe a transient uplift in 2004/2005 with comparable amplitude, although the time series at that time is much noisier. It is difficult to estimate the exact duration of this transient, in the absence of data from the earlier period. Deviations in the horizontal components are also visible during the same time period (Figure S3). Unfortunately, COPO was the only permanent station in operation in the region until 2015 and the survey network was installed starting in 2010.

Figure 3.

Comparison between observed vertical displacements, rainfalls, and displacements caused by hydrological, oceanic, and atmospheric loads. (a) Weekly vertical time series of COPO (black) and BN03 (red) represented in ITRF08. (b) Detrended cumulative rainfall time series (between 2002 and 2014 from Valdés‐Pineda et al., 2018, local meteorological data between 2009 and present day). (c) Cumulative loads (Estimating the Circulation and Climate of the Ocean, European Centre for Medium Range Forecasts Re‐Analysis Interim, and GRACE Mascons products) at COPO station. Green dashed lines correspond to the 2006 and 2015 swarms in Caldera, the 2010 M _w8.8 Maule, and 2015 M _w8.3 Illapel earthquakes, blue dashed lines to dates of major rainfalls. Duration of visible transient events is indicated by gray areas. COPO = Copiapó; GPS = Global Positioning System; GRACE = Gravity Recovery And Climate Experiment.

3. Assessment of Hydrological, Oceanic, and Atmospheric Loading Effects

In 2015, a very strong El Niño event was responsible for large rainfalls in Central North Chile. Smaller rainfalls occurred in May 2010 (probably also related to the strong El Niño event that year) and May 2014 (Figure 3b). Such events could generate variations of hydrostatic pore pressure in underlying aquifer, which could be measurable by GPS (e.g., Silverii et al., 2016). To assess if such effects could explain the observed transient, we gathered rainfall data set in Atacama, in particular at meteorological stations in the city of Copiapó (between 2002 and 2014 from Valdés‐Pineda et al., 2018, and from global weather website between 2009 and present day). Significant rainfalls have occurred several months before and after the beginning of the transient (in 2014 and 2015), but no clear temporal correlation is visible (Figures 3a and 3b). Moreover, any hydrological load should produce subsidence (e.g., Borsa et al., 2014; Wahr et al., 2013) while uplift is observed in 2014–2016.

Going further, we calculate the elastic response due to nontidal oceanic, atmospheric, and hydrological loading effects at our GPS sites (see, e.g., Petrov & Boy, 2004, for more details). Loading predictions are computed, respectively, using the Estimating the Circulation and Climate of the Ocean modeled ocean bottom pressure (ECCO, run kf080h; Wunsch et al., 2009), European Centre for Medium Range Forecasts Weather Re‐Analysis Interim surface pressure (ECMWF ERA‐interim; Dee et al., 2011), and Global Land Data Assimilation System/Noah v1 snow, soil moisture, and canopy water (GLDAS/Noah v1, Rodell et al., 2004). We also derive surface displacements due to hydrological loading from the National Aeronautics and Space Administration/Goddard Space Flight Center 1° iterated global mascons (NASA/GSFC, updates from Luthcke et al., 2013). Cumulative vertical displacements due to environmental loading contributions predict a vertical signal of less than 1 cm, without any marked transient signal in 2005, 2009, and 2015 (Figures 3c and S5 for the different contributions on the three components). We are therefore confident that the signal measured by GPS is not due to any nontectonic sources such as atmospheric, oceanic, or hydrological loads.

4. Evaluation of Seismotectonic Source

4.1. Postseismic Deformation Following the 2015 M _w=8.3 Illapel Earthquake

Megathrust earthquakes often trigger continental scale postseismic deformation (e.g., following the 2011 Tohoku earthquake or the 2010 M _w=8.8 Maule earthquake; Klein et al., 2016; Sun et al., 2015; Trubienko et al., 2014). As the Atacama region is located only ∼500 km north of the rupture zone of the 2015 Illapel earthquake (Klein et al., 2017; Ruiz et al., 2016), postseismic deformation following this event could potentially affect the area where we detect the transient. Nevertheless, this M _w=8.3 earthquake was significantly smaller than the Maule earthquake, consequently triggering significantly smaller postseismic deformation. In addition, postseismic deformation at such distances from the rupture zone is most likely due to viscoelastic relaxation in the asthenosphere, therefore generating a very long wavelength horizontal surface motion pointing toward the coseismic rupture zone, that is, southwestward (Klein et al., 2016). The short‐scale divergent pattern shown in Figure 1b at the latitude of Copiapó (27.5°S) is thus not consistent with such broad‐scale postseismic deformation. However, the southward horizontal motion observed around Vallenar (28°S) could possibly correspond to the postseismic signal of the 2015 Illapel earthquake. These two regions are separated by smaller displacements around 28.2°S, which clearly delineates two zones affected by distinct deformation sources.

4.2. Aseismic Slip on the Interface

After ruling out any strong postseismic signal in the region, an obvious tectonic source able to generate such a spatially coherent transient deformation is aseismic slip along the subduction interface. As presented in section 1, this area is characterized by relatively little seismic activity at depth larger than 40 km (Figure 1a). In addition, no significant earthquake in the Atacama region, able to induce such a deformation, occurred when the 2014 transient started. In the following, we use a Bayesian approach to explore how such slip at the interface can explain our observations.

4.2.1. Bayesian Modeling Strategy

We build a fault geometry with triangular patches based on the finite element mesh designed in Klein et al. (2016, 2017) from Slab1.0 (Hayes et al., 2012) between 28° and 26°S down to 70‐km depth.

The forward problem is defined as d=Gm with d, the vector containing GPS offsets, and m the slip model parameters. We assume a pure along‐dip faulting, and Green functions G are calculated at each node of the fault plane, assuming a homogeneous elastic half‐space (Okada, 1985). The model space is explored using AlTar, a parallel Monte Carlo approach (Duputel et al., 2015; Jolivet et al., 2015; Minson et al., 2013) to derive the posterior probability density function (PDF) of the model m given available observations d _obs

$$p(\mathbf{m}|{\mathbf{d}}_{\text{obs}})\propto p(\mathbf{m})\exp \left(-\frac{1}{2}{({\mathbf{d}}_{\text{obs}}-\mathbf{\text{Gm}})}^T{\mathbf{C}}_d^{-1}({\mathbf{d}}_{\text{obs}}-\mathbf{\text{Gm}})\right)$$ (1)

C _d is the covariance matrix describing observational uncertainties. p(m) is the prior PDF, defined as centered truncated Gaussian with a standard deviation of 25 cm bounded between −1 cm (small back slip is allowed to ensure correct sampling near zero) and 50 cm (Figure S6).

4.2.2. Results

Given large uncertainties (±2 cm) on vertical displacements, we first develop a solution based only on horizontal components. The resulting posterior mean slip distribution is presented in (Figure 4a). This distribution depicts significant slip between 40‐ and 60‐km depth with a peak slip larger than 50 cm. This deep slip patch is associated with a geodetic moment of 2.93·10¹⁹ N.m (M _w=6.9; see M _w PDF in Figure 4 and Text S1). As shown in Figure 4c, this solution reproduces the observed horizontal displacements and also the vertical ones not included in the inversion. Because displacements extracted from sGPS observations are associated with significant uncertainties (about 1 cm in horizontal, Figure 1b), there are significant posterior uncertainties on the slip distribution, of the order of 10 cm (Figure 4b and blue slip PDFs in Figure 4a). Both the posterior mode model (representing the maximum of each parameter marginal PDF) and the median model are presented in Figure S7 together with Movie S1 showing more than a thousand model realizations from the space of possible solutions. The variability of the slip at shallow depths reflects high uncertainties in this part of the model. Nonetheless, the deep patch of slip remains quite stable, especially at 27°S on almost every model.

Figure 4.

Bayesian modeling of deep aseismic slip. (a) Mean posterior slip distribution (color scale in centimeters). Insets show posterior marginal probability density functions (in blue) at three different locations of the subduction interface, indicated by arrows. Red probability density functions depict the model inverted with both horizontal and vertical components (cf., Figure S8b) 1σ posterior slip uncertainties represented with the same color scale as the slip distribution. (c) Predicted horizontal (yellow) and vertical (light blue) offsets and their associated 2σ uncertainties (filled gray ellipses) compared to measured horizontal (dark red) and vertical (dark blue) displacements.

We also conducted an inversion using both horizontal and vertical observations. The posterior mean slip distribution shown in Figure S8 presents a slightly smaller slip amplitude with an equivalent geodetic moment of 2.72·10¹⁹ N.m. In particular, approximately 15 cm of slip is needed to fit the data in the southern part of the fault (27.5°S), while more than 25 cm is required for the horizontal displacement based model. Nevertheless, the overall slip distribution is consistent for both models, with similar amplitudes in the main slip area (PDFs in Figure 4a). As expected, integrating more observations results in slightly smaller posterior slip uncertainties on the deep patch (Figure 4a, red PDFs, and Figure S8). Finally, we tested a model with broader priors, producing similar spatial slip distribution, M _w, and data fit (Figure S9). Therefore, we are confident that the overall pattern of slow slip distribution is fairly well defined.

5. Discussion and Conclusions

In this study, we report observations of a transient deformation signal, initially detected by the sGPS network, on the Atacama region along the Chilean subduction zone. We interpret this signal as a deep aseismic slip event on the subduction interface.

We carefully verified that observed GPS deviations cannot be explained by any nontectonic sources. Because the transient is revealed by both sGPS and cGPS observations which were processed independently, any instrumental errors are very unlikely. We also eliminated any oceanic, hydrological, or atmospheric sources. Although sGPS alone does not allow to accurately constrain the temporal evolution of this event, the compilation of cGPS time series in the region shows that this event started in September 2014 and lasted until mid‐2016 (i.e., duration longer than 1 year).

Using a Bayesian modeling approach, we demonstrate that deep slow slip along the subduction interface can explain the observed GPS deviations. The event released a scalar moment equivalent to a M _w=6.9, and its long duration (more than a year) is consistent with the duration versus magnitude scaling laws for slow earthquakes proposed by Ide et al. (2007). The inversion places the SSE downdip of the seismogenic zone (between 40‐ and 60‐km depths; Figure 1a) in the so‐called transition zone, which is commonly observed for SSEs in others subduction zones (Obara & Kato, 2016; Rolandone et al., 2018; Schwartz & Rokosky, 2007). This deep SSE in Atacama differs from the recently reported aseismic events in Chile on the shallower part of the interface preceding the occurrence of large earthquakes (Ruiz et al., 2017; Socquet et al., 2017).

Because of the sparse regional seismic network, it is difficult to discern nonvolcanic tremor activity, which is commonly observed in many subduction zones during SSEs (e.g., Cascadia, Mexico, and Japan; Obara, 2002; 2011; Payero et al., 2008; Rogers & Dragert, 2003; Zigone et al., 2012). Only one seismological station is available in the vicinity of the 2014–2016 SSE (GO03; Figure 2c). Some bursts of energy of a few decibels above the average could indicate tremor activities (for more details about the methodology, see Text S3; Husker et al., 2010; Kostoglodov et al., 2010; and Figure S10), but it is unclear whether such signal is tectonic or is associated with random fluctuations of high‐frequency noise. Additional seismic stations in the region would be necessary to further investigate the occurrence of the nonvolcanic tremor and its relationship with the deep slow slip deduced from GPS data.

Another interesting aspect is the possible interaction between SSE and repeated occurrences of seismic swarms offshore Caldera over the last 50 years (in 1973, 1979, and 2006; Holtkamp et al., 2011; and in 2015, CSN catalog, Figure 1b). Seismic swarms are often accompanied by significant magnitude earthquakes (M _w≥6.5 in 2006; Ducret, 2013; Figure S3) and can be markers of aseismic slip on the interface (e.g., Ruiz et al., 2017). Our tests (presented in Text S2 and Figure S11) show that the observed GPS transient is most likely not associated with seismic or aseismic slip in the area of the March 2015 swarm that occurred at shallow depth (∼20 km). In particular, shallow slip on the interface produces subsidence inland rather than uplift. Nonetheless, an indirect link between the deep transient event and the seismic swarms offshore Caldera should be more thoroughly investigated in the future.

Two additional transient signals observed on the COPO time series (Figure 3) suggest that similar SSEs might have occurred in 2005 and 2009. Based on this single cGPS station, the 2009 event seems to last for about 12 months, while the 2005 event is probably shorter (although it is hard to precisely define its beginning). Such recurrence is similar to long‐term SSEs reported in southwest Japan, New Zealand, and Mexico (Hirose & Obara, 2005; Radiguet et al., 2012; Wallace & Beavan, 2010), which typically occur every 3–4 years with a scalar moment equivalent to a M _w=7.

To our knowledge, this is the first observation of potentially recurrent deep SSEs reported in Chile. It raises key questions regarding the stress release mechanism along this segment of the Chilean subduction zone and potential interactions between slow slip asperities in the transition zone and megathrust asperities at seismogenic depths.

The shallower part of the interface is characterized by high coupling down to 40 km on the Atacama and Chañaral segments. These two segments are separated by a low coupled portion, the Baranquilla LCZ, which is located at the same latitude as the SSE reported in this study (27°S; Klein et al., 2018; Métois et al., 2013). Note that 2016 measurements have been excluded from the later interseismic model in order to capture the coupling of the subduction interface between SSEs. The modeled slow slip therefore lies below the coupled seismogenic zone with two main slip asperities that seems to be separated by a narrow creeping area (Figure 1a). The comparison of mean slip deficit in the SSE area with the slip inferred from this study is not entirely conclusive. The coupling ratio appears too low at these depths to accumulate in 4 years the 15 cm of slip released by the 2015 SSE (i.e., assuming that the 2009 SSE released all of the previously accumulated stress). However, both the coupling and the slow slip models are associated with significant uncertainties at such depths and the PDF of both models show a significant overlap (Figure S12). While we cannot completely rule out a small contamination of Klein et al.'s (2018) coupling model by measurements at the onset of the SSE in 2015, the above conclusion seems also consistent with the coupling values reported in models constrained before 2014 (Métois et al., 2013, 2016). Although it is difficult to conclude from a single GPS station, the difference in duration between the 2009 and 2014 transients could suggest that both events are different and therefore possibly not perfectly colocated. It could also suggest that the 2009 event did not release all of the previously accumulated stress where the 2015 SSE occurred. In any case, the interplay between interface locking and SSE in this region needs further investigation based on future observations.

The occurrence of deep SSEs also has impact in terms of seismic hazard. In particular, aseismic slip in the transition zone can potentially trigger earthquakes in the seismogenic zone. A similar mechanism has recently been proposed in Mexico where the 2014 SSE is believed to have triggered the M _w=7.3 Papanoa earthquake by either increasing the static stress in the coupled region or by helping the weakening of the earthquake hypocentral area (Radiguet et al., 2016). The occurrence of transient aseismic slip in Chile can thus potentially influence the seismic cycle by changing the stress level in the seismogenic zone, which may advance or delay the occurrence of the next earthquake in the region.

Given the limited amount of observations available until recently, it is obvious that denser deployments of geodetic and seismic instruments are necessary to better capture the behavior of the megathrust in the Atacama region. The launch of new Synthetic Aperture Radar satellite constellations such as the Sentinel‐1 mission can also help to monitor and detect transient deformations in sparsely instrumented regions (e.g., Rousset et al., 2016). Such improvements of our observational capabilities also open up new opportunities to investigate the potential impact of deep SSEs on the shallower part of the interface in Atacama where the last major event occurred almost a century ago.

Supporting information

Supporting Information S1

Movie S1

Klein, E. , Duputel, Z. , Zigone, D. , Vigny, C. , Boy, J.‐P. , Doubre, C. , & Meneses, G. (2018). Deep transient slow slip detected by survey GPS in the region of Atacama, Chile. Geophysical Research Letters, 45, 12,263–12,273. 10.1029/2018GL080613